Datasets:

CShorten

/

CORD-19-prototype

like 0

Predictive Power of Air Travel and Socio-Economic Data for Early Pandemic Spread

BACKGROUND: Controlling the pandemic spread of newly emerging diseases requires rapid, targeted allocation of limited resources among nations. Critical, early control steps would be greatly enhanced if the key risk factors can be identified that accurately predict early disease spread immediately after emergence. METHODOLOGY/PRINCIPAL FINDINGS: Here, we examine the role of travel, trade, and national healthcare resources in predicting the emergence and initial spread of 2009 A/H1N1 influenza. We find that incorporating national healthcare resource data into our analyses allowed a much greater capacity to predict the international spread of this virus. In countries with lower healthcare resources, the reporting of 2009 A/H1N1 cases was significantly delayed, likely reflecting a lower capacity for testing and reporting, as well as other socio-political issues. We also report substantial international trade in live swine and poultry in the decade preceding the pandemic which may have contributed to the emergence and mixed genotype of this pandemic strain. However, the lack of knowledge of recent evolution of each H1N1 viral gene segment precludes the use of this approach to determine viral origins. CONCLUSIONS/SIGNIFICANCE: We conclude that strategies to prevent pandemic influenza virus emergence and spread in the future should include: 1) enhanced surveillance for strains resulting from reassortment in traded livestock; 2) rapid deployment of control measures in the initial spreading phase to countries where travel data predict the pathogen will reach and to countries where lower healthcare resources will likely cause delays in reporting. Our results highlight the benefits, for all parties, when higher income countries provide additional healthcare resources for lower income countries, particularly those that have high air traffic volumes. In particular, international authorities should prioritize aid to those poorest countries where both the risk of emerging infectious diseases and air traffic volume is highest. This strategy will result in earlier detection of pathogens and a reduction in the impact of future pandemics.

Predicting the origin and emergence of new diseases is critical to preventing and controlling them [1, 2] . In particular, if the early spread of a newly emerging pathogen can be predicted and curtailed before it becomes pandemic, its impact on public health and global economies may be much reduced [3, 4, 5, 6] . In March and April of 2009, a novel H1N1 influenza A virus (2009 A/ H1N1) with gene segments from humans, swine, and birds led to the first pandemic of influenza in forty years [7, 8, 9, 10] . Current evidence points to a Mexican origin for the initial human-tohuman transmission of this virus, although preliminary genetic analyses suggest the virus has an older and highly-mixed lineage [8] . The virus' lineage and rapid spread suggest that global trade and travel may have played an important role in its early emergence [7, 8] . Here, we attempt to elucidate how these factors may relate to the emergence and spread of this newly detected virus. One unresolved question is to what degree does a country's development affects its ability to detect and respond to an emerging disease in a timely manner? Development may affect spending on healthcare infrastructure, and particularly, spending on the high cost, intensive public health surveillance needed during the early stages of a pandemic [11, 12, 13] . Socioeconomic factors will also likely affect individuals' abilities or desire to seek diagnosis or treatment, and a country's capacity to test and identify pathogens. Here, we analyze socio-economic and travel data to understand the initial spread of this virus. We focus on the early stages of the epidemic, when travel from Mexico was likely to be the dominant mode of viral spread. Finally, we examine poultry and swine trade data prior to the 2009 A/H1N1 pandemic to add to our understanding the processes that led to the emergence of this virus. As of May 8 th 2009, only two weeks after it was first reported, the 2009 A/H1N1 influenza strain had spread to 24 countries, 40 U.S. states (plus the District of Columbia) in the US, and 9 provinces in Canada ( Figure 1 ). This rapid spread resulted, in part, from the tight connectivity of the globe through air travel ( Figure 2) . A log-logistic survival analysis regression model was used to predict the time-to-reporting of the first confirmed 2009 A/H1N1 case to each country. Of all the models evaluated, a multivariate model with three predictors, (1) total country-level healthcare spending per capita, (2) estimated passenger volume arriving from Mexico via direct flights (direct flight capacity), and (3) passenger volume from Mexico via indirect, or two-leg, flights (indirect flight capacity), provided the best fit to the data using AIC, as detailed under Methods (Table 1 , DAIC = 0, overall x 2 = 54.33 on 5 degrees of freedom, p-value,0.0001). The correlation between total country-level healthcare spending and the flight data was low (r,0.4). Although the correlation between direct and indirect flight data was high for countries with direct flights (r.0.9), the indirect flight information provided critical additional information for areas without direct flights. The AIC scores demonstrated this, as the model that included only direct flight information and healthcare spending did not explain the data as well as the best fit model (DAIC = 9.044). Alternate socio-economic measures, even those directly related to healthcare, such as the number of physicians per capita, GDP, or population density were much less predictive than total healthcare spending per capita. Notably, out of univariate analyses, the model with healthcare spending per capita as the sole predictor fit better than models with flight information alone (Table 1) , demonstrating just how informative this data is in predicting the date of reporting. In the best fitting multivariate model, indirect flight capacity had the largest effect size, but including healthcare spending per capita substantially increased the fit to the data (Tables 1, 2 ). For Canadian provinces and American states, we conducted an analysis with just the flight data (Table 3 overall x 2 = 22.89 on 2 degrees of freedom, p-value ,0.001). While the direct flight information does not have a statistically significant effect, the indirect does, most likely because only a few key hubs had direct flights, and these hubs also have a large volume of indirect connections. For the country-level analysis, we compared the predicted reporting dates with the actual reporting dates, for countries where the disease arrived by May 8 th , 2009 ( Figure 3 , Supplemental Online Figure S1 ). We validated the model by determining how well a model fit to data up until May 8th predicted reporting dates for fourteen countries where the disease was detected between May 9 th and May 19 th (Supplemental Online Figure S2 ). The correlation between forward predicted and observed dates was 0.62, and the observed reporting date fell within the 95% confidence interval for all countries. Many of the actual reporting dates are earlier than predicted, which is expected due to the nonlinear nature a of log-log survival analysis regression. In particular, countries that had not reported disease by the cut-off date were included in the analysis by designating these as locations that ''survived'' the entire study period without acquiring the disease (i.e, censoring). This appropriately extends the predicted reporting dates by including information on both countries that had reported disease by the cut-off date as well as countries that had not. Using this methodology, we also estimated the reporting date of the disease in the remaining 103 countries and the 95% confidence intervals ranged from April 17 th to May 29 th , 2009 (Supplemental Online Figure S3 ). To elucidate the potential origins of this novel viral strain, and to shed light on targets for future surveillance and prevention programs, we analyzed global trade in live poultry and swine during the decade preceding the current pandemic [14] . We estimate the trade in live swine between Canada, the United States and Mexico to be over 1.75 million animals over the last decade, Previous studies suggest that data on air travel can be used to predict the spread of newly emerged human pathogens and better target public health measures [15, 16, 17, 18] . Our analyses support this, but demonstrate that the ability of a country to rapidly detect, diagnose, and report the new infection is a critical element that enhances our predictive power and control capacity. Other studies suggest that analysis of the underlying drivers of disease emergence (e.g. agricultural intensification, land-use change) can be used to predict the geographic origins of new emerging diseases [2] . The currently circulating pandemic influenza strain is a triple reassortment virus with closest known relatives from Europe, Asia, and North America, but there is uncertainty regarding its origin due to the large temporal separation between this pandemic 2009 A/H1N1 strain and the nearest ancestors (10-15 years) [7] . Our analyses of swine and poultry trade demonstrate an enormous potential for intercontinental mixing of potentially zoonotic pathogens, including influenza A viruses. Although artificial insemination is the predominant strategy for interbreeding of commercial swine, live swine are still routinely traded for breeding purposes [19] . Large numbers of poultry are also traded globally, and low pathogenicity influenza viruses are likely to spread unnoticed among poultry until they reassort or mutate to highly pathogenic forms, such as the A/H1N1v strain. This strain notably was the results of reassortment of several relatively low pathogenic influenza strains, as explained by Garten et al. [8] . In addition, as the recent cases of workers exposing a herd of pigs to the 2009 A/H1N1 virus makes clear [20, 21] , even dramatic reductions in the international live animal trade may not prevent the exposure of local livestock to novel viral types from distant locations [9, 10] . Although extensive trade of poultry and swine between continents and within the North American countries almost certainly contributed to the emergence of this virus, surveillance of influenza strains circulating among traded animals is poor [10] , so that it is impossible to designate any single country, trade connection or market as the key point at which the new strain evolved. Expanded surveillance for influenza in livestock populations may allow more of the markers of transmissibility and virulence to be identified, or factors driving higher virus transmission to be determined [9, 22] . In particular, we need to analyze all influenza strains, including the non-and low pathogenic influenzas, in addition to the highly pathogenic ones, with greater regularity. Only by this thorough surveillance can we begin to understand what differentiates the strains that cause pathogenesis in humans from those that do not. Such that eventually we may be able to predict viral emergence and develop vaccines against pandemic influenza viruses in advance of their spread. In order to develop such capability, we need to do more surveillance of livestock and wild influenza strains now. The speed at which 2009 A/H1N1 spread during the early phases of this pandemic is striking. It was detected in four continents within three weeks after Mexican authorities first reported it. In contrast, the 1918 Spanish flu took 3 years to circle the globe [23] . Our analyses of air-travel data support the WHO's decision to recommend against closing all air travel from Mexico, since the virus most likely had already spread to several other countries by the time it was first reported to be widespread in Mexico on April 29 th . In particular, cases had already been detected in the United States, which is a major hub for connecting flights [24] . Our current report is the first published analysis of H1N1 spread to include indirect flight data, and this significantly increased the predictive power of our model. Our analysis suggests that airports serving as major hubs could be targets for disease surveillance, and could become facilities that train people and stockpile medicines in preparation for pandemics. This approach differs from previous reports that focus on the role of travel restrictions at hubs [6, 17] . Our results further suggest a critical role for health care spending in determining a country's probability of detecting, confirming and reporting influenza cases in the early phases of a pandemic. The negative relationship between healthcare spending and detection of 2009 A/H1N1 influenza may be due to a delay in testing or in the collecting of specimens from individuals in countries lower healthcare resources. These countries likely have lower rates of health insurance, less healthcare infrastructure, lower self-reporting, and lower numbers of doctors per capita. One consequence of lower health care resources is that the threshold for detection (i.e., the number of cases that need to occur before a case is detected, tested and confirmed by medical authorities) is likely higher in lower-income countries that cannot afford to invest as much in public health and healthcare infrastructure. Similar socioeconomic factors have been shown to play an important role in determining spatiotemporal patterns of diseases such as tuberculosis, schistosomiasis, West Nile virus, and HIV/AIDS [12, 13, 25, 26] . We found that incorporating data on healthcare spending per capita significantly increased our power to predict the time of reporting of 2009 A/H1N1. This suggests important strategies for future disease control. During the early stages of a pandemic, countries with moderate to high air travel from a pandemic origin, but relatively low healthcare spending, are likely to significantly under-report cases. It is therefore in the best global health interest for intergovernmental and other aid agencies to specifically target these nations for assistance to test and report cases early in a new pandemic. We propose that subsidies for outbreak response to these nations with high connectivity and low resources would be the most effective strategy to reduce the spread and impact of a pandemic. Efforts to better target pandemics would be more effective in reducing disease spread if they were set up in advance of a pandemic [5, 6, 16] , as there is a very small window of opportunity in which to act once a new emerging disease is detected. Such efforts could be strategically positioned to target emerging disease 'hotspots' [2] that are also hubs of trade and travel for surveillance and prevention [16, 27] . For influenza viruses, any future identification of a spillover of a novel strain from poultry or swine to farm workers should be rapidly followed by analyses of the travel routes out of the country where the index case was discovered. At that point, intergovernmental agencies such as WHO could best target limited resources to the poorer countries that are most likely to receive high numbers of airline travelers from the pandemic origin. These are the countries where reporting is likely to be poorest, and where a significant, undetected caseload is likely to exist by the time resources are allocated. These at-risk countries are also the least capable of affording control measures. On the whole, this H1N1 strain appears to be relatively mild, although it is still inflicting additional morbidity and mortality. However, if a strain with a higher mortality rate, such as that observed with the H5N1 avian influenza subtype, were to spread in a similar fashion, the outcome would be catastrophic both in terms of human suffering and economic damage. For example, the impact of an H5N1 avian influenza outbreak, should the virus become easily transmissible between humans, on the United States economy has been estimated to be $71.3-$166.5 billion [28] . The measures we have proposed are likely to have economic benefits that far outweigh their costs. We compiled the data on international air travel from the IATA database, supplied by Diio, LLC through their APGdat service [29] . Similar to prior analyses [15, 16, 17, 18] , we used direct connection information with regards to aircraft type and passenger capacity to calculate the connectivity of Mexico with all airports included in the database, and summarized this information (as direct flight capacity) at the country level. Additionally, we estimated the number of connecting passengers (indirect flight capacity) by calculating the number of passengers (p i,j ) arriving at airport j from airport i, and then estimating the number of passengers (p j,k ) going from airport j to airport k, based on all flights reported in the database. We limited the potential connections (trip jRk) to flights that departed no sooner than one hour after the first trip (iRj), and no later than six hours after the arrival of the first trip. We also disallowed return of passengers to Mexico once they left the country, and the return of passengers to North America once they left that region. We thus obtained a quantity, x i,j,k , that estimates the total potential connections to airport k available to passengers from the first trip (iRj). Setting constant the fraction of all passengers that connect (x), we obtained an estimate of the number of passengers with two leg itineraries for each potential destination (iRk; Eq. 1): We summarized these connections at the country scale, thereby estimating connectivity for nearly every country on the globe with Mexico through either direct or indirect flights; the only countries excluded would require an overnight stay in a hub airport, or three or more connecting flights. We validated our algorithm (eq. 1) for connections within the contintental U.S.A. (the only data on actual itineraries, including connecting flight information, to which we had access). We randomly chose 50 connecting itineraries within the U.S.A. and compared our predictions to the actual routes. Our predictions were statistically significant, using a simple proportional model with log-normal errors, and explained over 60% of the variance in actual routes (F = 83.71, p,0.001 on 1, 49 d.f, adjusted R 2 = 0.6232). We determined the date a country reported its first WHOconfirmed 2009 A/H1N1 case through May 8 th , 2009. We chose this date in order to limit the analysis, as much as possible, to initial spread from Mexico, because it served as a natural breakpoint in the distributions of reporting dates, as well as being the date our initial analysis. We performed a survival analysis using R [30] , and used an accelerated life time model using a log-logistic distribution. We also examined using a scale-free exponential distribution, as opposed to a log-logistic distribution, which requires a scale parameter, but these models did not fit nearly as well, as measured by AIC. We followed Burnham and Anderson [31] , in using Akaike Information Criterion (AIC) to choose the model that best explains the data (i.e., the one with the lowest AIC, or equivalently DAIC, score). Additionally we provided the Akaike weights, which estimate the likelihood that a specific model is the true model, assuming that the true model is in the set of examined models [14] . Using this methodology, we choose to evaluate 22 models that made mechanistic sense including a null model for a reference. We did not include any models with only the indirect flight data, and without the direct flight data, because we feel that this does not make mechanistic sense. To reduce multicollinearity we included at most two socio-economic indicators. We evaluated four independent predictors for the date of first confirmed 2009 A/H1N1 case: the volume of (1) direct and (2) indirect passengers on international flights, (3) the country-specific Gross Domestic Product and (4) healthcare spending per capita, by both private and public entities, from 2006 (the most recent year with all data available) from World Bank estimates [32] . We also examined alternate socio-economic metrics as compiled by the World Bank [32] , such as the number of physicians, and average population density. However models including these predictors did not perform as well (as measured by AIC) and often had many more missing values if limited to most recent information. For all analyses, dates were transformed to Julian day since February 15 th , and all predictor variables were standardized (mean subtracted, then divided by standard deviation) in order make possible the direct comparison of coefficients. This standardization has the added advantage of canceling out the x factor in equation 1 for the statistical analysis; thus, our analyses do not require any assumptions about the number of passengers who make connecting flights. These statistical models were used to predict the expected time of detection for all countries in our database that had GDP, population density, healthcare, and flight data. Confidence intervals were constructed from the best model fit based on the variance of the data, using the ''predict'' functions in R [30] . We obtained United Nations Food and Agriculture Organization data on trade in Live Swine (commodity code HS96:S0103) and Live Poultry (S0105) from the U.N. Comtrade data portal [14] . We analyzed data from the last ten years (the approximate time since 2009 A/H1N1 diverged from the nearest sampled virus) [7] , and focused on trade to North America (Mexico, Canada and United States) from outside this region, as well as trade to Mexico within the North American region. Figure S1 Model predictions compared with actual case arrival dates. Dates of case arrivals (black diamonds) for cases that were reported before our cut off of May 8th. Grey whisker plots represent 95% confidence intervals for predicted arrival date, with interior grey bar as expected (mean) date of arrival from survival analysis. Found at: doi:10.1371/journal.pone.0012763.s001 (0.02 MB PDF) Figure S2 Forward prediction of future case arrival dates. Dates of case arrivals (black diamonds) for cases that were reported after our cut off of May 8th, but before May 19th. Grey whisker plots represent 95% confidence intervals for predicted arrival date, with interior grey bar as expected (mean) date of arrival from survival analysis. Found at: doi:10.1371/journal.pone.0012763.s002 (0.02 MB PDF) Figure S3 Forward prediction of future case arrival dates. Grey whisker plots represent 95% confidence intervals for predicted arrival date, with interior grey bar as expected (mean) date of arrival from survival analysis. Found at: doi:10.1371/journal.pone.0012763.s003 (0.03 MB PDF)

Evolutionary Entropy Determines Invasion Success in Emergent Epidemics

BACKGROUND: Standard epidemiological theory claims that in structured populations competition between multiple pathogen strains is a deterministic process which is mediated by the basic reproduction number ([Image: see text]) of the individual strains. A new theory based on analysis, simulation and empirical study challenges this predictor of success. PRINCIPAL FINDINGS: We show that the quantity [Image: see text] is a valid predictor in structured populations only when size is infinite. In this article we show that when population size is finite the dynamics of infection by multi-strain pathogens is a stochastic process whose outcome can be predicted by evolutionary entropy, S, an information theoretic measure which describes the uncertainty in the infectious age of an infected parent of a randomly chosen new infective. Evolutionary entropy characterises the demographic stability or robustness of the population of infectives. This statistical parameter determines the duration of infection and thus provides a quantitative index of the pathogenicity of a strain. Standard epidemiological theory based on [Image: see text] as a measure of selective advantage is the limit as the population size tends to infinity of the entropic selection theory. The standard model is an approximation to the entropic selection theory whose validity increases with population size. CONCLUSION: An epidemiological analysis based on entropy is shown to explain empirical observations regarding the emergence of less pathogenic strains of human influenza during the antigenic drift phase. Furthermore, we exploit the entropy perspective to discuss certain epidemiological patterns of the current H1N1 swine 'flu outbreak.

Recent years have seen an apparent acceleration in the rate of emergence of new infectious disease pathogens in the human population [1] . Some of these have their origins in animal (wild or domesticated) reservoirs [2] [3] [4] , and the years since 2003 have witnessed the appearance of SARS [5, 6] and swine flu [7] . The 20 th century saw, for example, the emergence of pandemic 'flu in 1918, 1957 and 1968 (with a limited H1N1 re-emergent outbreak in 1977) [8] , avian 'flu [9] and the rise of HIV in the 1980's [10, 11] . Additionally, antibiotic-resistant pathogens have become increasingly widespread in the past decade, particularly in healthcare settings [12] . Antigenically-variable pathogens are responsible for much of the burden of communicable disease in the world today. Therefore, developing an understanding of the factors that lead to the emergence and spread of novel pathogenic agents and strains is a topic of great interest. In recent months the emergence of a swine 'flu (H1N1 2009) with human-to-human transmission capability has re-focussed attention on this issue [7, 13] . Likewise, studies, such of those of Creanza et al. [14] who used a computational analysis of viral nucleotide and amino acid sequence data collected during seasonal 'flu epidemics show how diversity declines over the course of an epidemic. These observations underscore the role that ecological constraint play in the evolution of pathogens. For antigenically variable pathogens it is competition between strains that is the fundamental mechanism which determines the observed patterns of disease spread and prevalence. Diseases in this category include influenza A virus, meningococcal and pneumococcal bacterial infection, malaria and dengue fever, to name but a few. The principal epidemiological characteristics of such diseases are the absence of life-long protective immunity, cross-reactive immunity between strains and the potential for future re-infection. Each strain is in competition with the others for resources. In this case the resources are susceptible hosts, and dominance goes to those variants that are able to outpace their neighbours in their ability to infect susceptibles. From a Darwinian perspective it is the ''fittest'' strains that will dominate. Translating the qualitative notion of fitness into quantitative terms constitutes one of the fundamental problems in evolutionary epidemiology. Standard epidemiological theory as largely developed by Dietz [15] and Anderson and May [10] revolves around the basic reproduction number, R 0 , the number of secondary infectives, as the key parameter [15, 16] for analysing disease emergence, spread and vaccination strategy. In the case of structured populations, R 0 is defined as where the function V (x) is the infectious net-reproductive function. It is an ''infectious age''-dependent function that defines the rate at which an infected host generates secondary infections in the time interval following its initial infection (File S1 Section i). This theory has been extended to address competition between emergent pathogen strains using basic reproduction number as the metric for competitive dominance [16] . Selective advantage,s s, in the case of competing strains is now given bỹ where DR 0 is the difference in the basic reproduction number between the incumbent and invading strain [16] . The measure of selective advantage given by equation 2 implicitly assumes that the population is infinite -a mathematical idealisation. The fact that conditions of finite size may have an effect on the outcome of selection has been recognised in a population genetics context [17] but has not been explored analytically in multi-strain epidemic models [16] . These studies, however, assumed that the populations were unstructured or demographically homogeneous. The effect of finite size in studies of selection between competing types in structured populations was first developed in Demetrius [18] . The analysis focused on demographic structure and rested on the observation that in view of the heterogeneity in structure and the finite population size,fluctuations in population numbers will occur. The ergodic theory of dynamical systems was then exploited to generate a new family of demographic variables to describe the population dynamics and its fluctuations. A diffusion process was then applied to show that the outcome of selection will now be determined by the robustness or demographic stability of the population, and regulated by the population size and certain demographic parameters which characterise the geometric properties of the infectious net-reproductive function. Robustness, the rate at which the population returns to the steady state condition after a perturbation in age-specific fecundity and mortality variables, can be formalised in terms of the statistical measure evolutionary entropy. This macroscopic variable describes the uncertainty in the age of the mother of a randomly chosen newborn. The change in basic reproduction number, the classical criterion for selective outcome, was shown to be the limit, as population size tends to infinity, of the entropic selection principle. Hence the classical models of selection are limiting cases of the entropic models. This study of competition in age-structured populations was extended by Demetrius, Gundlach and Ochs [19] to the analysis of the dynamics of selection where the heterogeneity derived from individual variations in size, metabolic condition or spatial location. The entropy parameter in this general context describes the uncertainty in the state (size, metabolic condition or spatial location) of the ancestor of a randomly chosen individual. The results of this study formed the basis of a general model of the evolutionary process which is called directionality theory. We now apply this theory to analyse the effect of finite size in multi-strain epidemiological models where the heterogeneity derives from variability in infection age. The quantity V (x) in this class of model pertains to the product of the survivorship and infectivity of infectious individuals. We will apply the formalism described in [19] [20] [21] [22] to show that the invasion dynamics of competing strains in populations of finite size is predicted in terms of the macroscopic variable evolutionary entropy, S, which is given by The quantity p i is the probability that the parent of a randomly chosen infective is in the age class i. The statistical measure, S, describes the uncertainty in the age of an infected parent of a randomly chosen infective. The statistical parameter evolutionary entropy describes the rate at which the population returns to its steady state condition after a random perturbation in the age-specific fecundity and mortality variables. Entropy is analytically related to the generation time, T (the mean age of infection).We will use this analytical fact to show that entropy is also analytically related with the duration that the host organism is infected, and hence it can be regarded as a basic metric of pathogenicity. Directionality theory shows that entropy S predicts the outcome of competition between strains. The selective advantages s in the case of competing strains involves S and two additional macroscopic variables (W and c, the first and third moments of a random variable defined in terms of the net-reproductive function and the probability distribution p i ). The selective advantage is now [20] given bỹ Here N denotes the population size of infectives and DS is the relative evolutionary entropy of the incumbent and the invader. The quantities W (the reproductive potential) and c (the demographic index) are statistical parameters, and both are functions of the agespecific fecundity and survival functions which determine the infectious net-reproductive function V (x). The parameters W and c define different scenarios for the epidemiological population biology that prevail during the competitive invasion process. These quantities, in contrast to entropy, can assume positive or negative values contingent on the geometry of the infectious net-reproductive function. The demographic index, c, relates to the flatness or peakness of the infectious net-reproductive function V (x): generally speaking cv0 implies a peaked net-reproductive function, whereas cw0 implies it is flat. This term is less influential on the dynamics as it is scaled by 1=N, so when N (the number of infectives) is moderately largẽ s s~{WDS; so competitive advantage is now determined by the relative entropy and the reproductive potential. An equivalent formulation of the selective advantage,s s, can be written in terms of the growth rate r and a quantity called the demographic variance, s 2 [19] , which is the second moment of a random variable defined in terms of the infectious net reproductive function and the distribution p i , where Dr is the difference in growth rate between the incumbent and variant strains. Its equivalence to equation 4 is demonstrated in File S1, Section ii. It is evident that as N?large we recover equation 2, since Dr and DR 0 are positively correlated. We exploit this new theory of entropic epidemiology to explain detailed empirical observations regarding the emergence of less pathogenic strains of human influenza A virus, an issue that has remained elusive when viewed through the framework of classical epidemiological theory. Moreover, we discuss the current on-going swine 'flu (H1N1 2009) outbreak from the perspective of directionality theory, [18] [19] [20] [21] [22] . Much recent work on strain dynamics in multi-strain pathogens has focussed on the adaptations of basic epidemic models to deal with multiple strains with differing assumptions about host immune responses [23] [24] [25] [26] . This has led to progress in understanding issues such as strain clustering effects, for example, but at the price of intractability when large numbers of strains are considered. By contrast, the model presented here takes a different approach, as it focuses on the emergent properties of multiple competing strains without a detailed rendering of all biological features. The penalty for generating this alternative model is that the biological detail is presented more crudely than in the more detailed epidemiological approaches. Strain selection takes place at a number of levels ranging from within hosts all the way up to the population level. In such multi-scale systems a variety of modelling approaches are needed. However, we believe that the approach presented here complements existing formulations. To make the presentation concise we will summarise certain general results of the dynamics of competition in structured populations, as elaborated in directionality theory [18] [19] [20] [21] [22] , and apply them in an epidemiological context. Specifically, our model shows that in the context of emergence of new human 'flu strains in SE-Asia there will be a progressive shift to less pathogenic strains. This empirically observed pattern is consistent with our entropic perspective. Classical epidemiological theory associates increased competitive advantage with increasing basic reproduction number, R 0 [27, 28] . The argument is that a larger R 0 results in a faster rate of infection of susceptibles (as determined by the growth rate, r) thereby driving competitor pathogen strains with lower R 0 to extinction. The basic reproduction number varies for different pathogens (and their strains) and varies according to the circumstances (geographic location, age structure, population density, previous exposure etc) of the host population [10] . The claim that the outcome of competition between variant strains is a deterministic process mediated by R 0 is the analogue of the claim, which goes back to Fisher [29] , that the rate of increase of total population numbers -the Malthusian parameterdetermines the outcome of competition between an incumbent and a variant type. These studies, in both epidemiology and population genetics assume that populations have infinite size. Evolutionary entropy, in the context of epidemiological models, describes the uncertainty in the age of a parent of a randomly chosen infectives. This quantity is a demographic parameter that is positively correlated with the demographic stability. Directionality theory, a study of the dynamics of competitive invasion [18] [19] [20] [21] [22] 30] of structured populations when population size is finite predicts that the outcome of selection is a stochastic process determined by evolutionary entropy and contingent on population size and two other demographic variables which characterise the geometry of the infectious net-reproductive function. Evolutionary entropy in this more general context describes the uncertainty in the state of the ancestor of randomly chosen infective. (File S1, Section i). In this paper we exploit this general tenet to show that in finite populations with heterogeneity in age of infection the outcome of strain competition is a stochastic event determined by evolutionary entropy and contingent on the demographic parameters W and c. Selective advantage in this case is given by equation 4. One of the most significant parameters in epidemiology is the duration of host infection, D, which is taken to represent the period of time for which an infective is capable of transmission to susceptibles. This parameter is related to T, the generation time. D can be generally expressed in the form D~kT where k is a parameter dependent on the strain, but the characterisation of D will depend on the model system under consideration. In a basic S-I-R model, for example, the fecundity function is a constant independent of infectious age. Consequently the generation time will depend uniquely on the mortality rate (i.e. the rate of recovery from the pathogen). The generation time T will be inversely related to the rate at which individuals leave the infected class, denoted by n. Hence for this class of models D~kT~1=n. In the model described in this paper, survivorship and fecundity are functions of age, so the generation time involves both survivorship and fecundity components. The generation time T can be expressed in terms of the entropy function S. As shown by Demetrius et al. [31] , the evolutionary entropy can be shown to be analytically related to T by This equation asserts, therefore, that in systems with demographic heterogeneity the duration of infection will now be regulated by the pathogen entropy, S. This important fact underscores the fact that demographic heterogeneity dictated by V (x) will induce significant changes in the epidemiological dynamics. These results have particular significance in understanding the epidemiology of influenza A virus. Influenza A epidemics in humans are characterised by infrequent (typically on a decadal timescale), but nevertheless significant, genetic re-assortments that lead to pandemics of new sub-types (antigenic shift) as well as within-sub-type evolution (antigenic drift) on an annual timescale. In the antigenic shift scenario it is assumed that there is very little or no host cross-immunity with previous virus types, whereas within-sub-type drift can generate strains with varying degrees of host immune response. This difference is important in determining how variants invade host populations. Antigenic shift and antigenic drift can be characterised in terms of the demographic parameter W. This variable is analytically related to the population growth rate, r, and the entropy rate S/T by the identity r~S=TzW (where T is the generation time -see File S1 section i for parameter definitions). From this equation it follows that when Wv0, rvS=T. Wv0[rvS=T i.e. r small -corresponds to a relatively small/ negligible growth rate of incumbents. But when Ww0[rwS=T i.e. r large -corresponds to a relatively large growth rate of incumbents. We will consider these two notable epidemiological characteristics in turn: Antigentic Drift: Ww0. Recent work on the genetic and antigenic evolution of Influenza A H3N2 in humans has demonstrated that seasonal 'flu epidemics emerge from seed strains originating in countries of south-east Asia which subsequently spread sequentially through the global population [32] [33] [34] [35] . This suggests that novel H3N2 variants compete with existing strains within the E-SE Asia circulation network with the dominant strain being responsible for generating the next global seasonal 'flu strain. Once out of the seeding region there appears to be little subsequent viral evolution [32] , though as pointed out by Rambaut et al. [36] subsequent changes tend to be deleterious and so die out. When a new H3N2 variant is generated within the seeding region it competes for susceptibles with existing variants. The epidemiological picture in this densely populated SE-Asia seeding region is one of strong fluctuations of multiple circulating strains [32] . Much of this fluctuation is driven by the cross-reactive immunity and heterogeneity of host population immune response to the different strains and finite size population effects. The dynamics are characterised by repeated boom-bust cycles in the strain populations, so competition is taking place between a new variant and a rapidly growing incumbent pathogen population that has a large positive growth rate r, implying that the reproductive potential Ww0. Information on the infectious netreproductive function from which we would infer the value of c is less readily available [33] . It is plausible to assume, however, that the rate at which secondary infections are produced is broadly correlated with pathogen burden. This results in a netreproductive function for influenza [33, Figure 1 ] that is slightly skewed towards the early stages of host infection, implying that c is small and negative. From Table S1 , the constraints Ww0 and cv0 suggest that new variants will enjoy a selective advantage if their evolutionary entropy is lower than other competing strains (DSv0) during the competitive growth phase. In this model the invasion success of the new strain will be dependent on the relative evolutionary entropy of the variant contingent on the demographic variables W and c. The information given in Table S1 can be invoked to determine the selective outcome of competing strains. This information underscores the difference between the deterministic process which defines the classical models and the stochastic process which characterises entropic epidemiology. A variant strain has epidemic potential and will dominate existing H3N2 strains providing that DSv0. By contrast, in a deterministic model (equation 2) the outcome of competition between strains is a deterministic process predicted wholly by R 0 -the strain with the largest basic reproduction number will always dominate. In the entropic model there is a probability of a downward drift in the evolutionary entropy, S, of the dominant strain. Within the classical framework (infinite population size) the outcome is deterministically ordained, that is, we expect that those variants which produce more secondary infectives to dominate all others. By contrast, in a finite population (and contingent on the values of W, and to a lesser extent c) the outcome has an intrinsic stochastic component: it is the evolutionary entropy that determines success so there may be dominant variants which produce fewer secondary infectives (i.e. have a lower growth rate and R 0 ) than their co-circulating competitors. The analytic relation between evolutionary entropy and the duration of infection described above entails that lower entropy strains have a shorter duration of infection. Consequently, we expect new annual 'flu strains to have variable R 0 and short infectious duration relative to strains circulating in the SE-Asia seed network because of their smaller evolutionary entropy, S. This pattern is also seen in Table 1 where successful emergent epidemics have variable R 0 but short infectious durations. Additionally, these lower entropy strains will have a greater resilience in maintaining the chain of infection as the pathogen spreads because they have a selective advantage with respect to other strains with which they have to compete for susceptibles. Basic epidemic models [10] correlate the basic reproduction number R 0 with the duration of infection, D which implies that short duration infections will have smaller basic reproduction numbers and would be more likely to die out. By contrast, the analysis here suggests that in the context of emergent epidemics it is the minimisation of the evolutionary entropy (which is proportional to D -since S~log Dz(b{ log k)) that is the determinant of emergent epidemic success, Figure 1 shows the change in entropy over time for a simulation of the invasion process by variants, and it is clear that there is a tendency to decreasing entropy over time for each run. Classical epidemic models define the proportion of the population that need to be vaccinated to eliminate a disease in terms of R 0 , but the theory presented here shows that this is only true in the limit of infinite population which indicates that in future new vaccination criteria will be required for emergent epidemics which take into account finite population size, variability in infection profile and stochasticity effects. Antigenic Shift: Wv0. On longer timescales completely new influenza A virus sub-types occasionally emerge. These events are unpredictable and usually result in global pandemics with significantly elevated levels of 'flu-related mortality. These new sub-types are thought to arise from hybridisation of human 'flu viruses with those circulating in pigs and/or poultry [8] . Because the new virus is generated by a complete change in the HA antigenic subunit on the surface of the virus the entire global population is essentially susceptible to the disease thereby generating a pandemic. The last major antigenic shift event was in 1968 and it generated the H3N2 (Hong Kong 'flu) strain that has been in circulation since. Following antigenic shift the new influenza variant is in competition with an incumbent strain that is already at equilibrium in the population which suggests that the growth rate r is small, so Wv0. From the perspective of directionality theory (Table S1 ) the favourable condition for establishment of the new type is higher entropy relative to the existing circulating sub-types (DSw0), i.e. a longer duration of infection. In the antigenic shift case we are addressing infection in the entire global population so, in effect the population of infectives, N is very large. Therefore the role of the demographic index term c (though again it will be small and ,0, for influenza) is minimal. The larger entropy of the variant corresponds to longer infectious duration than the circulating incumbents. However, the mechanism of antigenic drift (as described above) then begins to operate on this newly established sub-type and so there will be a gradual decrease in infectious duration of the dominant circulating variant with time. Given that the new virus is likely to be a hybrid animal-human type, it is likely that it is less well adapted to human hosts so it might have a low R 0 . If, rather crudely, we associate infectious duration in the host with pathogenicity, directionality theory implies high pathogenicity immediately following establishment of a new sub-type followed by decreasing pathogenicity in subsequent years. That is, the new Influenza A sub-type (such as 1918, 1957 and 1968) appears to be highly pathogenic in the immediate interval following establishment, but there is a contribution to the decrease in 'flu mortality in the era following the antigenic shift by the action of competitive selection of lower entropy variants during the antigenic drift phase in the SE-Asia seeding region. When a new Influenza A virus sub-type is generated due to antigenic shift the cycle is repeated again. Clearly, there will be a decrease in the absolute influenza mortality figure (number of fatalities) during the drift phase because as time passes the overall population immunity level increases. However, the model suggests that the case fatality rate (CFR -number of fatalities per infected individuals) will decline during the drift phase due to declining pathogenicity. A decline in the CFR is observed empirically when comparing a pandemic attack year and the next subsequent epidemic [39] but there does not appear to be any detailed epidemiological analysis of long-term CFR trends during the drift phase. A recent detailed study of the epidemiology of Influenza A H1N1 in the era 1918-1951 [40] shows, Figure 2 , that during the H1N1 era there is evidence of a sustained decrease in mortality rate between 1918 (autumn wave) and 1924 and likewise between 1928 and 1944. Whilst some of this progressive weakening of the influenza epidemics is due to increasing host population immunity it is likely that there is also a contribution to declining pathogenicity resulting from the mechanisms proposed in the directionality theory. H1N1 (2009) Swine flu. In March 2009 the first reports of an epidemic of a novel influenza-like pathogen emerged in Mexico [13] . Analysis showed that the infectious agent in this on-going epidemic to be Influenza A H1N1 (swine flu) [7] . H1N1 has generated epidemics in humans in the past and was responsible for the 1918 influenza pandemic and a 1977 'flu epidemic. It is possible to explain the relentless spread of this current outbreak in directionality theory. This new variant has emerged from a 'flu type associated with pigs, but now has human-to-human transmission capability. This successful variant has a shorter infectious duration than other influenza A strains [13] which suggests that it may have emerged with a competitive advantage founded on its lower entropy DSv0 ð Þrelative to other currently circulating 'flu strains. The extent of pre-existing immunity to H1N1 (swine flu) is currently unknown, but our results suggest that in humans there may be some pre-existing protection from previous exposure to influenza virus. This possibility is also noted by Fraser et al. [13] . Alternatively it may simply be a novel pandemic strain with short infectious duration. SARS had a higher R 0 and longer infectious duration (Table 1 ) than swine flu yet did not have the same global impact, which reinforces the generic suggestion from entropic considerations that it is those emergent diseases with shorter infectious durations that appear to have the greater pandemic potential. Existing endemic infections. We noted above that many antigenically stable infectious diseases (measles, chickenpox for example) have comparatively long infectious durations compared with emergent infections (Table 1 ). Directionality theory suggests that this might be the consequence of a long evolutionary adaptation to humans by occasional mutations resulting in ever higher evolutionary entropy S. In this picture, for a mutation to dominate an established equilibrium incumbent strain it is necessary for the variant to have DSw0, so on evolutionary time scales we see an upward drift in entropy and, consequently, ever increasing duration of infectiousness. Although such diseases do have a seasonal component to their incidence they nevertheless exist at some stable mean prevalence within the host population. The next dominant strain measles has to compete against a long-established incumbent strain that is at overall equilibrium in the population. Figure 3 shows the change (upward drift) in entropy of the most frequent variants in the population using a simulation of the invasion process. In this picture the dynamics of invasion is considered at a global scale (number of infectives N large) so the conventional Malthusian picture re-asserts itself. Consequently in this situation there will be proportionality between the duration of infection and R 0 , so the concept of a basic reproduction number retains its conventional role as a measure of selective advantage and, hence, its usefulness as a metric for the amount of vaccination required to eliminate a given pathogen strain. It is apparent that within the approach presented here the epidemiological context (emergent pathogen versus established equilibrium) in which the competition takes place does matter to the outcome and the properties of strains that will dominate. In summary, directionality theory shows that during the fluctuating (opportunistic) competitive growth phase of multi-strain pathogen epidemic establishment it is short infectious duration (low entropy) strains that are favoured over longer infectious duration (higher entropy) strains. Moreover, they will be resilient to competition from other strains thereby giving them pandemic potential. By contrast, in established (equilibrium) populations it is longer infectious duration (high entropy) strains that have competitive advantage. This suggests that the epidemiological circumstances, opportunistic or equilibrium, that are prevalent in a host population during competitive emergence are critical in determining the properties of the dominant pathogen strain. To be clear, the stochasticity and fluctuations present in this model arise from consideration of an infection process in a finite population that has infection demographics defined by the function V (x). In this case the straightforward application of the concept of the basic reproductive number can be of limited usefulness as a key determinant of epidemic dynamics as there is no longer an automatic correlation between R 0 and competitive dominance. The model presented above has some explanatory power beyond that of conventional theory in that it suggests that for Influenza A new pandemic variants generated by antigenic shift will be more pathogenic (assuming pathogenicity correlates with infectious duration) than the subsequent seasonal strains generated by the process of evolutionary antigenic drift. From a public health perspective these results suggests that monitoring of those emergent strains with a shorter infectious duration is a better indicator of pandemic risk than focussing on just R 0 , as they present an elevated threat of triggering pandemics and may need to be the target of timely vaccine development. Moreover, these results suggest the calculations of R 0 may not provide a reliable guide to the vaccination effort required to eliminate an emergent pandemic strain. The limitation of a singular focus on R 0 has been highlighted by Meyers [41] in the context of epidemics on networks. However, further work is needed to develop the application of directionality theory to empirical epidemiological questions such as determining the optimal vaccination coverage. Table 2 contrasts the classical and entropic models in order to emphasize their fundamental differences in explanatory and predictive power. The reasons why some flu strains are more pathogenic than others is a complex issue involving specific details of host-virus interactions, but the model we propose has the attraction of capturing (on a simple criterion of pathogenicity, at least) evolution to generally less-pathogenic strains. The determinants that drive empirically observed patterns of emergence and spread of novel infectious pathogens are incompletely understood, turning as it does on the interplay of epidemiological, immunological and genetic considerations. No single model is able to capture the full complexity of this reality, but the work presented here is intended to shed some light on the criteria for invasion success and subsequent evolution of emergent strains. Our results show that conditions of demographics of the infection process, finite population size and consideration of the prevailing epidemiological dynamics against which strain competition occurs together impose limitations on the explanatory and predictive power of any analysis based solely on the basic reproduction number. The concept of evolutionary entropy provides a framework that is stochastic in its foundation for resolving these limitations. The population of infectives is divided into a number of discrete ''age'' classes. Each day every individual either moves up to the next age class or moves to the Recovered class with probability b i . An infective in age-class i produces on average m i infectives. These new infectives each begin their journey through the infective stage in age-class 1. Consequently, there are two functions, defined by l i and m i , (and hence V i~li m i see File S1 section i) that characterise a pathogen strain and its behaviour in the host. Transitions in the simulation are decided by a stochastic process. The simulation starts with N wild-type infectives. Each day each infective has its infectious age increased by 1. A random number in the interval (0,1) is then generated for each infective. If this number is ,current recovery probability b i then the individual recovers, otherwise it generates its quota of secondary infectives m i . New infectives begin with infectious age = 0. For each new infective a random number in the range (0,1) is generated. If it is ,mutation rate then this new infective will be a variant strain. Mutations are generated by a small perturbation to the function V i (see below ''Mutation and Competition''). Each day the number of each strain is calculated so that the dominant (highest frequency) strain can be identified. The entropy, S, of this strain is then calculated using the demographic parameters. To simulate the Ww0 scenario in Figure 1 (which corresponds to antigenic drift in the SE Asia region) requires rapid strain growth rates (r large). This is done by initially allowing the total population (of all strains) to grow rapidly. Once the supply of susceptibles becomes depleted the population collapses abruptly (resource availability variable). The supply of susceptibles is then re-instated and the boom-bust cycle repeats itself. New strain variants are generated through the cycle. The purpose of this is to mimic the conditions for emergence of new variants when there is competition and growth. In this scenario variants that compete against each other are not at equilibrium in the population so we are addressing a localised competitive situation. The total population size used for this simulation was 10,000 individuals with d(x)~a with a in the range +0:1 with a mutation rate of 10 {5 day -1 . To simulate the Wv0 scenario in Figure 3 the supply of susceptibles is controlled to maintain the total population of infectives at a broadly constant level (i.e. resource availability constant). This reflects low-to-minimal growth rate (equilibrium, r small) of the incumbent. The number of infectives fluctuates around an equilibrium level and new variants attempt to invade the system whilst it is in this configuration. In this scenario the incumbent is already established at equilibrium, so we are addressing a global competitive situation. These simulations were run for 200,000 days with d(x)~a with a in the range +0:1 and with a mutation rate of 10 {4 day -1 . Each variant is characterised by the net fecundity function V i . We assume that mutants are defined by V Ã (x)~V (x) 1zd(x) where d(x) is monotonic in x. As a consequence, mutants arise from translation on the function V (x) (corresponding to a change in the age of infector) or from a re-scaling of V (x) (corresponding to an increase or decrease in the net fecundity function). Monotonicity is imposed to preclude net-fecundity profiles that are large at early and late stages and low at intermediate stages. The genotypes of mutant and wild-types are constructed so as to have positive growth rates. Consecutive time-steps of evolution are simulated by generating random numbers to decide which individuals recover or continue as infectives. To simulate competition between strains some additional growth constraints have to be applied to each scenario. In the Ww0 case the initial population is allowed to grow rapidly from an initial starting number N init . Following exhaustion of the supply of susceptibles (i.e. resources are depleted) an extrinsic mortality 1{N init =N ð Þis applied to all individuals to reduce the population of infectives back down to its starting value. This process is repeated over many time-steps. In the Wv0 case if the total population size (N max ) of infectives is exceeded an extrinsic mortality 1{N max =N ð Þis applied probabilistically so that the population is maintained at a level that fluctuates around N max . In both scenarios, at each time step, the dominant (most frequent) genotype is determined and its entropy calculated from equation S4.The value of this entropy is recorded for the duration of the simulation. It should be noted that these simulations are not based on an elaboration of S-I-R models of the usual type where susceptibles and infectives interact via conventional mass-action terms. Here, the population of infectives is directly manipulated to reflect the kind of epidemiological dynamics that are typically seen in the emergent and equilibrium phases [42] . Table S1 Invasion criteria in the entropy model. *''a.s. = Almost surely'' refers to the fact the result is a stochastic process. The criteria for large and small population size are defined in more detail in Demetrius et al. [19] . The criteria noted in Table S1 have been tested against simulation where they have been shown to be replicated. Found at: doi:10.1371/journal.pone.0012951.s001 (0.07 MB DOC) Figure S1 Life cycle for an infective corresponding to the matrix in equation S1 with 4 infectious age classes.

The Nature of Protein Domain Evolution: Shaping the Interaction Network

The proteomes that make up the collection of proteins in contemporary organisms evolved through recombination and duplication of a limited set of domains. These protein domains are essentially the main components of globular proteins and are the most principal level at which protein function and protein interactions can be understood. An important aspect of domain evolution is their atomic structure and biochemical function, which are both specified by the information in the amino acid sequence. Changes in this information may bring about new folds, functions and protein architectures. With the present and still increasing wealth of sequences and annotation data brought about by genomics, new evolutionary relationships are constantly being revealed, unknown structures modeled and phylogenies inferred. Such investigations not only help predict the function of newly discovered proteins, but also assist in mapping unforeseen pathways of evolution and reveal crucial, co-evolving inter- and intra-molecular interactions. In turn this will help us describe how protein domains shaped cellular interaction networks and the dynamics with which they are regulated in the cell. Additionally, these studies can be used for the design of new and optimized protein domains for therapy. In this review, we aim to describe the basic concepts of protein domain evolution and illustrate recent developments in molecular evolution that have provided valuable new insights in the field of comparative genomics and protein interaction networks.

The protein universe is the collection of proteins of all biological species that exist or have once existed on Earth [1] . Our sampling and understanding of it began over half a century ago, when the first peptide and protein sequences were determined by Sanger [2, 3] and, subsequently, the sequencing of RNA and DNA [4] [5] [6] . In the meantime, the genome projects of the last decade have uncovered an overwhelming amount of sequence data and researchers are now starting to address a series of fundamental questions that should shed light onto protein evolution processes [7] [8] [9] [10] . For instance, how many gene encoding sequences are present in one genome? How many sequences are repetitive and are these sequences similar in the various organisms on Earth? Which genes were involved in the large scale genome duplications that we see in animals? A comparison of sequences for evolutionary insight is best achieved by looking at the structural and functional (sub)units of proteins, the protein domains. By convention, domains are defined as conserved, functionally independent protein sequences, which bind or process ligands using a core structural motif [11] [12] [13] . Examples of domain modes of actions in signaling cascades for instance, are to connect different components into a larger complex or to bind signaling-molecules [14, 15] . Protein domains can usually fold independently, likely due to their relatively limited size, and are well known to behave as independent genetic elements within genomes [16, 17] . The sum of these features makes protein domains readily identifiable from raw nucleotide and amino acid sequences and many protein family resources (e.g., Superfamily and SMART [see Table 1 ]) indeed fully rely on such sequence similarity and motif identifications [18, 19] . The algorithms that are used for domain identification are built around a set of simple assumptions that describe the process of evolution. In general, evolution is believed to form and mold genomes largely via three mechanisms, namely i) chemical changes through the incorporation of base analogs, the effects of radiation or random enzymatic errors by polymerases, ii) cellular repair processes that counter mutations, and iii) selection pressures that manifest themselves as the positive or negative influence that determines whether the mutation will be present in subsequent generations [20, 21] . By definition, each of these phenomena styles, reproductive strategies, or the lack of apparent polymerase-dependent proofreading such as in positivestranded RNA viruses [22] [23] [24] [25] . Consequently, substitution rates need therefore be calculated to correctly compare two or more sequences and hunt uncharted genomes for comparable domains. Particularly this last strategy, using general rate matrices like BLOSOM and PAM, is an elegant example of how new protein functions can be discovered [26] [27] [28] [29] [30] . Fast algorithms for pair-wise alignments can be found in the Basic Local Alignment Search Tool (BLAST), whereas multiple sequence alignments (MSAs, Fig. 1A) in which multiple sequences are compared simultaneously are commonly created with for example ClustalX and MUSCLE (see Table 1 ) [31] [32] [33] [34] . Close relatives, sharing an overall sequence identity above for example 50% and a set of functional properties, can also be grouped into families and subfamilies. In turn, these families share also evolutionary relationships with other domains and form together so-called domain superfamilies [18, 35] . Evolutionary distances between related domain sequences can easily be estimated from sequence alignments, provided that the correct rate assumptions are made. Subsequently, these can be used to compute the phylogenies of the domain that share an evolutionary history. These, often tree-like graphs (Fig. 1B) , depend heavily on rate variation models, such as molecular clocks or relaxed molecular clocks (e.g., Maximum Likelyhood and Bayesian estimation), which are calibrated with additional evidence Fig. (1A) . It was computed using Bayesian estimation and presents the best-supported topology for the alignment. Numbers indicate % support by the two methods used, while # indicates gene duplication events in the common ancestor and * marks a species-specific duplication event. For computational details, please see [42] . such as fossils and may therefore also provide valuable information on aspects like divergence times and ancestral sequences [36] [37] [38] . Commonly used phylogenetic analysis strategies are listed in Table 1 . A limitation of all inferred phylogenetic data is that it is directly dependent on the alignment and less so on the programs used to build the phylogenetic tree [39] . One of the shortcomings of automated alignments may thus derive from the fact that they commonly employ a scoring and penalty procedure to find the best possible alignment, since these parameters vary from species to species [22, 23] , as mentioned above. Careful inspection of alignments is therefore advisable, even though software has been developed that combines the alignment procedure and phylogenetic analysis iteratively in one single program [40] . Although sequence and phylogenetic analysis provide a relatively straightforward way for looking at domain divergence, comparison of solved protein structures has shown that protein tertiary organizations are much more conserved (>50%) than their primary sequence (>5%) [41] . For this reason, protein structures and their models provide significantly more insight into the relations of protein domains and how domain families diverged [16] . For example, the inactive guanylate kinase (GK) domain present in the MAGUK family was shown to originate from an active form of the GK domain residing in Ca2+ channel beta-subunits (CACNBs) through both sequence and structural comparison [42] . Furthermore, identification of functionally or structurally related amino acid sites in a fold sheds light on the complex, co-evolutionary dynamics that took place during selection [43] . As described above, the evolution of a protein domain is generally the result of a combination of a series of random mutations and a selection constraint imposed on function, i.e., the interaction with a ligand. The interaction between protein and ligand can be imagined as disturbances of the protein's energy landscape, which in turn bring about specific, three-dimensional changes in the protein structure [44, 45] . Binding energies however, need not be smoothly distributed over the protein's binding pocket as a limited number of amino acids may account for most of the free-energy change that occurs upon binding [45] [46] [47] . In these cases, new binding specificities (including loss of binding) may therefore arise through mutations at these hot spots. An example is a recent study of the PDZ domain in which it was shown that only a selected set of residues, and in particular the first residue of -helix 2 ( B1), directly confers binding to a set of C-terminal peptides [48] . The folding of a domain is essentially based on a complex network of sequential inter-molecular interactions in time [49] . This has of course significant implications for domain integrity, particularly if one assumes that the core of a protein domain is and has to be largely structurally conserved. Indeed, even single mutations that arise in this area may easily derail the folding process, either because their free energy contribution influences residues in the direct vicinity or disturbs connections higher up in the intermolecular network [49] . It is therefore hypothesized that protein evolution took place at the periphery of the protein domain core, and that gradual changes via point mutations, insertions and deletions in surface loops brought about the evolutionary distance we see among proteins to date [21, [50] [51] [52] . However, distant sites also contribute to the thermodynamics of catalytic residues. This is achieved through a mechanism called energetic coupling, which is shaped by a continuous pathway of van der Waals interactions that ultimately influences residues at the binding site with similar efficiency as the thermodynamic hotspots [53, 54] . Indeed in such cases, evolutionary constraints are not placed on merely one amino acid in the binding pocket, but on two or more residues that can be shown to be statistically coupled in MSAs [54, 55] . In addition to contributions to binding, these principles also explain why the core of a domain structure will remain largely conserved, while at functionally related places residues can (rapidly) co-evolve with an overall neutral effect [56] . Of course, these aspects of co-evolution are also of practical consequence for structure prediction and rational drug design [43] . Through selective mutation, protein domains have been the tools of evolution to create an enormous and diverse assembly of proteins from likely an initially relatively limited set of domains. The combined data in GenBank and other databases now covers over 200.000 species with at least 50 complete genomes and this greatly facilitates genome comparisons [57] [58] [59] . Following such extensive comparisons, currently > 1700 domain superfamilies are recognized in the recent release of the Structural Classification of Proteins (SCOP) [60] and it has become clear that many proteins consist of more than one domain [17, 61, 62] . Indeed, it has been estimated that at least 70% of the domains is duplicated in prokaryotes, whereas this number may even be higher in eukaryotes, likely reaching up to 90% [35] . There are various mechanisms through which protein domain or whole proteins may have been duplicated. On the largest scale, whole genome duplication such as those seen in the vertebrate genomes duplicated whole gene families, including postsynaptic proteins, hormone receptors and muscle proteins, and thereby dramatically increased the domain content and expanded networks [42, 63, 64] . On the other end of the scale, domains and proteins have been duplicated through genetic mechanisms like exon-shuffling, retrotranspositions, recombination and horizontal gene transfer [65] [66] [67] . Since the genetic forces, like exon-shuffling and genome duplication vary among species, the total number of domains and the types of domains present fluctuate per genome. Interestingly, comparative analyses of genomes have shown that the number of unique domains encoded in organisms is generally proportional to its genome size [60, 68] . Within genomes, the number of domains per gene, the socalled modularity, is related to genome size via a power-law, which is essentially the relation between the frequency f and an occurrence x raised by a scaling constant k (i.e., f (x) x k ) [69, 70] . A similar correlation is found when the multi-domain architecture is compared to the number of cell types that is present in an organism, i.e., the organism complexity or when the number of domains in a abundant superfamily is plotted against genome size (Fig. 2) [71, 72] . Given the amount of domain duplication and apparent selection for specific multi-domain encoding genes in, for example, vertebrates, it may come as little surprise that not all domains have had the same tendency to recombine and distribute themselves over the genomes [68, 73] . In fact, some are highly abundant and can be found in many different multi-domain architectures, whereas others are abundant yet confined to a small sample of architectures or not abundant at all [68, 70] . Is there any significant correlation between the propensity to distribute and the functional roles domains have in cellular pathways? Some of the most abundant domains can be found in association with cellular signaling cascades and have been shown to accumulate non-linearly in relation to the overall number of domains encoded or the genome size [70] . Additionally, the on-set of the exponential expansion of the number of abundant and highly recombining domains has been linked to the appearance of multicellularity [70] . A reoccurring theme among these abundant domains is the function of protein-protein interaction and it appears that particularly these, usually globular domains, have been particularly selected for in more complex organisms [70] . This positive relation is underlined by the association of these abundant domains with disease such as cancer and gene essentiality as the highly interacting proteins that they are part of have central places in cascades and need to orchestrate a high number of molecular connections [74, 75] . Their shape and coding regions, which usually lie within the boundaries of one or two exons, make them ideally suited for such a selection, since domains are most frequently gained through insertions at the N-or C-terminus and through exon shuffling [76] [77] [78] . From a mutational point of view, protein-protein interaction domains are different from other domains as well and this appears to be particularly true for the group of small, relatively promiscuous domains like SH3 and PDZ. These domains are promiscuous in the sense that they both tend to physically interact with a large number of ligands [79, 80] and are prone to move through the genome to recombine with many other domains. It has been found that particularly these domains evolve more slowly than non-promiscuous domains [70] . This likely stems from the fact that they are required to participate in many different interactions, which makes selection pressures more stringent and the appearance of the branches on phylogenetic trees relatively short and more difficult to assess when co-evolutionary data in terms of other domains in the same gene family or expression patterns is limited [42, 63] . Non-promiscuous domains on the other hand can quite easily evade the selection pressure by obtaining compensatory mutations either within themselves or their specific binding partner [70] . The overall phenomenon that the number of protein domains and their modularity increases as the genome expands has not been linked to a conclusive biological explanation yet. A rationale for the increase in interactions and functional subunits, however, may derive from the paradoxical absence of correlation between the number of genes encoded and organism complexity, the so-called G-value paradox [81] . There is indeed evidence that domains involved in the same functional pathway tend to converge in a single protein sequence, which would make pathways more controllable and reliable without the need for supplementary genes [73] . Additionally, the number of different arrangements found in higher eukaryotes is, given the vast scale of unique domains present, relatively limited. This in turn implies that evolutionary constraints have played an important role in selecting the right domain combinations and the right order from N-to C-terminus in multi-domain proteins [13, 82] . In fact, the ordering and co-occurrence of domains was demonstrated to hold enough evolutionary information to construct a tree of life similar to those based on canonical sequence data [70] . Furthermore, the increased use of alternative splicing and exon skipping in higher eukaryotes likely supplied a novel way of proteome diversification by restricting gene duplication and stimulating the formation of multi-domain proteins [83, 84] . In plants, however, the latter notion is not supported since both mono-and dicots show limited alternative splicing and a more extensive polyploidy [85] [86] [87] . It is clear that some of the above characteristics are underappreciated in the phylogenetic analysis of linear amino acid sequences. Moreover, the effects of evolution extend even further than these aspects and entail transcriptional and translational regulation, intramolecular domain-domain interactions, gene modifications and post-translational protein modifications [88] [89] [90] [91] [92] [93] [94] [95] [96] . New methods are thus being developed to take into account that when sequences evolve, their close and distant functional relationships evolve in parallel. Correlations of mutations have already been found between residues of different proteins [97, 98] and compensating mutational changes at an interaction interface were shown to recover the instability of a complex [99] . These observations are evidence for the current evolutionary models for the protein-protein interaction (PPIs) networks that are being constructed through large-scale screens [100] [101] [102] . In these, a gene duplication or domain duplication (depending on the resolution of the network) implies the addition of a node, while the deletion of a gene or domain reduces the amount of links in the network (Fig. 3) . In the next step, extensive network rewiring may take place, driven by the effect of node addition or node loss in the network (i.e., the duplicability or essentiality of a domain/protein) and mutations in the domain-interaction interface [67, 74, [103] [104] [105] . Beyond mutations at the domain and protein level, regulation of protein expression provides another vital mechanism through which protein networks can evolve. Microarray studies are now well under way to map genome-wide ex-pression levels of related and non-related genes under a variety of conditions [91, [94] [95] [96] . For example, transcriptional comparisons have investigated aging [106] and pathogenicity [107] . Unfortunately, given the highly variable nature of gene expression and the fact that different species may respond different to external stimuli, such comparisons can only be performed under strictly controlled research conditions. To date most studies have therefore focused on the embryogenesis, metamorphosis, sex-dependency and mutation rates of subspecies [94, [108] [109] [110] [111] . Other studies have revealed valuable information on promoter types and duplication events [91] [92] [93] [94] . To overcome the limitations mentioned in the previous paragraph, the analysis of co-expression data has been developed to supplement the direct comparison of individual gene expression changes [95] . In this procedure, a coexpression analysis of gene pairs within each species precedes the cross-comparison of the different organisms in the study. This approach thus primarily focuses on the similarity and differences of the orthologous genes within network, and is therefore ideally suited for the study of protein domain evolution and has already revealed that species-specific parts Fig. (3) . Evolutionary models for protein-protein interactions. The evolution of protein networks is tightly coupled to the addition or deletion of nodes. Additionally, events that introduce mutations in binding interfaces of proteins may result in the addition or loss of links in the network. Node addition may take place through e.g., domain duplication or horizontal gene transfer, while rewiring of the network is mediated by point mutations, alternative splice variants and changes in gene expression patterns. of an expression network resulted via a merge of conserved and newly evolved modules [95, 112, 113] . Finding evolutionary relationships protein domains is mostly based on orthology and thus commonly performed on best sequence matches. Identifying these and categorizing them depends largely on multiple sequence alignments and this will in most cases give good indications for function, fold and ultimately evolution. However, this approach usually discards apparent ambiguities that arise from speciesspecific variations (e.g., due to population size, metabolism or species-specific domain duplications or losses) and may therefore introduce significant biases [114] . Biases may also derive from the method of alignment, the rate variation model used to infer the phylogeny, and the sample size used to build the alignment [39, 40, 115] . Care should therefore be taken to not regard orthology as a one-to-one relationship, but as a family of homologous relations [91] , to select for appropriate analysis methods [39, 115] and extend comparative data to protein interactions and expression profiles [91] . Indeed, as our wealth of biological information expands, our systems perspective will improve and provide us with an opportunity to reveal protein domain evolution at the level network organization and dynamics. Large-scale expression studies are beginning to show us evolutionary correlations between gene expression levels and timings [94, 106, 107, 112, 116] , while others demonstrate spatial differences between paralogs or (partial) overlap between interaction partners [117] [118] [119] [120] . Indeed, when we are able to map the spatiotemporal aspects of inter-and intra-molecular interactions we will begin to fully understand the versatile power of evolution that shaped the protein universe and life on Earth [118] .

Quantitative Phosphoproteomics of Proteasome Inhibition in Multiple Myeloma Cells

BACKGROUND: The proteasome inhibitor bortezomib represents an important advance in the treatment of multiple myeloma (MM). Bortezomib inhibits the activity of the 26S proteasome and induces cell death in a variety of tumor cells; however, the mechanism of cytotoxicity is not well understood. METHODOLOGY/PRINCIPAL FINDINGS: We investigated the differential phosphoproteome upon proteasome inhibition by using stable isotope labeling by amino acids in cell culture (SILAC) in combination with phosphoprotein enrichment and LC-MS/MS analysis. In total 233 phosphoproteins were identified and 72 phosphoproteins showed a 1.5-fold or greater change upon bortezomib treatment. The phosphoproteins with expression alterations encompass all major protein classes, including a large number of nucleic acid binding proteins. Site-specific phosphopeptide quantitation revealed that Ser38 phosphorylation on stathmin increased upon bortezomib treatment, suggesting new mechanisms associated to bortezomib-induced apoptosis in MM cells. Further studies demonstrated that stathmin phosphorylation profile was modified in response to bortezomib treatment and the regulation of stathmin by phosphorylation at specific Ser/Thr residues participated in the cellular response induced by bortezomib. CONCLUSIONS/SIGNIFICANCE: Our systematic profiling of phosphorylation changes in response to bortezomib treatment not only advanced the global mechanistic understanding of the action of bortezomib on myeloma cells but also identified previously uncharacterized signaling proteins in myeloma cells.

The ubiquitin-proteasome pathway is responsible for proteolysis of eukaryotic cellular proteins related to cell cycle regulation, cell survival, and apoptosis [1] . Inhibition of proteasome activity is a novel therapeutic strategy against cancer cells. Bortezomib (formerly known as PS-341), a cell-permeable boronic acid dipeptide, is a specific inhibitor of the proteasome pathway [2] and received Food and Drug Administration (FDA) approval for the treatment of MM and mantle cell lymphoma [3] . Bortezomib has been reported to trigger pleiotropic signaling pathways in MM cells, including: (a) stabilizing cytoplasmic IkB and blocking NFkB nuclear translocation [4] ; (b) activation of stress response proteins such as heat shock proteins Hsp27, Hsp70, and Hsp90 [5] ; (c) up-regulation of c-jun NH2-terminal kinase [6] ; (d) induction of intrinsic cell death pathway [7] ; (e) activation of extrinsic apoptotic signaling through Bid and caspase-8 cleavage [8] ; (f) impairment of DNA repair machinery via inactivation of DNA-dependent protein kinase [9] ; (g) down-regulation of mitogen-activated protein kinase and phosphatidylinositol 3kinase/Akt signaling pathways [10] ; and (h) down-regulation of the p44/42 MAPK signaling cascade [11] . All these signaling events may collectively contribute towards the overall anti-MM activity of bortezomib. However, the exact number and identity of cellular signaling events involved in proteasome inhibition and the mechanisms underlying the associated apoptotic response in MM cells remain to be elucidated. Elucidation of cellular signaling networks requires methodologies for large-scale quantitative phosphoproteomic analysis that can reveal dynamic system-wide changes in protein phosphorylation. Recent technological advances in mass spectrometry-based proteomics have enabled us to make a large-scale identification of signaling molecules through the enrichment of phosphorylated proteins or peptides [12, 13] . One of the most widely used and well known strategies currently used in phosphoproteomic studies is stable-isotope labelling by amino acids in cell culture (SILAC). Although introduced relatively recently, SILAC has been used extensively in the proteomics community [14] . With SILAC, the entire proteome of a given cell population is metabolically labeled by heavy, non-radioactive isotopic variants of amino acids, thus making it distinguishable by MS analysis [15, 16] . Thereafter, two or more distinctly SILAC-labeled cell populations can be mixed and analyzed in one MS experiment, allowing accurate quantization of proteins from the different cellular states. By coupling with a phosphoprotein or phosphopeptide enrichment method, such as titanium dioxide (TiO 2 ) [17] , strong cation exchange (SCX) [18] , or the two in combination [19] , SILAC has been widely applied to profile dynamic phosphorylation changes in signal transduction [20, 21] . In this study, we investigated the differential MM phosphoproteome upon proteasome inhibition by using SILAC in combination with phosphoprotein enrichment and LC-MS/MS analysis. Many potential novel signaling proteins and associated signaling pathways were confidently identified. Our further functional results indicated that perturbations in stathmin phosphorylation play a significant functional role in mediating apoptosis in MM cells exposed to bortezomib and the bortezomib-induced changes in the MT stabilization can be attributed to the bortezomibinduced phosphorylation of stathmin. By correlating the phosphoproteomic data with functional studies, the current results provided novel insights into the mechanisms of bortezomib actions in MM cells. To obtain a global view of the changes of protein phosphorylation in bortezomib-treated myeloma cells, we compared the phosphoproteome of U266 cells treated with or without bortezomib. The workflow is outlined in Figure 1 . Cells in normal medium (light culture) were treated with bortezomib, and cells grown in medium containing stable isotopes (heavy culture) were treated with vehicle. These two populations of cells were lysed, mixed at a 1:1 ratio, and subjected to TiO 2 purifications followed by LC-MS/MS analysis. After LC-MS/MS analysis on the enriched phosphopeptides, all MS/MS spectra were searched, respectively, against the forward and reversed human protein sequence databases to estimate rates of false-positive matches. Search results were filtered based on peptide score of MASCOT and PTM score. In total 1024 phosphopeptides (redundant) from the target database passed our criteria, allowing 14 decoy matches. The phosphopeptide false-positive rate was therefore estimated to be 1.4%. Multiple filtering criteria were established to validate search results. For each of the phosphorylated peptides identified in this work, peptide sequences were manually confirmed. After validation we identified 418 unique phosphorylation sites from 244 unique phosphopeptides corresponding to 233 protein groups. This entire dataset is provided as Table S1 in which hyperlinks are built up to view all the MS/MS spectra. Using the PTM score, we could localize the phosphor groups with high confidence (Class I phosphorylation sites) in 197 cases (Table S1), indicating that the phosphorylation sites detected by current strategy are of high confidence. Figure 2A shows a representative MS/MS spectrum for a phosphosite-containing peptide in the detection, and all other MS/MS spectra are available via the hyperlinks in the Table S1 . To quantify the phosphorylation change for each phosphopeptide, we used the MSQuant software to calculate area ratio, defined as the ratio of the ''heavy'' peak area over the ''light'' peak area in the chromatogram (Fig. 2B ). All together, we have achieved quantification of 259 unique phosphorylation sites from 154 unique phosphopeptides corresponding to 132 protein groups. Based on a predefined threshold of 1.5-fold change, 131 phosphosites from 75 unique phosphopeptides corresponding to 72 proteins showed a 1.5-fold or greater change as listed in Table S2 . In other words, 31% of the total phosphosites detected showed significant alteration after bortezomib treatment, suggesting that dysregulated phosphorylation may play an important role in bortezomib-induced apoptosis. To further confirm the results from the quantitative phosphoproteome analysis, we chose stathmin for Western blotting verification using an anti-phospho-Ser38 stathmin antibody. As shown in Figure 2C , SILAC results were very much consistent with Western blotting analysis for this protein. Upon bortezomib treatment, phosphorylation of stathmin at Ser38 increased, whereas steady-state stathmin remained almost unchanged in Western blotting verification. To better characterize bortezomib-regulated phosphoproteins, we classified all the differentially expressed phosphoproteins (DEPPs) into 23 functional categories according to the PANTHER system. These proteins are implicated in a broad range of cellular activities (Fig. 3A ). Next to the unclassified proteins (17%), proteins involved in nucleic acid binding account for the second largest portion (14%). There are also a significant number of proteins involved in receptor (10%), regulatory molecule (6%), and kinase (5%). Previous studies indicated that bortezomib exerts its anticancer function by inhibiting protein degradation in cancer cells [22] . In this connection, many bortezomib-regulated phosphoproteins found in current study were nucleic acid binding (14%) and transcription factors (4%), as shown in Figure 3A . These data suggest that the cancer-inhibitory effect of bortezomib may also rely on its regulatory role in mRNA transcription. Table S2 only shows a list of all borterzomib regulated phosphoproteins in MM cells. It lacks the biochemical context. To create significance out of otherwise static proteomic data, we constructed an interaction networking of the DEPPs (Fig. 3B) . The DEPPs were linked by various evidences based on neighborhood analysis, experimental results or text-mining. These relationships are color coded, and the legends are provided next to the map. This does not mean that all the interactions took place within a single spatial and temporal situation, but it enables the identification of central nodes in these DEPPs. Notably, stathmin was a hub in this network, suggesting that stathmin may play an important role in mediating bortezomib-induced apoptosis in MM cells. Some proteins remained orphans because there is insufficient information in the database to link them to other proteins in the network. To further investigate the reliability of the results, PhosphoSite (http://www.Phosphosite.org) was used to distinguish known phosphorylation sites from novel phosphorylation sites. Among the identified 418 phosphorylated sites, 51% were also reported by others previously (for details, see Table S3 ). In other words, many of the phosphorylation sites were also determined by researchers with other cancer cells, further demonstrating that the phosphorylation sites detected by current strategy are reliable. To predict the kinase substrate relationships from the dataset, the computer algorithm SCANSITE was used. SCANSITE makes use of peptide library phosphorylation data to predict substrates recognized by specific kinases. Tables S4 & S5 show the results of phosphopeptides that were identified in this study and were predicted to be associated with a kinase binding motif by SCANSITE at the different stringency levels. It was found that most of the phosphorylation sites determined in this study were phosphorylated by acidophilic serine/threonine kinase and proline-dependent serine/threonine kinase. Notably, we found fifteen sites that can be phosphorylated by the Casein Kinase 2 (CK2). CK2 is a constitutively active protein kinase implicated in cellular transformation and the development of tumorigenesis [23] . Aberrantly active in MM cells, CK2 controls the cell survival [24] . Our SCANSITE analysis suggests that CK2 may play an important role in bortezomib-induced apoptosis and may represent a potential target in MM therapy. SILAC phosphoproteomic analyses and Western blotting revealed an increase of phosphorylation of stathmin at Ser38 and the unchanged steady-state stathmin in U266 cells upon proteasome inhibition (Fig. 2 ). It has been reported that the functional alteration of stathmin resulting from specific phosphorylation events may be involved in the process of apoptosis induced by proteasome inhibitors in proliferating cells [25] . To elucidate the differential phosphorylation of stathmin isoforms, Western blotting was performed to analyze the phosphorylation pattern of stathmin in bortezomib-treated U266 cells using stathmin antibody and specific phospho-antibodies against three known phospho-sites (Ser16, Ser25, Ser38) [26] . As shown in Figure 4A , in accordance with SILAC results, phosphorylation of stathmin at Ser38 was increased whereas steady-state stathmin remained unchanged. Furthermore, phosphorylation of stathmin at Ser16 was increased whereas Ser25 was decreased after bortezomib treatment (Fig. 4A ). It has also been reported that Ser16 is a target for calmodulindependent protein kinases (CamKII), Ser25 is specifically phosphorylated by mitogen-activated protein (MAP) kinase, and Ser38 is a target for cycline-dependent kinase-2 (CDK2) [26] . To test the activation status of upstream kinases catalyzing the incorporation of phosphoryl groups to each of these residues, specific antibodies against active forms of CaMKII, MAPK and CDK2 (responsible for the stathmin phosphorylation on Ser16, Ser25 and Ser38, respectively) were used. In accordance with the increase in p-Ser16 and p-Ser38 of stathmin in bortezomib-treated cells, a parallel activation of CamKII and CDK2 was detected after bortezomib treatment ( Fig. 4B ). At the same time, the decrease in phosphorylation levels of Ser25 was correlated with the inactivation of MAPK, a critical kinase for cell survival. Hence, these results suggest that the regulation of the phosphorylation profile of stathmin at the level of residues Ser16, Ser25, and Ser38 may participate in the response of myeloma cells to proteasome inhibitors, and that stathmin is a target for multiple protein kinases, which are regulated by multiple signal transduction cascades. The functional significance of stathmin phosphorylation in the response of MM cells to bortezomib was then examined more rigorously. To this end, stable U266 cell clones overexpressing the stathmin wild-type (U266-WT) or mutants (U266-S16A, U266-S25A and U266-S38A) were generated using His-tagged constructs. Figure 5A shows the expression of His-tagged target proteins as well as stathmin expression in these cells. The complete methods and characterization in terms of proliferation, cell cycle, colony-forming efficiency (CFE) and apoptotic ratio of these cells are described in the Supplemental Data S1. We observed that all these cells had a comparable growth pattern and CFE, similar proportions of cells in G1, S, and G2 plus M phases and similar apoptotic ratio (Table S6) . However, there were significant differences between these cells with regard to their sensitivity to bortezomib treatment. As shown in Figure 5B , cells transfected with wild-type stathmin (U266-WT) exhibited significant increase in bortezomib-induced cell death compared with parental cells (U266). In contrast, cells transfected with mutant stathmin (U266-S16A, U266-S25A or U266-S38A) were significantly less sensitive to bortezomib lethality than U266 cells (P,0.05 for U266-S16A and U266-S38A or P,0.01 for U266-S25A, respectively) (Fig. 5B) . It has been reported that phosphorylation turns off the microtubule destabilizing activity of stathmin [27, 28] and that proteasome inhibitors increase tubulin polymerization and stabilization in myeloma cells [29] . We thus tested whether bortezomib actually induced changes in the polymerization status of the MTs in these cells. Indeed, when these cells were treated with bortezomib for 24 hours, we observed an increase in the amount of tubulin in the polymerized 'P' fraction as compared with untreated cells (Fig. 5C ). The baseline proportion of a-tubulin in the polymerized fraction ranged from ,43% to 52%, while the polymerized proportion observed after bortezomib treatment was ,60-90% (Table 1) . To investigate whether these bortezomib-induced changes in the tubulin polymerization were mediated by phosphorylation of stathmin, we examined the tubulin polymerization in stable U266 clones that overexpressing WT stathmin and the phosphorylation site-deficient stathmin mutants S16A, S25A and S38A. As shown in Figure 5C , by comparing with U266 cells, overexpression of WT stathmin and phosphorylation site-deficient mutants resulted in a significant decrease in the percent of polymerized tubulin following treatment with bortezomib ( Fig. 5C and Table 1) . Thus, our findings support the notion that bortezomib induces tubulin polymerization and stabilization through the mediation by phosphorylation of stathmin and this may be a contribution factor to the mechanism of proteasome inhibition and toxicity in MM cells. Quantitative phosphoproteomic approaches offer great promise for rapid progress in the analysis of drug targets or mechanism of action, especially when combined with traditional biochemical approaches presently used for studying individual proteins [30] . Moreover, the ability to simultaneously measure changes in phosphorylation state of many proteins in a single experiment can provide unique information needed for quantitative modeling of signaling pathways. In the present study, we used a combined strategy that comprised phosphoprotein enrichment, SILAC, and LC/MS analysis to profile the differential phosphoproteome in bortezomib-treated MM cells. A total of 72 phosphoproteins were found to have a 1.5-fold or greater change upon bortezomib treatment (Table S2 ). According to their change pattern, these DEPPs can be categorized into up-(,0.75) or down-(.1.5) regulated groups (Table S2) . Notably, in comparison with the number of downregulated DEPPs, many more proteins are up-regulated in this process (70 versus 5). This unbalanced pattern has also been revealed by previous in vitro biochemical studies [22] . Many studies have demonstrated that bortezomib can trigger pleiotropic signaling pathways and suppress the proteasomal degradation of multiple phosphorylated singnaling molecules [8, 10, 31] . Therefore, we speculate that the up-regulation of phosporylation of these DEPPs may constitute one of the mechanisms of bortezomibinduced apoptosis in MM cells. The PANTHER classification system also revealed that the DEPPs implicated in a variety of molecular functions, such as nucleic acid binding, receptor, regulatory molecule and so on (Fig. 3A) , clearly showing that in addition to targeting proteins involved in apoptotic pathways, bortezomib also altered multiple signaling pathways to induce its anti-cancer effects. In particular, as suggested by protein network analysis (Fig. 3B) , stathmin may play a central role in mediating bortezomib-induced apoptosis in MM cells. Stathmin (also termed as p19, 19K, p18, prosolin, and Op18) is a ubiquitous 19 kDa cytosolic phosphoprotein that is highly expressed in a wide variety of cancers, including a subset of leukemias and breast carcinomas and is a key regulator in the control of proliferation and cell cycle [32, 33] . Stathmin is a phosphorylation responsive regulator of microtubule (MT) dynamics that increases the catastrophe rate of MTs (depolymerization or shrinkage phase of individual MTs) in a dose-dependent manner [34] . Four Ser residues, Ser16, Ser25, Ser38, and Ser63 in stathmin are subjects for phosphorylation in intact cells [26] . The regulation of stathmin phosphorylation is complex and, in all likelihood, multifactorial. For example, phosphorylations at all four Ser residues fluctuate during the cell cycle and CDK2 has been identified as the kinase system involved in cell cycle-regulated phosphorylation of Ser38. Besides, three distinct protein kinases have been identified to phosphorylate stathmin in response to external signals. These kinases are members of the MAPK family that phosphorylates Ser25, cyclic AMP-dependent protein kinase (PKA) that phosphorylates Ser63 and CamKII that phosphorylates Ser16 [26] . Because the kinases acting at these residues are distinct and may be functioning through different pathways, it is reasonable to speculate that regulation of stathmin phosphorylation can be achieved by multiple pathways, thus providing the cell with a finely tunable mechanism for controlling microtubule assembly and dynamics in relation to its needs. In the present study, the role of stathmin and its phosphorylation in bortezomib-induced cell death was further investigated by overexpression of the WT stathmin and phosphorylation site-deficient stathmin mutants S16A, S25A or S38A in myeloma cells. Overexpression of WT stathmin significantly increased bortezomib-induced cell death. On the contrary, overexpression of the phosphorylation site-deficient stathmin mutants S16A, S25A and S38A significantly decreased, but did not block, bortezomibinduced cell death. Importantly, increased levels of tubulin polymerization were observed upon bortezomib treatment in cells overexpressing WT or mutant stathmin, but to a lesser extent than parental U266 cells (Fig. 5C, Table 1 ). There are several possible explanations as to why bortezomib-induced cell death is not completely blocked by the stathmin mutants. First, endogenous WT stathmin is still present, and thus the mutant protein has to compete with the WT protein. Second, bortezomib can also modulate the activity of other MT regulatory proteins that may also contribute to the cell death [29] . Furthermore, the MT system is not the only event involved in bortezomib-induced cell death in myeloma cells [22, 31] . This suggests that bortezomibinduced cell death is the result of a concerted series of events. Therefore, we conclude that bortezomib-induced phosphorylation of stathmin promotes cell death and that phosphorylation on Ser16, Ser25 and Ser38 is necessary for this process. Furthermore, the bortezomib-induced changes in the MT stabilization can be attributed to the bortezomib-induced phosphorylation of stathmin, and MT stabilization is in fact responsible for the bortezomibinduced cell death-promoting activity of phosphorylated stathmin. Another protein of interest uncovered in this study is BCL2associated athanogene 3 (BAG3). In the current study, BAG3 was found to have increased phosphorylation at Ser377 upon bortezomib treatment. BAG3 belongs to the evolutionarily conserved BAG family of proteins that were originally isolated based on their ability to interact with the anti-apoptotic protein Bcl-2 [35, 36] . It is involved in a wide variety of cellular processes, including cell survival, cellular stress response, apoptosis and virus replication [16, 37, 38] . Recent evidence implicates an additional function of BAG3 in the regulation of the autophagy pathway. These findings indicate that autophagosome formation and turnover may depend on BAG3 and that BAG3 can stimulate autophagy processes [39, 40] . Autophagy is a major intracellular degradation system. Unlike the ubiquitin-proteasome system (UPS), autophagy is mainly responsible for the degradation of long-lived proteins and subcellular organelles [41, 42] . Autophagy plays important roles in development, cellular homeostasis and cell survival and is frequently activated in tumor cells exposed to chemotherapy or radiation and confers therapeutic resistance [43, 44] . The UPS and autophagy have been viewed as distinct degradation systems, but recent studies suggested that they are functionally coupled and that suppression of the proteasome promotes autophagy [45, 46] . However, the functional connection and the inter-regulation between the two systems are not well understood. Importantly, Zhu et al. has shown that proteasome inhibition activates autophagy through a phosphorylation of eIF2a-dependent mechanism to eliminate protein aggregates and alleviate proteotoxic stress [46] . However, their results also demonstrated that complicated mechanisms are involved in proteasome inhibition-mediated autophagy activation and the phosphorylation of eIF2a only partially accounts for this activation [46] . Thus, it is possible that the complexity of the process may be much higher than presently envisaged, and that other proteins may be as important in controlling autophagy activation as eIF2a. Based on the critical role of BAG3 in the stimulation of the autophagy pathway, it is tempting to suggest that bortezomibinduced phosphorylation of BAG3 might play an important role in autophagy activation. Therefore, the increase in BAG3 phosphor-ylation is likely part of the cell's response to bortezomib treatment and appears to represent a novel mechanism with a link between the two protein degradation systems. This speculative idea, however, is not yet supported by the current experimental data, and further investigations are undergoing to determine the functional implication of BAG3 in coordination between the proteasome and autophagy. In summary, we have, for the first time, performed quantitative phosphoproteomics to study the effects of bortezomib on MM cells. The current results expand the list of bortezomib-targeted phosphoproteins and their phosphorylation sites. Especially, our functional studies indicated that bortezomib-induced phosphorylation of stathmin promotes cell death and that the bortezomibinduced changes in the MT stabilization can be attributed to the bortezomib-induced phosphorylation of stathmin. Our comprehensive study of phosphorylation regulation in proteasome inhibition in MM cells may serve as a valuable resource for future research in the field and thus advance the general mechanistic understanding of bortezomib in MM. The human myeloma cell line U266 was purchased from American Type Culture Collections (Rockville, MD). Myeloma cells were routinely maintained in RPMI 1640 supplemented with 1% penicillin/streptomycin, 1 mmol/L L-glutamine, and 10% fetal bovine serum at 37uC, 5% CO 2 in air. Bortezomib was provided by Millennium Pharmaceuticals (Cambridge, MA). To differentially label bortezomib-treated and -untreated U266 cells, the SILAC Protein Quantitation Kit (Pierce Biotechnology, Rockford, USA) was used according to the manufacturer's instruction. In brief, cells were grown in SILAC RPMI 1640 Medium (Pierce Biotechnology, Rockford, USA) containing 10% v/v dialyzed FBS, and either 0.1 mg/mL heavy [ 13 C 6 ] or light [ 12 C 6 ] L-lysine (Pierce Biotechnology, Rockford, USA). To ensure full incorporation of the heavy and light labeled amino acids, cells were grown for at least six cell doublings prior to analysis. U266 cells were treated with 3 nM bortezomib for 24 h, according to the half-maximal inhibitory concentration (IC50) measured by Hideshima et al [11] . After treatment, cells were washed three times with ice-cold washing buffer (10 mM Tris-HCl, 250 mM sucrose, pH 7.0) and transferred to a clean 1.5 mL Eppendorf tube. Cells were lysed with RIPA lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 0.1% SDS, 1% NP-40, 0.5% sodium deoxycholate, 1 mM PMSF, 100 mM leupeptin, and 2 mg/mL aprotinin, pH 8.0). Cellular debris was removed by centrifugation for 30 min at 13, 200 g and at 4uC. Protein concentrations were measured in duplicate using RC DC protein assay (BioRad, Hercules, CA, USA) and confirmed by SDS-PAGE. Phosphopeptide enrichment was performed as previously described [17] . Briefly, equal amounts of proteins from untreated ( 13 C 6 -lysine) and bortezomib-treated ( 12 C 6 -lysine) U266 cells were mixed (1 mg in total) and subjected to disulfide reduction with 10 mM DTT (37uC, 3 h) and alkylation with 20 mM iodoacetamide (room temperature, 1 h in dark). The protein mixtures were mixed with four volumes of ice-cold acetone to precipitate proteins. Precipitated proteins were collected by centrifugation and washed with ethanol two times. The pellet was re-dissolved in 50 mM ammonium bicarbonate and then digested with sequencing grade modified trypsin (1:25 w/w) (Promega, Madison, WI) at 37uC for 20 h and then quenched by addition of TFA to a final concentration of 0.5%. The digests were evaporated to about 20 mL in SpeedVac centrifuge. The phosphopeptides from digested peptides were enriched by using Phosphopeptide Enrichment TiO 2 Kit (Calbiochem, San Diego, CA) according to the manufacturer's instruction with slight modifications. Briefly, the tryptic digest was dried, re-dissolved in 200 mL TiO 2 Phosphobind buffer containing 50 g/L 2,5-dihydroxybenzoic acid and then mixed with 50 mL TiO 2 Phosphobind Resin. After 30 min incubation, the supernatant was discarded, and TiO 2 was washed three times with the wash buffer. After that, 30 mL elution buffer was added two times to elute the phosphopeptides. The elutions were combined and acidified with 5 mL of 10% formic acid for SCX-LC-MS/MS analysis. All the buffers and the phosphopeptides purification resin were provided in the kit by the manufacturer. The enriched phosphopeptides were analyzed with a Finnigan Surveyor HPLC system coupled online with a LTQ-Oribitrap XL (Thermo Fisher Scientific, Waltham, MA) equipped with a nanospray source. The phosphopeptides were firstly loaded on a strong cation exchange (SCX) column using an autosampler, and the peptides were eluted by NH 4 Cl with different concentrations (1 mM, 10 mM, 100 mM, 200 mM, 1 M). Then, each fraction peptide was respectively loaded onto a C18 column (100 mm i.d., 10 cm long, 5 mm resin from Michrom Bioresources, Auburn, CA) using an autosampler. Peptides were eluted during a 0-35% gradient (Buffer A, 0.1% formic acid, and 5% ACN; Buffer B, 0.1% formic acid and 95% AcN) over 90 min and online detected in LTQ-Orbitrap using a data-dependent method [47] . The general mass spectrometric conditions were: spray voltage, 1.80 kV; no sheath and auxiliary gas flow; ion transfer tube temperature, 200uC. Ion selection thresholds were: 1000 counts for MS 2 and 500 counts for MS 3 . An activation q = 0.25 and activation time of 30 ms were applied in MS 2 acquisitions. The mass spectrometers were operated in positive ion mode, employing a data-dependent automatic switch between MS and MS 2 acquisition modes. For each cycle, one full MS scan in the Orbitrap at 1610 6 AGC target was followed by five MS 2 in the LTQ at 5000 AGC target on the five most intense ions. Selected ions were excluded from further selection for 90 s. Maximum ion accumulation time was 500 ms for full MS scans and 100 ms for MS 2 scans. All MS/MS spectra were collected using normalized collision energy (a setting of 35%), an isolation window of 3 m/z, and 1 micro-scan. The resolution used in the MS step in the Orbitrap is 60000. An extra DDNL (data-dependent neutral loss) MS 3 method was applied for phosphopeptide detection [17] . MS 3 was triggered if a neutral loss peak at 298.0, 249.0, 232.7 or 224.5 Da was observed in the MS 2 and that peak was one of the three most intense ions of the MS 2 spectra. Application of mass spectrometer scan functions and HPLC solvent gradients were controlled by XCalibur data system (Thermo Fisher Scientific, Waltham, MA). Peak lists for the database search were produced in the Mascot generic format using BioWorks 3.3.1 (Thermo Finnigan, San Jose, CA) and DTASuperCharge V 1.31 (SourceForge), and the derived peak lists were searched using the Mascot 2.2.04 search engine (Matrix Science, London, UK) against a real and false IPI human database (V3.56, including 153, 078 protein entries). The following search criteria were employed: full tryptic specificity was required; two missed cleavages were allowed; Carbamido-methylation was set as fixed modification, whereas Oxidation (M), Phospho (ST), and Phospho (Y) were considered as variable modifications. Precursor ion mass tolerances were 10 ppm for all MS acquired in the Orbitrap mass analyzer, fragment ion mass tolerance was 0.5 Da for all MS 2 spectra acquired in the LTQ. Mass spectra of identified phosphopeptides with peptide score .10 were further processed and validated with the MSQuant 1.5 software for post-translational modification (PTM) score analysis [48] . Two filters criteria for phosphopeptide identification were applied: 1) Peptide score threshold was 17; 2) The total threshold of PTM score and peptide score was 36. All fragmentation spectra were manually verified using the criteria as described by Mann et al [48] . For phosphopeptides with multiple potential phosphorylation sites, the probabilities for phosphorylation at each site were calculated from the PTM scores as described [48] . Phosphorylation sites that were occupied with probability .0.75 were reported as class I phosphorylation sites. For class II sites, localization probability was between 0.75 and 0.25. Phosphorylation sites with localization probability ,0.25 were discarded. The identified phosphopeptides were further processed with MSQuant 1.5 for statistics evaluation as well as quantization. Differentially expressed phosphoproteins (DEPPs) were classified based on the PANTHER (Protein ANalysis THrough Evolutionary Relationships) system (http://www.pantherdb.org), which is a unique resource that classifies genes and proteins by their functions [49] . The DEPP interaction network was build automatically by the STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) system with default setting except that organism, confidence(score), and additional (white) nodes were set to ''human'', ''0.20'', and ''10'', respectively [50, 51] . The gene name list of these proteins was input to search against the database which contains known and predicted protein-protein interactions. The retrieve included a detailed network which highlights several hub proteins. The identified phosphoproteins were compared to the public database of PhosphoSite (http://www.phosphosite.org/) to find out the novel phosphoproteins and phosphosites. Each confirmed phosphoprotein was searched with SCANSITE (http:// scansite.mit.edu) [52] for potential kinase motifs with high, medium, and low stringency. Protein extracts (30 mg) prepared with RIPA lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 0.1% SDS, 1% NP-40, 0.5% sodium deoxycholate, 1 mM PMSF, 100 mM leupeptin, and 2 mg/mL aprotinin, pH 8.0) were resolved by a 10% SDS-PAGE gel, and transferred onto Immobilon-P PVDF transfer mem-braneS (Millipore, Bedford, MA) by electroblotting. After blocking with 5% non-fat milk, the membranes were probed with rabbit anti-CaMKII polyclonal, rabbit anti-phospho-CaMKII (Thr286) polyclonal, goat anti-actin polyclonal antibodies (Santa Cruz Biotechnology, Santa Cruz, CA), rabbit anti-CDK2 polyclonal, rabbit anti-phospho-CDK2 (Thr160) polyclonal, rabbit antiphospho-stathmin (Ser25) polyclonal antibodies (Abcam Inc., Cambridge, MA), rabbit anti-His-tag polyclonal, rabbit antistathmin polyclonal, rabbit anti-phospho-stathmin (Ser38) polyclonal, rabbit anti-phospho-stathmin (Ser16) polyclonal, rabbit anti-p44/42 MAPK polyclonal, mouse anti-phospho-p44/42 MAPK (Thr202/Tyr204) monoclonal antibodies (Cell Signaling, Danvers, MA), and mouse anti a-tubulin monoclonal antibodies (DM1A, Sigma, St. Louis, MO). Blots were then incubated with peroxidase-conjugated anti-mouse, anti-rabbit or anti-goat IgG (KPL, Gaithersburg, Maryland) for 1 h at room temperature at a 1:2000 dilution and then developed by using the SuperSignal West Pico Kit (Pierce Biotechnology, Rockford, IL). Human stathmin was cloned into the NH 2 terminal His-tagged pReceiver-M01 expression vector (Genecopoeia, Rockville, MD). Construction of mutant stathmin cDNAs, where the codons for Ser-16, Ser-25 or Ser-38 are exchanged to Ala, was performed by site-directed mutagenesis using a QuikChange kit (Stratagene, La Jolla, CA) following the manufacturer's instructions. Primers used, with the introduced mutations underlined, were: (Ser16RAla) 59-AGCGTGCCGCAGGCCAG-39; (Ser25RAla) 59-GCTGATT-CTCGCCCCTCGGTC-39; (Ser38RAla) 59-CCCCCTTGCC-CCTCCAAAG-39. All mutant constructs were confirmed by DNA sequence analysis. The plasmids were introduced into U266 cells using the X005 mode of Nucleofector (Amaxa, Cologne, Germany), according to the Optimized Protocol for the U266B1 cell line. Electroporated cells were cultured in medium with the presence of 0.5 mg/ml G418 (Mediatech, Manassas, VA) for 14 days, and then cultured in the 96-well plates for dilution cloning. Finally, the clone selected from pReceiver-M01 blank vector transfected U266 cells was designated as U266-NC. The clones selected from wild type or mutant stathmin plasmids transfected U266 cells were designated as U266-WT, U266-S16A, U266-S25A or U266-S38A, respectively. The extent of apoptosis was evaluated by using Annexin V/PI staining and flow cytometry as described previously [53] . In brief, 1610 6 cells were washed once in 16PBS and were stained with Annexin V-FITC and PI (2 mg/mL) according to manufacturer's instructions. Samples were acquired on a FACScan flow cytometer (Becton Dickinson, San Jose, CA) and analyzed with the WinMDI 2.8 software program. Tubulin polymerization assay was performed essentially as previously described [29, 54] . Briefly, cells grown to confluency in 24-well plates were washed twice with 1X PBS. To separate polymerized (P) from soluble (S) tubulin, the cells were all incubated at 37uC for 5 min in the dark in hypotonic lysis buffer containing 5 mM paclitaxel, 10 mM Trichostatin-A (Calbiochem, San Diego, CA), 1 mM MgCl2, 2 mM EGTA, 0.5% Nonidet P-40, 2 mM phenylmethylsulfonyl fluoride, 200 units/ml aprotinin, 100 mg/ml soybean trypsin inhibitor, 5.0 mM e-amino caproic acid, 1 mM benzamidine, and 20 mM Tris-HCl, pH 6.8, vortexed vigorously and centrifuged at ,15,000 g at 22uC for 10 minutes. The supernatants containing soluble 'S' tubulin were transferred to another Eppendorf tube separating them from the pellets containing polymerized 'P' tubulin. Upon separation, tubes were placed on ice and pellets of polymerized 'P' tubulin were resuspended by sonication for 10-20 seconds in a volume of lysis buffer equal to the soluble 'S' fraction. Each had gel sample buffer added, equal aliquots were separated by 10% SDS-PAGE, and western blots using anti-a-tubulin antibody were obtained. The immunoblots were scanned, and densitometric analysis was performed using the public domain NIH Image program ImageJ (available on the Internet at http://rsb.info.nih.gov/nih-image/). The percentage of polymerized 'P' tubulin was determined by dividing the densitometry value of polymerized 'P' tubulin by the total tubulin content (the sum of the densitometry values of soluble 'S' and polymerized 'P' tubulin). An advantage of this assay is that the amount of total protein loaded for each sample is irrelevant since the 'P' and 'S' fractions are equalized for each pair, and it is the proportion of the polymerized to the soluble tubulin fraction that is measured. All data are expressed as mean 6 standard deviation. Statistical significance was determined by Student's t-test (two-tailed), while the significance of the differences was determined using the twotailed Mann-Whitney test. Statistical significance was assigned if P,0.05. Data S1 Found at: doi:10.1371/journal.pone.0013095.s001 (0.03 MB DOC) Author Contributions

Insights into the Evolution and Emergence of a Novel Infectious Disease

Many zoonotic, novel infectious diseases in humans appear as sporadic infections with spatially and temporally restricted outbreaks, as seen with influenza A(H5N1). Adaptation is often a key factor for successfully establishing sustained human-to-human transmission. Here we use simple mathematical models to describe different adaptation scenarios with particular reference to spatial heterogeneity within the human population. We present analytical expressions for the probability of emergence per introduction, as well as the waiting time to a successful emergence event. Furthermore, we derive general analytical results for the statistical properties of emergence events, including the probability distribution of outbreak sizes. We compare our analytical results with a stochastic model, which has previously been studied computationally. Our results suggest that, for typical connection strengths between communities, spatial heterogeneity has only a weak effect on outbreak size distributions, and on the risk of emergence per introduction. For example, if [Image: see text] or larger, any village connected to a large city by just ten commuters a day is, effectively, just a part of the city when considering the chances of emergence and the outbreak size distribution. We present empirical data on commuting patterns and show that the vast majority of communities for which such data are available are at least this well interconnected. For plausible parameter ranges, the effects of spatial heterogeneity are likely to be dominated by the evolutionary biology of host adaptation. We conclude by discussing implications for surveillance and control of emerging infections.

Zoonotic emergence of novel human infections poses a significant risk to global public health. For example, the 'Spanish flu' pandemic of 1918 probably originated in birds and caused millions of deaths worldwide [1] . While much less virulent, the subsequent influenza pandemics of 1957, 1968 and 2009 [2, 3] are potent reminders of the capacity of the influenza virus to cross the species barrier into humans. Many other pathogens share this capacity: the SARS outbreak of 2003 [4] [5] [6] has been linked to bats and palm civets [7, 8] . In 2008, a novel arenavirus which killed four out of five patients in South Africa was linked to rodents [9] . Previous work [10, 11] has studied models of within-host evolution and between-host transmission, in which an initially poorly transmitting pathogen acquires adaptations to human hosts, following repeated zoonotic introductions until it achieves pandemic potential. These make the natural, simplifying assumption that the host population is homogeneous, so that changes in infection parameters entirely reflect adaptations in the biology of host infection. In reality, however, factors such as human contact patterns [12] and other host heterogeneity [13, 14] may also shape the risk and speed of emergence events. We concentrate here on the heterogeneity in the spatial structure of the human host population, an area which has hitherto received little attention in the context of adapting pathogens. We model spatially separated communities with varying types and strengths of interconnections, for example between a village and a city. Our aim is to study under what regimes such 'ecological' structure could have a strong effect on emergence, in comparison with 'evolutionary' factors governing the biology of infection. In the following section we give an overview of the modelling approach. We then present new analytical results for the simple models studied previously, which ignored spatial host population structure. We use these expressions to provide useful answers to important questions: if we knew how a zoonotic pathogen would adapt to human physiology, could we anticipate its emergence? How reliable would such predictions be? Furthermore, can we predict which zoonoses will cause outbreaks which do not turn into epidemics? Next, we ask: how large does a single, finite host population have to be, for population size to have a negligible effect? We then incorporate spatial heterogeneity by separating the human host population into communities. We present a model in which a small village is connected, by human travel, with a large city as an example of the general case of two interconnected communities. We use this model to ask: how strong do community interconnections have to be for us to safely ignore the separation of a population into spatially structured communities, such as cities and villages? We review available commuting data to ask how these thresholds compare with typical human mobility patterns? We close with a discussion of implications for public health. We build on a model of evolution and emergence originally presented by [10] in which a zoonotic pathogen infects humans, and initially has very poor onward transmissability. Thus for people who are infected by animals the average reproductive number, R 0 , is well below one (R 0 %1). We call this the first reproductive number a 1 for the wildtype strain. Occasionally, during such zoonotic infections, the pathogen acquires genetic changes that increase its ability to pass to other humans. During any chain of transmission the pathogen might adapt sufficiently that it achieves such ease of human-to-human transmission that R 0 w1 and an epidemic becomes possible. Such a process can be characterised by a vector of reproductive numbers (a i with i~1, . . . , n), and a vector of mutation probabilities (m i with i~1, . . . , n) where n{1 denotes the number of adaptive steps necessary to reach the fully adapted strain n. In what follows we restrict our attention to the case of n~3, m 1,2~m and m 3~0 , allowing us to model two routes of adaptation with opposite and distinct characteristics while minimising the overall complexity in number of required strains. For both routes of adaptation, the first, wildtype strain has very low transmissibility, the third has pandemic potential, and the second strain has intermediate transmissibility. This intermediate transmissibility is not enough to sustain the novel pathogen within the human host population, but secondary infections by humans are possible. Thus we have a 1 va 2 v1. Finally, the human adapted strain has a reproductive number a 3 w1. Further, we assume an identical mean infectious period for all strains. Following [11] , we first distinguish two routes to adaptation: the 'punctuated' scenario has an evolutionary course a i~0 , 0:1, 2 ½ , while the 'gradual' scenario has a i~0 , 0:9, 2 ½ , the only difference being a 2 , the fitness of the intermediate stage. This leads to the following SIR-model, normalized with respect to the mean infectious period, where S is the number susceptible, I i is the number infected with strain i, and R the number of recovered or removed. We do not include births and deaths as we expect a zoonotic emergence, or extinction, to happen on a much shorter timescale than the human lifespan. We translate this model to a stochastic simulation of a multitype branching process, using the Gillespie algorithm [15] . The infection is seeded in a single random, susceptible host with the wildtype strain. In general, every introduction has only two possible outcomes: emergence or extinction. Extinction happens if the novel pathogen dies out because it fails to adapt for human transmission or just by stochastic extinction. Hence, the introduction only leads to a limited number of infectious hosts, which we refer to as the 'outbreak size'. Conversely, a novel pathogen of zoonotic origin emerges if it is sufficiently adapted for human transmission and begins to spread in a self-sustaining way. Formally, in an unlimited host population the cumulative number of infectious hosts is unbounded as time goes to infinity. Computationally we use a threshold of I 3~1 00 infectious hosts with the fully adapted strain to distinguish between emergence and extinction. This threshold ensures a probability of extinction less than (1=R 3 ) I3~7 :9 Ã 10 {31 [16] . Therefore, the number of falsely identified emergences, which would truly be extinctions, is negligibly small. Moreover, these arbitrarily small probabilities ensure that our simulation results are insensitive to the precise choice of threshold used. In situations where the host population is very small we relax our emergence threshold to smaller numbers of infectious hosts as some population sizes are below 100. In the special case N?? and S=N?1 it is possible to calculate the probability of emergence per introduction into the human host population, given the evolutionary course of the pathogen, and the mutation rate with which the pathogen adapts. Our derivations start with assuming one homogeneous human host population of infinite size. This assumption can be easily relaxed as we show later. To calculate the probability of emergence, we define next event probabilities of infection p i , mutation m i , and recovery r i for each individual infected host, therefore the probabilities for what type of event will come next for each infectious host are Note that in general, we can extend this adaptation process to arbitrarily many adaptive steps. The mutations are uni-directional towards the adapted strain. Using a branching-process approach similar to [10] , we derive the probability of emergence per introduction as follows (see Text S1, A.1.1, for more details) Author Summary Emerging infections are a continuing global public health issue, the most recent example being last year's 'Swine flu' influenza pandemic. However, for many zoonotic pathogens, some adaptation is required to cross the species barrier from an animal reservoir into humans and cause sustained transmission. Previous work has explored the relationship between the evolutionary biology of an adapting pathogen, and the epidemiology of cases that may arise before such a pathogen becomes pandemiccapable. Here, we extend this work to incorporate what is often an important host ecological feature, the spatial distribution of the host population. Many zoonoses occur away from large population centres. For example, HIV is thought to have entered the human population through bushmeat hunters in the sparsely populated jungles of Central Africa. We ask: when a pathogen is evolving to adapt for human transmission, under what circumstances does the spatial structure underlying the human population become important? We approach this question using mathematical models to explore regimes of connectedness between communities. Our results suggest that most communities are sufficiently interconnected to show no effect on the emergence process. We finish by discussing the implications of these findings for public health. The probability of emergence can be expressed by the next event probabilities and the probability of extinction q i given an index infection of strain i. This expression can be solved analytically for all possible routes of adaptation. Regardless of the underlying population structure and the pathogen's biology, we can make an estimate of the number of introductions needed before an emergence arises M, given the probability of emergence per introduction p em with 0ƒp em ƒ1 (see Text S1, A.1.3, for more details) Note that this is the average number of introductions without an emergence. The average number of introductions needed for an emergence event is SMTz1. In addition, the variance can be obtained in a similar way (see Text S1, A. This variance leads to a standard deviation of the same order as the average number of introductions, vMw. This makes the number of introductions before an emergence inherently unpredictable if the probability of emergence per introduction is small (p em %1). Again, in the special case S=N?1, the branching-process approach can be extended to derive the probabilities of outbreak sizes before emergence (see Text S1, A.1.2, for more details). In general, the probability of an outbreak of size x i is defined as g i (I i,0 , x i ) with i denoting the strain. The number of infected hosts to start with is denoted by I i,0 . Furthermore, the overall outbreak size probability can be derived using conditional probabilities p out (X )~g 1 (I 1,0 , x 1 ) X I 1,0 zx 1 ,...,I n{1,0 zx n{1 where X~I 1,0 z P n i~1 x i is the total outbreak size, and f i (I i,0 ) is the probability of getting I i,0 patient zeros to start with in strain i. The summation in the derivation of the overall outbreak size probability is over all possible subsets of infectious hosts to start with. We use a metapopulation model to explore the effects of spatial host heterogeneity, effectively dividing the human population into interconnected communities. As an example of the general case of spatially structured communities, we focus on a simple village -city model to approximate the spatial host heterogeneity in rural areas connected, by human mobility, to bigger cities. There are many different types of human mobility between communities such as villages and cities, including short-term commuting and long-term labour migration. Particularly in developing countries, however, information on dominant patterns is sparse. Nonetheless, anecdotal evidence from Vietnam [17] , for example, suggests that short term commuting plays an important role: here, a subset of the village residents collects agricultural produce for trading in local markets in the city, and travels to the city on a daily to weekly basis. Accordingly we present a model in which the residence time of villagers in the city is typically less than the infectious period. However, in the supplementary information we also present a model incorporating migration on longer timescales (see Text S1, A.2). These two different models illustrate that our results appear qualitatively robust to different types of human movement. As before, we have a wildtype pathogen capable of acquiring adaptations for human transmission. Assume a finite number of hosts in the village, and an effectively infinite number in the city. To allow for daily commuting, we label individuals in the city according to whether they are commuters from the village or not (neglecting commuters originating from the city and present in the village). The superscripts (v),(c) represent village and city inhabitants respectively, while (vm) denotes villagers in the city. Village members commute to the city at a per capita rate x 1 , and return at per capita rate x 2 . At any one time, a proportion w of villagers, the commuters, are in the city with w being set by x 1 and x 2 as described below. Further, we neglect susceptible village commuters acquiring infection in the city (this arises formally from the infinite number of hosts in the city). The number of village residents is fixed at N~N (v) zN (vm)~1 000, and we define the average number of commuters as Village, non-commuting : Village, commuting : City : Commuting equilibrium : Note that equation (11) arises from the fact that w~N (vm) = (N (v) zN (vm) ). To represent daily commuting, with an infectious period of 5 days, we set x 2~5 , and choose w to give the required average number of commuters vcw. To seed a wildtype infection in the village we set I (v) 1~1 . In the village-city model, an emergence event is defined as having 100 infectious hosts with the fully adapted strain in the city. We use the three strain model described before to study the impact of the mutation rate, m, and the average reproductive number of the intermediate strain, a 2 , on the probability of emergence per introduction in a single, infinite population. We assume 0vmƒ10 {1 as an illustrative spectrum of possible mutation rates. Figure 2 shows the probability of emergence for different mutation rates and average reproductive numbers of the intermediate strain. Not surprisingly, the probability of emergence grows non-linearly with a 2 and m. The probability of having no mutation in the second strain is (1zm) {x where x is the number of infected hosts with strain 2. While the intermediate reproductive number affects the exponent, the mutation rate has a direct influence on the base. We validate our analytical results by comparing them with the average probability of emergence of 10 3 simulated emergence processes, using one homogeneous population as described in (1). Figure 2 reveals an excellent agreement between the analytical results and simulations. For small host communities, the depletion of susceptible hosts can play a significant role in limiting an ongoing outbreak. What is the effect of a finite population size on these analytical results which assume an infinite host population? Figure 3A compares the simulated outbreak size distribution of different sized populations with our analytical predictions. Note that, for populations greater than 500, there is close agreement between numerical and analytical results. When considering populations of size 1,000 or more, we do not expect population size dependence to have a substantial effect. The village's number of residents in our commuting model is sufficient to avoid finite size effects on the outbreak size. Furthermore, it is independent of any spatial heterogeneity. The number of residents only has a limiting effect on the outbreak size distribution. Figure 3B shows the outbreak size distribution for our short-term commuting model. The average number of commuters ranges from 1 to 100. As we expect, no significant effect on the outbreak size distribution can be seen, even for vcw~1. It validates our assumption of the independence of infectious hosts, necessary for a branching-process formulation, as the simulations closely match the analytical predictions. It is noteworthy here that only the biology of the novel pathogen determines the emergence process, as outbreak sizes group according to the intermediate average reproductive number a 2 . The minimal deviation for vcw~100 commuters is based on the fact that the effective village size is only N (v)~9 00 due to the absent commuters. In Figure 4 , we extract the probability of emergence per introduction given a certain number already infected. These data are easily calculated using the outbreak size distribution and the probability of emergence per introduction. Assume an introduction has caused x infectious hosts already. The probability of extinction is the cumulative probability of getting an outbreak size equal to or larger than x, renormalized by all possible outcomes (extinctions and emergences) once x hosts are infectious. This yields the probability of emergence per number of infected. In Figure 4 , the effect of spatial heterogeneity can be seen directly. For vcw §10, the village-city simulations agree very well with the analytical results assuming a single, infinite population. But for vcw~1, the probability of emergence converges to approximately 0:42. While Figure 4 shows the probability of emergence as a function of the number infected, the actual outcome is highly unpredictable even if the probability of an event is known as the average waiting time to an emergence shows (see equations 4 and 5). It can be generalized for the probability of emergence given x infectious hosts. For example, the probability of emergence given five infectious hosts in the gradual route (II) is p em (x~5)~0:567. It follows on average every 1:736 times this happens an emergence will happen. The standard deviation is +1:161, which leads to the conclusion that even if the probability is known, it is inherently unpredictable when this will actually lead to an emergence. This confirms that a pathogen needs a sufficient connection between communities to emerge, despite its ability to cause outbreaks, regardless of the spatial structure. Hence we expect the existence of a threshold where spatial heterogeneity effectively does not matter any more. Previous research has shown that the effect of heterogeneity in spatially structured population models depends on the interconnectivity with a threshold effectively allowing the pathogen to spread between communities [18] [19] [20] . Our approach allows new insights, as we do not need to specify the actual number of infectious hosts migrating to a new community. We measure connectivity between communities in terms of the average number of commuters vcw for which rich empirical datasets can be found. Figure 5 presents illustrative examples of empirical data of commuting patterns in different parts of the world. Most data has been collected by Offices of Statistics of five countries on three continents [21] [22] [23] [24] [25] . A further, two independent studies have been used to estimate commuting patterns of towns in Indonesia [26] and China [27] . We next attempt to quantify the regimes in vcw for which spatial heterogeneity may be neglected. We approach this question using a simple analytical derivation for the effect of spatial heterogeneity, which considers only the adapted strain. Assume a connected community such as the village in our village-city model, with a fully adapted pathogen introduced into the village. Given an emergence and epidemic in the village, the probability that this causes an emergence and epidemic in the city is Therefore, H spatial is a spatial homogeneity coefficient, measuring the impact of spatial heterogeneity on an emergence process. It ranges from 0, leading to two isolated communities, to 1, effectively removing any spatial heterogeneity and forming one homogeneous population. f is the fraction of the village residents becoming infectious. It can be derived using [28] f~1{e The spatial homogeneity coefficient depends only on the connection strength expressed in commuters vcw and the average reproductive number R 0~a3 of the fully adapted strain. Though it only considers the fully adapted strain, we expect this coefficient to be a good approximation for a multi-strain model as the vast majority of infectious hosts will transmit the fully adapted strain in the case of an emergence. Figure 6 gives an overview of the influence of spatial heterogeneity as a function of vcw. Effectively, spatial heterogeneity is negligible once a critical number of ten commuters connect the two communities. This is a very low threshold, and empirical data shows that the probability of having a community with less than the critical number of ten commuters is approximately 1% for all our data combined. As illustrated by the close fit between analytical and numerical results in figure 6 , there is only a small error in the analytical expression arising from neglecting infections with mal-adapted strains. This error is greatest for the gradually adapting pathogen, because an intermediate strain with a 2~0 :9 tends to cause larger outbreaks than one with a 2~0 :1. Nevertheless, the deviation remains small. In light of this agreement, how does the critical average number of commuters vary with the adapted reproductive number? While for R 0~2 ten average commuters are sufficient to dissolve spatial heterogeneity, this changes dramatically for smaller average reproductive numbers (see Figure 7) . If a well-adapted strain is only just pandemic capable (i.e. R 0 just above 1) villages with only ten commuters are only 50% likely to seed an epidemic in their local city, and spatial structure becomes important again. For example, the critical number of commuters is close to 100 for R 0~1 :2. For R 0~2 the spatial homogeneity coefficient is approximately 0:42 for one commuter. This agrees with what we find in Figure 4 using simulated results. In this article we first present analytical results to calculate epidemiological parameters of a novel disease, adapting to humans. We explore the influence of spatial host contact structure, and validate our result with stochastic simulations of simple village-city models as an example of interconnected communities within a spatially structured population. Our study reveals that for plausible parameter ranges, spatial heterogeneity only has very limited impact on the probability of emergence, as well as the outbreak size distribution. Neither a change in strength of spatial heterogeneity (e.g. number of commuters), nor in its quality (e.g. short term versus long term commuting) shows a significant influence. Our results suggest that only the most remote rural communities would be subject to epidemiological isolation. In particular, the available empirical data suggests that communities tend to be highly interconnected with relatively high connection strengths. Of course, it is the most remote communities of the world for whom we have the least relevant data. More empirical research on spatial heterogeneity is needed to form a better understanding of its effect, and this need is greatest in developing parts of the world. In addition, population size becomes an important factor only when that population is relatively small { fewer than N&500 individuals. Only a small number of infectious hosts are actually involved in the emergence process, which relates to the small reservoir of susceptibles needed for a successful emergence. Moreover, biological processes such as the speed of evolution and the adaptive route show a strong influence on the overall emergence process. We show that epidemiological parameters such as the outbreak size group according to the evolutionary route. Previous research has shown the effect of the pathogen's route of evolutionary adaptation and mutation rate on the probability of emergence per introduction [10, 11] . Our theoretical derivation of the probability of emergence extends this and offers the benefit of being analytically solvable for any possible route of adaptation and any mutation probabilities. We note here that previous work has highlighted the role of other, significant types of heterogeneity in emergence of a novel infection. For example, [29] describe the effect of the pathogens life history, such as the length of infection, on the emergence of a novel pathogen. Further, heterogeneity in human-to-human transmission within a population may have an influence on the course and probability of emergence and outbreaks [13, 14] , usually lowering the probability of emergence. While we have concentrated here on simple types of spatial heterogeneity, a significant question for future research is the role of mixed heterogeneities, for example spatially structured populations with additional heterogeneity in the human-to-human transmission. We also find that the waiting time for an outcome of a novel pathogen's introduction is highly unpredictable, even if the probability for such an event is known. Conversely, this means that an estimate of the underlying epidemiological parameters from observed data will be highly uncertain. Unfortunately, a large number of observations will be necessary to achieve confidence in the parameters, and even a large number of introductions gone extinct do not rule out the possibility of emergence for a pathogen. We came to a similar result [11] using a measurement on the upper bound of the probability of emergence. Moreover, the probability of emergence given a certain number of infectious hosts can be surprisingly low. Even a comparatively large number of infectious hosts can end in extinction, especially for low mutation rates and intermediate-stage average reproductive numbers just below one. Figure 5 . Data of commuting patterns in different parts of the world. Shown is the cumulative fraction of all communities with equal or less than the specified number of commuters. The data was mostly collected by the National Statistical Offices of the respective countries. The gold line represents commuting data from Brazil [21] , the red line data from the USA [22], the blue line data from the UK [23] , the brown line data from Japan [25], the cyan line data from Hong Kong [24] , and the green line data from two independent sources. The green line has orange [27] and pink data points [26] , corresponding to its data sources. Our data represents the commuting flows between administrative units. The definition of administrative units varies highly between countries. For example, the US data is on a granularity of 3,141 counties, while the data from Japan is based on its 47 prefectures. However, heterogeneity can also be found within countries datasets. The Brazilian data is on a level of 5,507 municipalities with resident sizes ranging from 1,166 to 10, 435, 548. doi:10.1371/journal.pcbi.1000947.g005 Figure 6 . Deviation between simulated and analytical predicted probability of emergence. The deviation is a function of the average number of commuters vcw. The deviation is defined as Dp (ana) em {p (sim) em D=p (ana) em . The analytical probability of emergence p (ana) em is for an infinite ulation without spatial structure. The simulated probability of emergence p (sim) em is for short-term commuting with the blue data points representing the gradual route (II) of adaptation, and the orange data points representing the punctuated route (I). The solid black line is 1{H spatial (1=a 3 ) vcwf , as defined in equation 12 in the main text. It is the analytical expected deviation for spatial heterogeneity as a function of the spatial homogeneity coefficient. The simulations agree very well with the analytical expected deviation. The gradual route (II) is slightly more off from the theoretical prediction as a result of the small but significant number of infected with the intermediate strain. Effectively, it lowers the number of commuters infected with the fully adapted strain and therefore the probability of transmission from the village to the city. Nevertheless, the analytical prediction as well as the simulations show no significant impact of spatial heterogeneity from a critical commuter threshold of vcw~10. doi:10.1371/journal.pcbi.1000947.g006 Figure 7 . Impact of spatial heterogeneity on disease transmission between communities (I). The impact is measured with the spatial homogeneity coefficient with 0ƒH spatial ƒ1. Given H spatial~1 every emergence in the village automatically leads to an emergence in the city, and H spatial~0 represents no chance of successfully transmitting the pathogen into the city. The figure reveals that spatial structure becomes especially important for small average reproductive numbers. In addition, the average number of commuters needed to show an effect of spatial heterogeneity is surprisingly small. doi:10.1371/journal.pcbi.1000947.g007 Our work has relevance for important public health issues: if a novel disease is detected in a rural setting, and it appears to be spreading, how feasible is it to contain infection by restricting movements to and from the village? Our results suggest that first, an infeasibly tight level of quarantine would be required for any chance of containment, corresponding to enforcing a low level of vcw in Figure 6 . To all intents and purposes isolation would have to be absolute to be effective. In most circumstances such extreme intervention would not be acceptable. Second, given typical mobility patterns, it is likely that once there is a detectable number of cases in the village, there may already be a significant number of cases outside of it. Therefore quarantining interventions are likely to come too late. Our work raises important questions for future research: where should surveillance be focused to detect an emergence as early as possible, especially if resources are limited? Given emergence of a novel infection in a rural setting, how much time can we buy through limiting travel to and from major urban centres? These and other questions will undoubtedly benefit from more systematic studies of emergence in the context of population distributions. Nonetheless, theoretical models such as those presented here can offer useful, fundamental insights to guide such studies. Text S1 Supplementary material with figures for 'Insights into the Evolution and Emergence of a Novel Infectious Disease'. Found at: doi:10.1371/journal.pcbi.1000947.s001 (0. 16 MB PDF) Author Contributions

Role of Host Immune Response and Viral Load in the Differential Outcome of Pandemic H1N1 (2009) Influenza Virus Infection in Indian Patients

BACKGROUND: An unusually high number of severe pneumonia cases with considerable mortality is being observed with the pandemic H1N1 2009 virus infections globally. In India, all mild as well as critically ill cases were admitted and treated in the government hospitals during the initial phase of the pandemic. The present study was undertaken during this early phase of the pandemic. METHODOLOGY: The role of viral load and host factors in the pathogenesis were assessed by examining 26 mild (MP), 15 critically ill patients (CIP) and 20 healthy controls from Pune, India. Sequential blood and lung aspirate samples were collected from CIP. Viral load and cytokines/chemokine levels were determined from the plasma and lung aspirates of the patients. TLR levels were determined by staining and FACS analysis. Gene profiling was done for both cells in the lung aspirates and PBMCs using TaqMan Low Density arrays. Antibody titres and isotyping was done using HA protein based ELISAs. PRINCIPAL FINDINGS: 13/15 critically ill patients expired. All plasma samples were negative for the virus irrespective of the patient's category. Sequential lung samples from CIP showed lower viral loads questioning association of viral replication with the severity. Anti-rpH1N1-09-HA-IgG titres were significantly higher in critically ill patients and both categories circulated exclusively IgG1 isotype. Critically ill patients exhibited increase in TLR-3, 4, 7 and decrease in TLR-2 expressions. The disease severity correlated with increased plasma levels of IL1RA, IL2, IL6, CCL3, CCL4 and IL10. Majority of the immune-function genes were down-regulated in the PBMCs and up-regulated in the cells from lung aspirates of critically ill patients. No distinct pattern differentiating fatal and surviving patients was observed when sequential samples were examined for various parameters. CONCLUSIONS: Disease severity was associated with pronounced impairment of host immune response.

The first pandemic of this century was unexpectedly caused by a novel swine-origin H1N1 virus, the pandemic H1N1 (2009) virus (p-H1N1-09). Mexico was the first country to be affected in early March with reports of mild respiratory infection as well as severe pneumonia cases and considerable mortality [1, 2] . Several countries were subsequently affected reporting variable mortality, smoking, pregnancy and obesity being important risk factors for severe disease [3, 4, 5, 6] . On 1 st August 2009, a 14 year-old girl without history of known risk factors succumbed to p-H1N1-09 infection in Pune, western India representing the first fatality from the country. As of 21 st April 2010, India has reported 1483 deaths during the pandemic (http://pib.nic.in/h1n1/h1n1.asp), Pune contributing to 173 cases (http://www.maha-arogya.gov.in/march-april%202010.htm). During the initial phase of the pandemic, designated wards in the government hospitals admitted every mild case, treated with Oseltamivir and discharged after recovery. A special intensive care unit treated the critically ill patients. The present study was undertaken during this very early phase of the pandemic. To understand the basis of differential disease presentation/outcome, we investigated 26 mild cases and 15 critically ill patients during 1 st August -19 th September 2009. This report provides comparative data on viral load, cytokines, gene-profiling, Toll-like-receptor (TLR) levels, antibody titres and antibody isotypes. Ethical clearance for the study was obtained from, 'Institutional Human Ethical Committee' as part of the pandemic influenza investigations. Written consent was obtained from all the participants involved in the study. For minors and critically ill patients it was obtained from parent/guardian. Patients confirmed to have p-H1N1-09 infection by a positive real time PCR test (http://www.who.int/csr/resources/publications/swineflu/CDC RealtimeRTPCRprotocol_SwineH1Ass-2009_20090428) were studied. These included 15 patients admitted to Intensive Care Unit and requiring mechanical ventilator support and 26 suffering from mild respiratory symptoms. All the mild cases were ambulatory patients admitted to a designated hospital. In the initial phases of the pandemic during which this study was performed, patients suggestive of Influenza-like illness were admitted to designated ward of a Corporation hospital. Throat swabs were collected and sent to the National Institute of Virology for diagnosis. Osletamivir treatment was initiated immediately after the confirmation of diagnosis. These patients were discharged after the completion of the treatment. On the contrary, majority of the severe patients were admitted only after the development of serious respiratory consequences. Same protocol was followed for diagnosis and antiviral treatment. A single blood sample was collected from mild cases, 1-3 days after the development of respiratory symptoms. As controls, blood samples from 20 apparently healthy individuals were collected. Sequential blood/lung aspirate samples (standardized tracheal aspirates) were collected from the critically ill patients. The first sample was collected within 3 days for 13 patients; while one each was collected on 4 and 8 (pregnant woman) days after the appearance of symptoms (Table 1 ). Blood and lung aspirates were transported to the lab within half an hour of collection. Lung aspirates obtained from severe cases were immediately aliquoted and frozen at 280uC. An aliquot was mixed in 1:3 proportions with the RNAlater and stored at 280uC for gene analysis. Hundred microlitre of the blood sample from every patient was processed immediately for TLR staining as described below. In parallel, blood samples were immediately processed for the isolation of Peripheral Blood Mononuclear Cells (PBMCs) by density gradient centrifugation using Ficoll-Hypaque (Sigma). Plasma layer was removed and stored at 280uC in aliquots; cell pellets were stored in 500 ml RNALater solution (Ambion) at 280uC. A highly sensitive and specific ELISA was carried out by coating the wells with purified recombinant HA protein (expressed in baculovirus system) of p-H1N1-09 virus, sera at 1:100 dilution and anti-human-IgG-HRP conjugate as the detector antibody [7] . Isotyping was done as described earlier [8] . Anti-human antibodies for TLR 2, 3, 4 and 9 (eBioscience, USA) and TLR 7 and TLR 8 (Imgenex, USA) were used for the staining. For surface staining of TLR 2 and 4, 100 ml of whole blood was lysed with BD FACS lysis solution (BD Biosciences, USA), washed, fixed and processed for staining with anti-human TLR 2 (FITC conjugated) and TLR 4 (PE conjugated) antibodies respectively. For intracellular staining, 100 ml of the whole blood was fully lysed with BD FACS lysing solution (Becton Dickinson), washed twice with Perm-wash buffer (BD bioscience) and processed for staining with the following anti-human antibodies: PE-TLR 3, FITC-TLR 7, FITC-TLR8 and PE-TLR 9 respectively. Stained cells were resuspended in 500 ml 1% paraformaldehyde, analysed on FACScalibur flow cytometer (Becton Dickinson).On the basis of forward and side scatter plot lymphocytes and monocytes were gated and data analysis was done using BD FACSDivasoftware.TLR levels were expressed as median fluorescence intensity (MFI). Concentrations in the plasma/lung aspirates were determined for seventeen cytokines/chemokines (IL1a, IL1b, IL6, TNFa, Frozen lung aspirates were thawed, centrifuged to pellet down the cells and pellets were used to isolate RNA. Total RNA was extracted from PBMCs and lung aspirate cells by using Ribopure Kit (Ambion) as per the manufacturer's instructions. RNA was eluted in 100 ml elution buffer, quantitated using Nanodrop (ND-1000) and processed for quality check in Agilent bioanalyzer (Agilent, U.S.A). Equal quantities of RNA (500 ng) with $7 RIN value were processed further for cDNA synthesis using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, U.K.). All cDNAs were tested in real time PCR assay using TaqMan primers and probe for 18S rRNA (Applied Biosystems, U.K.) to ensure efficient cDNA synthesis. cDNAs were mixed with equal volumes of TaqMan 2X PCR master mix from one-step RT-PCR kit (Applied Biosystems) and 125 ng (RNA equivalent) cDNAs were loaded per port of the TaqMan Low Density Array card (TLDA) of the Human Immune panel (Applied Biosystems, U.K.) and run on 7900HT Fast Real-Time PCR system (Applied Biosystems, U.K.). Relative gene expression values were obtained employing comparative Ct method using Applied Biosystems' Relative Quantification (RQ) Manager Software v1.2. cDNAs from six healthy individuals processed similarly were considered as calibrators. 18s RNA was used as an endogenous control. Relative quantitation values of each study group were used to calculate mean RQ values. For cluster analysis, relative quantitation values were log2 transformed and hierarchically clustered with analysis software (Cluster 3.0) [9] . Viral RNA was extracted from 140 ml plasma/lung aspirates using QIAamp Viral RNA Mini Kit (Qiagen, Germany) as per the manufacturers' instructions. For viral RNA quantitation, CDC primers and probe were used. The target HA gene from an Indian isolate was cloned to obtain in vitro transcripts, serial 10-fold dilutions of the RNA were used to generate a reference curve. A linear relationship was obtained from 10 10 -10 2 starting copies/ reaction (r 2 = 0.99), the detection limit being 100 copies. Levels of cytokines and chemokines were analyzed after log transformation and a value of 0.2 pg/ml was used in the case of undetectable concentration of cytokine or chemokine in the tested samples. The Mann-Whitney U or Fisher exact tests were used for group comparisons of numerical and categorical data respectively. For all analyses, a P value of less than 0.05 derived from a two tailed test was considered significant. All statistical analyses were performed with 'SPSS11.0 for Windows' software (SPSS Inc.). All the 26 mild cases (Male: Female ratio 12:14, age range 6-51 yrs) had fever/history of fever in last 3 days. The other symptoms included sore throat/cough (14/26), nasal discharge (6/ 24), headache/bodyache (5/26), diarrhea (2/26) and breathlessness (4/26). Radiological examination was not indicated and hence not performed. Nasal oxygen was not required. Of the 15 critically ill patients, 2 survived (Table 1 ). The male:female ratio was 6:9, the age range being 13-53 years. Associated co-morbidities were present in 1/2 surviving and 7/13 fatal patients. Importantly, radiological findings showed that all the fatal cases had $3/6 lung zonal involvement and required ventilator assistance. Both the surviving patients initially needed nasal oxygen but were put on ventilator subsequently. The patients were either treated with Osletamivir alone (n = 7) or in combination with Zanamivir (n = 8). From the endotracheal tubes of 4 patients Acenotobacter was isolated. None of the 26 mild cases had recognizable risk factors. S21, a 36 year female suffering from severe seasonal influenza survived after 39 days of hospitalization. Plasma samples from both patient categories were negative for pH1N1-09 influenza virus RNA. Sequential lung aspirate samples of 6 patients (2-5 samples) demonstrated gradual/continued lower or no viral load. Of the single lung aspirate from 3 patients with rapid death, two were negative for viral RNA while one exhibited 1.51610 5 copies/ml ( Figure 1 ). Serum samples available from 4 critically ill patients exhibited HI titres of 40-320. At admission, the geometric mean titres (GMT) of anti-rpH1N1-09-HA-IgG antibodies (ELISA) were significantly lower in the mild infections (800, 95% CI values 571-1120) than the critically ill patients (4996, 95% CI values 3970-6288) (p,0.0001). The last blood sample collected 4-11 days later from 11 critically ill patients showed .4 fold rise in antibody titres. IgG1 was the exclusive isotype in both the categories. Comparison among healthy controls and disease categories showed no difference in the levels of IFNc, IL12p70 and IL2RA were significantly lower in mild cases and higher in critically ill patients ( Table 2 ). All other molecules showed significantly higher levels in both patient categories. The overall pattern reflects pandemic-H1N1-09 influenza infection. The disease severity correlated with significant increase in IL1RA, IL2, IL6, CCL3, CCL4 and IL10 levels in critically ill patients. Analysis of gene expression profiles revealed significant findings; please see Table S1 : Comparison of controls with disease categories. Tables 2 and S1 represent the status of cytokines and chemokines at protein and gene levels respectively in the lung aspirates/cells of lung aspirates of critically ill patients. In the absence of similar samples from mild cases/healthy controls for ethical reasons, no comparison was possible. Comparison of plasma and lung aspirates of critically ill patients identified higher levels of CXCL8 and IL12p70 in the lung aspirates. No significant difference was noted for the other molecules. None of the plasma markers of severity exhibited higher levels in the lung aspirates. Expression patterns of genes were also analyzed by hierarchical clustering. Results are graphically represented by assigning a specific color to each cell on the basis of the expression levels. Genes showing no changes in expression levels (as compared to controls) are shown as black. Upregulated genes are shown in red with increasing intensities in proportion to the expression levels and in different intensities of green indicating levels of downregulation. Figure 3a shows cluster image of samples from mild and critically ill cases taken on admission. Seasonal flu sample (S21) showed a distinct gene expression pattern and formed a separate cluster (cluster 6) from all other samples. Three lung samples from critically ill patients also formed a separate cluster as most of the proinflammatory genes were upregulated (cluster 2). Remaining samples formed mixed clusters (cluster 4 and 5) which included both mild and critically ill patients. Overall, no distinct clustering pattern for the patients was observed for the severe and mild cases. With respect to gene clustering, clusters A and B contained genes associated with inflammation. Comparisons of genes of lungs and PBMCs of the same 3 patients or all the critically ill patients as a group yielded similar results. Comparison of lung cytokines and chemokines at protein and gene levels showed comparable elevated levels of IL12B, CXCL8, IL1B, CCL3, IL1A and IL6 while IL17, TNF and IL10 were elevated only at the gene level. IFNG was at basal level both at gene and protein levels. Variable numbers of sequential blood samples were available from critically ill patients. Comparisons of TLRs did not show any distinguishable pattern (data not shown). For evaluating gene expressions, a separate cluster analysis was carried out for the sequential PBMC samples of critically ill patients (Figure 3b ). Samples formed two distinct clusters. Samples from S1, S2, S3 and S4 formed a single cluster (cluster 1) showing significant Figure 2 . TLR levels in the blood lymphocytes and monocytes. TLR levels of peripheral blood from the patients and healthy controls were determined separately by either surface staining or intracellular staining of the cells using TLR specific antibodies followed by FACS analysis. Scatter dot plots and histograms: A representative sample each from different categories, a) healthy individual, b) mild case, c) critically ill case, d) TLR levels: Expressed as median fluorescence intensities (MFI) for different categories of patients. doi:10.1371/journal.pone.0013099.g002 downregulation of most of the immune function genes. Cluster 2 contained all sequential samples from the seasonal flu case (S21) and one pH1N1 case (S7). Though most of the gene expression levels were similar to survived case, S7 did not survive. Genes formed four clusters, A, B, C and D. Cluster C contained majority of the analyzed genes (73/87) and all were downregulated in the cluster 1. The dynamics of cytokines/chemokines at protein levels (6 patients) are shown in Figure 4 . Data for the single critical patient suffering from seasonal influenza (S21) is also presented. Comparison of plasma cytokines and chemokines did not show any distinct pattern in the sequential samples of the fatal or survived cases. Patient S6 did exhibit different pattern when compared to others. This first comprehensive study addresses important issues of the identification of markers for severity of p-H1N1-09 infection and dynamics of immune responses in severe disease by evaluating several parameters. The investigations were initiated during the early phase of the pandemic when isolation of all the Figure 3 . Gene expression analysis. Gene expression profiles of total PBMCs from blood samples of mild and critically ill cases and lung aspirate cells from critically ill cases were determined using TaqMan Low Density immune panel arrays. PBMCs from six healthy individuals were taken as controls, which were treated as replicate arrays to calculate the mean baseline expression level for each gene. The fold changes in the gene expression levels were calculated in relation with the controls. Values for eighty-seven genes were hierarchically clustered on log2 transformation. The corresponding gene of each cluster is listed by a gene symbol on the right-hand side of the images. a) Cluster image of gene expression analysis of PBMCs from mild and critically ill cases and lung aspirate cells from critically ill cases: PBMC samples from mild cases are denoted as M (M1, M2, M13). First sample taken after admission from each critically ill case was taken for the comparison and are denoted as S1-1, S2-1, S21-1. Single samples from critically ill patients are denoted as S8, S12 etc. Lung aspirate samples are denoted as L. b) Cluster image of gene expression analysis of sequential blood samples from critically ill cases: Total PBMCs from the sequential blood samples of severe cases obtained on different post admission days are denoted as S1-1, S1-2, …S21-8). doi:10.1371/journal.pone.0013099.g003 p-H1N1-09 mild infections in designated wards was obligatory. Therefore, we could collect samples before the initiation of the antiviral therapy and the data truly represents acute phase of mild disease. On the other hand, critically ill patients were given Oseltamivir immediately after the confirmation of the infection and the samples were collected subsequently. As underlying medical conditions were present in one of the 2 survivors and 7/ 13 critically ill patients, association of co-morbidities alone does not seem to be responsible for complications or aberrant immune response. The absence of viremia in both patient categories and relatively low viral load in the lung aspirates of the critically ill patients suggest that enhanced replication of the virus may not be an important contributor to the pathogenesis (Figure 1) . The viral load in lung aspirates was independent of fatality. In contrast, among the Spanish patients [10] , 93% and 57% of the mild and critical cases respectively were positive for serum viral RNA, with no significant difference in the viral load. Both studies used CDC primers for real time PCR (http://www. who.int/csr/resources/publications/swineflu/CDCRealtimeR TPCRprotocol_SwineH1Ass-2009_20090428) and the critical cases were bled when already on Oseltamivir treatment, negating sensitivity of the PCR, effect of antiviral therapy or delay in collection of samples to be responsible for different results. The absence of uniform mutations in the fatal casesderived Indian p-H1N1-09 isolates suggests limited/no role of mutant virus in the pathogenesis [11] . Viremia was associated with the outcome of H5N1 infection, 9/11 fatal and 0/5 nonfatal cases being viremic [12] . Serum HI-antibody positivity was noted in 1/15mild, 2/10 critically ill Spanish patients and 4/4 critically ill patients from India. ELISA could detect antibodies in every patient. The absence of viremia among the Indian patients may be due to an early antibody response. The presence of IgG-anti-p-H1N1-09 antibodies in all the severe Indian patients (HI antibodies in all the 4 screened) does indicate switch from the innate to adaptive immunity. Poor outcome despite the switch may probably be attributed to the timing of the shift or role of antibodies in disease severity. Significantly higher titres of IgG-anti-p-H1N1-09 antibodies in the critically ill patients supports the role in pathogenesis. This observation is in sharp contrast with that for the SARS patients examined from Canada [13] documenting significantly lower anti-SARS CoV spike antibody titres in the critically ill patients. Though the presence of neutralizing antibodies protect against Influenza virus infection, clearance of the infection is mediated by cellular immunity. The exclusive presence of IgG1 antibodies in both patient categories document Th2 bias with significant enhancement with severity. This observation correlates with the elevated levels of TLR3, 4, and 7, Th2 cytokines (IL6 and IL10) at protein level and IL4, IL5, IL6, IL10 at gene level. We identified higher plasma levels of IL1RA, IL2, IL6, CCL3, CCL4 and IL10 as markers of severity. Of these, IL6 is a known marker of influenza disease severity with probable involvement in tissue damage [12, 14] . CCL3 and CCL4 are important mediators of virus induced inflammation. IL10 is recognized as a regulatory (anti-inflammatory) cytokine and can act on multiple cell types to regulate immune and inflammatory responses [15] . IL6 is known to be responsible for regulating plasma levels of IL1RA and IL10 [16] . Among Spanish patients, serum IL15, IL12p70 and IL6 were recognized as the hallmark of severity [10] . Significant increases in TLR levels without corresponding rise in cytokines suggest aberrant immune response in critically ill patients. The possibility of viral proteins diminishing cytokine production cannot be ruled out. Similar to in-vitro studies [17, 18] , cytokine storm associated with pathogenesis of H5N1 infection [12] was not the feature of p-H1N1-09 infection. The observations of no increase in the levels of plasma IFNG in both patient categories when compared to controls as well as basal gene expression in the lungs of critically ill patients point out lack of co-ordination in the modulation of innate and adaptive immune responses. At PBMC gene level, down-regulation of a large number of genes was associated with disease severity (Table S1, Figure 3b ). This profile is indicative of massive infiltration of monocyterecruited neutrophils, DCs/macrophages to the target organ for mounting immune response/tissue repair and/or viral proteininduced shut down of the cellular genes, as these cells are known to efficiently replicate the virus [19, 20] . Additional studies are required to examine if the virus specifically down regulates host antiviral genes or dictates generalized shut down of host mRNA synthesis [19, 21] , especially in relation to the disease severity. Paradoxically, constant inflammatory signals provided by the significant increase in several chemokines (CCL2, CCL3, CCL19, CXCL8, CXCL10, CXCL11) and pro-inflammatory cytokines (IL6, IL17, IL1A, IL1B, TNF) in the lung cells may have resulted in massive infiltration of leucocytes and excessive tissue damage. This is to our knowledge the first report on p-H1N1-09-induced gene expression in human lung cells. As against differential cytokine levels and viral load observed in a male and pregnant female fatal H5N1 case [21] , except for higher down-regulation of IL1B/IL2, the pregnant woman (S3) in our series was similar to others. It was intriguing to note that despite attempts of mounting immune response similar to mildrecovered patients, the patient S7 did succumb to the disease. It is important to recognize that the current study examines virus-host interactions throughout the course of severe disease, limitation of the study being absence of similar data from mild cases. We could investigate a critical case suffering from seasonal influenza (S21). As against similar patterns recorded for pandemic influenza cases, distinct differences were recorded for this case (Figure 3a, 3b) . In the ferret model with 27% mortality based on the .20% weight loss following infection with A/California/07/2009 pandemic influenza virus [22] , sequential lung sampling documented decreased gene expression of CCL2, CXCL10, TNFA and IL1B on day 7 when animals show highest weight loss. Comparison of the human data presented in our study (Table S1) shows that in the lungs of the critically ill patients all these and several other chemokines/inflammatory cytokines are overexpressed. Thus, the gene expression profiles at the time of overwhelming symptoms in the lungs of the infected ferrets (decreased expression) do not match the data on severe human cases from the present series (elevated expressions). While finalizing the manuscript, we came across two studies from Hong Kong [23, 24] . The mild patients (n = 22) were non viremic while 13% of the 23 fatal cases were viremic. Data from Spain, Hong Kong and India suggests the role of host genetics in immune response to the pandemic-causing virus. In conclusion, our data confirms earlier findings of dysregulated host response in severe infections with H5N1 [12] and 1918 influenza viruses [25, 26, 27] . However, the mechanisms leading to similar end results seem to vary with the type of viral infection. Indepth studies to understand the role of virus/viral proteins in modulating host response need to be undertaken on priority. Identification of specific pathways might provide clues to include immunomodulators for the treatment of severe cases, albeit in conjunction with potent antivirals.

Generation of Human Antigen-Specific Monoclonal IgM Antibodies Using Vaccinated “Human Immune System” Mice

BACKGROUND: Passive transfer of antibodies not only provides immediate short-term protection against disease, but also can be exploited as a therapeutic tool. However, the ‘humanization’ of murine monoclonal antibodies (mAbs) is a time-consuming and expensive process that has the inherent drawback of potentially altering antigenic specificity and/or affinity. The immortalization of human B cells represents an alternative for obtaining human mAbs, but relies on the availability of biological samples from vaccinated individuals or convalescent patients. In this work we describe a novel approach to generate fully human mAbs by combining a humanized mouse model with a new B cell immortalization technique. METHODOLOGY/PRINCIPAL FINDINGS: After transplantation with CD34(+)CD38(−) human hematopoietic progenitor cells, BALB/c Rag2(−/−)IL-2Rγc(−/−) mice acquire a human immune system and harbor B cells with a diverse IgM repertoire. “Human Immune System” mice were then immunized with two commercial vaccine antigens, tetanus toxoid and hepatitis B surface antigen. Sorted human CD19(+)CD27(+) B cells were retrovirally transduced with the human B cell lymphoma (BCL)-6 and BCL-XL genes, and subsequently cultured in the presence of CD40-ligand and IL-21. This procedure allows generating stable B cell receptor-positive B cells that secrete immunoglobulins. We recovered stable B cell clones that produced IgM specific for tetanus toxoid and the hepatitis B surface antigen, respectively. CONCLUSION/SIGNIFICANCE: This work provides the proof-of-concept for the usefulness of this novel method based on the immunization of humanized mice for the rapid generation of human mAbs against a wide range of antigens.

Hyper-immune sera containing polyclonal immunoglobulins (Igs) have been widely used in both therapeutic and prophylactic clinical settings [1] . However, the use of polyclonal sera was associated with several problems, such as the stimulation of allergic reactions, low reproducibility between clinical batches and high off-label use, which finally caused a decline in their use [2] . The advent of technologies to make monoclonal antibodies (mAbs) derived from animals, especially mice, has overcome many of the problems associated with the use of polyclonal sera. The technology to make monoclonal cell lines of antibody-producing cells by fusing antibody producing plasma cells with myeloma cells was described for the first time in 1975 by Milstein and Kohler [3] . The therapeutic potential of mAbs was immediately recognized and in 1980 the first mAb, OKT3, was approved for therapeutic applications. This antibody inactivates T cells, thereby preventing rejections of organ transplants [4] . However, because of the animal origin of the first generation of mAbs that were used in clinical trials, human subjects treated with these antibodies developed vigorous immune reactions against the animal proteins, which were thereby eliminated preventing their therapeutic actions [5] . To overcome these problems technologies were developed to diminish the immunogenicity of mouse antibodies by replacing part or the complete mouse antibody backbone by its human equivalent, first generating chimeric, and subsequently fully humanized antibodies [6] . In a parallel approach transgenic mice bearing the human Ig region were created to obtain fully human antibodies following immunization. The use of these mice obviates the elaborate molecular engineering of antibodies that is needed to humanize antibodies generated in wild-type mice, however, the maturation process of the mouse B cells expressing human Igs is different from that of fully human B cells [7] . Immortalization of B cells from immune humans seems to be the logical strategy to avoid these problems. However, the methods to achieve this goal have showed low efficiencies, although some progress has recently been reported [8, 9] . Nevertheless, the major disadvantage of human B cells immortalization is the need for cells from either vaccinated individuals or patients who had recovered from an infection. Thus, to fully exploit the Ig repertoire of human B cells in an in vivo setting, we explored the possibility to raise mAbs following de novo induction of human B cell responses in mice carrying elements of the human immune system (HIS). HIS mice are generated by engrafting immunodeficient mice with human hematopoietic stem cells (HSC) with or without human lymphoid tissues from fetal origin [10, 11, 12] . In particular, mice deficient for the recombinase activating gene-2 (Rag2) and the common gamma chain of the IL-2 receptor (Il2rg) on a BALB/c or a non-obese diabetic (NOD) background are permissive for human HSC xenografts. Inoculation of newborn mice from these strains with human HSC of fetal or umbilical cord blood origin gives rise to robust engraftment of a number of immune cells, including T, B, NK and dendritic cells. In this work, we describe a convenient approach to generate fully human mAbs based on the immunization of BALB/c Rag2 2/2 IL-2Rcc 2/2 engrafted with human CD34 + CD38 2 HSC [13, 14] . To this end, HIS mice were immunized with commercial vaccines against hepatitis B virus (HBV) and tetanus. Following immunization, human CD19 + B cells were sorted based on surface CD27 expression, as a marker of memory phenotype, and the isotype of surface Igs. The sorted B cell populations were immortalized in vitro by retroviral transduction with human B cell lymphoma (BCL)-6 and BCL-XL genes and antigen-specific B cell clones were established and characterized. The obtained results provided the proof-of-concept for the usefulness of this generic approach based on HIS mice combined with immortalization of human B cells for the rapid and inexpensive development of human mAbs against a wide range of antigens. The use of fetal liver tissue obtained from elective abortions with gestational age ranging from 14 to 20 weeks was approved by the Medical Ethical Committee of the AMC-UvA and was contingent on informed written consent. BALB/c Rag2 2/2 IL-2Rcc 2/2 mice were bred and maintained in individual ventilated cages, and fed with autoclaved food and water. HIS mice were generated as previously described [13, 14, 15, 16] , with the approval of the Animal Ethical Committee of the AMC-UvA (permit number DHL-100970). In brief, human fetal livers were obtained from elective abortions with gestational age ranging from 14 to 20 weeks. Magnetic enrichment of CD34 + cells (.98% pure) was performed by using the CD34 Progenitor Cell Isolation kit (Miltenyi Biotech), after preparation of single cell suspensions and isolation of mononuclear cells by density gradient centrifugation over Lymphoprep (Axis Shield). Finally, newborn (,5 days old) sub-lethally irradiated (3.5 Gy) BALB/c Rag2 2/2 IL-2Rcc 2/2 mice were injected via intra-hepatic route with 5-10610 4 sorted CD34 + CD38 2 human fetal liver hematopoietic stem cells in 30 ml. All manipulations of HIS mice were performed under laminar flow. Cell suspensions were prepared in RPMI medium supplemented with 2% fetal calf serum (FCS). Twelve to sixteen weeks after CD34 + CD38 2 HSC engraftment, HIS mice were killed and single cell suspensions of splenocytes were prepared. Red cells lysis was performed in 1 ml of red cell lysis buffer (Sigma) for 10 min. Splenocytes were washed, resuspended in 600 ml of RLT lysis buffer (Qiagen) and homogenized by passing through a 21-gauge needle several times using RNase free syringes. RNA was prepared using RNeasy mini kits (Qiagen) according to manufactures instructions. BCR V H immunoscope was performed as previously described [17] . Briefly, cDNA was prepared and real-time PCR performed by combining primers for the different V H chains (V H 1-7) and specific fluorochrome-labeled probes against the different constant regions (C H m, C H a and C H c). An additional four PCR cycles 'run-off reactions' were then performed on the PCR products using fluorescent primers specific for the constant regions (Fcm, Fca and Fcc). Products were gel separated to determine CDR3 lengths. Analysis of six individual HIS-mice containing greater than 30% human chimerism in the spleen was performed. The number of human CD19 + B cells in chimeric spleens ranged from 5-12610 6 . Eight weeks after HSC transplantation, blood was taken from HIS mice to verify the level of engraftment by flow cytometry, as described elsewhere [18] . HIS mice with a good level of human reconstitution (.20% hCD45 + cells) were immunized by intramuscular route (biceps femoris) using a 29G needle, three times on weeks 14, 16 and 18 with either 100 ml of the HBV vaccine (Engerix-B, GlaxoSmithKline) or 50 ml of tetanus toxoid (TT) containing vaccine (Tetanus vaccine, The Netherlands Vaccine Institute). These amounts correspond to 1/10 of the normal human dose. Negative controls received the same volume of PBS buffer. Two weeks after the last immunization, HIS mice were exsanguinated under isofluran/oxygen narcosis. Spleens and mLN were removed aseptically and cellular suspensions were prepared. The BM cells were isolated from the femur and tibia. from Beckman Coulter; CD3 (SK7), CD4 (SK3), CD8 (SK1), CD19 (HIB19), CD38 (HIT2), CD45 (2D1 and HI30), CD45RA (HI100), CD138 (MI15), IgM (G20-127), IgD (IA6-2), IgG (G18-145) and CCR7 (3D12) from BD Biosciences; CD27 (LT27) from AbD-Serotec; CD27 (LG.7F9) from eBioscience. TT-specific B cells were also occasionally stained with PE-coupled TT, kindly provided by Dr. Andreas Radbruch (German Rheumatism Research Center, Berlin, Germany). Dead cells were excluded based on DAPI incorporation. All washings and reagent dilutions were done with PBS containing 2% FCS and 0.02% NaN 3 . Stained cells were analyzed with an LSR-II interfaced to a FACS-Diva software system (BD Biosciences). Cell sorting of B cell subsets were performed on HIS mouse BM and spleens using a FACS-Aria cell sorter interfaced to a FACS-Diva software system (BD Biosciences). For these experiments, all washings and reagent dilutions were done with 2% FCS supplemented PBS without NaN 3 . The human BCL6 [19, 20] and BCL-XL [21] encoding cDNAs were further cloned in a LZRS retroviral expression vector, around a T2A cleavage-promoting peptide sequence and upstream a cassette containing an internal ribosome entry site (IRES) and the gene encoding GFP. We therefore obtained a LZRS vector in the following configuration: BCL6-T2A-BCLXL-IRES-GFP [9] . Transfection of Phoenix-GALV packaging cells and virus production were performed as previously described [22] . Before retroviral transduction, memory B cells were activated on c-irradiated (50 Gy) mouse L cell fibroblasts stably expressing CD40L (CD40L-L cells) in the presence of 25-50 ng/ml recombinant mouse interleukin-21 (rmIL-21, R&D systems) for 36 h [19] . The B cells were washed, mixed with retroviral supernatants in Retronectin-coated plates (Takara), centrifuged at room temperature for 60 min at 360 g, and subsequently incubated with the retroviruses at 37uC, 5% CO 2 for 6-8 h. Transduced B cells were maintained in co-cultures using CD40L-L cells (10 5 cells/ml) and in standard IMDM (Gibco) culture medium supplemented with 8% fetal bovine serum (FBS; HyClone), penicillin/streptomycin (Roche) and 25 ng/ml rmIL-21. The analysis of human Ig-V H sequences was performed as follows. Total RNA was isolated from approximately 5610 5 monoclonal B cells with Trizol (Invitrogen). The cDNA was generated and subjected to PCR with primers specific to the different V H family members. PCR products were sequenced to determine the CDR3 region of the different clones. Sequence analysis was performed using BigDye Terminator chemistry (Applied Biosystems Inc.) and CodonCode Aligner software. The plasma harvested from HIS mice (7 days after the first and second immunization; 10 days after the third immunization) and B cell clone culture supernatants were screened by ELISA for the presence of total human IgM, total human IgG and antigenspecific antibodies. Measurement of total IgM and IgG was performed by coating 96-well plates either with AffiniPure F(ab') 2 fragment goat anti-human IgM (Fc5m-specific, Jackson Immu-noResearch) or AffiniPure goat anti-human IgG (Fcc fragmentspecific; Jackson ImmunoResearch). Control human serum protein calibrator (Dako) with known IgM (0.8 mg/ml) and IgG (10.4 mg/ml) concentrations was used as a standard to be compared to the samples. For the detection of antigen-specific antibodies, 96-well plates were coated either with tetanus vaccine (Nederlands Vaccin Instituut) or Engerix B (GlaxoSmithKline) (106 diluted in PBS) for 1 h at 37uC or overnight at 4uC. Alternatively, Ridascreen Tetanus IgG ELISA plates (Biopharm) were also used to screen for TT-specific antibodies. After coating, the plates were washed in ELISA wash buffer (PBS, 0.5% Tween-20). A PBS solution containing 4% of milk was used as a blocking agent, before adding serial dilution of HIS mouse plasma (starting at a dilution of 1:5) or cell culture supernatants (starting at a dilution of 1:2). Enzyme-conjugated detection antibodies were added at a dilution of 1:2500 for HRP-conjugated anti-IgG and a dilution of 1:5000 for HRP-conjugated anti-IgM (both from Jackson ImmunoResearch). Then, TMB substrate/stop solution (Biosource) was used for the development of the ELISA assay. Statistical analyses were performed using GraphPad Prism version 5.02 for Windows (GraphPad Software). Data were subjected to two-tailed unpaired Student t test analysis. The obtained p values were considered significant when p,0.05. We have generated HIS mice by transplanting human HSC into alymphoid BALB/c Rag2 2/2 IL-2Rcc 2/2 newborn mice ( Figure 1A ). As reported previously, multilineage human hematopoietic reconstitution is observed in HIS mice, which demonstrate human thymopoiesis, B cell differentiation, NK cell and plasmacytoid dendritic cell development, and myelopoiesis [10, 13, 14, 15, 16] . Human immune cells accumulate in lymphoid tissues, and several B cell subsets are observed in HIS mice ( Figure 1B) . We analyzed the human B cell repertoire present in naive HIS mice by using B cell receptor (BCR) immunoscope analysis based on quantitative PCR of Ig variable (V H ) and constant (C H ) region gene segments [17] . Due to the lack of human spleen samples, the cells isolated from HIS mouse spleens, which contained sufficient numbers of human B cells to perform the immunoscope analysis, were compared to control human peripheral blood mononuclear cells (PBMC) samples, which were considered acceptable for the purpose of the performed comparison. We observed that IgM-expressing B cells as well as Ig isotype-switched B cells are found in naive HIS mice ( Figure 1C) . The vast majority of B cells of HIS mice expressed an IgM (97.961.0%), whereas IgG (1.861.0%) and IgA (0.0760.04%) expressing B cells represented minor populations. Only the frequency of IgA-expressing B cells was found significantly higher in control human PBMC samples (p,0.0001). At 10-14 weeks post-transplantation (i.e. in steady state conditions), the human Ig concentrations in the blood were 12268 mg/ml (IgM) and 143612 mg/ml (IgG) ( Figure 1D ), as previously reported [8, 16] . In comparison, the normal range for Ig concentration in healthy humans is 400-3100 mg IgM/ml and 7200-14700 mg IgG/ml. In brief, despite a low frequency of IgG-expressing cells, both human IgM and IgG accumulated in the plasma of ,3 month-old HIS mice to levels representing around 10% and 1% of adult human IgM and IgG concentrations, respectively. We further examined the antigen receptor repertoire diversity in HIS mice, by determining the length of CDR3 hypervariable regions for each Ig-V H gene family. The analysis of CDR3 length distribution of individual HIS mouse splenocytes showed that IgM repertoires are undistinguishable from normal human PBMC IgM repertoires, as measured by the BCR immunoscope analysis ( Figure 1E ). This observation suggests that HIS mice contain a broad variety of naive IgM + B cell clones. The V H -family usage was large and similar to control human PBMC ( Table 1) . The BCR immunoscope analysis was also performed for IgG and IgA repertoires and we observed more restricted repertoires, as expected from B cells undergoing clonal selection and Ig class switch recombination ( Figure S1 ). Immunization of HIS mice with HBV and tetanus vaccines results in the generation of antigen-specific antibody responses Since HIS mice contained broad naïve B cell repertoires, we analyzed the induction of human antigen-specific B cell responses after immunization with commercially available human vaccines. We designed a vaccination protocol based on repeated intramuscular immunizations (3 injections with 2-week intervals) of 10-14-week old HIS mice with vaccines containing hepatitis B surface antigen (HBsAg) or TT. Seven days after the last immunization mice were sacrificed, the blood and the lymphoid organs were harvested, and the phenotype and function of human cells was analyzed. All HIS mice showed human reconstitution (.20% hCD45 + cells) in the blood before starting the immunization protocol, which correlated with human engraftment in lymphoid organs. Overall, 42% of HBsAg-vaccinated (8 out of 19 vaccinated animals) and 40% of TT-vaccinated (6 out of 15) HIS mice showed significant production of antigen-specific IgM antibodies, as detected by ELISA (Figure 2A) . We performed a kinetic monitoring of antigen-specific plasma Ig levels in individual HBsAg-vaccinated responder HIS mice and we observed that after the first immunization antigen-specific Igs were rarely detected. In contrast, after the second immunization antigen-specific IgM was detected, which steadily increased after the third immunization with approximately 25-40% of responder mice also showing an antigen-specific IgG response (Figure 2A) . This suggests that repeated vaccination leads to enhanced antigen-specific antibody production. The responder mice exhibited higher total IgM (173641 mg/ml) and total IgG (4596140 mg/ml) concentrations in their plasma, as compared to PBS-injected (IgM: 37612 mg/ml; IgG: 191658 mg/ml) and non-responder vaccinated (IgM: 44611 mg/ml; IgG: 192667 mg/ml) animals ( Figure 2B) . At the end of the immunization protocol, vaccinated animals showed significantly higher numbers of hCD45 + cells in all organs (i.e. spleen, bone marrow (BM) and mesenteric lymph nodes (mLN)) in comparison to mock-injected control mice. Responder HIS mice exhibited higher numbers of human T and B cells in the spleen, as well as T cells in the BM ( Table 2; Table S1 ), suggesting that the vaccination protocol had a positive impact on the accumulation of human B and T cells. Moreover, the mLN isolated from vaccinated HIS mice contained 4 to 5-fold more hCD45 + cells than those of control animals ( Figure 2C ), suggesting that the mLN structure might play a role in eliciting an immune response in the HIS mice. In humans, the CD27 + memory B cell population contains the majority of antigen-experienced B cells [23, 24] , and we reasoned that the same should be true in vaccinated HIS mice. We therefore cell sorted several different CD19 + CD27 + B cell subsets from individual HIS mice. We used two strategies to isolate the following human B cell (CD45 + CD19 + ) subsets from BM and spleens of vaccinated HIS mice: (i) CD27 hi CD38 hi , (ii) CD27 + CD38 lo/int IgD + , and (iii) CD27 + CD38 lo/int IgD 2 on the one hand ( Figure 3 - , although the number of cells was increased for each of these subpopulations in the vaccinated animals, as expected from the enhanced number of total B cells ( Table 2) . We only observed a significant increase in the frequency of IgG + B cells within the CD27 hi CD38 hi plasmablast population of vaccinated responder HIS mice, as compared to PBS-injected animals (13.262.7% and 4.361.8% of CD27 hi CD38 hi B cells, respectively; p = 0.0311). In order to identify, isolate and immortalize the antigen-specific antibody-producing B cells, the aforementioned B cell subsets were transduced immediately after cell sorting with a retroviral vector encoding both human BCL6 and BCL-XL [9, 19] . By ectopically expressing BCL6 and BCL-XL in splenic or peripheral blood memory B cells and culturing them with factors produced by follicular helper T cells (CD40L and IL-21), we generated highly proliferative, BCR positive B cell lines that secrete Igs. Since these cells express BCL6, the differentiation of memory B cells to terminal plasma cells is blocked [28, 29, 30] . Therefore, the resulting B cells can expand extensively in vitro for long periods of time in presence of CD40L and IL-21, and provide a tool to generate antigen-specific human BCR-positive, antibody-secreting B cell lines. The number of isolated cells from spleen and antigen-specific B cell clones that were generated with the BCL6/BCL-XL transduction approach is provided in the Table S2 . Since the frequency of antigen-specific B cell clones was unknown, we started with microcell cultures ranging from 0.6 to 640 cells per well. The wells containing antigen-specific B cells -as determined by HBsAgspecific or TT-specific ELISA -were subsequently cultured by limiting dilution to obtain monoclonal B cell lines. Overall, we generated 15 anti-HBsAg IgM + B cell clones from 3 HIS mice vaccinated with HBsAg, and 18 anti-TT IgM + B cell clones from 3 HIS mice vaccinated with TT ( Table S2 ). The estimated frequency of HBsAg-specific B cells (clones) in the HIS mice after vaccination was 1/350. The IgM secretion level of the B cell clones were in the range of 1 mg per 10 5 cells over 3 days in culture, which was in a similar range of secretion (0.6-5 mg/10 5 cells/3 days) to what was previously reported for B cell clones generated from human blood [9] . The IgM V H regions of the BCR of the antigen-specific IgM + B cell clones were sequenced. Overall, the BCR of HBsAg-specific and TT-specific B cell clones exhibited a V H sequence close to the germ-line sequence, although limited frequencies of somatic hyper-mutations were observed (Table S3 and Table S4 ). Somatic hyper-mutations were occasionally detected in all framework regions (FR) and complementary determining regions (CDR), and most of the BCR diversity was the result of Nadditions in the CDR3 region. Based on the BCR sequence, we observed that 12 out the 15 anti-HBsAg IgM + B cell clones were unique, as well as 5 out the 18 anti-TT IgM + B cell clones ( Table S2, Table S3 and Table S4 ). The supernatants of TT-specific B cell clones were further tested for their capacity to recognize different antigens by ELISA. We observed that IgM mAbs did not cross-react with unrelated antigens (i.e., HBsAg and respiratory syncytial virus (RSV) antigens) ( Figure 4A) . The TT-specific B cell clones were also screened by flow cytometry for direct binding of the TT antigen labeled with a fluorochrome ( Figure 4B) . Interestingly, three types of clones that produced antibodies that gave a similar signal in ELISA were detected, with high, intermediate and low binding of the fluorescent TT antigen. In the present work we established a new approach to generate fully human mAbs. We immortalized B cells from vaccinated HIS mice by transduction with BCL6 and BCL-XL followed by expansion in presence of IL-21 and CD40L. Antigen-specific B cell clones were obtained that expressed the BCR on their cell surface and secreted antigen-specific antibodies. Similarly to methods based on the immortalization of human memory B cells from individuals that were either vaccinated or exposed to pathogens, our strategy exploits the antibody repertoire of human B cells which is likely to be different from that of B cells of mice expressing human Ig gene segments. Naïve HIS mice display an extensive human IgM-expressing B cell repertoire. Based on the analysis of the length of the CDR3 regions, this IgM B cell repertoire is similar to the repertoire of healthy individuals. Thus, HIS mice have no obvious limitations for the generation of human IgM mAbs against any possible antigen. Upon intramuscular vaccinations with either TT or HBsAg, approximately 40% of the HIS mice were able to mount an antigen-specific antibody response. Human IgM-producing B cell lines against both antigens were obtained after isolation of memory B cells followed by ex vivo differentiation into plasmablastlike cells. It is important to highlight that the selection of the antigen-specific human B cell clones relied on relevant bioassays (e.g., ELISA or neutralization test). In contrast to EBV-based approaches, human B cell immortalization using transduction with BCL-6 and BCL-XL preserves the expression of the BCR at the surface and antigen-specific B cell clones can also be selected by binding of labeled antigen to the BCR of immortalized memory B cells (e.g. by using a labeled antigen). Even when IgG was used as a selection criterion, we were unable to establish antigen-specific IgG + human B cell clones. The reason for this might be that T cell help in this system is suboptimal as indicated by the absence of antigen-specific T cell responses after vaccination (not shown). We also observed that the BCR of the B cell clones had a close to germ-line sequence, suggesting that also the induction of somatic hyper-mutation is sub-optimal in HIS mice. In our hands the great majority of the vaccinated HIS mice showed a defective formation of germinal centers [14, 16] , which further explains the absence of antigenspecific Ig-class switched B cells. So far, humanized mouse models based on the transplantation of human HSC only -i.e. without additional human tissues -share these limitations, and immunization strategies result in the limited generation of class-switched antigen-specific B cell responses [14, 31, 32] . Similar patterns are observed in human HSC-transplanted immunodeficient mice infected with lymphotropic pathogens, such as HIV [33] or EBV [34] , although Dengue virus infection in HIS mice was reported to induce an IgG response in a majority of the responder animals [35] . It is not clear why IgG antigen-specific responses are limited while serum IgG can accumulate efficiently, considering the low frequency of IgG + B cells in HIS mice. It remains to be determined whether this apparent discrepancy might be explained by the conjunction of particularly effective IgG production on a cell basis by IgG + B cells (which might occur in a T cell independent manner, such as in the case of the IgG3 subclass), long-term stability of human IgG in the HIS mouse serum as compared to human IgM, and/or defective survival of IgG + B cells under specific conditions (e.g. after antigen-specific triggering of the BCR). Although IgM mAbs might already be useful for some specific applications or could be modified by Ig class swapping to obtain IgG mAbs [36] , optimized humanized mouse models with improved B cell function are highly desirable. One reason for the suboptimal interaction of T and B cells may be the poor survival resulting in a high turnover of human T cells (discussed in [10, 37] ), making it very likely that procedures leading to improved accumulation of human T cells may promote B cell responses and isotype switching. It was already shown that human B cells undergoing isotype switching can be obtained in humanized mice, provided that a human environment supporting this process is present, e.g. in SCID mice transplanted with human fetal bones, thymus and lymph nodes [38] . Consistent with this notion, enhanced human peripheral T cell accumulation was observed in NOD/SCID mice transplanted with human bone marrow HSC, fetal liver and fetal thymus tissues (referred to as BLT mice), as compared to conventional humanized mouse systems [39] . Interestingly, BLT mice consistently generated an antigen-specific IgG response after HIV-infection [40] . Although it is yet unknown whether the isotype switch observed in BLT mice is truly T cell dependent, those data might support the idea that improved T cell homeostasis has a positive impact on B cell responses. To obtain humanized mouse models with improved B and T cell homeostasis, alternative strategies not relying on the transplantation of human fetal tissues -which are not necessarily easy to access to, for ethical, legal or practical reasons -will likely be favored in the future. The replacement of mouse genes involved in the hematopoietic system by their human equivalent is a valuable strategy to improve development, maintenance and/or function of several hematopoietic cell subsets in humanized mouse models, as shown with cytokines, such as IL-7 and Il-15 [15, 21, 41, 42] , or MHC molecules (N.D.H and J.P.D., manuscript submitted) [43, 44] . The fact that the human CD47 was shown to be unable to properly interact with the mouse SIRPa indicates that reintroducing a functional phagocyte inhibition mechanism via the CD47/SIRPa signaling axis is another strategy of potential interest [45] . In conclusion, our results show using two standard vaccine antigens the general applicability of an innovative B cell immortalization method in combination with the HIS mouse model to generate human mAbs. Similarly to methods based on the immortalization of human memory B cells from vaccinated or convalescent individuals [9] , our approach exploits the broad antibody repertoire of human B cells, overcoming the potential limitations of conventional humanized murine mAbs such as laboriousness or impaired biological properties, synthetic antibody libraries that require a known target antigen, and transgenic mice bearing the human Ig locus that have limited B cell repertoires. In addition, our method enables to exploit experimental infection models and immunization regimes that would be unethical or untenable in humans. Considering the upcoming advances in HIS mice models [37] , this new approach will provide a powerful tool to generate human mAbs for either diagnostic or therapeutic purposes. Figure S1 IgG/IgA B cell repertoire in naïve HIS mice. Similarly to Figure 1E , the naive IgG (A) and IgA (B) B cell repertoires of HIS (BALB-Rag/c) mice were evaluated on splenocytes by performing a BCR immunoscope for each V H family. The profiles obtained with control human PBMC are also shown. Found at: doi:10.1371/journal.pone.0013137.s001 (1.61 MB TIF) Figure 3 . Limited dilutions of B cells transduced with BCL-6 and BCL-XL were performed with 6.4 and 0.64 cells/well. After sub-cloning of the positive wells, we generated 15 IgM + anti-HBsAg mAbs, of which 13 are unique (as determined by Ig-V H sequence, see Table S3 ), and 18 IgM + anti-TT mAbs, of which 5 are unique (see Table S4 ). In the case of HBsAg vaccination, the number of screened B cells was ((192*6.4)+(96*0.64))*3 = 3870, which eventually suggests that the frequency of HBsAg-specific B cells is at least 1/350 B cells. Found at: doi:10.1371/journal.pone.0013137.s003 (0.13 MB DOC) IgM V H amino-acid sequence of generated HBsAgspecific B cell clones. The germ-line sequence is given for each V H family, with indication of framework regions (FR) and complementary determining regions (CDR). Highlighted amino-acids correspond to N-additions (in the CDR3 region) and somatic hyper-mutation events, whether it results in a silent mutation (green) or not (red). Clones with identical BCR sequences are grouped together. Found at: doi:10.1371/journal.pone.0013137.s004 (0.02 MB PDF)

Reflections on Pandemic (H1N1) 2009 and the International Response

Gabriel Leung and Angus Nicoll provide their reflections on the international response to the 2009 H1N1 influenza pandemic, including what went well and what changes need to be made in anticipation of future flu pandemics.

There is general consensus that the only predictable characteristic of influenza viruses and pandemics is their unpredictability [1] . Given such uncertainty, reasonable application of the precautionary principle should prevail in the responses. Indeed many of the initial responses to the 2009 pandemic went well. Once isolated, the pandemic virus strain was shared immediately, specific diagnostic assays were produced and distributed worldwide, antivirals were available in many countries, vaccine development started promptly, and clinical trials demonstrating vaccine safety and immunogenicity were conducted rapidly. There were many inherently favourable features of the pandemic itself, not all of which were immediately apparent (Table 1) . This was not 1918 Spanish flu. The impact has been mostly confined to the health sector. But that impact has been significant and heterogeneous, with pressure experienced by primary and hospital care (especially intensive care and paediatric services). Distilling descriptions of the impact of a complex public health threat like a pandemic into a single term like ''mild,'' ''moderate,'' or ''severe'' can potentially be misleading [2] . Certainly the experience of hospital clinicians indicated that this pandemic, sometimes described as ''mild to moderate,'' was not limited to only mild or moderate illness. Many patients were severely ill and died, and undoubtedly, high-quality clinical management of patients with severe complications in intensive care units saved many lives of the critically ill, who often required prolonged hospitalisation [3] . The epidemiology of this pandemic is different than for seasonal influenza epidemics, but not unlike previous pandemics. Young people have been disproportionately affected in terms of hospitalisation and deaths compared to seasonal influenza in which complications and mortality are predominantly borne by the elderly [4] . Similarly, the risk to pregnant women has been higher than for seasonal influenza [5, 6] , which was also noted in previous pandemics. The attributable premature mortality may remain unclear for some time. Recent American analyses have estimated many more deaths than those officially reported with laboratory confirmation of infection and that years of life lost were equivalent to the 1968 pandemic. The lower bound of such estimates is equivalent to the annual burden caused by a typical H3N2 seasonal epidemic in temperate climates [7, 8] . The years-of-life-lost metric captures the impact of a different agespecific mortality pattern which death counts cannot. Deaths involving the young and healthy incur many more potential years of life lost compared to those of older adults and of chronically ill individuals. There are also a number of ''firsts'' for the 2009 pandemic after an interpandemic period of more than four decades (Box 1). These brought both opportunities and challenges. Under the auspices of the World Health Organization (WHO), the process of a global review by public health specialists from around the world has recently begun. They were nominated by national authorities and are led by an elected chair who assessed the handling of the 1976 swine influenza event among US military personnel at Fort Dix [9] . Here we offer some initial reflections on the first 12 months of the present pandemic. Considerable effort in recent years had been dedicated to preparing for surveillance during a pandemic and to incorporating modelling in planning in some countries. The pandemic virus was detected and isolated reasonably early, although too late for any attempt at containment. It remains unclear precisely when or where it first emerged, but the earliest human infections were detected in North America and the best estimates of the timing of emergence are variously mid-February from field epidemiology in southeast Mexico or mid-January from a molecular clock model [10] . Situational awareness during the early phase allowed quick assessment by countries, notably those affected first (Mexico, US, Canada, and Southern hemisphere temperate countries). The integration of clinical, laboratory, and epidemiologic data proved essential and gave important insights into disease severity, transmission dynamics, and anticipated impact of interventions. Focused local or national studies with analyses shared through WHO or regional bodies proved more valuable than relying on collection of primary data for analysis in some regions [2] . Although there were modelling efforts underway, only a few governments incorporated such data for policy decisions. Data from seroepidemiological studies have been limited, primarily due to the lack of routine influenza serosurveys, and The Essay section contains opinion pieces on topics of broad interest to a general medical audience. technical challenges with the assays, interpretation, and validation of results. Available serological data on prevalence or seroincidence of humoral immunity yielded age-specific attack rates that indicated a substantial proportion of asymptomatic infections and mild illnesses, similar to or greater than past pandemics and seasonal outbreaks. This was confirmed by a recent Hong Kong study showing the proportion of asymptomatic infection, secondary attack rates, viral shedding, and course of illness among household members were largely similar between infections with seasonal and pandemic influenza virus strains circulating during 2009 [11] . The few published serosurveys revealed heterogeneities in infection rates among different groups and between different places [12] [13] [14] . In particular there appears to be serological evidence of substantial preexisting humoral immunity among older adults, ranging from 23% (1:32 titre by haemagglutination inhibition in those 65 years or over) [14] to 34% (1:80 titre by microneutralisation assay in those 60 years or over) [15] in different studies. Further data on population susceptibility by age or the availability of a rapid and accurate serological test could allow health services to further target vaccine efforts for subsequent waves, as has been done in a few countries [14] . Early on, some airports installed thermal screening and others asked travellers to declare fever or respiratory symptoms at disembarkation. The utility of these interventions has been repeatedly challenged [16] , although if executed well could delay the start of community transmission by a few weeks [15, 17] (Table 2) . Similarly, during the early stages of global or local spread, quarantine, isolation, school closures, and other social distancing measures were variously implemented in some populations (e.g., Mexico [18] , western Japan [19] , UK), although most have not yet been formally evaluated and published [20] . Two exceptions are in Hong Kong and the UK. In the former, it was estimated that transmission fell by 25% when schools closed [21] . In settings like Hong Kong, with the infrastructure and resources to implement such measures and N Decisions regarding pandemic response during the exigencies of a public health emergency must be judged according to the best evidence available at the time. N Revising pandemic plans-to be more flexible and more detailed-should wait for WHO leadership if national plans are not to diverge. Surveillance beyond influenza should be stepped up, and contingencies drawn up for the emergence or re-emergence of other novel and known pathogens. [22] but some countries attempted containment in Phases 5 and 6. Some countries even instituted a ''containment phase'' using case-finding and various measures such as isolation and antiviral treatment of ill suspected and confirmed cases, and quarantine of exposed persons with or without antiviral chemoprophylaxis, while others never attempted or quickly moved from resource-intensive containment to mitigation [22] . A preliminary evaluation of intensive containment undertaken in parts of the UK during its spring/summer wave of 2009 demonstrates how resource-and labour-intensive community containment could have been and also how even with a lot of resources the measures had to be abandoned [23] . It is now recognised that the phrase ''containment'' was unfortunate and potentially misleading since at best the actions were only mitigating impact [24] . This pandemic virus transmitted efficiently among children and at least one study has shown that school closures were associated with reduced population transmission when implemented early [21] . Closures appear to have stopped school outbreaks in western Japan and might have also mitigated impact initially on the local communities [25] . However, decisions on this intervention were contextually specific, dependent on feasibility and their potential downsides [26] . In Europe and the US the judgement was generally that proactive school closures would not be justified as a community mitigation intervention in the context of a perceived mild-to-moderate pandemic among the general population, and reserve plans for widespread closure have not been activated in most jurisdictions. However, local decisions were made to close schools in some areas as a response to prevent transmission and high attack rates among schoolchildren or simply where there was too much illness and absenteeism to sustain teaching [21, 27] . Personal protective interventions such as face masks, hand hygiene, and early isolation may have been beneficial in reducing transmission at the individual level in the home [27, 28] , although household secondary attack rates during the pandemic were similar to those with seasonal influenza [13, 29] . Their population level impact remains to be assessed. There was much debate over whether to use conventional masks or respirators in health care settings. One well-conducted Canadian trial on seasonal influenza virus transmission published during the pandemic suggested no additional advantage from N95 respirators [30] . Oseltamivir and zanamivir (and later peramivir in some countries) played a role in the mitigation effort, sometimes drawing on national stockpiles. Except for Japan, widespread use of antivirals had not been the norm previously. It became standard to recommend neuraminidase inhibitors for treatment of inpatients and high-risk outpatients, and in restricted circumstances for chemoprophylaxis. Innovative delivery schemes were sometimes developed. Those who fell sick in England could have a telephone assessment (taking pressure off primary care) and then if appropriate receive empiric oseltamivir treatment from a local pharmacist. In Norway oseltamivir was made available ''over the counter.'' However in many European settings, reluctance remained among primary care providers to prescribe a drug they were unused to. Another controversy was whether to offer oseltamivir to all those with symptoms or target those at higher risk for complications. The observational data so far suggest that early treatment with neuraminidase inhibitors have worked to reduce severe disease and have not been linked to significant adverse risks [31, 32] . Late clinical presentation and delayed initiation of antiviral treatment have been implicated with more severe complications worldwide, indicating gaps in identifying and treating patients before disease severity increases. While sporadic cases of oseltamivir resistance have been reported in association with a specific mutation (H275Y in neuraminidase), such oseltamivir-resistant viruses have rarely transmitted [3] . Indeed, the pandemic virus has remained genetically and antigenically stable so far. The core pharmaceutical preventive intervention was vaccines and this has Box 1. A Series of ''Firsts'' about Pandemic (H1N1) 2009 N The first pandemic to emerge in the twenty-first century. It has been more widespread and remains ongoing, compared to SARS. N The first pandemic to occur after major global investments in pandemic preparedness had been initiated. N The first pandemic for which effective vaccines and antivirals were widely available in many countries, thus requiring public health authorities to earn and retain the confidence of health care providers through whom such are usually distributed. N The first influenza pandemic to coincide with the ongoing HIV/AIDS pandemic and for which preliminary data do not suggest a substantial, disproportionate impact on HIV-infected patients. N The first pandemic that took place within the context of a set of International Health Regulations and global governance, which had not been widely tested until the present. N The first pandemic with early diagnostic tests that led to rapid diagnosis but also an early obsession in the media and of policymakers with having reports of the numbers of those infected. N The first pandemic with antivirals available in many countries that led to a hopeful expectation that the pandemic might be containable, leading to the preparation for and implementation of a ''containment phase'' in some places. N The first pandemic in which intensive care was available in many countries to treat critically ill patients, fostering an expectation that everyone could be treated and cured. N The first pandemic with a ''blogosphere'' and other rapid social media messaging tools that challenged conventional public health communication. been a particular focus for critics citing the uneven and suboptimal uptake across countries. Development of a pandemic vaccine was a scientific success, but limited availability until after the autumn/winter wave had nearly peaked in the Northern Hemisphere contributed to lower coverage than anticipated [33] . Vaccination coverage depended on many factors, including availability, preordering, licensing and bureaucratic hurdles, logistics, convenience, and, most crucially, public and professional perceptions. This pandemic presented a particular risk communication challenge, since while infection usually results in mild illness, occasionally it is lethal, even in the young and previously healthy despite optimal treatment [34] [35] [36] . In the absence of any excess risk of serious side effects compared to annual seasonal vaccines [37] (despite the intensive effort to look for such) the benefits of immunisation far outweighed any potential down-sides at the individual level, particularly for those at higher risk for complications. Notwithstanding such evidence, the cost of pandemic vaccines was considerable and a loss of public confidence has sometimes been triggered by unsubstantiated media reports of serious side effects with a ''new vaccine'' that utilised the same manufacturing technology as for years of seasonal vaccines. Uptake among health care providers as role models has been mixed, as has their expression of the need for vaccination at all. This sometimes cast doubt in the minds of the public. Conversely, pandemic deaths in young healthy people abruptly changed public perception (such as in Canada, Romania, and Finland); supply and organisational issues then became crucial. Another more fundamental criticism challenges whether vaccines should have been procured at all given an eventual surplus in the developed North. The unexpected finding that a single dose was immunogenic among all persons except for younger children, which reduced the required number of doses by half from the projected number needed in most countries, but this was not known in advance of countries placing vaccine orders. Had there been ''overpreparation''? The prior worry had been the reverse -would there be sufficient production capacity to meet needs [38] ? Even in retrospect, and with the observed burden of the pandemic, a vaccine was clearly justified for countries where annual vaccines for seasonal influenza are routinely recommended. Field and pharmacovigilance data so far have shown that these vaccines were immunogenic, effective, and very safe [39] . However, the frailty was timing and availability. Generally supplies came in later and in smaller amounts than forecasted, in part due to lower yield in growth of the vaccine virus strain than expected. Reduce and delay community spread somewhat at the earliest stage to allow better preparation for mitigation response [15] Completely prevent entry of infected individuals due to suboptimal sensitivity and asymptomatic (including infected and within incubation period) or subclinical presentation [16] Many countries did not attempt these measures because of logistics, stage of pandemic [22] or other cost-benefit considerations [16] China Hong Kong SAR Japan Personal protective measures (e.g., face masks, hand hygiene, cough etiquette, early self-isolation when ill) Reduce risk of infection to self and close contacts (if self is ill and infected) [27, 28] Have not been evaluated whether they can provide significant populationlevel protection Virtually all countries implemented these measures to varying degrees in health care settings according to the risk of the situation. Almost all encouraged hand hygiene, cough etiquette, and early self-isolation Most countries recommended adoption of hand hygiene, cough etiquette, and early self-isolation when ill, but use of face masks in the community was uncommon except in East Asia. Antivirals for treatment and chemoprophylaxis [21, 22] Mitigation: reduce illness severity and complications if administered early; reduce transmission from those receiving treatment; sometimes also used as chemoprophylaxis in high-risk circumstances Provide significant population-level protection or allow containment Attempts at source containment were not possible, as the pandemic was effectively already in WHO Phase 5 when what became the pandemic virus was first identified [22] . Initial observational studies suggest antivirals were successful when early treatment was administered Canada Germany Hong Kong SAR Japan UK US (these populations attempted the intervention initially but effort was not sustained towards the later stages of the pandemic) Vaccines Mitigation (a) at individual level by conferring immunity to infection in those at higher risk of severe disease or (b) at a population level by immunizing population groups especially those who are transmitting most (i.e., children) In most countries vaccine was not available early enough and/or arrived in insufficiently large amounts to achieve mitigation at a population level. Greater population benefit may occur in the next season Most of the orders arrived after the peak of the autumn/winter wave in the geographic north (whose countries had received most vaccines). Therefore, judgement on their impact in averting serious morbidity and deaths may come only after the second winter. Perhaps then, differential use by countries will allow for comparisons where there is good surveillance for severe disease and deaths. There have been claims of extraneous influence on the independent and objective judgment of expert advice that in turn influenced decision-making [40] . These claims have been robustly countered as they relate to WHO's advisory and decision-making process [41] . As Harvey Fineberg, chair of WHO's external review, pointed out, when assessing any allegations of impropriety or bias, or the perception of such, it would be important to distinguish between financial or other conflicts with potential pecuniary gains versus predispositions arising from an individual's background and experience. Rather than aiming for a complete purge of any and all experts who had worked with vaccine manufacturers and received sponsorship, as these are often the very same group who possess the most relevant and useful expertise precisely because they have been closely involved in the research and development process, the focus should be on making the declaration of such interest wholly transparent and comprehensive according to a set of robustly established procedures that can withstand the strictest scrutiny. It is reassuring that WHO has honoured its commitment to making public the names and declarations of interest of the pandemic Emergency Committee when the pandemic was declared over on 10 August 2010. Additionally, receiving advice should be differentiated from making decisions. The people entrusted with undertaking the latter task should then judge the validity of the advice rendered by experts, having taken into account their interest declarations. The decision makers should also be prepared to justify their actions. It is important to learn from our experience through the first year and beyond as we move into the new seasonal influenza [42, 43] . It is theoretically possible, although unlikely, that the second winter of this pandemic will be worse than the first, as happened for the 1968 pandemic when transmissibility increased [44] . Equally, if the pandemic virus outcompeted the A(H3N2) virus strains responsible for more intense seasonal epidemics, there may even be a diminution of disease burden in older people. As of this writing, seasonal influenza A (H3N2) and B virus strains continue to cocirculate. Antigenic drift in the 2009 H1N1 virus is expected to occur in the future, especially under the pressure of so many people now being immune through infection or immunisation, although the timing is unpredictable. The pandemic virus is included in the trivalent seasonal influenza vaccine composition for both hemispheres. Clear communication of public health messages will remain a particular challenge and not confusing what could happen (and should be prepared for) with what is most likely to happen. In assessing the pandemic response, decisions made during the exigencies of a public health emergency must be judged according to the best evidence available at the time. Hindsight always gives perfect vision and using post-hoc information to evaluate prior decisions at best confuses and often produces unfair conclusions. Preparedness plans will have to be revised in due time, after the many lessons learned have been gathered. This should be done quickly in case the worst is not yet over [45] . However, rewriting plans should best wait for WHO leadership if national plans are not to diverge. A strong argument exists for making future plans more flexible and having extra descriptions including the many aspects of severity when a pandemic is emerging that then determine the consequential public health actions [2] . Broadening surveillance for a range of influenza A viruses among a wide range of animals (e.g., swine), not just in avian species, as well as strengthening the monitoring of seasonal influenza virus infections in humans will facilitate identification of novel influenza A viruses of pandemic potential, and earlier detection of the emergence of a pandemic virus. More broadly we should look beyond influenza and draw up contingencies for the emergence or re-emergence of other novel and known pathogens [45] . One challenge faced initially in this pandemic was for timely collection and sharing of clinical data to inform optimal management of critically ill patients worldwide. Establishing clinical research infrastructure prior to a pandemic and a central institutional review board will facilitate data collection and analyses [46] , whether for the next influenza pandemic, SARS outbreak, or next novel respiratory pathogen of global importance. Clinical management of severe influenza disease should not be limited to the current antiviral regimen, and include the development of other therapeutics (e.g., novel antivirals and immunotherapy). Ongoing improvements in the routine and timely monitoring of hospital admissions and deaths attributable to influenza, as well as representative serological surveys at regular intervals can provide epidemiological data with which to reduce uncertainty around the true burden of influenza and thus inform policy choices [47] . Assessment of the humoral and cellular immune response over time in a subset of vaccinated individuals could reveal how vaccine-induced immunity differs from natural infection, and whether cross-reactive responses to other influenza virus strains are modulated by the two types of immunological response [48] . The latter could become important as the pandemic strain has already been cocirculating with other interpandemic influenza A virus strains in some parts of the world. Greater access to antivirals and influenza vaccines worldwide is an ongoing challenge. Although WHO secured pledges of 200 million vaccine doses and monies for operations, and more than 80 less-resourced countries have signed agreements with WHO for supply of vaccines, this gap remains. It is an indefensible fact that these vaccines started to flow to the poorer countries well after they began going to the countries with advance purchase arrangements. Delivering timely pandemic influenza vaccination in countries without existing seasonal vaccine programmes is proving difficult. The longterm solution has to be improved surveillance, expanded monitoring of disease burden, and better prevention and control of influenza, including the development of seasonal vaccine use and production in all regions of the world [49] . Increased coverage of available bacterial vaccines (Hib, pneumococcal) will help prevent secondary invasive bacterial coinfections with either seasonal or pandemic influenza. Finally accusations of ''overreaction'' can be countered by the observation that investment in fire services or insurance is usually judged against their ability to respond to conflagrations. If the first test is a lesser fire, that experience should be used for improvements rather than as a reason to scrap the fire engines and cancel the insurance [40] .

Development of a large-scale isolation chamber system for the safe and humane care of medium-sized laboratory animals harboring infectious diseases

The close phylogenetic relationship between humans and non-human primates makes non-human primates an irreplaceable model for the study of human infectious diseases. In this study, we describe the development of a large-scale automatic multi-functional isolation chamber for use with medium-sized laboratory animals carrying infectious diseases. The isolation chamber, including the transfer chain, disinfection chain, negative air pressure isolation system, animal welfare system, and the automated system, is designed to meet all biological safety standards. To create an internal chamber environment that is completely isolated from the exterior, variable frequency drive blowers are used in the air-intake and air-exhaust system, precisely controlling the filtered air flow and providing an air-barrier protection. A double door transfer port is used to transfer material between the interior of the isolation chamber and the outside. A peracetic acid sterilizer and its associated pipeline allow for complete disinfection of the isolation chamber. All of the isolation chamber parameters can be automatically controlled by a programmable computerized menu, allowing for work with different animals in different-sized cages depending on the research project. The large-scale multi-functional isolation chamber provides a useful and safe system for working with infectious medium-sized laboratory animals in high-level bio-safety laboratories.

In high-level bio-safety laboratories, animals infected with Risk Group 3 pathogens (as defined by the World Health Organization) must be housed in isolation chambers (World Health Organization, 2004) . In an effort to minimize the risks for scientists working with these animals, a new isolation chamber was designed in a bio-safety lab. The chamber was designed mainly to separate the internal controlled environment from the external environment, and the operator from the experiment and the experimental products. The primary aim was to prevent crosscontamination from the internal to the external environment or vice versa (Kruse et al., 1991; Wathes and Johnson, 1991; Huang, 2005; Tattershall, 2006) . Isolation chambers were designed to improve experimental safety by preventing this cross-contamination, reducing the likelihood of operator error, and minimizing the contaminated area. In addition, the comfort level of scientists working with the isolation chambers was taken into account. The safety level of these chambers allows for the operator to avoid wearing too many protective suits, which improves both the comfort and flexibility of the operator (Sawyer et al., 2007) . There are two main types of isolation chambers: positive pressure and negative pressure. In general, positive-pressure isolation chambers are used to ensure that specified-pathogen-free (SPF) laboratory animals housed in the chambers are protected against outside contaminants (Hurni, 1981; Clough et al., 1995) . Negative-pressure isolation chambers are used with infectious animals housed in the chambers to prevent migration of hazardous contaminants to the outside (Wathes and Johnson, 1991) . Both rigid and soft barriers are commonly used for physical separation in the construction of isolation chambers (ISO 14644-7:2004) . Rigid barriers can be made of many different materials, and the more rigid the material, the more reliable the physical barrier. Construction of these rigid barriers usually makes use of plastic enclosures, metal profile enclosures, or hot-worked metal enclosures (ISO 10648-1:1997) . The isolation chamber is designed to house a living animal, and therefore continuous airflow inside the enclosure is needed to drive out heat and moisture generated by the animal's metabolism and to decrease the concentration of odor, dust, and infectious substances (Hillman et al., 1992) . The resulting exhaust gas is subject to a filtering system designed to prevent pathogen contamination of the external environment (Institute of Laboratory Animal Resources Commission on Life Sciences, National Research Council, 1996) . The aerodynamics is allowed for one-way flow or turbulence of the airflow inside the isolation chamber, under negative pressure relative to the environment. Supply and exhaust air can be passed through high-efficiency particulate air (HEPA) filters to prevent the formation of aerosols that could potentially escape into the environment (Runkle et al., 1969) . A double port transfer exchange (DPTE) system is commonly used in isolation chambers to allow the transfer of experimental materials from one chamber to another without exposing the experimental material to the outside environment (Allen et al., 2009) . The acronym DPTE was originally derived from the French phrase 'double porte de transfert etanche', meaning double door sealed transfer or double door transfer port. A newly validated rapid transfer port boasts bi-directional transfer as one of its features, a system also known as an α-β transfer port or rapid transfer port (RTP) (Michael et al., 2004; Tsai et al., 2009) . Non-human primates with a close phylogenetic relationship to humans are often susceptible to a number of diseases and parasites found in humans (Tauraso et al., 1969) . To establish an unknown pathogen as the cause of a disease Koch's postulates should be followed (Fouchier et al., 2003; Pan and Sun, 2004; Conly and Johnston, 2008) . The similarity of genetic, physiological, and behavioral characteristics between humans and non-human primates, and the occurrence of similar pathological changes upon infection, have led to non-human primates becoming an irreplaceable animal model for the study of pathogen infection, the screening of anti-pathogen drugs, and vaccine evaluation (Conly and Johnston, 2008) . Because of the critical role that non-human primates play in the study of these pathogens, it is critical to find safe and reliable methods for their physical containment. In the present study, a negative-pressure isolation chamber was designed and an initial attempt at building the chamber was completed. Two extraction blowers are used, one working and one on standby, to ensure the isolation chamber internal pressure is both stable and secure. The isolation chambers contain two DPTE systems for the transfer of experimental materials and the collection of excrement into a small container. The sterilization chain for the isolation chamber is composed of a peracetic acid sterilizer and its associated pipeline. A programmable logic controller (PLC) system and WINCC 6.0 system are used to automatically control and monitor the isolation chamber. Rapid pressure changes and the exhaust system are adjusted via a variable frequency drive inlet air blower and extraction blower. Isolation chamber parameters are menu-managed by the automatic control system allowing the researcher to set different isolation environmental parameters depending on experimental requirements. 2 Negative-pressure isolation chamber system design The structure of the negative-pressure isolation chamber system includes a top ventilation unit, an isolation chamber working zone, a lower part of the control system, and an isolation chamber support frame (Fig. 1) . The isolation chamber working zone is composed of a stainless-steel box and two front doors. The welded box was manufactured using dumb-gloss stainless-steel 316L with a thickness of 3 mm. The interior chamber dimensions are 400 cm (L)×120 cm (W)×120 cm (H). The isolation chamber front door includes two damping braces on each front side with a dual-piston mechanism for holding the front door securely open to let the animal cages in, two operation panels with stainless-steel 316L framework, and two door hinges connected to the stainless-steel box. The operation panel is made of transparent polymethylmethacrylate (PMMA). The transparent front door allows for visualization of the contents of the isolation chamber. Silicone seals around the PMMA panel ensure that the system is air tight. The front door is manually fastened onto the box framework with a hammer bolt. There are seven circular polyethylene (PE)-machined glove ports on the operation panels. The port inner diameter is 300 mm and the center-tocenter spacing of each port is 450 mm. The glove port assembly includes a glove port ring, glove port gasket, pressure ring, and glove port inner-securing ring. The glove port ring and glove port inner-securing ring are jointed with a thread connection. The glove port ring edges are fixed on the PMMA panel by compressing the glove port gasket and pressure ring on each side of the PMMA panel by tightening a fixing screw. The changeable sleeve and glove combination is mounted on the glove port through a sleeve fixing ring that secures the elastic Hypalon sleeve onto the glove port inner-securing ring. A glove port bung connects the glove and sleeve. Neoprene glove shapes are ambidextrous. The transfer system for the isolation chamber is composed of two DPTE systems. One round, 270-mm diameter DPTE system with an α transfer door is built into the left wall of the isolation chamber using a transfer port assembly kit. The transfer port assembly kit includes the DPTE transfer port external flange, DPTE transfer port external flange sealing ring, and DPTE transfer port internal flange. The DPTE transfer port external flange is fixed onto the inside isolation chamber wall by compressing the DPTE transfer port external flange sealing ring on the outside of the isolation chamber wall with a tightening fixing screw. The animal transfer container is autoclavable and contains a β door that can be manually docked to the port. The transfer container is 50 cm deep and is capable of transferring a 6-kg monkey into the isolation chamber. The transfer container can be autoclaved without compromising its containment and can be opened with a specialized tool to remove the sterilized waste. This system works very well for rapidly and safely transferring experimental materials and animals. Another 270-mm diameter DPTE system with an α transfer door is mounted on the bottom of the isolation chamber. The autoclavable canister using a β door can dock to the port and with a depth of 20 cm is more suitable for transferring smaller materials such as animal blood samples. This type of short bottom transfer canister can be opened with a specialized tool in the biological safety cabinet, and the samples can then be removed for further stages of analysis (e.g., centrifugation), while the transfer canister is closed and put into a plastic bag for autoclaving. The animal excrement collection system is composed of a disinfection pool, animal excrement collection DPTE transfer container, and scrubbing tools. The disinfection pool is oblong, has a slightly tilted bottom and a perimeter wall, and is close to and higher than the bottom transfer port. A drain valve is installed in the disinfection pool outfall. One end of a flexible diversion silicone tube is connected to the drain valve, and the other end touches the autoclave plastic bag inside the animal excrement collection DPTE transfer container docked to the bottom transfer port. The transfer container depth is 50 cm. There are stainless-steel lockable wheels under the bottom of the animal excrement collection DPTE transfer container. One gas-tight water service valve with a serrated hose is mounted on the left-side of the interior. A spray gun is connected to the serrated hose for cleaning the isolation chamber. Four primate cages can be placed on the stainless-steel-cage slideways in the isolation chamber. The slideways are attached to the isolation chamber bottom in a manner that allows cage movement in a direction along the axis perpendicular to the axis of the isolation chamber front door. If the primate cages are removed, other animal cages such as rabbit cages and guinea pig cages, can also be placed on the detachable frame, but each time only one kind of animal species cage can be inside. The isolation chamber is supported by two type 304 stainless-steel isolation chamber support stands. The control cabinet stainless-steel shelf and fasteners are mounted to the right-side tubular frame of the base stand and provide a work surface to support the control cabinet. Stainless-steel slideways are also mounted on the top of the isolation chamber box. The pipes, air blower, valve, and adjustable illumination lamp are fastened to the reserved mounting holes or mounting plates of the slideways by a fixing screw. Two extraction blowers share one exhaust port in which an anemometer is installed. Each of the extraction blowers is connected with its own coupling clamp to the outlet ventilation pipe, exhaust ventilation pipe, and exhaust electronic control ball valve to form an exhaust channel. The two exhaust channels have a parallel connection. Two sets of sterilizing agent bypass tubes have a series connection with an ipsilateral sterilizing agent bypass electronic control ball valve and a parallel connection with the same side of the exhaust ventilation pipe on two ends of the exhaust electronic control ball valve. All of the valves are automatically controlled by the PLC. The airflow direction through the isolation chamber via the air inlet and outlet is shown in Fig. 1 . Room air is drawn into the interior of the isolation chamber through the inlet pre-filter, inlet air blower, inlet ventilation pipe, inlet air electronic control ball valve, and inlet HEPA filter in the animal breeding mode. The air from the isolation chamber is drawn out of the exhaust export through the exhaust pre-filter, exhaust HEPA filters A and B arranged in series, exhaust ventilation pipe, and exhaust electronic control ball valve via an extraction blower. The HEPA filters are arranged in series to ensure that if one fails, the other can still ensure exhaust security and prevent pathogens from being discharged into the atmosphere. During the sterilization process, the inlet air blower, inlet air electronic control ball valve, exhaust electronic control ball valve, and extraction blower are all closed. Following this closure, the sterilizing agent by-pass electronic control ball valve is opened. The sterilizing agent in the peracetic acid sterilizer evaporation tank is heated and its vapors are pushed by compressed air into the volume to be sterilized by a sterile connecting hose, a sterilization reagent import, a coupling clamp for the inlet ventilation pipe and an inlet HEPA filter. The exhaust sterilizing vapors escape from the interior of the isolation chamber through an exhaust pre-filter, two-in-series HEPA filters, a sterilizing agent by-pass tube, a sterilizing agent by-pass electronic control ball valve, and an extraction blower to the exhaust export. The isolation chamber interior pressure is controlled by automated instrumentation that is connected to the supply and exhaust ventilation system. The automatic pressure regulation system is capable of maintaining the relative pressure inside the isolation chamber via the exhaust ventilation system, which can account for transient volume changes such as glove entry or withdrawal. A videotape system mounted on the control touch panel stainless-steel box on the side of the isolation chamber box includes a rotatable camera and camera-installation base. The camera installation base is fastened to the control touch panel stainless-steel box by fixing screws. The position of the rotation camera can be adjusted by using the telescopic locking nut and rotating locking nut. This system enables recording of both the scientist's experimental procedures and the status of animals living in the isolation chamber. The video is displayed on the personal computer (PC) screen and saved automatically in the central control room through the control interface connected to the videotape system. To ensure the comfort and welfare of animals in the isolation chamber, chambers are equipped with an automatic light control system and a television entertainment system. The automatic light control system includes adjustable illumination lamps and a lampshade. The adjustable illumination lamps are composed of three cold light lamps and their conditioners. The illumination system can be used to meet the needs of the animal's physiology, as well as experimental requirements. The television entertainment system consists of a flat television and two transparent television installation boxes fastened to the front door by fixing screws. The animal is able to watch television programs that are controlled by the central control room. The purpose of this design is to reduce depression associated with the space constraints faced by the animals and to ensure the ethical treatment of the animals. The composition of the isolation chamber control system is depicted in Fig. 1 . This system includes an alarm indicator light tower, liquid crystal display (LCD) touch screen, and control cabinet. The LCD touch screen is fastened on the outside of the isolation chamber by a fixing screw. The PLC is built into the control cabinet. The control cabinet, which has a fan and a filter cover, is mounted onto the stainless-steel shelf of the isolation chamber support frame through fasteners and a fixing screw. Step 7 software installed in the PLC central processing unit (CPU) and through the LCD touch screen enables users to automatically control a variety of options. The animal breeding mode program, containment test program, sterilization program, auto/manual control mode, maintenance mode, and custom programs can all be automatically controlled by the PLC CPU and associated touch screen. The management system for the isolation chamber touch screen was developed by Simatic WINCC flexible software. The operation can be recorded and printed, as can any system failures. The data interchange between the PLC and touch screen is made possible through an industrial trunk Profibus decentralized periphery (DP). The temperature, humidity, illumination, atmospheric pressure, and air flow velocity are measured by appropriate sensors, and the values are imported to the PLC through control lines. Normal value ranges for each parameter can be programmed into the PLC, and if the parameter values deviate from the set upper and lower limits, the PLC automatically adjusts the interior environment of the isolation chamber to match the programmed values. For example, the levels of humidity, illumination, and ventilation can all be controlled by the PLC to adjust values back to pre-determined normal levels. If the PLC is unable to bring the parameter back into a normal range, the digital output module in the PLC lights an alarm bulb and sounds a buzzer, as all alarms are indicated with both a warning light and sound. The control system controls alarms for a variety of isolation chamber problems including major equipment error alarms (such as the air blower or HEPA), major parameter alarms when values are out of the desirable ranges (such as temperature or humidity), an alarm when switching to the uninterruptible power supply (UPS)/emergency power supply (EPS), and alarms for experimental failure or error (such as negative-pressure breeding mode procedures or pressure test procedures). To control pressure, a micro-differential pressure sensor is mounted on the side of the first exhaust filter. The analog module in the PLC compares the values of program settings with the values collected from the micro-differential pressure sensor, and automatically conducts proportional-integral-differential (PID) regulation. The adjusted output values are used to control blower velocity through the output module of the PLC, regulating the isolation chamber internal pressure. If a plug or leak occurs, the microdifferential pressure sensor transmits the detected signal to the PLC. If the detected values are beyond the scope of the pre-loaded high and low pressure settings, the exhaust electronic control ball valve and inlet air electronic control ball valve automatically shut down to maintain the isolation chamber as a fully-contained environment and to prevent the escape of pathogens into the outside environment. At the same time an alarm indicator light tower would start to sound an alarm, and the touch-screen would display information on the alarm. The alarm information would then be transmitted to the lab server through the industrial Ethernet module in the PLC. The alarm message displayed is recorded onto the lab server for analysis at a later date. The blower rotation rate and frequency are automatically controlled by the PLC system to ensure that the airflow velocity, air exchange rate, and atmospheric pressure match the set values. If one exhaust blower fails, the PLC system responds by switching to another backup exhaust blower to ensure ventilation safety and the internal negative-pressure state of the isolation chamber. The cold light lamp regulator is controlled by the PLC digital output module to automatically adjust the illumination time (12 h/12 h or 10 h/14 h light/dark cycle) according to animal behavior. The illumination time and intensity (5-10, 15-20, and 100-200 lx) can be set from the touch screen by the operator and automatically executed. The lamps also can be switched on or off manually to meet different lighting requirements during an experimental procedure. Temperature and humidity sensors are fitted within the isolation chamber. The isolation chamber internal temperature is maintained at 18-22 °C, and relative humidity is kept within 40%-70%. The isolation chamber internal temperature and humidity readings are collected by the PLC analog module, visualized as the project value (actual values of temperature and humidity), and automatically displayed and recorded on the touch screen. The values are also recorded on the lab server. To operate the isolation chamber, the main power on the control cabinet is first turned on. The current operating parameters are displayed on the touch screen interface and can then be adjusted by the human operator by following the interface prompts. The isolation chamber should be monitored for 48 h to ensure that it is running normally. This includes supplying filtered air to the isolation chamber and ensuring that the exhaust air is cleaned by the double in-line HEPA filters and passed through the exhaust air system into the open air. Fresh air exchanges should be conducted at a rate of about 36 times per hour. Following 48 h of monitoring, the inside temperature is 22-23 °C, the relative humidity is 60%, the working negative pressure in the isolation chamber is adjusted to −100 Pa with respect to the laboratory. Appropriate anesthesia should be used on healthy medium-sized animals to pass them through the quarantine system and into the isolation chamber via the DPTE system. The animal experiment should be performed according to bio-safety operation standard procedures while the animal is still under the appropriate anesthesia. If the gloves are removed during the operation, a low pressure audible/visual alarm system is activated, and a minimum velocity value of 0.65 m/s in the center of the glove port is maintained. If a glove is damaged by a needle, the blowers are capable of maintaining the isolation chamber pressure at −100 Pa, but the alarm system would not be activated because the micro-differential pressure sensor is not sensitive enough for a leak of this size. Proper procedure dictates that the small hole be labeled by the operator and a new glove exchanged. All of the feeds, experimental material, waste, feces, and other materials can be transferred into or out of the isolation chamber by the DPTE system. There are several isolation chambers in one room, and the autoclave does not connect with any of them. Instead, DPTE β canisters are filled with items and are then sterilized in the autoclave. After sterilization, the sterile items are removed after opening the β door with specialized tools. The sterile items are then sent to a centralized disposal center for medical waste. Following the completion of studies with single animals, each animal is treated as dictated by bio-safety operation standards and general animal welfare. The cadavers are autoclaved in β canisters and sent to animal carcass disposal sites. The isolation chamber is connected to the peracetic acid sterilizer to sterilize its interior. The newly designed multi-functional isolation chamber system reported here achieves multiple technical improvements: (1) By using variable frequency drive blowers as the inlet air and extraction blowers, adjustments for rapid pressure changes (e.g., insertion of gloves) can occur automatically without breaching the inert atmosphere. The extraction blowers contain an integrated backup system with one blower running at full strength and another on standby to act as a backup. Negative or positive pressure state is kept at a stable and safe level through the automatic control system. The pressure is adjusted depending on different requirements for different animals and/or experimental conditions. (2) A DPTE 270 α door mounted on one side wall of the isolation chamber allows for the transfer of experimental materials. Another DPTE 270 α door is mounted on the bottom of the isolation chamber near the disinfection pool with a slightly tilted bottom. Animal excrement can be discharged to the DPTE container via the flexible silicone diversion tube. The excrement collection package is then minified. An experimental sample can also be transferred to the bio-safety cabinet through a shorter DPTE container. (3) The control cabinet installation is comprised mainly of the PLC, electric element (which includes the voltage transformer, secure alternating current contactor, circuit breaker, electric cable, indicating lamp, and button), printer port (for data output), and industrial Ethernet interface (which allows for the remote data control of multiple isolation chambers by the WINCC 6.0 program system). Automatic control and monitoring of the isolation chamber and sterilization system are achieved by the exchange of data between the touch screen and control cabinet through the industrial bus Profibus DP to meet different laboratory animal research project parameter requirements such as pressure, humidity, temperature, illumination, and disinfection. A human operator can set the isolation chamber environment parameters according to the requirements of the infectious animals or for cleaning animals, allowing for the acquisition of adequate and authentic data. (4) Animal welfare is ensured by installing adjustable illumination lamps, a rotatable camera, a flat television, a micro-differential pressure sensor, and a temperature humidity sensor to maintain comfortable living conditions for the animal. In the future we will report on the use of this isolation chamber in specific experiments to further demonstrate its versatility and usefulness.

Proteomics: Challenges, Techniques and Possibilities to Overcome Biological Sample Complexity

Proteomics is the large-scale study of the structure and function of proteins in complex biological sample. Such an approach has the potential value to understand the complex nature of the organism. Current proteomic tools allow large-scale, high-throughput analyses for the detection, identification, and functional investigation of proteome. Advances in protein fractionation and labeling techniques have improved protein identification to include the least abundant proteins. In addition, proteomics has been complemented by the analysis of posttranslational modifications and techniques for the quantitative comparison of different proteomes. However, the major limitation of proteomic investigations remains the complexity of biological structures and physiological processes, rendering the path of exploration paved with various difficulties and pitfalls. The quantity of data that is acquired with new techniques places new challenges on data processing and analysis. This article provides a brief overview of currently available proteomic techniques and their applications, followed by detailed description of advantages and technical challenges. Some solutions to circumvent technical difficulties are proposed.

The term proteomics describes the study and characterization of complete set of proteins present in a cell, organ, or organism at a given time [1] . In general, proteomic approaches can be used (a) for proteome profiling, (b) for comparative expression analysis of two or more protein samples, (c) for the localization and identification of posttranslational modifications, and (d) for the study of protein-protein interactions. The human genome harbors 26000-31000 protein encoding genes [2] ; whereas the total number of human protein products, including splice variants and essential posttranslational modifications (PTMs), has been estimated to be close to one million [3, 4] . It is evident that most of the functional information on the genes resides in the proteome, which is the sum of multiple dynamic processes that include protein phosphorylation, protein trafficking, localization, and protein-protein interactions [5] . Moreover, the proteomes of mammalian cells, tissues, and body fluids are complex and display a wide dynamic range of proteins concentration one cell can contain between one and more than 100000 copies of a single protein [6] . In spite of new technologies, analysis of complex biological mixtures, ability to quantify separated protein species, sufficient sensitivity for proteins of low abundance, quantification over a wide dynamic range, ability to analyze protein complexes, and high throughput applications is not yet fulfilled [7] . Biomarker discovery remains a very challenging task due to the complexity of the samples (e.g., serum, other bodily fluids, or tissues) and the wide dynamic range of protein concentrations [8] . Most of the serum biomarker studies performed to date seem to have converged on a set of proteins that are repeatedly identified in many studies and that represent only a small fraction of the entire blood proteome [9] . Processing and analysis of proteomics data is indeed a very complex multistep process [10, 11] . The consistent and transparent analysis of LC/MS and LC-MS/MS data requires multiple stages [12] , and this process remains the main bottleneck for many larger proteomics studies. To overcome these issues, effective sample preparation (to reduce complexity and to enrich for lower abundance components while depleting the most abundant ones), state-of-the-art mass spectrometry instrumentation, and extensive data processing and data analysis are required. A wide range of proteomic approaches are available such as gel-based applications include one-dimensional and twodimensional polyacrylamide gel electrophoresis [13, 14] , and gel-free high throughput screening technologies are equally available, including multidimensional protein identification technology [15] , isotope-coded affinity tag ICAT [16] ; SILAC [17] ; isobaric tagging for relative and absolute quantitation (iTRAQ) [18] . Shotgun proteomics [19] and 2DE DIGE [20] as well as protein microarrays [21, 22] are applied to obtain overviews of protein expression in tissues, cells, and organelles. Large-scale western blot assays [23] , multiple reaction monitoring assay (MRM) [24] , and label-free quantification of high mass resolution LC-MS data [25] are being explored for high throughput analysis. Many different bioinformatics tools have been developed to aid research in this field such as optimizing the storage and accessibility of proteomic data or statistically ascertaining the significance of protein identifications made from a single peptide match [26] . In this review we attempt to provide a overview of the major developments in the field of proteomics, some success stories as well as challenges that are currently being faced. About 20-30% of all genes in an organism encode integral membrane proteins, which are involved in numerous cellular processes [27] . Membrane proteins constitute 30% of the typical proteome, yet their propensity to aggregate and precipitate in solution confounds their analysis [28] . The target residues for tryptic cleavage (i.e., lysine and arginine) are mainly absent in transmembrane helices and preferentially found in the hydrophilic part of these lipid bilayer-incorporated proteins. Because of the protein aggregation step of IEF, 2DE is unsuitable for the separation of integral membrane proteins and is limited to detection of membraneassociated proteins and membrane proteins with a low hydrophobicity [29] . Membrane solubilization methods have been deployed to analyze enriched membrane fractions and address the solubility issue by using detergents [30] , organic solvents [31] , and organic acids [32] compatible with subsequent proteolytic digestion/chemical cleavage, separation and analysis by LC/MS. In this approach, (1) an enriched yeast membrane fraction is solubilized with 90% formic acid in the presence of cyanogens bromide. The concentrated organic acid provides the solubilization agent, and cyanogen bromide, functional under acidic conditions, allows many embedded membrane proteins to be cleaved, (2) a membrane-enriched microsomal fraction is solubilized by boiling in 0.5% SDS and, following isotope-coded affinity tag (ICAT) labeling, is diluted to reduce the concentration of SDS, and (3) by using an enriched membrane sample, the proteins are thermally denatured and sonicated in 60% organic solvent (methanol) in the presence of trypsin. The resultant peptide mixture is then analyzed by LC/MS. All three of these methods are effective and optimize the identifications of membrane proteins. Another method using high pH and protenase K is optimized specifically for the global analysis of both membrane and soluble proteins [33] . High pH favors the formation of membrane sheets, while proteinase K cleaves exposed hydrophilic domains of membrane proteins. Commercially available nonionic detergents, dodecyl maltoside, and decaethylene glycol mono hexadecyl are proved most efficient membrane protein solubilizers [34] . Another more successful approach to isolate membrane proteins relies on cell surface labeling in combination with high resolution two-dimensional (2D) LC-MS/MS [35] . In addition, improved analytical tools should be developed, that is, multidimensional liquid chromatography of peptide mixtures generated from membrane proteins, nanoflow chromatographic techniques for hydrophobic transmembrane peptides, and native electrophoresis of membrane protein complexes, which, in combination with mass spectrometry, should lead to the identification of the majority of proteins in the membrane proteome of simple microorganisms. It is important to quantify not only the identified membrane proteins but also to determine the levels of interacting partners. Subcellular fractionation techniques that employ a combination of centrifugation steps are a common choice for preparing plasma membrane-(PM-) enriched fractions including detergent-resistant membrane fractions, commonly known as lipid rafts. These methods can offer a significant improvement in specificity for PM proteins over approaches that do not perform any subcellular fractionation, but rather use whole-cell or tissue preparations [36] . Chemical-tagging methods [37] have been a more applied technique used to enrich for PM proteins and are often used in conjunction with physical separation strategies. This method allows for a specific class of protein or modification of interest to be physically separated from other nontagged proteins. Importantly, when chemical tags are attached to the extracellular domain of PM proteins on intact cells, they offer an unrivaled specificity for PM proteins, because they offer a manner to distinguish true PM proteins from intracellular contaminants. Cell-surface biotinylation, the covalent attachment of a biotin tag to the extracellular domain of PM proteins, is also a popular choice [38] [39] [40] . Serum is a complex body fluid, containing a large diversity of proteins. More than 10000 different proteins are present in the human serum and many of them are secreted or shed by cells during different physiology or pathology processes [41] . Serum is expected to be an excellent source of protein biomarkers because it circulates through, or comes in contact all tissues. Consequently, serum proteomics has raised great expectations for the discovery of biomarkers to improve diagnosis or classification of a wide range of diseases, including cancers [42] . However, serum has been termed as the most complex human proteome [43] with considerable differences in the concentrations of individual proteins, ranging from several milligrams to less than one pictogram per milliliter [44] . The analytical challenge for biomarker discovery arises from the high variability in the concentration and state of modification of some human plasma proteins between different individuals [45] . Albumin is a protein of very high abundance in serum (35-50 mg/mL) that would be a prime candidate for complete selective removal prior to performing a proteomic analysis of lower abundance proteins. Thus, removal of albumin from serum may also result in the specific removal of low abundance cytokines, peptide hormones, and lipoproteins of interest. Immunoglobulins, and antibodies are also abundant proteins in serum that function by recognizing "foreign" antigens in blood and initiating their destruction [46] . The presence of higher abundance proteins interferes with the identification and quantification of lower abundance proteins (lower than ng/mL in serum). Complexity and dynamic range of protein concentrations can be addressed with a combination of prefractionation techniques that deplete highly abundant proteins and fractionate. Heparin chromatography coupled with protein G appears to be an efficient and economical strategy to pretreat serum for serum proteomics [47] . Protein prefractionation by immunodepletion and reversed-phase separation of the depleted plasma on mRP-C18 column provide methods compatible with LC-MS-based analysis. A polyclonal antibody-based system to rapidly deplete multiple high abundant proteins in serum, plasma, CSF, and other biological fluids. Individual antibody materials are mixed in selected percentages and packed into a column format. Albumin can be removed by immunoaffinity columns [48] , isoelectric trapping [49] , dye-ligand chromatography [50] , and peptide affinity chromatography [51] . Another approach involves the removal of IgG by affinity chromatography using immobilized protein A or protein G [52] . A recently developed depletion method that mixes 6 high-specificity polyclonal antibodies (MARS) to remove the top 6 proteins in a single purification step is commercially available [53] . Human-14 multiple affinity removal column depletes the top 14 abundant proteins from human serum, plasma, CSF, and other biological fluids. To address 2D limitations several types of mass spectrometry, in conjunction with various separation and analysis methods, are increasingly being adopted for proteomic measurements [54] . In contrast, 2D-PAGE analysis, SELDI-TOF MS is a rather new method which is especially valuable for the identification of serum-derived biomarkers [55] . This method is based on ProteinChip Arrays which carry various chromatographic properties, such as anion exchange, cation exchange, and hydrophilic or hydrophobic surfaces [56] . For the analysis of serum, only 5-10 μL of serum sample is applied to these surfaces; after washing off unbound material, the protein fingerprint can be determined and visualized by time-of-flight mass spectrometry. The advantages of this method are the low amount of sample necessary for analysis, its speed, and high throughput capability. Many different groups have used this method and related methods based on prefractionation of serum proteins by beads and subsequent MALDI analysis for the identification of biomarkers in serum, urine, pancreatic juice, and other biological fluids [57] . The necessity of this removal or separation is also illustrated that many proteins found useful as biomarkers [58] . Different fractionation steps (such as electrophoresis, SELDI, and liquid chromatography) have been developed to reduce the complexity of serum proteome and to allow the detection and the identification of single proteins [59] . 2DE and MALDI MS had applied to identify candidate biomarkers at early and late stages of lung cancer disease. This method identified 46 proteins in tumor bearing mice this included disease regulated expression of orosomucoid-8, a-2-macroglobulin, apolipoprotein-A1, apolipoprotein-C3, glutathione peroxidase-3, plasma retinol-binding protein, and transthyretin [60] . Recently 1065 proteins were identified by stable isotope labeled proteome (SILAP) standard coupled with extensive multidimensional separation with tandem mass spectrometry of which 121 proteins were present at 1.5-fold or greater concentrations in the sera of patients with pancreatic cancer [61] . Specimen collection (Blood, serum, plasma samples) is an integral component of clinical research. Access to high-quality specimens, collected and handled in standardized ways that minimize potential bias or confounding factors, is key to the "bench to bedside" aim of translational research [62] . Variables that may impact analytic outcomes include (1) the type of additive in the blood collection tubes; (2) sample processing times or temperatures; (3) hemolysis of the sample; (4) sample storage parameters; (5) the number of freeze-thaw cycles [63, 64] . The key variable in any analysis is that the case and control samples are handled in the exact same manner throughout the entire analytical process from study design and collection of samples to data analysis [63, 65] . These types of differences between samples could have a significant impact on the stability of proteins or other molecules of interest in the specimens. Small differences in the processing or handling of a specimen can have dramatic effects in analytical reliability and reproducibility, especially when multiplex methods are used. A representative working group, standard operating procedures internal working group, comprised of members from across early detection research network should be formed to develop standard operating procedures (SOPs) for various types of specimens collected and managed for biomarker discovery and validation work. Limitations of Two-Dimensional Electrophoresis. Figure 1 gives the general work flow in proteomics and Table 1 addresses their strengths and limitations. Two-dimensional electrophoresis (2DE) was developed two decades before the term proteomics was coined [66, 67] . The 2DE entails the separation of complex protein mixtures by molecular charge in the first dimension and by mass in the second dimension. 2DE analysis provides several types of information about the hundreds of proteins investigated simultaneously, including molecular weight, pI and quantity, as well as possible posttranslational modifications. 2DE is extensively used but mostly for qualitative experiments and this method falls short in its reproducibility, inability to detect low abundant and hydrophobic proteins, low sensitivity in identifying proteins with pH values too low (pH < 3) or too high (pH > 10) and molecular masses too small (Mr < 10 kD) or too large (Mr > 150 kD) [2] [3] [4] [5] . Poor separations of basic proteins due to "streaking" of spots and membrane proteins resolution [68] are limiting factors in 2DE. However, 2DE is the only technique that can be routinely applied for parallel quantitative expression profiling of complex protein mixtures such as whole cell and tissue lysates [69] and most widely used method for efficiently separating proteins, their variants and modifications (up to 15000 proteins). There are two ways to study posttranslational modifications by means of 2DE. First, posttranslational modifications that alter the molecular weight and or pI of a protein are reflected in a shift in location of the corresponding protein spot on the proteomic pattern. Second, in combination with Western blotting, antibodies specific for posttranslational modifications can reveal spots on 2DE patterns containing proteins with these modifications [70] . Protein extraction and solubilization are key steps for proteomic analysis using 2DE, highly hydrophobic proteins tend to precipitate during isoelectro focusing (IEF), low copy number and the insolubility of transmembrane proteins renders quantitative analysis of these peptides and polypeptides are very challenging [71] . In order to enhance protein extraction and solubilization, different treatments and conditions are necessary to efficiently solubilise different types of protein extracts [72, 73] . The major challenge for protein visualization in 2DE is the compatibility of sensitive protein staining methods with mass spectrometric analysis. Therefore, several fluorescent staining methods have been developed for the visualization of 2DE patterns, including sypro stainings and Cy-dyes [74] . Although sypro ruby [75] and silver staining [76, 77] have a comparable sensitivity, sypro ruby staining allows much higher reproducibility, a significantly wider dynamic range and less false-positive staining. In addition, sypro ruby allows for the detection of lipoproteins, glycoproteins, metalloproteins, calcium-binding proteins, fibrillar proteins, and low molecular weight proteins that are less "stainable" using other methods. Finally, a large number of protein spots on 2DE patterns contain several proteins with a similar pI. A pH gradient with a narrow range allows zooming into different proteins with the same molecular weight. Increased separation distance 40 × 40 cm gels using CA-IEF [78] could increase the proteome coverage up to 5000 proteins. Use of overlapping narrow range IPGs "Zoom" gels and increase in separation area could yield better membrane protein separation [79] . This technology, however, is biased against certain classes of proteins including low abundance and hydrophobic proteins. Proteins can also be fluorescently labelled with Cy2, Cy3, or Cy5 prior to 2DE [80] . CyDyes are cyanine dyes carrying an N-hydroxysuccinimidyl ester reactive group that covalently binds the e-amino group of lysine residues in proteins. During DIGE [81] , proteins in each of up to 3 samples can be labelled with one of these fluorescent dyes, and the differentially labelled samples can be mixed and loaded together on one single gel, allowing the quantitative comparative analysis of three samples using a single gel ( Figure 2 ). The DIGE technique has exhibited higher sensitivity as well as linearity, eliminated postelectrophoretic processing (fixing and destaining) steps and enhanced reproducibility by directly comparing samples under similar electrophoretic conditions [81, 82] . The resulting images are then analyzed by software such as De-Cyder which are specifically designed for 2D-DIGE analysis [83] . The major advantages of 2D-DIGE are the high sensitivity and linearity of its dyes, its straightforward protocol, as well as its significant reduction of intergel variability, increasing the possibility to unambiguously identify biological variability, and reducing bias from experimental variation. Moreover, the use of a pooled internal standard, loaded together with the control and experimental samples, increases quantification accuracy and statistical confidence [84] . The DIGE technique has dramatically improved the reproducibility, sensitivity, and accuracy of quantitation; however, its labeling chemistry has some limitations; proteins without lysine cannot be labeled, and they require special equipment for visualization, and fluorophores are very expensive [83, 85] . Tag (ICAT) . Gel-free, or MS based, proteomics techniques are emerging as the methods of choice for quantitatively comparing proteins levels among biological proteomes, since they are more sensitive and reproducible than two-dimensional gel-based methods. ICAT is one of the most employed chemical isotope labeling methods and the first quantitative proteomic method to be based solely on using MS [86, 87] . Each ICAT reagent consists of three essential groups: a thiol-reactive group, an isotope-coded light or heavy linker, and a biotin segment to facilitate peptide enrichment. In an ICAT experiment, protein samples are first labeled with either light or heavy ICAT reagents on cysteine thiols. The mixtures of labeled proteins are then digested by trypsin and separated through a multistep chromatographic separation procedure. Peptides are identified with tandem MS, and the relative quantifications of peptides are inferred from the integrated LC peak areas of the heavy and light versions of the ICAT-labeled peptides [88] . The ICAT concept has been widely used after its introduction [89] [90] [91] . Different software programs were developed to analyze ICAT labeled MS data (e.g., proICAT from Applied Biosystems, spectrum Mill from Agilent Technologies, and Sashimi from the Institute of System Biology [92] ). ICAT is extremely helpful to detect peptides with low expression levels, which is one of the bottleneck issues in analytic protein techniques [93, 94] . However, major limitations of this technique include selective detection of proteins with high cysteine content and difficulties in the detection of acidic proteins [95, 96] . The methods for direct comparison of DIGE and ICAT for the identification and quantification of proteins in complex biological mixtures are also being considered [97] . While the ICAT reagent only interacts with the free sulfhydryl of homocysteine and 8% protein is noncysteine, the SILAC has emerged as a valuable proteomic technique [98] which becomes more common for cell types and have been applied in many fields [99] [100] [101] . The SILAC technique can be effectively expanded to compare the differential expression levels of tissue proteome at different pathological states, which allows to identify new candidate biomarkers [102] . Compared with the ICAT, a popular in vitro labeling, SILAC as an example of in vivo coding requires no chemical manipulation, and there is very little chemical difference between the isotopically labeled amino acid and its naturally occurring counterpart [103] . In addition, the amount of labeled proteins requires for analysis using SILAC technique is far less than that with ICAT. Therefore, the SILAC-based method has broadly applied in many areas of cell biology and proteomics. Except that the SILAC-based quantitative method is powerful in comparative/differential proteomics, it has been widely used in analyzing protein posttranslational modification, such as protein phosphorylation, detection of protein-protein or peptide-protein interactions and investigating signal transduction pathways [104, 105] .Though there are numerous advantages for using SILAC-based methods compared to chemical labeling, a major drawback of SILAC is that it cannot be applied to tissue protein analysis directly. To overcome this shortcoming, SILAC has been successfully applied to tissue proteome based on 15 N isotope labeling [106] . Microorganisms such as malaria parasite can be labeled with isoleucine [107] . Latterly the culturederived isotope tags (CDITs) method was developed as an alternative quantitative approach for studying the proteome of mammalian tissues based on the application of SILAC [108] . 18O Stable Isotope Labeling. Differential 16O/18O coding relies on the 18O exchange that takes place at the Cterminal carboxyl group of proteolytic fragments, where two 16O atoms are typically replaced by two 18O atoms by enzyme-catalyzed oxygen exchange in the presence of H218O [109] . The resulting mass shift between differentially labeled peptide ions permits identification, characterization, and quantitation of proteins from which the peptides are proteolytically generated. In contrast to ICAT, 18O labeling does not favor peptides containing certain amino acids (e.g., cysteine), nor does it require an additional affinity step to enrich for these peptides [110] . Unlike iTRAQ, 16O/18O labeling does not require a specific MS platform nor does it depend on fragmentation spectra (MS2) for quantitative peptide measurements. It is amenable to the labeling of human specimens (e.g., plasma, serum, tissues), which represents a limitation of metabolic labeling approaches (e.g., SILAC). Taken together, recent advancements in the homogeneity of 18O incorporation, improvements made on algorithms employed for calculating 16O/18O ratios and the inherent simplicity of this technique should result in increased use of 18O labeling [111] . In general, 18O labeling suffers from two potential drawbacks, inhomogeneous 18O incorporation and inability to compare multiple samples within a single experiment. A dual 18O labeling using a non-gel-based platform has been developed to overcome the major problems of existing proteolytic 18O labeling methods [112] . (iTRAQ). The iTRAQ reagent is well known for relative and absolute quantitation of proteins. The iTRAQ technology offers several advantages, which include the ability to multiplex several samples, quantification, simplified analysis and increased analytical precision and accuracy [113] [114] [115] . The interest of this multiplexing reagent is that 4 or 8 analysis samples [116] can be quantified simultaneously. In this technique, the introduction of stable isotopes using iTRAQ reagents occurs on the level of proteolytic peptides ( Figure 3 ). This technology uses an NHS ester derivative to modify primary amino groups by linking a mass balance group (carbonyl group) and a reporter group (based on N-methylpiperazine) to proteolytic peptides via the formation of an amide bond [117] . Due to the isobaric mass design of the iTRAQ reagents, differentially labelled peptides appear as a single peak in MS scans, reducing the probability of peak overlapping. When iTRAQ-tagged peptides are subjected to MS/MS analysis, the mass balancing carbonyl moiety is released as a neutral fragment, liberating the isotope-encoded reporter ions which provides relative quantitative information on proteins. An inherent drawback of the reported iTRAQ technology is due to the enzymatic digestion of proteins prior to labelling, which artificially increases sample complexity and this approach needs a powerful multidimensional fractionation method of peptides before MS identification. Prefractionation of proteins based on electrokinetic methodologies in free solution essentially relaying on the isoeletric focusing (IEF) has gained wide acceptance. Many commercial devices are now constructed to take the advantage of this principle ( Table 2 ). Reproducible fractionation steps will break down the sample complexicity while concentrating low abundant species, resulting in more confident protein identifications and quantification by 2D gels, mass spectrometry, and protein arrays. A good example of a innovation is liquid-phase isoelectric focusing (IEF) as a prefractionation tool before the first dimension of 2D gel electrophoresis [118, 119] . For more consistent pI separation, the Zoom IEF fractionator [120] and multicompartment electrolyser (MCE) [121] are being used to prefractionate the proteins. The fractionated samples can be directly applied on standard narrow range IPG strips for 2D electrophoresis. This allows at least 10000 to 15000 separate proteins to be analyzed, including proteins of very low abundance. IEF, a highresolution electrophoresis technique, has been widely used in shotgun proteomic experiments [122] . IEF runs in a buffer-free solution containing carrier ampholytes or in immobilized pH gradient (IPG) gels. The use of IPG-IEF for the separation of complex peptide mixtures has been applied to the analysis of plasma and amniotic fluid [123, 124] as well as to bacterial material [125] . The IPG gel strip is divided into small sections for extraction and cleaning up of the peptides. This technique recovers the sample from the liquid phase and was demonstrated to be of great interest in shotgun proteomics [126] . IEF is not only a high resolution and high capacity separation method for peptides, it also provides additional physicochemical information like their isoelectric point [127, 128] . The pI value provided is used as an independent validating and filtering tool during database search for MS/MS peptide sequence identification [129] . The recent introduction of commercially available OFFGEL fractionator system by Agilent Technologies provides an efficient and reproducible separation technique [130] . This separation is based on immobilized pH gradient (IPG) strips and permits to separate peptides and proteins according to their isoelectric point (pI) but is realized in solution [131] . Its micropreparative scale provides fraction volumes large enough to perform subsequent analyses as reverse phase (RP)-liquid chromatography (LC)-MALDI MS/MS. The combined use of iTRAQ labeling and OFFGEL fractionation methods for the proteomic study of complex sample is also being considered [132, 133] . In this procedure, a large well is used to separate the sample by PAGE and lanes are created on the membrane containing immobilized protein with the use of a manifold [134] . Compatible combinations of primary antibodies are predetermined, with the criterion of being able to identify proteins that do not comigrate. Different combinations of primary antibodies are added to each well, with appropriate dilutions of each primary antibody so that expressed proteins are detected in a single condition. The scalability of the system depends on defining suitable combinations of primary antibodies, with up to 1000 antibodies in 200 lanes being used in the largest screens. Detection software is used to identify proteins based on their expected and observed gel mobility. Unlike 2D PAGE and HPLC-MS/MS, large-scale western blotting only identifies proteins for which antibodies are already available. While this is not an appropriate screen for identifying uncharacterized proteins, it greatly simplifies the verification and functional analyses of proteins that are detected. In addition, this approach is highly flexible, and can be focused to particular sets of proteins or protein function, such as cell signaling molecules. Importantly, the foundation of this approach is the large amount of data on individual antibodies, which are already available and characterized in the literature [135] . Another approach to analyse proteomes without gels is "shotgun" analysis using MudPIT [136] . In the MudPIT approach, protein samples are subject to sequencespecific enzymatic digestion, usually with trypsin and endoproteinase lysC, and the resultant peptide mixtures are separated by strong cation exchange (SCX) and reversed phase (RP) high performance liquid chromatography (HPLC) [137, 138] . Peptides from the RP column enter the mass spectrometer and MS data is used to search the protein databases [138] . The MudPIT technique generates an exhaustive list of proteins present in a particular protein sample, it is fast and sensitive with good reproducibility however, it lacks the ability to provide quantitative information [139] [140] [141] . A combination of HPLC, liquid phase isoelectric focusing, and capillary electrophoresis provides other multimodular options for the separation of complex protein mixtures [142] . High throughput production of human proteins using different methods is being developed to make protein array approach more practical. Recently simple and efficient production of human proteins using the versatile gateway vector system has been developed [143] . In this approach, protein expression system is applied to the in vitro expression of 13364 human proteins and assessed their biological activity in two functional categories and developed "human protein factory" infrastructure which includes the resources and expression technology for in vitro proteome research. In another approach, DNA array to protein array (DAPA) is utilized, which allows the "printing" of replicate protein arrays directly from a DNA array template using cellfree protein synthesis [144] . Based on the nucleic acid programmable protein array (NAPPA) concept, high-density self-assembling protein microarray is developed to display thousands of proteins that are produced and captured in situ from immobilized cDNA templates [145] . This method will enable various experimental approaches to study protein function in high throughput. The adventage of protein-based microarrays allows the global observation of biochemical activities on an unprecedented scale, where hundreds or thousands of proteins can be simultaneously screened for protein-protein, proteinnucleic acid, and protein-small molecule interactions, as well as posttranslational modifications [146, 147] . The microarray format provides a robust and convenient platform for the simultaneous analysis of thousands of individual protein samples, facilitating the design of sophisticated and reproducible biochemical experiments under highly specific conditions [148] . The principal challenges in protein array development are 3-fold: (1) creation of a comprehensive expression clone library; (2) high-throughput protein production, including expression, isolation, and purification; (3) adaptation of DNA microarray technology to accommodate protein substrates [149] . Functional protein microarrays differ from analytical arrays in that functional protein arrays are composed of arrays containing fulllength functional proteins or protein domains (Figure 4) . These protein chips are used to study the biochemical activities of an entire proteome in a single experiment. They are used to study numerous protein interactions, such as protein-protein, protein-DNA, protein-RNA, proteinphospholipid, and protein-small molecule interactions [150, 151] . Companies have introduced protein arrays aimed not only at proteomic analysis but also functional analyses of proteins (e.g., Biacore AB, Ciphergen Biosystems Inc., Phylos Inc.). Affinity proteomics aim to produce antibodies to every protein expressed by the human genome and these will be characterized against purified antigens and tested on tissue arrays to collect information about their specificity for tissue antigens [152] . Companies are focused to produce various binding partners, for example, affibodies, monoclonal antibodies, and their fragments [153] . Protein chips will likely be the next major manifestation of the revolution in proteomics and offer another solution to analyze low abundant proteins and have the potential for high throughput applications to identify biomarkers [154] . Protein chips differ from previously described methods; whereas screening by 2DE or LC MS/MS can potentially detect any protein, and protein chips can only provide data on set of proteins selected by the investigator [155] . The development and application of high throughput, multiplex immunoassays that measure hundreds of known proteins in complex biological matrices, is becoming a significant tool for quantitative proteomics studies, diagnostic discovery, and biomarker-assisted drug development. Two broad categories of antibody microarray experimental formats have been developed [156] , direct labelling, single antibody experiments [157] , dual antibody, sandwich immunoassays are described [158, 159] . In the direct labelling method, all proteins in a complex mixture are tagged, providing a means for detecting bound proteins following incubation on an antibody microarray. In the sandwich immunoassay format, proteins captured on an antibody microarray are detected by a cocktail of detection antibodies, each antibody matched to one of the spotted antibodies. In addition, a variety of microarray substrates have been described, including nylon membranes, plastic microwells, planar glass slides, gel-based arrays and beads in suspension arrays. Much effort has been expended in optimizing antibody attachment to the microarray substrate. Finally, various signal generation and signal enhancement strategies have been employed in antibody arrays, including colorimetry, radioactivity, fluorescence, chemiluminescence, quantum dots and other nanoparticles, enzyme-linked assays, resonance light scattering, tyramide signal amplification, and rolling circle amplification. Each of these formats and procedures has distinct advantages and disadvantages, relating broadly to sensitivity, specificity, dynamic range, multiplexing capability, precision, throughput, and ease of use. In general, multiplexed microarray immunoassays are ambient analyte assays [160] . Given the heterogeneity of antibody array formats and procedures currently in use in proteomics studies, and the absence of a "gold standard," there exists an urgent need for development and adoption of standards that permit platform comparisons and benchmarking. Regardless of the choice of a given proteomic separation technique, gel-based or gel-free, a mass spectrometer is always the primary tool for protein identification. During the last decade, significant improvements have been made in the application of MS for the determination of protein sequences [161] . Mass spectrometers consist of an ion source, the mass analyzer, and an ion detection system. Analysis of proteins by MS occurs in three major steps (a) protein ionization and generation of gas-phase ions, (b) separation of ions according to their mass to charge ratio, and (c) detection of ions [162] . In gel-free approaches such as ICAT and MudPIT, samples are directly analyzed by MS whereas, in gel-based proteomics (2DE and 2D-DIGE), the protein spots are first excised from the gel and then digested with trypsin. The resulting peptides are then separated by LC or directly analyzed by MS. The experimentally derived peptide masses are correlated with the peptide fingerprints of known proteins in the databases using search engines (e.g., Mascot, Sequest). There are two main ionization sources which include matrix assisted laser desorption/ionization (MALDI) and electrospray ionization (ESI) and four major mass analyzers, which are time-of-flight (TOF), ion trap, quadrupole, and fourier transform ion cyclotron (FTIC) which are currently in use for protein identification and characterization [163] . A combination of different mass analyzers in tandem such as quadrupole-TOF and quadrupole-ion trap has combined the individual strengths of different types of mass analyzers and greatly improved their capabilities for proteome analysis [162] . Simple mass spectrometers such as MALDI-TOF are used for only measurement of mass, whereas tandem mass spectrometers are used for amino acid sequence determination [164] . In MALDI the sample of interest is crystallized with the matrix on a metal surface and a laser ion source causes excitation of matrix along with the analyte ions, which are then released into the gas phase. MALDI measures the mass of peptides derived from a trypsinized parent protein and generates a list of experimental peptide masses, often referred to as "mass fingerprints" [165, 166] . In ESI, the analyte is ionized from a solution and transferred into the gas phase by generating a fine spray from a high voltage needle which results in multiple charging of the analyte and generation of multiple consecutive ions. Tandem mass spectrometry or MS/MS is performed by combining two different MS separation principles. In tandem MS, individual trypsin-digested peptides are fragmented after a liquid phase separation. Tandem MS instruments such as triple quadrupole, quadrupole ion trap, fourier transform ion-cyclotron resonance, or quadrupole time-of-flight are used in LC-MS/MS or nanospray experiments with electrospray ionization (ESI) to generate peptide fragment ion spectra [167] . Ion mobility spectrometry (IMS) has been utilized as a rapid gas-phase separations strategy for biomolecular ions [168, 169] . The strategy provides high sensitivity because the gas-phase dispersion of peptide ions separates features corresponding to low abundance species from interfering chemical noise [170] . Reduced spectral congestion also allows for the use of shorter experimental run times (LC separations) without sacrificing throughput; short analysis time scales are key to measuring the large numbers of samples required to determine normal protein variability prior to realizing individual plasma profiling. Additionally, mobility-dispersed ions can be fragmented and mobility linked to fragment ions without ion loss from precursor mass selection [171] . These advantages have been demonstrated in head-to-head comparisons with conventional LC-MS/MS technology using rapid (21 minutes) LC gradients [169] . Accurate mass and time (AMT) tag approach [172] addresses an analogous situation in LC-MS-based proteomics studies. In this approach, initial LC-MS/MS analyses are performed on prefractionated peptide samples in order to provide peptide sequence identifications. These experiments are relatively low throughput because the peptide prefractionation can be quite extensive and require separate LC-MS/MS analyses for each fraction. The high-throughput accurate mass and time (AMT) tag proteomic approach was utilized to characterize the proteomes for cytoplasm, cytoplasmic membrane, periplasm, and outer membrane fractions from aerobic and photosynthetic cultures of the gram-negative bacterium Rhodobacter sphaeroides 2.4.1. There has been a recent trend in proteomics toward the development and application of technologies for the targeted analysis of proteins within complex mixtures [173] . Selected reaction monitoring (SRM) is a powerful tandem mass spectrometry method that can be used to monitor target peptides within a complex protein digest [174, 175] . The specificity and sensitivity of the approach, as well as its capability to multiplex the measurement of many analytes in parallel, has made it a technology of particular promise for hypothesis driven proteomics. The use of tandem mass spectrometry data acquired on an LTQ ion trap mass spectrometer can accurately predict which fragment ions will produce the greatest signal in an SRM assay using a triple quadrupole mass spectrometer [176] . One of the biggest benefits of a targeted assay on a triple quadrupole mass spectrometer is high throughput. Using the selectivity of multiple stages of mass selection of a tandem mass spectrometer, these targeted SRM assays are the mass spectrometry equivalent of a Western blot [173] . An advantage of using targeted mass spectrometry-based assay over a traditional Western blot is that it does not rely on the creation of any immunoaffinity reagent. While its application is novel in the proteomics community, SRM has been utilized for several decades in the toxicology and pharmacokinetics disciplines [177] . Peptidebased immunofractionation methods show potential for proteome wide screening approaches but are limited by the availability of antibodies [178, 179] . The stable isotope standards with capture by antipeptide antibodies (SISCAPA) approach is based on the addition of stable isotope labeled standard peptides to the digested clinical sample followed by immunoaffinity enrichment of standard and analyte peptide by highly specific antipeptide antibodies [180, 181] . This approach enables the absolute quantification of selected diagnostic peptides from digested clinical samples down to physiologically relevant analyte concentrations (ng/mL) at high precision (10% CV) and accuracy [178, 179] . Further improvement of MRM-based biomarker quantification should be possible if whole sets of analyte peptides can be enriched by immunofractionation. Since this method relies on one specific antibody per target protein/peptide the generation of more than 10000 antibodies is necessary for proteome wide screening approaches. Novel peptide affinity enrichment strategies enabling proteome wide analyses of signature peptides may provide an important addition to future proteome workflows. Undoubtedly, the accuracy, high throughput, and robustness of MS technologies have made the characterization of entire proteomes a realistic goal [180, 181] . The major bottlenecks in proteomics research today are related to data analysis to create an environment where computer scientists and biologists and the people who collect data can work closely together, so they can develop the necessary analytical tools that will help interpret the data [182] [183] [184] . Processing and analysis of proteomics data is indeed a very complex multistep process ( Figure 5 ). The meaningful comparison, sharing, and exchange of data or analysis results obtained on different platforms or by different laboratories remain cumbersome mainly due to the lack of standards for data formats, data processing parameters, and data quality assessment. Accurate, consistent, and transparent data processing and analysis are integral and critical parts of proteomics workflows [185] . We can now generate huge amounts of data, and currently there is an enormous challenge to figure out how to actually analyze this data and generate real biological insights. The necessity of an integrated pipeline for processing and analysis of complex proteomics data sets has therefore become critical. Validation. This step consists of the assignment of MS/MS spectra to a database search using one of several engines available (e.g., Sequest, Mascot, Comet, X!tandem, etc.). One of the difficulties related to the use of sequest for peptide identifications is the lack of methods to globally evaluate the quality of data and the lack of methods to access global changes created by filtering schemes and/or database changes [186] . Most approaches are matching and scoring large sets of experimental spectra with predicted masses of fragment ions of peptide sequences derived from a protein database. Results are scored according to a scheme specific to each search engine that also depends on the database used for the search. Usually tools are linked to one specific platform or were optimized for one instrument type. The various search engines do not yield identical results as they are based on different algorithms and scoring functions, making comparison and integration of results from different studies or experiments tedious [187, 188] . Peptide identification via database searches is very computationally intensive and time-demanding. High quality data allow more effective searches due to tighter constrains, that is, tolerance on precursor ion mass and charge state assignment, which will drastically reduce the search time in case of an indexed database. In addition, accurate mass measurements of fragment ions further simplify the database searches and add confidence to the results. The association of identified peptides with their precursor proteins is a very critical and difficult step in shotgun proteomics strategies as many peptides are common to several proteins, thus leading to ambiguous protein assignments. Therefore it becomes critical to have an appropriate tool that is able to assess the validity of the protein inference and associate a probability to it. Protein Prophet database tool combines probabilities assigned to peptides identified by MS/MS to compute accurate probabilities for the proteins present [189] . 5:1921-1926, 2006. impossible. The lack of common standards and protocols has led to this situation and often resulted in duplication of efforts. Results were usually reported as a set of identified proteins (i.e., list of peptides identified and associated proteins) with minimal supporting data. Obviously the large volume of such data sets has made publication of detailed results using classical mechanisms very challenging. Sharing and exchange of data and results requires the definition of standard formats for the data at all levels (including raw mass spectrometric data, processed data, and search results) as well as a better definition (and/or standardization) of the parameters used for the data processing or the database searches. Organellar proteomics aims to describe the full complement of proteins of subcellular structures and organelles. Identification of the proteins contained in subcellular organelles has become a popular proteomics endeavor [190] . When compared with whole-cell or whole-tissue proteomes, the more focused results from subcellular proteomic studies have yielded relatively simpler datasets from which biologically relevant information can be more easily extracted [191] . Subcellular fractionation consists of two major steps, disruption of the cellular organization (homogenization) and fractionation of the homogenate to separate the different populations of organelles. Such a homogenate can then be resolved by differential centrifugation into several fractions containing mainly (1) nuclei, heavy mitochondria, cytoskeletal networks, and plasma membrane; (2) light mitochondria, lysosomes, and peroxisomes; (3) golgi apparatus, endosomes and microsomes, and endoplasmic reticulum; (4) cytosol. Each population of organelles is characterized by size, density, charge, and other properties on which the separation relies [192] . Analyzing subcellular fractions and organelles allows tracking proteins that shuttle between different compartments, for example, between the cytoplasm and nucleus. A high dynamic range of proteins can be partially achieved by fractionation of the proteome into subproteomes by applying affinity purification may allow proteomic analysis of low copy number proteins [193] . The nuclear, chloroplast, amyloplast, plasma membrane, peroxisome, endoplasmic reticulum, cell wall, and mitochondrial proteomes were successfully characterized in Arabidopsis [194] . Several groups have taken advantage of this approach to recover a higher percentage of membrane proteins from subcellular extracts using various nonionic and zwitterionic detergents or phase-partitioning methods. These efforts resulted in the successful determination of the protein complement of the thylakoid and envelope membrane systems of the chloroplast [195] . By enriching for the protein class of interest based on a particular chemical/physical characteristic(s), offer the advantage of reducing sample complexity and access to lower abundance proteins in a discoverydriven experimental approach [196] . Free flow electrophoresis (FFE) utilizes differences in electrophoretic mobility rather than density to separate cells or subcellular organelles [197] . FFE has previously been used in separating endosomes from hamster ovary cells [198] , plasma membrane from human platelets [199] , and insulin transporting vesicles in liver cells. The separation is based on the electrophoretic motility of cells or cell organelles suspended in a vertical free flowing buffer film on which an electric field is applied at a right angle to the flow direction. FFE has been a most valuable tool in the investigation of the composition of secretory vesicles and in addition, it has clarified how the membrane of plasma membrane vesicles is oriented after nitrogen disruption of human neutrophils [200] . Importantly, subcellular fractionation is a flexible and adjustable approach that may be efficiently combined not only with 2D gel electrophoresis but also with gelindependent techniques. However, they do have limitations of considerable cross-contamination with other subcellular organelles. PTMs of proteins are considered to be one of the major determinants regarding organisms complexity [201] . To date, at least more than 200 different types of PTMs have been identified of which only a few are reversible and important for the regulation of biological processes. Specific functions are usually mediated through PTMs, such as phosphorylations, acetylations, or glycosylations, which places additional demands on the sensitivity and precision of the method [202] . One of the most studied PTMs is protein phosphorylation, because it is vital for a large number of protein functions that are important to cellular processes spanning from signal transduction, cell differentiation, and development to cell cycle control and metabolism. Enzymes and receptors can be switched "on" and "off " by phosphorylation and dephosphorylation. It was estimated that 10-50% of proteins are phosphorylated. Phosphorylation often occurs on serine, threonine, and tyrosine residues in eukaryotic proteins [203] . Analysis of the entire cellular phosphoproteome has been an attractive study subject since the discovery of phosphorylation as a key regulatory mechanism of cell life. Unfortunately, phosphoproteins analysis is not straightforward for five main reasons. First, the stoichiometry of phosphorylation is generally relatively low, because only a small fraction of the available intracellular pool of a protein is phosphorylated at any given time as a result of a stimulus. Second, the phosphorylatation sites on proteins might vary, implying that any given phosphoprotein is heterogeneous (i.e., it exists in several different phosphorylated forms). Third, many of the signaling molecules, which are major targets of phosphorylation events [204] , are present at low abundance within cells and, in these cases; enrichment is a prerequisite before analysis. Fourth, most analytical techniques used for studying protein phosphorylation have a limited dynamic range, which means that although major phosphorylation sites might be located easily, and minor sites might be difficult to identify. Finally, phosphatases could dephosphorylate residues unless precautions are taken to inhibit their activity during preparation and purification steps of cell lysates. In addition, various methods for protein phosphorylation site determination have been developed, yet this task remains a technical challenge [205] . Western blot has been widely used to determine the presence of PTMs. However, this technique relies on the prior knowledge of the type and position of specific modifications and the availability of antibodies. It has low throughput and not ideal for studying highly complicated samples. Specific chemical or affinity enrichment steps are usually incorporated into the sample preparation or fractionation stages of the general scheme of proteomic studies [206, 207] . Well established methods involving the analysis of 32P-labeled phosphoproteins by Edman degradation and two-dimensional phosphopeptide mapping have proven to be powerful but not without limitations. Consequently, mass spectrometry (MS) has emerged as a reliable and sensitive method for the characterization of protein phosphorylation sites [208] and may therefore represent a method of choice for the analysis of protein phosphorylation [209] . Immobilized metal affinity chromatography (IMAC), Metal oxide affinity chromatography (MOAC), and covalent methods are all capable of selectively enriching phosphopeptides [210] . MOAC based on adsorption to TiO2 is especially attractive, but as with all techniques, loading, rinsing, and elution solutions must be carefully selected to minimize nonspecific adsorption and to maximize the detection of both monophosphorylated and multiphosphorylated species. IMAC might not provide the selectivity available with TiO2 enrichment, but with appropriate reagents, IMAC can be selective and sensitive for monophosphorylated and tetraphosphorylated peptides. However, some buffers and reagents such as EDTA are not compatible with IMAC, so HPLC purification may be needed prior to this technique [211] . When trying to isolate and identify as many phosphoproteins as possible in a cell lysate, chromatographic column-based methods are required. Multiple elutions from IMAC or MOAC columns or even gradient elutions can help to simplify fractions of proteins and reveal more peptides [212, 213] . A combination of techniques can reveal large numbers of phosphopeptides in complex samples, but comprehensive phosphoproteomics is still not possible. For the highest protein coverage, future phosphoproteomic techniques will likely employ multiple enrichment techniques along with two-dimensional separations, but such studies are time consuming. Combinations of affinity-based enrichment and extraction methods, multidimensional separation technologies, and mass spectrometry are particularly attractive for systematic investigation of posttranslationally modified proteins in proteomics [214] . Organisms. The application of proteomics and related technologies for the analysis of proteome is severely hampered by the lack of publicly available sequence information for most of the unsequenced organisms [215] . Despite the precision of the mass information yielded by the SELDI technique, a significant number of proteins were found to have no similarity to known peptides, an aforementioned weakness of proteomics studies in nonmodel organisms [216] . In order to circumvent this limitation, different strategies and tools were developed to make unsequenced organisms amenable to high-throughput proteomics [217] (Figure 6 ). However, an evaluation of their performance in an integrated proteomics strategy using high-throughput shotgun MS data is currently missing. In principle, two different approaches can lead to an increase in protein identifications from unsequenced organisms. In the first approach, MS/MS data are searched against a protein database of an evolutionarily closely related organism. However, as a matter of principle of database-dependent searches, only proteins can be identified that contain at least one peptide with exactly the same sequence as the peptide from a protein in the database. With increasing evolutionary distance this will be an increasingly severe restriction [218] . In the second approach, the amino acid sequence of a peptide is extracted from the MS/MS spectrum for de novo sequencing, that is, in a fully databaseindependent manner using exclusively the information contained in the MS/MS spectrum. Several software tools for peptide de novo sequencing are now available and some of them provide sufficiently good results when applied to high-quality spectra [219] . A basic limitation of MS de novo sequencing methods is the necessity for backbone cleavage between each pair of adjacent amino acids; a mass value representing a terminal fragment containing only one of the two residues is a first requirement for ordering of a specific pair [220, 221] and this limitation urged the need for bioinformatics approaches that can help interpret the proteomics data [219] . In the past several years there have been very important extremely useful advances in proteomics methods based on bottom-up display and bottom-up identification using peptides [222] . These methods offer more sensitivity, greater rapidity and greater proteome coverage are often made with the explicit or implicit assertion that these methods are bound to replace more traditional methods based on topdown analysis, especially using 2D gels [223, 224] . The combination of bottom-up display and bottom-up identification has achieved very important successes in detecting the presence of large numbers of different proteins in cells or subcellular organelles [225, 226] . The use of specific fractionation schemes and prudent adoption of methods to increase the number of proteins able to be identified and quantified is enabling significant biological advances to be made. Further technological developments that enable a larger proportion of the proteome to be visualized will further enhance our ability to characterize biological systems. As such, these advances in proteomics will impact not only academic pursuits but also pharmaceutical, biotechnology and diagnostic research and development [227] . In the future gel-free techniques MudPIT, iTRAQ and 18O stable isotope labeling could be expected to gain more importance as they become more established. Sample prefractionation system provides a highly valuable tool to fractionate proteins and peptides from complex eukaryotic samples like plasma. This approach has a positive influence on the number of proteins identified compared to SCX method [228] . iTRAQ is a very powerful tool, recognised form its ability to relatively quantify proteins. iTRAQ reagent improves MALDI ionisation, especially for peptides containing lysine. Although SILAC labelling is easy for any laboratory that uses cell culture, the MS technology that is required is still beyond the capabilities of most groups. One of the factors that contributed to the rapid acceptance of the SILAC technology was the availability of an open-source program, MSQuant, for interpreting results. Protein microarrays offer the ability to simultaneously survey multiple protein markers in an effort to develop expression profile changes across multiple protein analytes for potential use in diagnosis, prognosis, and measurement of therapeutic efficacy [229] . This technology is an excellent high-throughput method used to probe an entire collection of proteins for a specific function or biochemistry. It is an exceptional new way to discover previously unknown multifunctional proteins, and to discover new functionalities for well-studied proteins [230] . A systematic and efficient analysis of vast genomic and proteomic data sets is a major challenge for researchers today. To overcome limitations of current proteomics strategies in regard to the dynamic range of peptides detected and alternative mass spectrometrybased approaches are being explored. Targeted strategies exemplified by multiple reaction monitoring detect, quantify, and possibly collect a product ion spectrum to confirm the identity of a peptide with much greater sensitivity because the precursor ion is not detected in the full mass spectrum [231] . A systematic and efficient evaluation of large-scale experimental results requires (1) automatic retrieval of user defined information to construct a customized, queryable database; (2) an intuitive graphical and query platform to display and analyze experimental data in the context of the customized database; (3) efficient utilization of webbased bioinformatics software tools for data interpretation, prediction of function, and modeling; (4) scalability and reconstruction of the database in response to changing user needs and an ever-expanding base of knowledge and bioinformatics tools [232] . Creating a software tool to encompass the four crucial features outlined above is a challenging and ongoing task, particularly with respect to the ever-expanding publicly available base of knowledge and bioinformatics tools. The data processing and analysis bottleneck can be overcome through integration of the entire suite of tools into one linear pipeline. The good news is that all of the various proteomics strategies are in phases of very rapid technological development and that important advances in sensitivity, throughput, and proteome coverage can be expected in the near future for all of them. Post translational modifications 2DE: Two-dimensional gel electrophoresis DIGE: Fluorescence 2D difference gel electrophoresis ESI: Electrospray ionization FTIC: Fourier transform ion cyclotron HPLC: High performance liquid chromatography ICAT: Isotope-coded affinity tag iTRAQ: Isobaric tags for relative and absolute quantitation IPG: Immobilized pH gradient LC: Liquid chromatography MALDI: Matrix-assisted laser desorption/ionization MS: Mass spectrometry MudPIT: Multidimensional protein identification technology TOF: Time of flight PAGE: Polyacrylamide gel electrophoresis SCX: Strong cation exchange MRM: Multiple reaction monitoring assay SELDI: Surface-enhanced laser desorption/ionization IMAC: Immobilized metal affinity capture.

A Systematic Molecular Pathology Study of a Laboratory Confirmed H5N1 Human Case

Autopsy studies have shown that human highly pathogenic avian influenza virus (H5N1) can infect multiple human organs other than just the lungs, and that possible causes of organ damage are either viral replication and/or dysregulation of cytokines and chemokines. Uncertainty still exists, partly because of the limited number of cases analysed. In this study, a full autopsy including 5 organ systems was conducted on a confirmed H5N1 human fatal case (male, 42 years old) within 18 hours of death. In addition to the respiratory system (lungs, bronchus and trachea), virus was isolated from cerebral cortex, cerebral medullary substance, cerebellum, brain stem, hippocampus ileum, colon, rectum, ureter, aortopulmonary vessel and lymph-node. Real time RT-PCR evidence showed that matrix and hemagglutinin genes were positive in liver and spleen in addition to positive tissues with virus isolation. Immunohistochemistry and in-situ hybridization stains showed accordant evidence of viral infection with real time RT-PCR except bronchus. Quantitative RT-PCR suggested that a high viral load was associated with increased host responses, though the viral load was significantly different in various organs. Cells of the immunologic system could also be a target for virus infection. Overall, the pathogenesis of HPAI H5N1 virus was associated both with virus replication and with immunopathologic lesions. In addition, immune cells cannot be excluded from playing a role in dissemination of the virus in vivo.

Influenza pandemic are characterized by the worldwide spread of novel influenza strains for which most of the population lacks substantial immunity [1, 2] . Pandemic viruses typically cause heightened morbidity and mortality [1] . The continued circulation of highly pathogenic avian influenza (HPAI) viruses H5N1 has resulted in occasional coincident infections among humans. Since late 2003, when widespread H5N1 virus poultry outbreaks were reported in multiple countries in Asia, there have been 467 laboratory confirmed human cases in ten countries reported to the World Health Organization as of December 2009 with a mortality rate of about 60% [3, 4] . Global public health concerns surrounding H5N1 viruses include not only individual transmission events between infected poultry and individual humans, but also their pandemic potential, should these viruses acquire genetic changes that result in sustained human-to-human transmission. To date, several case clusters of H5N1 infections have been reported [5] and limited epidemiologic information has suggested person-to-person transmission of H5N1 in a few instances, usually involving family members. Of additional concern to both human and animal health, is the extensive geographic spread of HPAI H5N1 viruses in recent years and their isolation from multiple species of wild birds and mammals [6] [7] [8] [9] . Despite the recent emergence of the 2009 H1N1 pandemic [10] , the pandemic threat from HPAI H5N1 viruses has not diminished [11] . Human H5N1 disease is clinically and pathologically distinct from that caused by seasonal human influenza A H3N2 or H1N1 viruses [12] . The majority of confirmed human HPAI H5N1 virus infections have been characterized by a severe clinical syndrome including a rapid progression of lower respiratory tract disease, often requiring mechanical ventilation within days of admission to a hospital [13] [14] [15] [16] [17] [18] . In addition to pulmonary complications, other clinical manifestations of H5N1 virus infections may include severe lymphopenia, gastrointestinal symptoms, and liver and renal dysfunction [14, 16, 17, 19, 20] . Reactive hemophagocytosis in multiple organs, and occasional detection of viral antigen or viral RNA in extrapulmonary organs suggest a broader tissue distribution of H5N1 viruses compared with seasonal viruses in fatal human cases [21, 22] . Patients with severe H5N1 disease have unusually higher serum concentrations of proinflammatory cytokines and chemokines. Levels of plasma macrophage attractant chemokines CXCL10 (IP-10), CXCL9 (MIG), and CCL-2 (monocyte chemoattractant protein 1, MCP-1) and of neutrophil attractant interleukin-8 (IL- 8) were substantially higher in patients with H5N1 disease compared with those experiencing seasonal influenza virus and were significantly higher in H5N1 patients who died compared with those who recovered [23, 24] . The elevation of plasma cytokine levels was positively correlated with pharyngeal viral load [23] and may simply reflect more extensive viral replication, and consequently, direct viral pathology rather than being causative of the pathology observed in H5N1-infected patients. Compared with human H1N1 and H3N2 influenza viruses, infection of human primary macrophage cultures in vitro with H5N1 viruses also lead to the hyper-induction of proinflammatory cytokines [25] . Most studies on H5N1 pathology describe pulmonary features of human disease. Although H5N1 virus infection of humans is primarily one of the lower respiratory tract, more recent reports suggested that influenza A H5N1 may in rare, severe cases, disseminate beyond the lungs and infect brain [26, 27] , intestines [20, 27] and lymphoid tissues [27] , and result in extra-pulmonary clinical manifestations including encephalopathy or encephalitis [15, 28] . This extrapulmonary dissemination of HPAI H5N1 virus contrasts with seasonal influenza virus infection of humans which, even in fatal cases, is restricted to the respiratory tract. However, there have been relatively few reports describing histopathology and virus distribution in H5N1 cases [5, 16, 21, 24, 26, 27] . To better understand the pathogenesis of human H5N1 virus infection, and investigate the route of virus dissemination in vivo, we report on the use of different techniques to detect virus distribution and infection of 5 organ systems in a laboratory confirmed fatal human H5N1 virus infection, and analyze the relationship between viral load in tissues and host response. Our results suggested that the virus can infect multi-organs besides pulmonary. High viral load is associated with increased host response though the viral load is significantly difference in various organs. Cells of immunologic system could not be excluded to play a role in dissemination of the virus. Virus culture, real-time RT-PCR, IHC and ISH were used to identify virus distribution in different tissues. As shown in Table 1 , live virus was recovered from respiratory tissues including lung, trachea, bronchus and aortopulmonary vessel. In the digestive system, virus was isolated from tissues collected from the ileum, colon and rectum, but not the stomach, duodenum or liver. Of note, virus culture was also positive on tissues collected from brains, ureter and axillary lymph-node. Sequencing results showed that the sequences of isolates are identical. Real time RT-PCR results were consistent with the detection of virus by culture, except for the liver and spleen tissues which were positive by real time RT-PCR but negative for virus isolation. The tissue distribution of viral RNA or antigen detected by ISH and IHC stains respectively, was also generally consistent with virus isolation by culture or real time RT-PCR result. However, several exceptions were noted. There was a lack of detectable staining by either method in bronchus tissue, despite virus detection. On the other hand, both staining methods detected viral product in the kidney, although virus isolation and real time PCR were both negative. The viral load is associated with host response figured out by proinflammatory factors. The relative H5N1 viral load in different tissues by determining the ratio of viral HA copy number relative to the copy number of the beta actin gene for a given tissue sample. Tissues of the respiratory system yielded higher copy number ratios than tissues from all to other organ systems with the lower left lung lobe yielding the highest viral load, overall. Interestingly, ureter tissue had the highest viral load of non respiratory system tissues. All other tissues had lower viral loads that were not. Among the digestive system tissues, the viral load in the liver was the highest (See Fig. S1 ). Additionally, we detected mRNA copies of macrophage attractant chemokine CXCL10 (IP-10), macrophage inflammatory protein 3b (MIP-3b), RANTES, tumor necrosis factor (TNF)-related apoptosis-inducing ligand (TRAIL) and TNFa in tissues of brain, respiratory organ, spleen and lymph-node. RANTES and TRAIL can be detected in all selected tissues. MIP-3b is positive in selected tissues except right upper lobe lung with positive b-actin. IP-10 can be determined copies in cerebral context, left upper lobe lung, left lower lobe lung, right middle lobe lung, right lower lobe lung, and aortopulmonary vessel. TNF-a can be showed copies in left upper lobe lung, left lower lobe lung, right middle lobe lung, right lower lobe lung, aortopulmonary vessel and spleen. The correlation among levels of viral load and were showed by Fig. 1 . The Pearson's cross correlation analysis (See Table S1 ) showed that high viral load is associated with high host response figured out by proinflammatory factors (p,0.05) except TNF-a (P.0.05), and the correlation is significant between proinflammatory factors except between TRAIL and TNF-a (p.0.05), and between TRAIL and IP-10(p.0.05) as well. The histopathologic features in different organs are shown in Fig. 2 . Lung showed diffused alveolar damages including intraalveolar edema, focal intra-alveolar hemorrhage, necrosis of alveolar line cells, focal desquamation of pneumocytes in alveolar spaces, interstitial mononuclear inflammatory cell infiltrates, and extensive hyaline membranes. Trachea showed focal denudation of the epithelium with edema, and mononuclear inflammatory cell infiltrates. Spleen showed depletion of lymphocytes with congestion and organized infarcts. Axillary lymph-node was congested with depletion of lymphocytes. The central nervous system showed extensive edema with focal neuronal necrosis in hippocampus. Diastem between Purkinje cells layer and particle cells layer showed focal augmentation in cerebellum. Liver was congested with edema and focal fatty degeneration. Kidney was congested with edema. Other selected tissues showed no significant histological changes. (Fig. 3I ). As influenza virus is a negative-strand RNA virus, ISH stain with sense and antisense probes detected both virus RNA (with sense probe), mRNA and cRNA(with anti-sense probe). ISH stain showed that the two sets of probes (for hemagglutinin and nucleoprotein) generated similar staining results. In the selected 24 autopsy tissue from 5 organ systems, positive staining of ISH was detected in all samples except Bronchus, stomach and duodenum. In the lung tissue samples, sense probes extensively hybridized in the nuclei of pneumocytes (Fig. 4A) , whereas antisense probes hybridized in both the cytoplasm and nuclei. In all other organs with positive staining, both sense and antisense were present mainly in the cytoplasm of infected cells. In respiratory system, the positive staining was found in epithelial cell of lung ( We systematically studied the tissue tropism of H5N1 virus in a fatal human case, based on 24 autopsy tissue samples from 5 organ systems, and analyzed the relationship between viral load and host response level in selected tissues. We presented evidence suggesting that the H5N1 virus can infect selected tissue of selected 5 organ systems including respiratory, digestive, nervous, urinary and lymphoid system. High viral load can induce high proinflamatory factors level in tissues, and immune cells could be the target of the virus. There is a substantial amount of evidence that HPAI H5N1 virus can infect extrapulmonary organ tissues [16] [17] [18] 27] and precede other clinical manifestation [20, 29] . Our results presented that the viral antigens or viral RNA can be found in trachea, lung, brain, intestines, liver, spleen, lymph-node and kidney which were reported as same before [20, 26, 27, 30] , as well as in aortopulmonary vessel and ureter which were not reported before. Notably, the virus can be found in tissues of lower gastrointestinal tract including small intestine and large intestine but in stomach and duodenum. The origin of infection in the extrapulmonary organs could be blood-borne, which is supported by previous studies showing live H5N1 virus can be isolated from the serum [20] and plasma [31] . Unfortunately, we haven't obtained blood samples to [25, 30, [33] [34] [35] [36] and T lymphocytes [37] . The viral load shows various levels in different organs although the virus can be found in multiple organs. Quantitative detection of viral gene showed that viral load is the highest in tissues of respiratory system, especially, in left lower lobe of lung. The liver had the highest viral load in tissues of digestive system. Interestingly, renal duct showed a high viral load although PCR detection of kidney was negative. And viral load was high in spleen and lymphonode. Despite of the viral cells tropism, possible reasons of different viral load could include activation of low PH [38] , viral receptor distribution [39] , cell N-linked glycoprotein distribution [40] or other unknown mechanisms. On other hand, non-permissive or abortive infection is extra possible in some of the tissues where the virus cannot replicate effectively. Additionally, it should be possible that phagocytosis of viral antigen from pathologically significant lytic/replicative infection caused positively viral detection in these tissues. High viral load is associated with increased host responses. To be different with seasonal influenza virus, the H5N1 virus caused intense transcription induction and secretion of inflammatory cytokines/chemokines, as documented in human cases [15, 22] , and increased induction of cytokines and chemokines has been demonstrated in H5N1-infected mice [41] . Moreover, High virulence of H5N1 virus is associated with increase with increased host responses [42] . However, the relationship between viral load and host immuno-response has never been reported. Our study suggested that the high viral load is correlated with proinflamatory factors including IP-10, RANTES, MIP-3b and TRAIL. This finding is particularly relevant to the mechanism of H5N1 pathogenesis which is associated not only with the virus but also with the host responses. Inflammatory protein can be produced by almost any infected cells and by immune cells, including alveolar macrophages [43] . However, type II pneumocytes have a marked ability to secrete large amounts of cytokines, such as TNF-a, GM-CSF, MCP, and IL-8, in response to various insults [44, 45] and can be induced to secrete IL-1a, IL-6, RANTES, and MCP-1, in response to TNF-a [45, 46] , the last being produced by alveolar macrophages during H5N1 infection. Therefore, it is possible that the much greater cytokine induction by the H5N1 virus was as much a consequence of the response of individual, infected cells as it was a consequence of the numbers of infected cells. Immune cells could be a target of the virus infection. Previous findings showed that viral sequences and antigens have been detected in lymphocytes in lymph node tissue, as well as in Hofbauer cells (macrophages of the placenta), Kupffer cells (macrophages of the liver), and mononuclear cells in the intestinal mucosa [27] . This is consistent with our find. Our presented evidence shows that virus can be detected in macrophage. Moreover, our finding showed that viral RNA can be detected in CD3+ T lymphocytes, progenitor cells and follicular dendritic cells of spleen. The spleen is a complex organ with several functions, including the removal of senescent or aberrant red blood cells from circulation, as well as the removal of circulating pathogenic organisms [47] . So circulationg immune cell with H5N1 virus could play a role in dissemination of virus when those cells with virus can not remove the virus in cells completely. We unprecedentedly investigated 24 autopsy tissues of 5 organ systems from a dead patient of H5N1 infection. Our study results indicate that the H5N1 virus could be found in multiple organs. Viral load in infected tissues is correlated with host response. And circulating immune cell can not be excluded to play a role in dissemination of virus in vivo. The patient was laboratory confirmed as an H5N1 infection by China CDC on February 20, 2008. He is a 42 years-old Chinese male living in Guangxi, China with a history of 6 days of fever, cough and dyspnea. Two weeks before hospital admission, he bought 3 hens at an open-air market, one of which died later at same day. The remaining birds exhibited symptom of chicken attack next day. The patient killed and cooked the birds, and ate them together with other family members who later did not experience any clinical symptom. On admission, the patient was febrile with a temperature of 40.3uC, had infiltration of lower left lung lobe based on chest radiography, bilateral lower lung moist rales, and substantially reduced oxygen saturation. He was placed on a ventilator and treated with antibiotics, spasmolysis, corticosteroids, and the dissipatation of phlegm and fluids. Despite treatment, the patient presented with function damage of multiple organ (lung, heart, liver and kidney) on the second day after admission, and died 59 h after admission, and 8 days after the onset of symptoms. Throat swabs were collected and performed RT-PCR and rRT-PCR detection at the day patient died. No antiviral drug treatment was given since the patient died before finally laboratory diagnosis. After informed consent was obtained, the cadaver was stored at 4uC and underwent autopsy about 18 h after death. The autopsies were done following conventional protocols with strict adherence to biosafety procedures [48] . Twenty-four tissues were collected from respiratory, digestive, nervous, urinary and lymphatic organ systems. Duplicate tissue samples were collected; one sample was fixed in diethylpyrocarbonate (DEPC) treated 10% formalin for pathologic analyses, while a second sample was frozen at 280uC for virus isolation and molecular analyses. Frozen tissues were thawed and homogenized before inoculation into embryonated eggs for viral culture. Briefly, 1.0 ml of phosphate-buffered saline (PBS) was added to 100-150 mg of tissue, which was homogenized, and, then centrifuged; 0.1 ml of the recovered supernatant was injected into the allantoic cavity of 11 days-old specific pathogen-free embryonated chicken eggs. The allantoic fluids were tested for haemagglutation activity with turkey red blood cells after 72 h incubation at 35uC. Samples containing virus were subjected to RT-PCR and sequencing. Three blind passages were performed on hemagglutinationnegative samples to confirm the absence of infectious virus. All steps were performed in BSL-3 containment laboratory. Total RNA from the tissues samples was extracted using an Axygen total RNA extraction kit (Axygen, USA) as described in the manufacturer's instructions. Briefly, 400 ml of lysis buffer was added to 30-40 mg of ground tissue. The nucleic acids were eluted in 50 ml nuclease-free water and stored at 220uC. For every five samples, one negative control (water) was included to detect any possible contamination. Control RNA products derive from in vitro transcription of the matrix (M) gene RNA of A/Anhui/1/2005(H5N1), human house keeping gene beta-actin and human cytokines/chemokines (including IP-10, RANTES, TRAIL, MIP-3b and TNF-a) genes were used as positive controls and to establish the detection limit of the assay. Recombinant plasmids with entire M gene, beta-actin gene segment and cytokines/chemokines gene segment were linearised by restriction enzyme, and then purified using a DNA clean-up kit. DNA concentration was measured as OD units at 260 nm. One mg of linearized plasmid DNA was transcribed using Riboprobe in vitro transcription system kit (Promega, USA) according to the manufacturer's instructions. The transcribed RNA was purified using phenyl/chloroform solution and was quantified by spectrophotometer. RNA copy number was then determined following the method of Fronhoffs [49] . To analyze the H5N1 viral load and quantify proinflammatory factors in different tissues, real-time RT-PCR was performed with a Strategene detection system using a fluorescently labeled TaqMan probe to enable continuous monitoring of amplicon formation. The primer and probe of H5 HA gene is from the WHO-released primer sets [50] . The primers and probes of proinflammatory factors and beta-actin gene were obtained from the literature [51, 52] . The concentration of primer and probe used was 40 mM and 10 mM, respectively. The reaction was completed in a total volume of 25 ml performed by QuantiTect Probe PCR Kit (Qiagen, Germany). The reaction mixture was incubated with 5 ml DNase-treated total RNA (the template which was used to amplify b-actin gene was performed by 100-fold dilution) at the following temperature cycles. First, the reverse transcription reaction was completed by 1 cycle at 50uCfor 30 min. Next, H5N1 HA gene, cytokine/chemokine gene and house keeping (bactin) genes were amplified by 1 cycle at 94uC for 15 min and 45 cycles at 94uC for 15 s, 55uC for 30 s, and 72uC for 30 s each. As described in previous report [53] , the standard curve was generated using serial dilution of in vitro transcribed standard RNA (from ,10 to 10 7 copies). The viral load and cytokines/ chemokine levels are presented as the log 10 value of the ratio between copies of the target gene and b-actin gene. Immunohistochemical stain was performed on 5 mm thick deparaffinized sections using monoclonal antibodies against the nucleoprotein of influenza A (Serotec, UK) by a two-step peroxidase method. For controls, we used unrelated antibodies in place of the primary antibody. Briefly, sections were deparafinized by 2 washes in xylene and were rehydrated through decreasing concentrations of ethanol. After washing in PBS at pH 7.6 for 5 min at room temperature (RT), sections were heattreated with antigen-retrieval solution (TRIS/EDTA buffer, pH 9, Dako, Denmark) for 10 min using microwave antigen retrieval method. After blocking with 10% normal horse serum for 10 min at RT, the sections were incubated with the specific antibody (1:100) for 30 min at RT. Unbound antibody was removed by 3 washes in PBS before adding HRP-labelled polymer for 30 min at RT (Dako CSA Detection System, Denmark). After washing unbound labeled polymer in PBS 3 times, peroxidase staining in tissue sections was revealed by DAB solution (CSA Detection Systems, DAKO, Denmark). After stopping the reaction in running water, sections were counter-stained with a quick rinse in Mayer's hematoxylin solution. After dehydrating with increasing concentrations of ethanol and xylene, the sections were mounted with DPX and examined by light microscopy (Olympus BX51, Japan). The development of probes was based in analysis of the full haemagglutinin and nucleoprotein gene sequences of all Chinese human isolates of H5N1. Oligonucleotide DNA probes representing conserved gene regions were used. Probes were labeled by digoxigenin-UTP (Roche Diagnostics, Penzberg, Germany) and tested for specificty using human biopsy of gut tissue with in vitro infection of H5N1 virus. Since H5N1 is a negative-stranded RNA virus, sense probes were defined as the probes that detect the viral RNA (negative-stranded), whereas antisense probes detected mRNA and complementary RNA (cRNA), which are both positive-stranded. Briefly, before hybridization, all solutions were prepared with DEPC-treated water. After deparaffinization and rehydration, tissue sections of 5 mm thickness were treated with proteinase K digestion for 30 min. Tissue sections were then incubated with a hybridization cocktail containing 25 mg/mL of probes at 37uC for 16,18 h. All sense and antisense probes were applied separately on consecutive tissue sections. After blocking with normal horse serum (1:100), sections were incubated with alkaline phosphatase-labelled digoxigenin antibody (1:1000, Roche Diagnostics, Penzberg, Germany) for 1 h, and the reaction products were colorised with nitroblue tetrazolium/5-bromo-4choloro-3-indolyl phosphate (Roch Diagnostics, Penzberg, Germany) for 1-1.5 h, and counterstained in 2% nucleic fast red liquid for 1 min. As a positive control, we used human lung and gut biopsy tissues with in-vitro infection of H5N1 virus. Negative controls also included an unrelated antisense probe against the fragment of the heamaglutinin gene of the seasonal influenza virus H3N2 as well as H5N1 in-situ hybridization probes to tissues. After completeing the colorization reaction of the ISH as outlined above, sections were incubated with 3% hydrogen peroxide to quench endogenous peroxidase activity. The sections were then blocked with 10% normal horse serum for 10 min at RT, before incubating with monoclonal antibodies (against CD3 + , CD68 + , CD34 + , D35 + ) for 1 h at RT. Unbound antibody was removed by 3 washes in PBS before the addition of HRP-labeled polymer for 30 min at RT (Dako CSA Detection System, Denmark). After washing 3 times with PBS to remove unbound labeled polymer, peroxidase staining in tissue sections was revealed by DAB solution (CSA Detection Systems, DAKO, Denmark). After stopping the reaction in running water, sections were counter-stained by 2% nuclear quick red solution for 30 s. For each tissue, double-labeled stain with IHC and ISH was repeated three times for data analysis. To strengthen further the results of colocalization studies, we performed ISH and IHC on consecutive sections. Tissue sections showing ISH-positive cells were carefully compared with consecutive tissue sections on which IHC with antibodies against specific cell markers was applied. Co-localization of a specific cellular marker and viral genome was clearly identified. The correlation between viral load and quantitative proinflammatory factors profile was analyzed by Pearson's correlation test using Instat software (Vision 5.0, GraphPad prism). Differences were considered significant at p,0.05. Figure S1 The distribution of viral load in selected tissue samples. The viral HA gene and b-actin gene copies in tissues were determined by quantified real-time RT-PCR. The ratios between HA and b-actin gene copies which was showed by logarithm presented the viral-load level in different tissue. Found at: doi:10.1371/journal.pone.0013315.s001 (0.11 MB TIF)

Situational Awareness and Health Protective Responses to Pandemic Influenza A (H1N1) in Hong Kong: A Cross-Sectional Study

BACKGROUND: Whether information sources influence health protective behaviours during influenza pandemics or other emerging infectious disease epidemics is uncertain. METHODOLOGY: Data from cross-sectional telephone interviews of 1,001 Hong Kong adults in June, 2009 were tested against theory and data-derived hypothesized associations between trust in (formal/informal) information, understanding, self-efficacy, perceived susceptibility and worry, and hand hygiene and social distancing using Structural Equation Modelling with multigroup comparisons. PRINCIPAL FINDINGS: Trust in formal (government/media) information about influenza was associated with greater reported understanding of A/H1N1 cause (β = 0.36) and A/H1N1 prevention self-efficacy (β = 0.25), which in turn were associated with more hand hygiene (β = 0.19 and β = 0.23, respectively). Trust in informal (interpersonal) information was negatively associated with perceived personal A/H1N1 susceptibility (β = −0.21), which was negatively associated with perceived self-efficacy (β = −0.42) but positively associated with influenza worry (β = 0.44). Trust in informal information was positively associated with influenza worry (β = 0.16) which was in turn associated with greater social distancing (β = 0.36). Multigroup comparisons showed gender differences regarding paths from trust in formal information to understanding of A/H1N1 cause, trust in informal information to understanding of A/H1N1 cause, and understanding of A/H1N1 cause to perceived self-efficacy. CONCLUSIONS/SIGNIFICANCE: Trust in government/media information was more strongly associated with greater self-efficacy and handwashing, whereas trust in informal information was strongly associated with perceived health threat and avoidance behaviour. Risk communication should consider the effect of gender differences.

Pandemic influenza A/H1N1 has a clinical profile similar to seasonal influenza, despite initially appearing more severe [1] . Respiratory infectious diseases (RIDs) such as influenza are a major public health issue best dealt with by prevention, ideally vaccination. However, in the first six-months or so of a newlyemergent RID epidemic/pandemic vaccines are generally unavailable and non-pharmacological interventions can play a major role in minimizing RID spread [2] [3] [4] . Government health education messages are a major source of information for promoting self-protective practices against RIDs. These preventive messages generally emphasize improved hygiene, face-mask use by infected persons, and social distancing measures, including avoiding crowds during epidemics [5] [6] [7] . Predictors of population uptake of health protective behaviours in RID epidemics have begun to be studied [8] [9] [10] [11] [12] [13] , yet related theory remains nascent and this is problematic: to effectively predict behaviour during future epidemics robust theory is critical. Effective models that enable comprehensive prediction of health protective behaviours remain limited mainly to two overlapping theoretical paradigms: the Theories of Reasoned Action/Planned Behaviour (TPB) [14] [15] [16] and Bandura's concept of self-efficacy [17] [18] [19] (the belief that one can successfully execute some behaviour), particularly regarding the core TPB concept of perceived behavioural control, which controversially is claimed by some to be largely synonymous with self-efficacy [19] [20] [21] and by others to be indistinguishable from intent [22] (the intention to execute a particular behaviour), the key predictive element of TPB [16] . When used to account for health-related behaviours TPBbased models typically account for ,35% of variance in outcomes [16] , while self-efficacy accounts for ,25% of variance in outcomes [23, 24] . However, neither TPB nor Self-efficacy allow for the social and affective influences that might be expected logically to be important in RID [25, 26] . We report on a theoretical model that incorporated elements of influenza causal knowledge, perceived self-efficacy and also social and affective influences ( Figure 1 ) because these latter variables have been less frequently studied in combination, but have theoretical and logical support for their potential importance in the context of RIDs. We tested this model against data collected in the early phase of the influenza A/H1N1 pandemic (Table S1 ) to examine how levels of trust in formal and informal sources of risk/prevention information associated with hand washing and social distancing. Ethics approval was obtained from the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster. For this telephone interview, written informed consent was waived by the IRB but verbal consent was required from all the respondents and agreement to participate in the interview was taken as further consent. Before the interview began, a brief introduction about the study aims and interview contents was given and then respondents were asked whether the interview could start. If approval was received this was recorded and the interview performed. If not, respondents were thanked and the call was terminated. More than 98% of Hong Kong households have landline telephones and all local calls are free. Random-digit dialled telephone numbers and within-household random-sampling grids (Kish grids) are a cost-effective way to survey highly representative random population samples. Kish grids are matrices containing random numbers for different sized households that facilitate random selection of individuals within households and help minimize sampling bias. The number of eligible household residents, ''n'', is determined by asking the person of first contact in the household. The Kish grid provides a randomly generated number ''k'' between 1 and ''n'' which is used by the interviewer. Ordering by age and starting from the oldest eligible member in the household, the k'th member is then invited to participate in the survey. Different grid values are used for each household. As part of a series of surveys to monitor A/H1N1 epidemic activity, a commercial polling organization administered the questionnaire using this telephone-survey methodology, targeting 1,000-1,500 participants on each occasion, a sample size calculated to give an estimate of A/H1N1 health protective behaviours with a precision of 63%. The survey with the largest sample was selected for this analysis. Sampling was performed during the evening to minimize exclusion of young working adults. Data on attitudes, knowledge, situational awareness, risk perception and preventive behaviours (Table S1) were collected by household telephone interviews, based on random digit dialling. One Cantonese-speaking adult (age$18) who lives .4 nights per week in each household was selected using a Kish grid. All interviews were conducted between 8:30pm-10:45pm from 23 rd -25 th June, 2009, two weeks after the first community transmission had been identified in Hong Kong. Existing theoretical frameworks of behaviour change have been adapted to predict health-related behaviour-change for chronic, non-communicable diseases [15, 16] , but we lack a comprehensive evidence-based model of protective behaviour against RID threat [11] . A recent review of 26 papers on RID prevention behaviours concluded that 23 lacked a theoretical basis [13] . Existing applications of health behaviour change models in communicable disease are almost exclusively limited to HIV/AIDS research [24, 27] and to a lesser extent hepatitis B and C, which share the same transmission pathways as HIV. There are good reasons why sexually-transmitted diseases embody a different set of influences than do RIDs. For example people are highly motivated to seek sexual contact (or injection drug use) and have a high degree of potential control (e.g. condom use) over the nature of these encounters, even though they may be situationally constrained from executing that control, and are infected only by direct exchange of bodily fluids. In contrast, one can acquire an RID transmitted by air droplets, hand contact or fomites for up to 72 hours after the person who is the source departs [28] , or immediately by being sneezed on. Infection is much more casual. Clearly, the controllability of RIDs requires different behavioural imperatives to those in STDs and hence different psychological influences should be considered. Attempts by the TPB to accommodate social influences had relied on incorporating social norms [14] , the behavioural expectations within a group. However, norms, and hence theoretical models reliant on norms to account for social processes, cannot accommodate the fact that communicable respiratory diseases make other humans ambiguous sources of threat: one can usually control sexual encounters but not who shares public transport. In this respect social factors in communicable respiratory disease differ significantly from those in non-communicable diseases and warrant greater consideration than existing HBC models allow. Outbreaks of new infectious diseases constitute situations that are uncertain, dynamic, and embody highly personal threat, requiring rapid decisions on appropriate action [29] . Under such circumstances timely and relevant information on the best preventive actions become critical to such decision-making. Hence, health protective behaviour during the early stages of a novel epidemic would be more likely to resemble situational reactions using established or known default actions such as avoiding crowds (social distancing), rather than intention-based planning before any behavioural change, such as deciding to consult a doctor to administer a vaccination. Later in the epidemic as threat familiarity increases, different factors such as planned behaviour may become important. Reporting delays, uncertainty and other biases affect publicly available information on the characteristics of newly-emergent communicable diseases, such as A/H1N1 lay knowledge of infection-related risks can be limited. The resulting uncertainty about disease severity and transmissibility at the epidemic onset extends to the utility and timing of adopting preventive measures. Information cues to individuals about initiating protective action must therefore be synthesized from various sources. Perceived information reliability or trustworthiness influences decisions to utilize any given information source [30] to inform awareness of the situation. More trustworthy sources are therefore likely to be more influential. Epidemic situational awareness is likely derived from formally-announced public information like news items, government press releases and health education messages, and also from informal, social sources [25, 29] ; observation of other peoples' behaviour and communications from family, peers and neighbours. Noting how others behave informs action decisions in the observer [19] . If those around you are wearing masks, this indicates others might have knowledge you do not possess, and that the threat level might be locally high and imminent, suggesting prudent precautionary or RID preventive behaviour. Observers are also subject to social conformity influences that can help adoption of group patterns of behaviour. Maintaining situational awareness, involving elements of perception, comprehension and prediction [31] , during epidemics probably relies on these two types of information. However, when uncertainty is high and widespread, or when there is low confidence in social and other information sources then individuals' HPBs might be expected to be more independent of formal and informal information sources. Perceived risk is influenced by several stimulus characteristics, including unfamiliarity, invisibility, dreadfulness and inequity [32] , and by recipient characteristics, including demographics and trust in information source and content [33] . Perceived risk is an important determinant of protective behavioural responses [12, 34, 35, 36] , but is subject to optimistic bias, where for example people distort their risk of contracting influenza downwards relative to others [35, 37] . Nonetheless, susceptibility to risk remains an important measure in understanding variation in behavioural responses to threat and reflects the key element of perceived risk in an epidemic/pandemic situation. Worry is a cognitive process linked to anxiety [26, 38] and reflects negative affectivity, interacting with perception of susceptibility to risk [26, 39] and may also influence RID protective responses such as social distancing [13] . Because data were collected using telephone interviews we had to adapt measures to suit a brief format in order to avoid people hanging up mid-way or providing invalid answers to hurry the interview, a problem encountered with this data collection method. We therefore used parsimonious measure to minimize assessment fatigue and low response rates which threaten representativeness. Trust in government/media (formal) information: We asked about respondents' agreement with three statements (Table S1 ). Responses were made on categorical five-point scales ranging from ''strongly disagree'' to ''strongly agree''. Scalability of these three items was assessed using Cronbach's alpha, which at 0.61 indicated that the internal consistency between items was low, but acceptable. However, to minimize potential measurement error arising from the low internal consistency, this construct was treated as a latent variable in the subsequent analysis [40] . A latent variable is a concept opposed to an observed variable. A latent variable can not be measured directly but is inferred from one or more variables that are directly measured (observed variable) while an observed variable can be directly measured with a specific question or item or observed by the researchers. For example, an ''attitude'' is a concept that is difficult to measure directly with single items but can be inferred from various questions asking about different aspects of that attitude. Then within the analysis ''attitude'' is treated as the latent variable while the questions used to infer it are the observed variables. Trust in interpersonal (informal) information: Respondents' agreement with two statements (Table S1 ). Responses were made on categorical five-point scales ranging from ''strongly disagree'' to ''strongly agree''. Scalability of these two items was assessed using Cronbach's alpha, which at 0.50 indicated that scalabilty was unsuitably low for two items. This suggests that these two items measure different aspects of social information. Again to minimize potential measurement error this construct was treated as a latent variable in the subsequent analysis. Understanding cause of A/H1N1 (''I understand how Swine flu is caused'') and self-efficacy (confidence in one's ability to act in a way that achieves desired future outcomes) for A/H1N1 prevention (''I am confident that I can protect myself against Swine flu''): Each was assessed using responses on 5-point scales of agreement with these two single item statements (Table S1) . Perceived personal susceptibility: Two items, one assessing absolute susceptibility (perceived absolute probability of developing A/H1N1) and another assessing relative susceptibility (perceived probability of developing A/H1N1 relative to peers) formed a latent variable for perceived personal susceptibility (Table S1 ). The Cronbach alpha of these two items was 0.66. Worry about contracting H1N1. Respondents were asked to indicate their level of worry over the past one week about contracting influenza A/H1N1. Responses were 5-point scales of worry ranged from ''never thought about it'' to ''extremely worried'' (Table S1 ). Hand hygiene. Respondents were asked to indicate frequencies of use of four hand hygiene practices over the three days prior to interview: hand washing after sneezing, coughing and touching nose; hand washing after returning home, use of liquid soap for hand washing, and hand washing after touching common objects. Responses were on a 4-point scale of frequency: 1 ''never'', 2 ''sometimes'', 3 ''usually'' and 4 ''always''. Cronbach's a was 0.62 (Table S1 ). Social distancing behaviours: a. Social Avoidance. Respondents were asked to indicate if they had adopted any of four avoidance behaviours due to influenza A/H1N1 in the past 7 days: avoiding eating out, avoiding using public transport, avoiding going to crowded places, and rescheduling travel plans Responses were coded as 1 ''yes'' and 0 ''no''. Cronbach's a was 0.61 (Table S1 ). We first compared the demographic structure of the sample against that of the general population derived from the Hong Kong government General Household Survey to identify any sample differences. Our model proposes that trust in formal (government and media sources) and informal (from other people) information affects RID epidemic health protective behaviours, the former by informing about generic risk and response characteristics for dealing with a potential threat (causes and protective responses), the latter about threat imminence, severity and response effectiveness (seeing how others behave). We refer to the product of these combined processes as situational awareness, and propose that rather than driving behaviour directly information acts through altering the cognitive/affective domain of situational awareness. Thus the model is predicated on several premises: that understanding of the disease and perceived personal susceptibility influence self-efficacy [17, 18, 31] ; that the effect of perceived susceptibility to influenza on HPBs acts through increasing worry about the disease [26, 33, 38, 41, 42] ; and that more worry from perceived susceptibility prompts HPBs [39, 41, 42] . These cognitive/affective processes are represented in the hypothesized model ( Figure 1 ). Structural Equation Modeling (SEM) is a method for simulating and testing multiple and interrelated causal relationships simultaneously in statistical data, making it suitable for theory development and testing [40] . SEM was applied to test the hypothesized model. SEM is usually performed when a model contains latent variables assessed with specified measurement models. Despite including estimations of a series of multiple regression equations, SEM differs from regression analysis in several ways, which make it advantageous for this kind of analysis. First, SEM is usually theoretically based because it is performed after researchers specify the hypothesized model. Second, it can be used to refine the hypothesized model by estimating the measurement model and structural model simultaneously. Finally SEM analysis can accommodate measurement errors of the constructs in the model [40] . In our hypothesized model, trust in formal information, trust in informal information, perceived personal susceptibility, hand washing and social distancing behaviours were entered as latent (inferred) variables while other constructs were entered as observable (directly measured) variables because they were assessed with only one item. Two different health protective behaviors, hand washing and social distancing, were entered as the HPB outcomes because we hypothesized that different influences may act on each of these. We assumed that the ''disturbances'' of the two health behavior outcomes were correlated. Disturbance represents the unexplained variances of the latent variables predicted by the specified independent variables [40] . In making this assumption, we assumed that unexplained variance in the outcome variables could be correlated and the variables in question jointly influenced by other unknown factors, and so allowed for such constraints within the model by using more conservative criteria. Previous studies have shown that hand hygiene and social distancing behaviours during a pandemic could be influenced by some common causes which were not fully explored in our study such as current health, past experience of disease and cues to action [14] . In particular, in our study, the two kinds of health protective behaviours occurred in the same situation of the 2009 influenza pandemic, and so it is sensible and reasonable to assume that they could be influenced by some common causes which were not fully explored in our studies. Adequacy of the measurement models was tested before testing the full structural model. To test the full structural model, all constructs ( Figure 1) were entered into the model and all factor loading, specified paths, covariance, measurement errors and disturbances were estimated simultaneously. Since the model contained categorical variables, Weighted Least Square with mean-and variance-adjusted estimation (WLSMV) was used to estimate the standardized parameter (b) for each path [43] . With this kind of estimation, chi-square difference testing is inappropriate. We therefore used the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI) and Root Mean Squared Error of Association (RMSEA) to evaluate the model fit to the data. A CFI.0.95, TLI.0.95 and RMSEA,0.05 indicate a good fit of data to the model [43] . The analysis was conducted in Mplus 6.0 for Windows [43] . The proportion of missing values ranged from 0.1% for ''In the past one week, have you ever worried about catching influenza A/H1N1'' to 10.1% ''did you wash hands after sneezing, coughing or touching nose in the past 3 days''. Missing data were handled with multiple imputation to generate 10 datasets which were summarized into one for subsequent analysis. Multiple imputation was performed in AmeliaView [44] . Responses are likely to differ by sociodemographic factors [13] . We therefore stratified the sample by gender and by age (,45 years old vs. .45 years old). Education is also likely to have a significant effect but there are difficulties in education stratification in Hong Kong. The age cut-off of 45 years was adopted to account for the introduction in Hong Kong of 6-year compulsory education in 1971 and 9-year compulsory education in 1978 [45] . This means that people aged 45 or above are much less likely to have a tertiary (college/university) level education and less secondary (high school) education than people aged ,45 years old [45] . Moreover, in traditional families in China, a son (who lived with his parents after marriage) was usually more educationally-favoured over daughters (who moved to their in-laws' home on marriage) to ensure support for the parents in their old age, so males usually obtained more education than females [45] . These distinctions were somewhat evidenced by our data which showed that 98% of the respondents aged ,45 compared to 71% of the respondents aged 45 or above (x 2 = 147.69, p,0.001), and 89% of male compared to 80% of female respondents, obtained at least secondary education (x 2 = 17.05, p,0.001). Since the numbers of tertiary educated respondents and primary (elementary) educated respondents were too small to produce stable models, we limited stratification to gender and age only and acknowledge that this also incorporates indefinable education and income effects. Consequently, we used a multi-group SEM to assess the invariance of the model (Figure 1 ) across gender and age group (respondents aged 18-44 and aged 45 or above). We tried to test the model by stratifying the sample into four subgroups (female aged 18-44, female aged 45 or above, male aged 18-44 and male aged 45 or above). However, the sample size for males aged 18-44 was relative small (Table S2) . Moreover, all the model variables were treated as categorical variables and we used the WLSMV method to estimate the model. This method requires that each subsample covers all the categories of each variable. In the case of one category, younger males, not all variable values were present. To meet the assumptions for analysis we would need to recode all variables, intrinsically altering the model. In order to avoid this, we relinquished a combined four-group comparison and instead compared the model across gender and the two age groups separately. To perform multigroup comparison we first ran a model with all parameters unconstrained. We then identified factor loadings that were not significantly different (p$0.05) and set these as equal, while loadings that were significantly different were allowed to vary, and finally paths that did not differ significantly were constrained to be equal while those that differed significantly were allowed to vary and estimated separately by groups. The ''DIFFTEST'' option in MPlus 6.0 was used to obtain a correct chi-square difference test for the WLSMV estimators and was used to estimate the differences between the least constrained model (with all the paths freely estimated) and the most constrained model (with all the paths constrained to be equal) as well as the partially constrained model (with some of the paths freely estimated and others constrained to be equal) [43] . A p-value.0.05 for the ''DIFFTEST'' indicate a nonsignificant difference between the models. Finally, to help interpret these multigroup SEM comparisons, we performed a post-hoc examination of the model variable means for different gender and age groups and tested differences using the Mann-Whitney test, which tests differences between two groups on ordinal scales of measurements. A total of 1,001/1,449 (69.1% response rate) Hong Kong adults successfully completed the interview. The characteristics of the sample were compared against the Hong Kong 2006 by-census population data [46] , showing respondents to be better educated and more likely to have been born in Hong Kong compared to the general population (Table 1 ) but otherwise representative. Both formal and informal information trust were correlated with all situational awareness variables except worry about contracting A/H1N1 (''Worry''), while formal information trust was also independent of perceived personal susceptibility (''Susceptibility''). In turn, understanding of H1N1 cause (''Understanding'') and Perceived self-efficacy (''Self-efficacy'') were significantly associated with hand washing while Worry and Susceptibility were significantly associated with social distancing (Table S3 ). The SEM model fitted well to the data with CFI = 0.977, TLI = 0.969 and RMSEA = 0.026. Standardized coefficients indicated two primary features in the model; the first one linking Formal information and hand hygiene and a second linking Informal information and Social distancing ( Figure 2 ). Paths were seen via Formal information trust and Self-efficacy (b = 0.25) and Self-efficacy and hand hygiene (b = 0.23), and via Formal information trust and Understanding (b = 0.36), and Understanding and hand hygiene (b = 0.19) while Understanding and Selfefficacy were independent. These associations formed the first feature. Marginal associations between Worry and hand hygiene and between Self-efficacy and social distancing were seen, but the small standardized coefficients of b = 0.13 suggest that these paths are minor. Susceptibility and Worry were associated, but otherwise were functionally independent, both upstream from formal information trust, and downstream from hand hygiene. The second feature of the model is reflected in a different set of paths associating informal information trust with social distancing. Trust in informal information sources was inversely associated with Susceptibility (b = 20.21), which was associated positively with Worry (b = 0.44), and inversely with Self-efficacy (b = 20.42). However, more confidence in informal information sources was associated with more Worry (b = 0.16) and finally, only Worry was associated with social distancing (b = 0.36). Trust in informal information was independent of Understanding and Self-efficacy. The only remaining notable feature of the model was a strong inverse association (b = 20.42) between Susceptibility and Selfefficacy. This suggests some interaction between these two variables that could strongly influence both sets of paths mentioned so far. Overall, the model explained 11.3% of the variance in hand hygiene and 16.1% of the variance in social distancing behaviors. Across gender, both the least constrained model and the most constrained models fit the data well with CFI.0.970, TLI$0.970, RMSEA = 0.025. The most constrained model did not differ significantly from the least constrained model (x 2 for ''DIFFT-EST'' = 29.30, d f = 19, p = 0.061). However, three sets of associations differed significantly between females and males: those between Formal information trust and Understanding, from Informal information trust to Understanding, and from Understanding to Self-efficacy. These paths were set free and estimated separately in female and male. The model with these paths freely estimated fit well to the data with CFI = 0.978, TLI = 0.976 and RMSEA = 0.023, and did not differ significantly from the least constrained model (x 2 for ''DIFFTEST'' = 15.07, d f = 16, p = 0.519). Figure 3 presents the results of multigroup comparison of the model applied to males and females with the three path parameters unconstrained. For a given path, if the path coefficients did not differ significantly between males and females, only the path coefficient for males is presented; if the path coefficients differed significantly between males and females, the path coefficients for both genders are presented with the coefficients for males presented on the left of the slashes and for females presented on the right of the slashes. By comparison, the model shows that for both genders while the association between Formal information trust and Understanding was positive this association was stronger amongst females (b = 0.50) than males (b = 0.25); the association between Informal information trust and Understanding was weakly positive in males (b = 0.12) but weakly negative in females (b = 20.14), and; the association between Understanding and Self-efficacy was positive (b = 0.12) in males but non-significant in females (b = 20.01). Across the two age groups, both the least constrained model and the most constrained model fit well to the data with CFI.0.960, TLI$0.950, RMSEA#0.030. The most constrained model did not differ significantly from the least constrained model (x 2 for ''DIFFTEST'' = 15.85, d f = 19, p = 0.667). No path was found to be significantly different between the two age groups. Means and standard deviations for all model variables by gender and age group showed differences (Table S4 ). All the constructs did not differ by gender except for hand hygiene and social distancing with female being more likely to wash their hands and adopt social distancing behaviours. Trust in formal and informal information sources, Self-efficacy, and Hand hygiene significantly differed by age groups, with respondents of older age group being more likely to trust the information from both sources, perceive higher self-efficacy and wash their hands. We tested a hypothesized model of associations between trust in (formal/informal) information, situational awareness variables (causal understanding, self-efficacy, susceptibility and worry) and different types of health protective behaviours (hand hygiene and social distancing) for influenza protection. The model suggested that two different sets of influences relate trust in information to hand hygiene, and to social distancing respectively. The strongest associations observed were between Susceptibility and Self-efficacy Neither age nor gender contributed significant variation to the association between Trust in Formal information and Self-efficacy, and Self-efficacy and hand hygiene. These findings are consistent with other studies showing self-efficacy is enhanced by procedural information [18, 19, 47, 48] and that attitudinally and actionoriented interventions are more successful in changing behaviour for communicable disease protection, such as in the case of HIV [24] . Similarly, exposure to relevant media stories during the 2009 A/H1N1influenza pandemic was associated with higher efficacy beliefs regarding hygiene, which in turn was associated with greater frequency of reported tissue access and sanitising gel purchase among British people [49] . However, there is evidence that coping style interacts with the ability of procedural information to enhance self-efficacy and under circumstances of high threat, such as during SARS-type epidemics where mortality is high, procedural information might be counter productive for some segments of the community who use an information avoidance (''blunting'') coping style [50] . Self-efficacy was only weakly associated with social distancing. People are limited in their ability to avoid crowds in Hong Kong, one of the most densely populated cities on earth, despite the Hong Kong government recommending this in order to limit the pandemic [51] . However, the relatively mild impact of A/H1N1 meant that people saw no reason to jeopardize their economic well-being and curtail other social activities, given such a low perceived threat [49, 52] . Hand washing was probably seen as sufficient protection. The association between Trust in Formal information and Understanding of influenza cause differed by gender but not age, with females showing a stronger association. Men tend to have poorer health knowledge than women [53] . We found that females were more likely to wash their hands than were males. Older respondents reported significantly greater trust in formal information, marginally-significantly better understanding of influenza cause and were more likely to wash their hands. This is consistent with other studies reflecting that preventive practice is enabled by knowledge of causes [49, 54] . However, increasing knowledge is not itself sufficient to always ensure preventive behaviour [55] . In this context, Understanding has an independent contribution to hand washing practice only. Trust in Informal information seems to be associated with less perceived susceptibility to health threat. This may reflect rational processes or cognitive bias. Trusting social cues involves comparison and conformity influences, and can enhance optimistic bias (the tendency to view oneself less likely to experience negative events but more likely to experience positive events) in personal risk estimates [56] , thereby reducing perceived Susceptibility. Conversely, others' behavioural cues about health threat proximity can arouse motivating worry and anxiety producing protective action [17, 29] . We found Trust in informal information was independent of both Understanding of influenza cause and Self-efficacy. However, when stratified by gender, the Trust in informal information-Understanding association was positive among males but negative among females. Education is probably an important influence in understanding and may have a bearing on these patterns which await clarification. Susceptibility was strongly associated with both Self-efficacy (negatively) and Worry (positively). Neither Worry nor Susceptibility varied significantly by gender or age group. This is plausible and theoretically consistent [26, 34, 35, 39] . Worry was strongly associated with social distancing, again consistent with British data [49] . Although Worry was also significantly associated with hand hygiene, the association was weak. Elsewhere, using a generic measure of personal hygiene practices we have found a stronger association between disease worry and hygiene, suggesting a moderate effect of level of disease worry [57] . The model tested explained only a modest proportion of the variance in adoption of HBPs, suggesting that there are significant theoretical gaps that remain to be filled. These await further research. Social distancing is unassociated with formal HPB messages, suggesting potential susceptibility to a ''herd-like'' response in this Chinese community, particularly if confidence in formal (government or doctors) information is low. Voerten and colleagues describe such a pattern of response in the early stages of SARS [25] . These models support the hypothesis that social distancing is more likely to occur when perceived health threat is high [25] . Logically, when others seem to be behaving in a way that is informed and probably consistent then their actions provide clear information. If mixed social messages occur signalling uncertainty then the utility of social information will fall. This is likely to be associated with increase perceived susceptibility, and possibly greater worry and distancing behaviour. This pattern of responses would be most likely early in a novel RID epidemic where disease characteristics and behaviour are often uncertain. High threat uncertainty then drives social avoidance of potentially high-risk others. High levels of worry are associated with greater social distancing. Around 50% of 997/14,297 (response rate 7%) British respondents agreed that social avoidance would minimize risk of A/H1N1 infection, and respondents reporting more anxiety were more likely to engage in preventive actions; severity and likelihood of infection were the most important determinants of preventive action [12] . Further research on social influences on HPB during epidemic and pandemic RIDs is warranted. Providing more knowledge about disease causes can improve hand hygiene but is unlikely to influence social avoidance, which appears less amenable to formal health messages. However, as formal messages achieve acceptance across the population, and uptake of HPBs increases, then under circumstances where a critical mass of the population are practicing precautions trust in informal information should increase, reducing susceptibility and worry and leading to declines in social avoidance. Because others are likely adopting HPBs this makes them less of a contagion risk. Conversely maintaining a high level of hand washing practices may require sustained public education activities. Finally, different segments of the population probably communicate different types of information with their peers. Self-efficacy in preventing A/H1N1 influences hand hygiene but has little influence on social distancing. Formal health education messages that focus on enhancing the public's sense of their ability to protect themselves by adopting hygiene practices would seem to be the most effective to improve hand hygiene, but where the practice is already established, high levels of trust in these messages are not likely to significantly increase hand hygiene. This study is limited in being cross-sectional and relying on hypothesized modeling to infer causality. This is potentially errorprone and can only be confirmed by specific longitudinal tests of the hypotheses proposed above. There are potential limitations related to measurement imposed by the need to be parsimonious in questioning due to use of telephone interviews. Where this is not done refusal rates would have been unacceptably high [12] raising serious questions about representativeness. As a consequence, construct validity for some latent variables was weaker than expected, for example, only two items were scaled to measured trust in informal information giving a low internal consistency. We re-ran the SEM treating the two trust items as separate which gave almost identical associations with different situation awareness variables, so we entered their combined score as a latent variable in the final model. Only one item measured self-efficacy. This is generally not considered adequate but does have precedent indicating it is valid for predicting behavioral change [9] . Finally, this random sample, closely representative of the population of Hong Kong and collected early in the epidemic phase, nonetheless was slightly older and less-well educated than the general population. This was likely due to unavoidable sampling bias from surveying in the early evening to 10pm. Many young adults do not return home from work until after this time and were thereby not sampled. The results may in part reflect this bias. Otherwise the response rate was high at 69% and excellent compared to similar studies [12] . Some of these above limitations may also have contributed to the low explained variance of the model. Many factors influence RID protective behaviour. This study has examined a very limited number of these. Confidence in formal information such as health education messages is associated with greater compliance to recommended preventive measures for influenza A/H1N1 [12] . However, the mechanisms for this were unclear. We have shown that this probably involves different mechanisms for hand washing and social distancing, and suggest how these might function. Formal messages may not reduce social distancing behaviours until such time that preventive behaviours are widely adopted in the community. Social distancing seems more likely to occur when there is high influenza-related worry and uncertainty, such as in the initial stages when epidemic circumstances are unknown, or if an epidemic is severe and appears poorly controlled, as during early SARS. This would seem to be largely worry/affect-driven. If so, then social distancing is likely to occur irrespective of government messages as population anxiety about an epidemic increases. Susceptibility may also increase and this may inhibit self-efficacy regarding hand washing. Finally, high levels of community uncertainty or rumour are likely to increase distancing by exacerbating perceived susceptibility and worry. A simple version of our findings can be found it the supporting file (Text S1). Table S1 Found at: doi:10.1371/journal.pone.0013350.s001 (0.05 MB DOC) Text S1 This is a simple version of our study findings for nonspecialists. Found at: doi:10.1371/journal.pone.0013350.s005 (0.03 MB DOC)

Human Anti-Plague Monoclonal Antibodies Protect Mice from Yersinia pestis in a Bubonic Plague Model

Yersinia pestis is the etiologic agent of plague that has killed more than 200 million people throughout the recorded history of mankind. Antibiotics may provide little immediate relief to patients who have a high bacteremia or to patients infected with an antibiotic resistant strain of plague. Two virulent factors of Y. pestis are the capsid F1 protein and the low-calcium response (Lcr) V-protein or V-antigen that have been proven to be the targets for both active and passive immunization. There are mouse monoclonal antibodies (mAbs) against the F1- and V-antigens that can passively protect mice in a murine model of plague; however, there are no anti-Yersinia pestis monoclonal antibodies available for prophylactic or therapeutic treatment in humans. We identified one anti-F1-specific human mAb (m252) and two anti-V-specific human mAb (m253, m254) by panning a naïve phage-displayed Fab library against the F1- and V-antigens. The Fabs were converted to IgG1s and their binding and protective activities were evaluated. M252 bound weakly to peptides located at the F1 N-terminus where a protective mouse anti-F1 mAb also binds. M253 bound strongly to a V-antigen peptide indicating a linear epitope; m254 did not bind to any peptide from a panel of 53 peptides suggesting that its epitope may be conformational. M252 showed better protection than m253 and m254 against a Y, pestis challenge in a plague mouse model. A synergistic effect was observed when the three antibodies were combined. Incomplete to complete protection was achieved when m252 was given at different times post-challenge. These antibodies can be further studied to determine their potential as therapeutics or prophylactics in Y. pestis infection in humans.

Yersinia pestis (Y. pestis) is the causative agent of plague that has killed over an estimated 200 million people in previous pandemics [1] . The current incidence of plague is low but the animal reservoirs for Y. pestis exist worldwide. Sporadic cases have been reported recently with an average case number of 2,500 worldwide [2] . Y. pestis can be rendered airborne and its potential use as a bioweapon is recognized [3] as a category A agent on the NIAID list of biodefense-related pathogens. Current treatment for plague consists of antibiotics, while a live attenuated vaccine against plague is used in the former Soviet Union for prevention [4] . Nevertheless, these live attenuated whole-cell vaccines or killed whole-cell vaccines have adverse effects to varying degrees [4] . Though both types of treatment are efficacious, there is a need for an alternative treatment for plague [5] . A multiple-antibiotic-resistant isolate of Y. pestis has been isolated, and drug resistance was shown to be mediated by a self-transferable plasmid [6, 7] . A subunit vaccine, which consists of two virulent factors, the F1 protein and V-antigen, is currently in human clinical trials [8] [9] [10] . Studies involving the vaccine antigens in various formats have provided the proof-of-concept data that humoral response can be efficient in protection against Y. pestis [11, 12] . There are multiple reports that mouse anti-plague monoclonal antibodies (mAbs) against a Y. pestis challenge can passively protect a mouse against plague [13] [14] [15] . Therefore, mAb therapy may be an attractive alternative to the existing treatments for plague. Despite the promising possibilities, there remains a major hurdle in the treatment against plague and that is the possible immune response of humans to the mouse mAbs that are currently available. One possibility to ameliorate the immune response against the mouse mAb is to humanize the mAb for use in humans, or another alternative is to develop new and fully human anti-plague monoclonal antibodies for clinical usage [16] . We describe here the isolation of three mAbs from a large naive human phage-displayed Fab library. One, designated as m252, is against the F1-antigen and the other two (m253, m254) are against the V-antigen. When used alone, m252 displayed good protective effects, whereas m253 and m254 did not. However, a clear synergistic effect was found when they were used together. Maximum protection by m252 alone could be achieved by altering the antibody administration schedule. This is the first report describing the isolation of fully human anti-plague mAbs that show efficacy in a mouse model of plague. These antibodies represent a significant breakthrough toward possible adjunctive therapeutic treatment of Y. pestis infection in humans. With the F1 antigen, only the plate format yielded positive Fab clones after four rounds of selection with the F1 antigen. Sequencing of the clones confirmed that they were identical and designated as m252. With the V-antigen, the plate and bead format each yielded two positive Fab clones after four rounds of selection with each format. One clone from each format, designated as m253 and m254, respectively, was selected for further analysis. Sequence analysis revealed that m252 has heavy and light chains originated from germlines IGHV1-2*02 and IGKV1-16*01 respectively. M253 originated from IGHV1-18*01 and IGKV1-9*01, while m254 was from IGHV3-43*01 and IGKV1-27*01. The mutational rate ranged from zero to less than 10%. This is typical for antibodies isolated from naïve human libraries by panning against viruses causing acute infection in contrast to neutralizing antibodies selected from immune human libraries by panning against HIV-1, which causes a chronic infection [17] . Each of the clones was then transformed into HB2152 cells and the respective Fab was expressed and purified ( Figure 1a ). After conversion to IgG1 expressing clones, the three antibody clones were transiently transfected into Freestyle HEK 293F cells, and the expressed IgG1s were purified ( Figure 1c ). To determine both the specificity and affinity of the selected antibodies, ELISA with both Fab and IgG formats were conducted as described in the methods. All Fabs and IgGs bound to their respective antigens specifically without cross-reaction to other antigens tested (Figure 1b and d) . Anti-F1 Fab and IgG have apparent affinities in the low and sub-nM range, respectively. Both m253 and m254 Fabs have apparent affinities of approximately 100 nM (Figure 1b and d) . Their IgGs however have sub-nM apparent affinities (avidities). The avidity effect is very pronounced for all three antibodies. Low level of competition between the human anti-F1 and anti-V Fabs and mouse anti-F1 and anti-V mAbs The three human anti-plague Fabs were used in competition-ELISAs against a panel of mouse anti-plague mAbs. The mouse anti-plague mAbs included the anti-F1 mAb F1-04-A-G1, and five anti-V mAbs, which included the anti-V mAb 7.3 m that was highly protective. We found no apparent competition between the human anti-V m253 Fab and the mouse anti-V mAbs (Figure 2a ). However, we observed some weak competition between the human anti-V m254 Fab antibody and some of the mouse anti-V mAbs (7.3 m, 10-1 m, and 74-1 m) that we did not see with the human anti-V m253 Fab antibody (Figure 2b ). The competition between the human anti-F1 m252 Fab antibody and the mouse anti-F1 mAb was also minimal (Figure 2c ). We ran two controls in the competition studies with the human and mouse mAbs. For one control we did not add a primary antibody (labeled NC, Figure 2a -2c), and for a second control we used a nonspecific mouse isotype IgG1 mAb in the competition assay (labeled Bm, for a Burkholderia mallei IgG1 mAb). There was also a lack of competition between the human anti-V m253 and m254 Fab antibodies, suggesting that these two human anti-V Fabs recognize different epitopes (one which may be conformational) on the V-antigen (Figure 2d ). Of note, however, is that when the competition-ELISA was performed in a different fashion, namely when the human anti-F1 or anti-V mAbs were allowed to bind to the respective antigens before adding the mouse anti-F1 or anti-V mAbs, moderate competition was detected between the human anti-F1 m252 and the mouse anti-F1 mAbs, as well as between the human anti-V m253 and mouse anti-V 84-1 mAbs (data not shown). To characterize the binding of the human anti-F1 and anti-V mAbs to the F1-and V-antigens, respectively, more closely, we examined the binding of the human mAbs to two separate panels of overlapping peptides. One panel covered the full-length of the F1 antigen (27 peptides) and the other -the V antigen (53 peptides). For the human anti-F1 m252 mAb, there was a weakmoderate binding signal with peptides 1 and 2, which are located at the N-terminus of the F1-antigen (Figure 3a ). This suggests that the m252 mAb may also recognize a conformational region that involves peptides 1 and 2. The binding by human anti-V m253 mAb resulted in a strong signal with peptide 2 and a weak signal with peptide 1 (Figure 3b ). The anti-V m254 mAb, did not bind to any of the peptides, suggesting that its epitope may be conformational ( Figure 3c ). In initial studies with m254 we saw some weak binding to peptides 36 and 42, but when we repeated the binding studies with the human anti-V mAbs we did not see binding to peptides 36 and 42. This might explain its weak competition with mouse antibodies that recognize diverse epitopes on the V-antigen ( Figure 2b ). The positive signals seen with the m253 and m254 mAbs with V-antigen peptides (numbers 19, 20, 27, and 28) are nonspecific signals that are generated by the secondary antibody (Amemiya et al. unpublished). The epitopes of the mouse antibodies have also been determined by the peptide binding assay (Amemiya, et al. unpublished). The data is consistent with the competition-ELISA presented in this study. To test if the human anti-F1 and anti-V IgGs bind their respective targets on bacterial cells, we performed flow cytometry analysis. Both the mouse anti-F1 F1m mAb and human anti-F1 m252 mAb bound specifically to Y. pestis grown at 37uC and not to the same strain grown at 26uC. Neither did they bind to a control E. coli strain grown at 37uC (Figure 4 , bottom panel). This is consistent with previous reports that the expression of the F1-antigen is regulated by temperature (37uC), and it is not expressed at RT. The human anti-V m254 mAb showed minor binding to the Y. pestis grown at 37uC, although we do not normally see binding with the human anti-V m253 or mouse anti-V mAbs, which included 7.3 m, 74-1 m, and 141-1 m (Amemiya, unpublished). A mouse IgG1 isotype control mAb, which did not show any binding to the wholecells, was included to show binding by the mouse and human anti-F1 mAbs was antibody specific. To further confirm the binding data we used immunofluorescence technique. The results were consistent with the flow cytometry data, where only the mouse and human anti-F1 mAb bound to Y. pestis whole-cells ( Figure 5 ). Neither the human anti-V m254 mAb ( Figure 5D ) nor mouse anti-V mAbs, like 7.3 m (data not shown) nor control samples ( Figure 5A , nonspecific mouse IgG1 isotype; Figure 5E , no primary mAb) showed significant binding to Y. pestis. Human anti-F1 and V mAbs protect synergistically against a Y. pestis challenge in a bubonic plague model The ability of the human anti-F1 and anti-V mAbs to passively protect mice against a Y. pestis infection was evaluated in a bubonic plague model. The human anti-F1 and anti-V mAbs were used either separately or together in different combinations. When mAbs m252 and m253 and m254 were given to mice separately before challenge with Y. pestis CO92, only the human anti-F1(m252) mAb showed some efficacy. The mean-time-to-death (MTD) in the m252 mAb-treated mice was shifted to 13.0 days (1/ 6 survivors) when compared with mice given normal mouse serum (NMS), which had a MTD of 7.0 days (0/6 survivors) ( Figure 6A ). Unlike m252, however, the human anti-V mAbs, m253 and m254, did not show any significant protection [mean-time-to-death (MTD) of 6.7 days (0/6 survivors) and 7.3 days (0/6 survivors), respectively] when compared to the NMS -treated mice. The mouse anti-F1 (F1m) and anti-V (7.3 m) control mAbs both passively protected all (6/6) mice under the same challenge conditions (MTD of 21 days). When both human anti-V mAbs were given to mice passively ( Figure 6B ), no improvement in protection after challenge was observed (MTD of 6.8 days, 0/6 survivors), which was similar to that seen with the NMS-treated mice (MTD of 7.3 days, 0/6 survivors). However, when the human anti-F1 m252 mAb (MTD of 11.3 days, 2/6 survivors) was given together with the two human anti-V m253 and m254 mAbs, a greater number of mice were passively protected (MTD 14.0, 5/ 6 survivors) than when the antibodies were used separately, suggesting a synergistic effect. Although we saw some protection with the human m252 mAb, and a synergistic protective effect when the human m252 mAbs was given with the human m253 and m254 mAbs in the bubonic plague model, we wondered if there was any effect (positive or negative) with a nonspecific human IgG1 mAb by itself or when combined with the human anti-F1 m252 mAb in the number of survivors in the mouse model of plague. To answer this question, we injected five groups of mice with the following mAbs: one group of mice with only a nonspecific human IgG1 mAb (Hu-IgG1, 1500 mg); another group of mice with the mouse anti-F1 mAb (F1m, 500 mg); another group of mice with the human anti-F1 mAb (m252, 500 mg). Two other groups of mice were included, that were given either F1m (500 mg) or m252 (500 mg), with the nonspecific Hu-IgG1 mAb (1000 mg) one day before challenge with Y. pestis CO92 ( Figure 7 ). All mice in the group that received only the nonspecific Hu-IgG1 mAb died by day 8, which was similar to the control antibody mouse groups in Figure 6A and 6B. The same number of survivors was obtained (6/6) whether mice were given only mouse F1m mAb or F1m combined with the nonspecific Hu-IgG1. As we have seen previously ( Figure 6A ), only 1/6 mice survived in the group that received only human m252 mAb. When the human m252 mAb was combined with the nonspecific Hu-IgG1 mAb, we obtained one more survivior (2/6) than we obtained without the nonspecific Hu-IgG1 mAb. This variation in the number of survivors was not different than we saw previously ( Figure 6B ). In addition, the MTD was not affected by the presence of the nonspecific Hu-IgG1 (15.8 days) when given with the Hu-antiF1 mAb (15.8 days). These results suggest that the nonspecific Hu-IgG1 mAb had little effect on the survival of mice given the human m252 mAb or mouse F1m mAb. Delaying time of delivery of human anti-plague mAbs provided better protection against a plague challenge One possible reason we observed less protection with the human anti-F1 and anti-V mAbs in the mouse plague model was that the level of the human IgGs may not have been sustained in the mouse over time compared to the mouse IgG mAbs, We tested this hypothesis in two separate studies. We first examined the concentration of the human antibody in mice directly, by measuring the level of m252 (anti-F1) and m253 (anti-V) in serum after they were given the human mAbs by i.p. injection. Mouse sera were collected at different time points after the initial dosing and human IgG levels were monitored by direct ELISA. As seen in Figure 8A , the anti-F1 m252 mAb appeared to have a half-life of approximately 8 days, and the half-life of m253 mAb was approximately 10 days. After 21 days, the levels of these two human antibodies were undetectable. In contrast, the level of both the mouse anti-F1 (F1m) and anti-V (7.3 m) mAbs may have decreased initially like the human anti-plague mAbs, but after 21 days, the levels were still approximately 40-50% of the initial concentration ( Figure 8B ). The half-live of human IgG mAbs in mice reported here is similar to what was found in another study where human mAbs were used against another biothreat agent [18] . In contrast, a human IgG molecule would have an average serum half -life of 21 days in a human [19] . Because of these findings we then administered the anti-F1 m252 mAb at different time points relative to the time of challenge. While the original regimen provided consistently modest protection, administration of the human m252 mAb 24 and 48 hours postchallenge provided increasing protection with the 48 hours schedule provided complete protection ( Figure 9 ). Antibody administration at even later time points was not performed since mice began to die 3-4 days after challenge without any treatment. However, we did evaluate the effect of a second dose of antibody at a later time point. In this group, mice first received an initial dose of human anti-F1 m252 mAb 24 hour before challenge as was done with the earlier protocols. These mice then received a second dose of the human anti-F1 m252 mAb 5 days after challenge. There was an increase in both the number of survivors (5/6) and MTD (20 . Epitope mapping of the human anti-F1 and anti-V antibodies by peptide binding assay. A. Each of twenty-seven peptides that covered the full length of the F1-antigen were used to coat an ELISA plate (0.05 ml of a 25 mg/ml solution of each peptide), and binding by the human anti-F1 m252 mAb (0.05 ml of a 10 mg/ml solution) was analyzed. The sample labeled F1 was the full-length antigen used to coat the plate as the positive control (0.05 ml of a 2 mg/ml solution), and the sample labeled Media was the negative control with no primary antibody added. B and C. Each of fifty-two peptides that covered the full length of the V-antigen was used to coat an ELISA plate (0.05 ml of a 25 mg/ml solution), and the human anti-V m253 (B) and m254 (C) mAbs were used (0.05 ml of a 10 mg/ml solution), respectively, to analyze for binding. The sample labeled V was the positive control (0.05 ml of a 2 mg/ml solution), and the sample labeled Media was the negative control without the primary antibody. doi:10.1371/journal.pone.0013047.g003 days) approaching the efficacy displayed by a single dose administered 48 hour after challenge. These data suggest that the optimum serum concentration of the human IgG1s was critically dependent on the time of administration, and that the optimum concentration of the human anti-plague IgG1s in turn determined the outcome of the treatment protocol. Antibiotics have been at the forefront of combating bacterial infection for decades with great success. However, the develop-ment of new antibiotics is struggling to keep pace with the emergence of drug resistant bacterial strains, for example as in Y. pestis [6] . There has been an intense interest in developing antibody-based therapies as an alternative method of treatment [5] . Initially, antibody-based therapy was mostly limited to treating cancer or immune disorders. However, because of a better understanding of the pathogenesis of infectious agents, and the advancement in the development of protective or neutralizing antibodies, the use of antibody-based therapy against infectious agents has become more frequent [20] . In this report we described the first isolation of fully human mAbs against the Y. pestis Figure 6 . Human anti-F1and anti-V mAbs show synergistic protection when used together in the murine bubonic plague model. The human and mouse anti-F1 and anti-V mAbs were given i.p. to mice 24 hrs before parenteral challenge with Y. pestis CO92, and the number of surviving mice for each treatment group was monitored for 21 days after challenge. 6A. The following antibodies and amounts were used: normal mouse serum (NMS, 500 mg); mouse anti-F1 (F1m, 500 mg); mouse anti-V (7.3 m, 100 mg); human anti-F1 (m252, 500 mg); human anti-V (m253 and m254, 500 mg each). 6B. The following mAbs and amounts were used: NMS, 500 mg; F1m, 500 mg; m252, 500 mg; m253+ m254, 500 mg each; m252+ m253+ m254, 500 mg each. doi:10.1371/journal.pone.0013047.g006 virulence factors F1-and V-antigens. Previous studies have shown that mouse mAbs can be effective in protecting mice against Y. pestis ( [13] [14] [15] . However, because they are mouse mAbs they are not safe to use in their present form in humans [21] . Of particular concern is the immune reaction against mouse primary antibody sequences in human system. This may lead to severe adverse effects and at the same time reduce the potential benefits. It is highly desirable to have fully human antibodies for these reasons. The fully human anti-F1 (m252) reported here displayed moderate to good protection against a bubonic plague challenge with Y. pestis CO92. On the other hand, the two anti-V mAbs (m253, m254) when used separately did not show any efficacy, but when they were used together with m252, the combination of the human anti-plague mAbs resulted in better protection overall, suggesting a synergistic effect between the antibodies. A similar effect was reported in studies using mouse anti-F1 and anti-V mAbs in a mouse model of plague [15] . Further in our case with the human anti-F1 mAbs, when we gave mice the human anti-F1 mAb 1-2 days after challenge, we saw a greater protection against a plague challenge could be achieved. This suggested that the maintenance of serum concentration of the human mAbs in the mouse was possibly one critical factor for better protection. Kinetic studies revealed that indeed the serum concentration of the human antibodies dropped further than the mouse anti-plague mAbs over the course of the study. It is also plausible that the human antiplague mAbs might bind to other mouse antigens nonspecifically, thus decreasing the amount of free circulating human anti-plague antibody in the mouse. Figure 7 . A nonspecific human IgG1 antibody (Hu-IgG1) had little effect on the mean-time-to-death (MTD) and number of surviving mice that received the human anti-F1 mAb. The nonspecific human IgG1mAb, and the human and mouse anti-F1 and anti-V mAbs were given i.p. to mice 24 hrs before parenteral challenge with Y. pestis CO92, and the number of surviving mice for each treatment group was monitored for 21 days after challenge. Mouse or human mAbs and their amounts were given to the following groups of mice: nonspecific human IgG1 (Hu-IgG1, 1,500 mg); F1m (F1m, 500 mg); F1m (500 mg) + Hu-IgG1(1,000 mg); m252, 500 mg; m252 (500 mg) + Hu-IgG1 (1,000 mg). doi:10.1371/journal.pone.0013047.g007 Another important underlying factor for efficient protection by antibodies is the epitopes the antibodies recognize. Although the human anti-F1 (m252) mAb appeared to bind to the same region as the mouse antiF1, which was at the amino-terminal end of the F1antigen (Amemiya et al., unpublished), we could not demonstrate direct competition between these two antibody species. This observation may be the result of the nature of the antibody binding site or epitope, because several mouse mAbs isolated independently recognized the same region or epitope on the F1-antigen, behaved in the same manner (Amemiya et al., unpublished). It may be that once these mAbs bound to the amino-terminal end of the F1antigen, they may not readily come off the protein or may dissociate very slowly. Whether this is because the binding site involved both linear and conformational sites is not known, but both the mouse and human anti-F1 mAbs bound to the whole anti-F1 antigen very well, but only weakly -moderately to the 59-peptides. Nevertheless, the human m252 mAb was as protective as the mouse anti-F1 mAb when the human mAb was given after challenge. It has also been reported that a neutralizing epitope on the V-antigen was located in a region spanning amino acids 135 to 275, and a possible minor, secondary neutralizing epitope exists near the amino-terminal region of the V-antigen [14] . Neither of our human anti-V antibodies reported here competed with the mouse anti-V antibodies efficiently. The minor competition between the human anti-V Fab antibody and the mouse anti-V antibodies suggests that the recognition site of the human anti-V antibodies is slightly different or they may partially share conformational binding site. These differences might be one reason for their inability to protect as efficiently as the human anti-F1 m252 mAb. Exactly how the human anti-F1 252 m mAb is able to protect mice may be directly related to the presence of F1 antigen on the surface of the plague organism. The F1-antigen has been reported to be anti-phagocytic [22, 23] . The ability of macrophages to take up the plague organism is directly related to the lack of the F1antigen, and resistance to phagocytosis is related to the presence of the F1-antigen. The binding of the human anti-F1 252 m mAb to the surface of the plague bacilli or opsonization may trigger phagocytosis of encapsulated bacilli into macrophages, thereby allowing phagocytic cells to clear the host of the pathogen. The exact mechanism by how the V-antigen exerts it virulence is not completely known. There are reports showing that Vantigen is secreted into the growth medium and the secretion is important for virulence [24, 25] . The secretion of V-antigen in the medium has been described to be dependent on contact with the host cell, and could also be directed into the host cell by a Yersinia outer proteins (Yops) dependent secretion (Ysc) type III system (TTSS) [26] [27] [28] . It also has been suggested that free V-antigen may enter the cell by endocytosis besides being injected into the cell by the Ysc TTSS [29] . Once inside the host-cell, we do not know exactly what host proteins interact with the intracellular Vantigen [29] . Nevertheless, there is some evidence that anti-V antibodies enhance phagocytosis through possibly the Fc receptor, and thereby block Yop delivery into the host cell, and thus preventing Ysc dependent TTSS injection of V-antigen into the host cell [30, 31] . In this study, however, we were not able to detect binding of the human anti-V mAbs on the surface of Y. pestis cells by flow cytometry or fluorescent microscopy, suggesting that the Vantigen was not on the bacterial cell surface under the conditions used in our studies or the expression level of V-antigen was below the level of detection. As has been discussed previously, however, the presence of V-antigen on the surface of the cell may be dependent on contact with the eukaryotic host cell. The highly specific region of the neutralizing epitope (s) on the V-antigen suggests a possible ligand-receptor interaction between the Vantigen and a cellular factor. This indicates that perhaps the Vantigen exerts its biological effect through mechanisms other than mediating the TTSS pathway. In conclusion, the human anti-plague antibodies reported here represent perhaps the ones that are closest to practical clinical use. They may be safer and more efficient in the human system due to their fully human nature, and a likely longer half-life in humans. Also, intravenous application of the antibodies in humans may be a rapid delivery system that may augment antibiotics treatment in plague-exposed individuals. Finally, the affinity of all three antibodies can be further increased using readily available techniques, may reduce the dose required for efficient protection. The successful development of these three human anti-plague Figure 9 . Post-challenge administration of the human anti-F1 m252 mAb conferred better protection. The human anti-F1 m252 mAb was administered before or after Y. pestis challenge, and mice were monitored for 21 days after challenge. The mouse anti-V 7.3m mAb (100 mg) was used as a positive control mAb. Normal human serum (NHS, 500 mg), which was used as a negative control, had only 5 mice per group. The numbers behind each antibody represent the time in days in which the antibody was administered (500 mg) to mice relative to the day of challenge (day 0). The two numbers after the human anti-F1 m252 (21 and +5) represent two different days when the mAb (500 mg) was added to the same group of mice relative to the day of challenge. doi:10.1371/journal.pone.0013047.g009 antibodies in this model suggests that new and more potent anti-V antibodies can be potentially developed using the same approach but with restricted V-antigen subunit fragments containing the critical neutralizing epitopes, and perhaps other virulent factors. The Y. pestis CO92 strain used in the challenge studies was originally obtained from T. Quan, Centers for Disease Control and Prevention, Fort Collins, Co. It was isolated from the sputum of a human case of pneumonic plague [32] . The Y. pestis CO92 was grown and inoculum prepared for challenges essentially as described previously [33] . A Y. pestis pgmstrain, which was originally isolated from Y. pestis CO92, and used in the antibody binding studies described below was obtained from Susan L. Welkos (USAMRIID, Frederick, MD). Y. pestis purified F1,V, and F1-V [34] protein antigens were obtained from Brad Powell (USAMRIID, Fort Detrick, Frederick, MD). The 27-peptide array that covered the F1-antigen were 14to 17-mers with 11 amino acid overlaps; they were obtained from the Biodefense and Emerging Infections Research Resources Repository (BEI)(Manassas, VA). The 53-peptide array that covered the V-antigen were 15-to 17-mers with 11 or 12 amino acid overlaps and were obtained from BEI. The anti-F1 mouse mAb F1-04-A-G1 (or mF1) was provided by George Anderson (USAMRIID) [13] , and anti-V mouse mAbs 10-1 m, 74-1 m, 84-1 m, and 141-1 m as well as the control IgG1 mouse anti-Burkholderia mallei (Bm) antibody were obtained from Sylvia Trevino (USAMRIID) and anti-V mouse mAb 7.3 (7.3 m) was obtained from Jim Hill (Porton Down, Wiltshire, UK) [14] and used in competition ELISAs and as positive controls in mouse passive protection experiments. All mice mAbs were IgG1 isotypes. The human IgG1 control mAb used in the passive protection study was an anti-human IGFII mAb. Purified F1-and V-proteins were either coated directly to Maxisorp plates (Nunc, Denmark) in PBS buffer at 4uC, overnight for plate format panning or were biotin-labeled first with EZ-link Sulfo-NHS-LC-Biotin (Pierce, Rockford, IL) for streptavidinconjugated magnetic bead format panning. The labeling was performed according to the manufacture's recommended protocol. For the plate format, approximately 10 12 Fabs displayed on the surface of phage amplified from a large naive library [35] were suspended in PBS with 2% dry milk and applied to wells coated with the F1-or V-proteins. After incubating for 2 hours at room temperature, each well was washed 5 times for the first round and 10 times for the subsequent four rounds before the phage were rescued with TG1 cells at the exponential growth phase. For the bead format, biotin-labeled F1-and V-antigens were first incubated with the same amount of phage as in the plate format in 1 ml of PBS+2% dry milk suspension at room temperature for 2 hour. Fifteen ml of Dynabeads MyOne Streptavidin T1(Invitrogen Dynal AS, Oslo, Norway) pre-blocked with PBS+2% dry milk was then added to the antigen/phage mixture for one hour at room temperature. The beads were then washed 5 times with PBS for the first round and 10 times for the subsequent four rounds of selection. Phage were then rescued with TG1 cells. A total of four rounds were performed for each antigen with each format. Monoclonal ELISA was then performed to select for positive clones. One hundred clones were screened for each antigen from each format. Only clones displaying an OD405.2.0 were selected for plasmid preparation and sequencing. Expression, purification, conversion to IgG1, and generation of stable clones Clones selected as described above were transformed into E.coli strain HB2151 for expression [36] . Briefly, a single clone was inoculated into 2YT supplemented with 100 units of ampicillin and 0.2% glucose and incubated at 37uC with shaking. When the OD600 reached 0.6-0.9, IPTG was added to achieve a final concentration of 1 mM and the culture was shifted to 30uC with shaking and incubated overnight. Cells were then collected, and lysed with polymyxin B (Sigma, St Louis) in PBS, and mixture subjected to Ni-NTA agarose bead (Qiagen, Hilden, Germany) purification. For IgG1 production, the heavy and light chains of the respective Fabs were cloned into the bi-cistronic expression vector pDR12 kindly provided by Dennis Burton (Scripps Research Institute, La Jolla, CA). For small scale IgG1 production, transient transfection and expression in Freestyle HEK 293F cells (Invitrogen, Carlsbad, CA) were used. For large scale production, stable clones were generated using CHO-K1 (ATCC, Manassas, VA) cells. Briefly, the heavy and light chains of the three human anti-plague IgG1s were cloned into pDR12 vectors and transfected into CHO-K1 cells. One day after transfection, the cells were replated and subjected to selection in GMEM medium supplemented with 25 mM MSX. Two weeks later, the MSX resistant clones were amplified further. The clones were tested for the expression of respective IgG1s and then adapted to growth in serum-free medium HyQSFM4CHO (HyClone, Logan, UT) supplemented with 30 mM MSX. The serum-free growth medium was then collected and passed through a protein A-sepharose resin column for IgG1 purification. An ELISA assay was used to assess the binding ability of the Fabs and IgG1s. Briefly, F1-and V-antigens were coated to a Costar high binding 96-well plate (Corning, Corning, NY) and incubated overnight at 4uC. The next day, the plate was blocked with 2% dry milk in PBS before serial dilution of Fabs or IgGs were applied to the plate. After an incubation of the plate at 37uC for one hour, anti-His-horse radish peroxidase (HRP) for Fab detection or anti-human-Fc-HRP (for IgG detection) in PBS+2% dry milk was added to each plate and incubated for another hour at 37uC. The plates were then washed four times, and the ABTS substrate (Roche, Mannheim, Germany) was added. After approximately 10 min at room temperature, the OD405 was taken. For competition studies between the mouse mAbs and human Fabs the antigen at 2 mg/ml (F1-protein or V-antigen) was used to coat 96-well plates (Immulon 2HB, Thermo Electron, Milford, MA), and the plates were incubated overnight at 4uC. After washing the plates, a blocking solution (1% bovine serum albumin with 0.05% Tween 20 in PBS) was added to the plates, and plates incubated for 1 hr at 37uC. The flag-labeled human Fabs, and biotinylated-mouse mAb were allowed to bind to the antigen simultaneously for 1 hr at 37uC. To detect the presence of the human Fab, an anti-flag-M2-peroxidase conjugate was used, and to detect the amount of biotinylated mouse mAb present, a streptavidin-conjugated HRP was added for 1 hr at 37uC, before adding a hydrogen peroxidase-3,39,5,59-tetramethylbenzidine solution. The color reaction was allowed to develop at room temperature for 15 min and read at 450 nm. For analysis of IgG1 binding to F1-or V-antigen peptides, the peptides were added to the plates in 0.05 ml per well at 25 mg/ml. and the plates incubated overnight at 4uC in Immunlon 2HB 96well plates. After the washing and blocking steps as described above, binding of the human and mouse mAbs IgGs was detected as described above. The binding of all mAbs to the peptides was evaluated at 10 mg/ml. Flow cytometry was used to analyze the binding of the human anti-F1 and anti-V antigen IgG1s to Y. pestis pgmwhole-cells. Y. pestis pgmwas first streaked onto a sheep-blood agar plate and grown at room temperature (RT) for 3 to 4 days until colonies were readily visible. A single colony was picked and inoculated into 10 ml of Heart Infusion Broth (Remel, Lenexa, KS) containing 0.2% xylose and 2.5 mM CaCl 2 , and cells grown overnight (o/n) at RT with rigorous shaking. The next day, an aliquot of the bacteria culture was shifted to 37uC, and the other remained at RT, and the cultures were allowed to continue for another 3 hours before the bacteria were collected and suspended in PBS. Ten mL of bacterial cells was mixed with 90 mL of Fc Block solution [PBS with 1% FCS and 10 mg/ml FcBlock (BD Bioscience)] to achieve a final density of 1610 7 cells/ml. Human or mouse IgGs were added to the bacterial cell suspension to a final concentration of 10 mg/ml. After incubating at 4uC for 30 min, the bacterial cells were collect by centrifugation and suspended in 1 ml of PBS, and cells washed twice. The bacterial cells were then suspended in the same Fc Block solution, and secondary antibodies, which included either goat anti-mouse IgG-FITC (Pierce, Rockford, IL) or goat anti-human IgG-FITC (Southern Biotech, Birmingham, AL) were added to the cells at a dilution of 1:100. After 30 min at 4uC, the cells were then washed three times with PBS and subjected immediately to FACS analysis using a FACSCalibur (BD Bioscience, San Diego, CA), after the cells were fixed with 4% paraformaldehyde. For detection of antibody binding to whole-cells by immunofluorescent microscopy, the growth of Y. pestis pgmand the sample preparation for binding by human or mouse anti-F1 or anti-V mAbs was identical as that for the FACS analysis, except a Nikon T-2000 fluorescent microscope (Nikon Instruments Inc., Melville, NY) was used for detection. Passive protection by human or mouse anti-F1 or anti-V mAbs and Y. pestis challenge studies Antibodies to be evaluated for their efficacy against a plague challenge were given intraperitoneal (i.p.)(500 mg per mouse, except when stated differently in the figure legend) to 6-10 week old BALB/c mice 24 h before they were challenged or at time points post-challenge as indicated in the figure legend. The challenge dose was prepared from frozen stocks of Y. pestis CO92 that were streaked on tryptose blood agar slants and incubated at 28uC for 48 h. After the incubation period, the slants were rinsed with 10 mM potassium phosphate buffer, pH 7.0, and cell density adjusted to the required density with the same buffer. Mice were given the challenge dose of LD 50 ,25-40, subcutaneously in 0.2 ml, where 1 LD 50 is equal to 1.9 cfu [37] and observed for at least 21 days. Research was conducted in compliance with the Animal Welfare Act and other federal statues and regulations relating to animals and experiments involving animals and adheres to principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 1996. The facility where this research was conducted is fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. The studies involving mice were approved by the IACUC at the U.S. Army Medical Research Institute of Infectious Diseases, animal protocol number AP-07-040. Half-life of human or mouse anti-F1 and anti-V mAbs given passively to mice An ELISA as described above was used to detect the amount of human or mouse anti-F1 and anti-V mAbs present in mice over time. Briefly, F1-V protein (2 mg/ml in 0.2 M carbonate buffer, pH 9.4)) was used to coat a 96-well plate (Immulon 2HB) overnight at 4uC before washing and blocking. Two-fold dilutions of mouse serum taken retro-orbitally after i.p. administration of the mAb were made in 1X PBS with 1% BSA and 0.05% tween-20, added to plates and incubated for 1 hr at 37uC before washing. The amount of human or mouse anti-F1 or anti-V mAb binding to the antigens was detected by the addition of goat-anti-human or goat-anti-mouse IgG conjugated to HRP (Southern Biotechnology). The results from 3 mice at each time point for each mAb was performed in triplicate and were reported as the mean of the reciprocal of the highest dilution giving a mean OD of at least 0.1, which is at least twice the standard deviation (SD).

Diagnostic value of triggering receptor expressed on myeloid cells-1 and C-reactive protein for patients with lung infiltrates: an observational study

BACKGROUND: Differential diagnosis of patients with lung infiltrates remains a challenge. Triggering receptor expressed on myeloid cells (TREM)-1 is a neutrophil and monocyte receptor up-regulated during infection. The aim of this study was to evaluate the diagnostic accuracy of TREM-1 and of C-reactive protein (CRP) from patients with lung infiltrates to discern community acquired lung infections. METHODS: 68 patients admitted to a medical ward with acute respiratory illness were enrolled in the study. Neutrophil and monocyte TREM-1 expression were measured by flow cytometry, sTREM-1 by an enzyme immunoassay and C-reactive protein by nephelometry. Clinical pulmonary infection score was recorded. RESULTS: 34 patients were diagnosed with bacterial community acquired pneumonia (group A) and 34 with non-bacterial pulmonary disease (group B). Median serum TREM-1 concentration was 102.09 pg/ml in group A and lower than 15.10 pg/ml (p < 0.0001) in group B. Mean±SE neutrophil TREM-1 expression was 4.67 ± 0.53 MFI in group A and 2.64 ± 0.25 MFI (p = 0.001) in group B. Monocyte TREM-1 expression was 4.2 ± 0.42 MFI in group A and 2.64 ± 0.35 MFI (p = 0.007) in group B and mean±SE CRP was 18.03 ± 2 mg/ml in group A and 7.1 ± 1.54 mg/ml (p < 0.001) in group B. A cut-off of 19.53 pg/ml of sTREM-1 with sensitivity 82.6% and specificity 63% to discriminate between infectious and non-infectious pulmonary infiltrates was found. sTREM-1 at admission greater than 180 pg/ml was accompanied with unfavourable outcome. CONCLUSION: TREM-1 myeloid expression and sTREM-1 are reliable markers of bacterial infection among patients with pulmonary infiltrates; sTREM-1 is a predictor of final outcome.

Early diagnosis of lung infections remains a challenge. There is no gold standard for diagnosing microbial infection as clinical and laboratory signs are neither sensitive nor specific enough, and microbiological studies often remain negative. The presence of a new infiltrate on plain chest radiograph is considered indicative for diagnosing pneumonia, especially when is supported by clinical and laboratory findings. However it is difficult to differentiate a chest infiltrate of bacterial origin from a chest infiltrate of non-bacterial origin solely based on radiological criteria [1] . The diagnosis of infection is not always clear in the acute setting in patients with respiratory tract disease and a surrogate marker of infection would be a major benefit in the diagnostic armamentarium. Many inflammatory mediators and acute phase reactants, like C-reactive protein (CRP) and procalcitonin, have been described as reliable markers of infection; however none are specific enough, since they are also increased in non-infectious inflammatory conditions [2] . Triggering receptor expressed on myeloid cells (TREM)-1 is a recently described receptor on neutrophils and monocytes. It behaves like a pattern recognition receptor (PRR) since its activation leads to the release of pro-inflammatory cytokines, namely of tumour necrosis factor-alpha (TNFα) and of interleukin (IL)-8. Although its ligand is still unknown, activation is mediated by bacteria and fungi [3, 4] . A soluble form of TREM-1, namely sTREM-1, is increased in the bronchoalveolar lavage (BAL) of patients with ventilator associated pneumonia (VAP) [5, 6] , and in the serum of patients with sepsis, with bacterial meningitis and with acute pancreatitis [7] [8] [9] [10] [11] [12] . This same soluble form of TREM-1 seems to be increased in patients bearing noninfectious processes like peptic ulcer, inflammatory bowel disease, viral infections, malignant pleural effusions and chronic obstructive pulmonary disease (COPD) but also among patients after cardiac surgery or cardiac arrest. Increase of sTREM-1 seems particular prominent when the latter non-infectious states are complicated with systemic inflammatory response syndrome (SIRS) without infection [13] [14] [15] [16] [17] [18] [19] . Several published studies yielded contradictory results for the diagnostic and prognostic usefulness of TREM-1 and of sTREM-1 for infections [5, 7, [20] [21] [22] . The created impression is that more data are necessary to yield definitive results for its usefulness as a diagnostic and prognostic marker of community acquired pneumonia (CAP). The aim of the present study was to define whether expression of TREM-1 on cell membranes of neutrophils (nTREM-1), of monocytes (mTREM-1) and serum sTREM-1 may help in the diagnosis of acute bacterial infections for patients admitted with a new pulmonary infiltrate or pleural effusion. In this observation trial, all consecutive admissions to the Department of Critical Care and Pulmonary Services on predetermined and randomly selected emergency duty days were eligible. Inclusion criteria were: i) age above 16 yrs, ii) written informed consent; iii) acute respiratory illness and iii) presence of new pulmonary infiltrates or pleural effusion on chest x-ray or lung computed tomography. Exclusion criteria were: i) Human immunodeficiency virus (HIV) infection, ii) documented extrapulmonary infection, iii) neutropenia; and iv) oral intake of corticosteroids defined as any more than 1 mg/kg of prednisone for more than 1 month. The study protocol was approved by the Ethics Committee of the hospital and written informed consent was obtained from all patients within the first 12 hrs after admission. Clinical, laboratory, and imaging data were recorded for each patient including: i) clinical presentation; ii) body temperature, iii) arterial blood gas, iv) peripheral blood cell counts, v) gram stains and cultures of all biological fluids obtained (blood, sputum, bronchial secretions, BAL, and pleural fluid); vi) imaging findings, vii) antigen serology (Legionella spp and Streptococcus pneumonia urinary antigen, serological testing for Legionella pneumophila, Mycoplasma pneumoniae, Chlamydia pneumoniae) and viii) in-hospital mortality. The severity of illness was assessed by calculating Acute Physiology and Chronic Health Evaluation (APACHE) II, Sequential Organ Failure Assessment (SOFA) and Clinical Pulmonary Infection (CPIS) scores at admission [23] . A diagnosis of community-acquired pneumonia (CAP) was established in any patient presenting with a combination of fever, cough and purulent sputum, shortness of breath, chest pain, and new consolidation on chest X-ray or computed tomography. The severity of pneumonia was assessed the first 24 hours of admission according to Confusion, Urea nitrogen, Respiratory rate, Blood pressure (CURB) index. Patients having two or more criteria were identified to have severe pneumonia [24] . Sepsis, severe sepsis and septic shock were defined according to current recommendations [25] . Pneumonia was considered to be absent when: i) an alternative cause for pulmonary infiltrate was established (e.g. pulmonary embolus) and ii) full recovery was achieved without antimicrobial therapy. Pulmonary embolism was diagnosed according to current recommendations [26] . Lung cancer was ruled out based on histology and/or cytology specimens. Congestive heart failure was diagnosed according to American Heart Association [27] , and interstitial lung disease according to American Thoracic Society guidelines [28] . All cases were evaluated by two clinicians blinded to TREM-1 and sTREM-1 results. Agreement about the diagnosis was achieved in all cases. Patients with CAP were classified as having bacterial respiratory infection (group A). All other patients were classified as having non-bacterial respiratory disorders (group B). All patients assigned to group B were subject to chest computed tomography. For the measurement of sTREM-1, mTREM-1, nTREM-1 and CRP 10 ml of peripheral venous blood were sampled after venipuncture of the antecubitul vein under sterile conditions on the day of admission and on days 3 and 7 of hospitalization. Seven ml were centrifuged and serum was stored in -80°C until assayed for sTREM-1. Three ml were collected into EDTA-coated tubes (Vacutainer, BD) for estimation of nTREM-1 and mTREM-1 expression. Briefly, red blood cells were lysed by ammonium chloride. White blood cells were labelled by phycoerythrin-conjugated anti-TREM-1 monoclonal antibodies (R&D InC, Minneapolis, USA) for 30 minutes in the dark. nTREM-1 and mTREM-1 expression were assessed after passage of labelled cells through a flow cytometer (Epics XL/MSL, Beckman-Coulter Co, Miami Florida) and expressed as the mean fluorescence intensity (MFI) with gating for neutrophils and for monocytes by their characteristic FS/SS scattering. Determination of sTREM-1 was performed in duplicate by a developmental enzyme-linked immunoabsorbent assay according to the instructions of the manufacturer (R&D Inc, Minneapolis, USA). The lower detection limit and inter-day variation of the assay were 15.1 pg/ml and 5.23% respectively. Measurement of serum CRP was performed by an immunoturbidimetric assay on Roche automated clinical chemistry analyzers and was expressed in mg/ml. CRP was used as a comparator due to its universal application in all studies of evaluation of biomarkers. Asumming that measured parameters between groups A and B differed by 50%, it was calculated that 30 to 40 patients should be assigned into each group to yield a difference at the 5% level with 80% power. Values for nTREM-1, mTREM-1 and CRP are presented as mean ±SE; those of sTREM-1 are presented as medians and 95% confidence intervals (CI) or interquartile range (IQR). Comparisons between groups for nTREM-1, mTREM-1 expression and for CRP were done by ANOVA, followed by the Tukey's test for multiple comparisons. Comparisons of sTREM-1 between groups were done by Mann-Whitney U test after Bonferroni corrections for multiple comparisons. Comparisons of sTREM-1 between consecutive days within one group were done by Wilcoxon's signed rank test. Receiver Operator Curves (ROC) were designed to asses sensitivity, specificity, positive and negative predictive values for the estimated parameters to disclose infectious from non-infectious infiltrates. Patients were divided into two categories according to serum levels of sTREM-1 upon admission: those with sTREM-1 below or equal to 180 pg/ml; and those with serum sTREM-1 greater than 180 pg/ml. This concentration has been proposed as a threshold defining final prognosis in septic populations [22, 29] . Since CAP is a common cause of sepsis, this threshold was considered of merit. Survival was assessed by Kaplan-Meier and comparisons were done by log-rank test. Correlations between severity scores and measured parameters were done according to Spearman. Probability values less than 0.05 were considered statistically significant. All statistics and graphs were done using the Statistical Package for the Social Sciences software version 17.0.0 (SPSS Inc, Chicago, IL). The study flow-chart is shown in Figure 1 . Demographic and clinical data of the patients are summarized in Table 1 . Patients suffering from tuberculosis and enrolled in group B were presented with pleuritis. Group A (n = 34) consisted of patients with community acquired pneumonia (CAP) likely to be caused by extracellural bacteria. Seventeen had microbiological evidence of pulmonary infection, with isolation of the offending pathogens from sputum, blood or BAL samples (when bronchoscopy was performed). Seventeen patients were diagnosed with CAP on the basis of typical clinical and radiological presentation and good response to antibiotic therapy. Main radiological Group B (n = 34) consisted of patients with non-bacterial respiratory disorders. Diagnoses were: lung cancer (13 patients); pulmonary embolism (six patients); interstitial lung disease (six patients); heart failure (n = 5); pulmonary tuberculosis (two patients); rheumatoid pleuritis (one patient); and Q-fever (one patient). Main radiological findings were: right pulmonary infiltrate (six patients); left pulmonary infiltrate (three patients); bilateral pulmonary infiltrates (11 patients); right pleural effusion (four patients); left pleural effusion (one patient); both right lung infiltrate and right pleural effusion (four patients); both left lung infiltrates and left pleural effusion (two patients); bilateral pulmonary infiltrates and left pleural effusion (one patient); and left pulmonary infiltrate and bilateral pleural effusions (one patient). Among patients from group A with CAP nine (n = 9) died; six patients were admitted to the ICU and three were not admitted to the ICU due to relatives' denial. Mean age of patients not admitted to ICU was 80 years; the first two patients had a case-history of stroke and chronic heart failure; the third patient had a case-history of lung cancer. All three died from severe sepsis and multiorgan dysfunction syndrome (MODS). Mean age of patients admitted to ICU was 70 years; two patients had a case-history of aortic valve stenosis; two patients were under chronic intake of receiving corticosteroids; the fifth patient suffered from end-stage renal disease; and the sixth patient was suffering from hepatic failure due to alcohol intake. All six patients died from severe sepsis and multiorgan dysfunction syndrome (MODS). All patients in the ICU accomplished the clinical and radiological criteria for acute respiratory distress syndrome (ARDS) and were ventilated with the strategy of low tidal volume ventilation, according to current guidelines [30] , with volume limited mode ventilation, low tidal volumes (about 6 ml/kg ideal body weight), a maximum of 25-30 breaths per minute, high positive end-expiratory pressure (PEEP 10 cmH 2 O) and a goal plateau airway pressure < 30 cmH 2 O. Among patients admitted in the ICU, two died on the second day post-admission; one died on the third day post-admission; one on the seventh day post-admission; one the eighth day postadmission; and one on the twentieth day post admission. Concentrations of sTREM-1 and of CRP in sera of both groups and expression of nTREM-1 and mTREM-1 are given in Table 2 . All four parameters were significantly greater in group A than group B. ROC of sTREM-1, nTREM-1, m-TREM-1 and CRP to differentiate whether a chest X-ray infiltrate is due to CAP or to a non-infectious process is shown in Figure 2 . Area under curve (AUC) of sTREM-1 was 0.771 ± 0.068 (95%CI: 0.63 -0.9, p = 0.001). Sensitivity and specificity to diagnose between a pulmonary infiltrate of infectious origin and a pulmonary infiltrate of non-infectious origin were 82.6% and 63% respectively at concentrations above 19.53 pg/ml. AUC of nTREM-1 and mTREM-1 were 0.778 ± 0.063 (95%CI: 0.65 -0.9, p = 0.001) and 0.712 ± 0.07 (95%CI: 0.56 -0.86, p = 0.009) respectively. Sensitivity and specificity to diagnose between a pulmonary infiltrate of infectious origin and a pulmonary infiltrate of non-infectious origin were 78.3% and 58.6% for nTREM-1 above 2.55 MFI. Sensitivity and • Lung cancer (13) • Staphulococcus aureus (3) • Pulmonary embolism (6) • Haemophilus influenzae (2) • Congestive heart failure (5) • Pseudomonas aeruginosa (2) • Interstitial lung disease (6) • Other (4 specificity to diagnose between a pulmonary infiltrate of infectious origin and a pulmonary infiltrate of non-infectious origin were 82.6% and 75.9% respectively for mTREM-1 above 3.05 MFI. AUC of CRP was 0.789 ± 0.065(95%CI: 0.66 -0.9, p < 0.001). Sensitivity and specificity to diagnose between a pulmonary infiltrate of infectious origin and a pulmonary infiltrate of non-infectious origin were 78% and 76% respectively at concentrations above 8.7 mg/ml. Positive correlations were found between APACHE II scores and expression of TREM-1 on monocytes on day 1 (r s : +0.363, p: 0.010); and between APACHE II scores and sTREM-1 on day 1 (r s : +0.262, p: 0.043). No significant correlations were found between APACHE II scores and expression of TREM-1 on neutrophils on day 1 as well as between SOFA scores and any of the measured parameters on day 1. Correlations between serum levels of sTREM-1 and CRP and expression of TREM-1 on monocytes and neutrophils in relation to the identified causative pathogen of CAP are shown in Figure 3 . Serum levels of sTREM-1 were greater among patients with CAP caused by Gram (+) cocci and Haemophilus influenzae than among patients with CAP caused by other pathogens. Death occurred in three out of 17 patients were no pathogen was defined (17.6%); in nil out of three patients infected by atypical pathogens (0%); in three out of seven patients (42.9%) infected by Gram-negative bacteria; and in three out of nine patients (33.3%) infected by Gram-positive cocci or H. influenzae (p: 0.034 between grouping according to pathogen). Survival of patients with sTREM-1 on day 1 below or equal to 180 pg/ml was prolonged compared with patients with sTREM-1 on day 1 above 180 pg/ml ( Figure 4) . The results of the present study indicate that TREM-1 can be used as marker of bacterial infection in patients with lung infiltrates. sTREM-1, nTREM-1, mTREM-1 and CRP were comparable to their discriminating ability between a pulmonary infiltrate of infectious origin and a pulmonary infiltrate of non-infectious origin. sTREM-1 levels were decreased within the first 48 hours in patients with CAP with favourable outcome probably after the initiation of appropriate therapy followed by improvement of clinical symptoms. Finally, sTREM-1 levels above 180 pg/ml were an accurate independent predictor of in-hospital mortality from CAP. Discrimination of the infectious or non-infectious origin of a pulmonary infiltrate remains an everyday clinical problem. CPIS was introduced for that purpose helping considerable in cases of ventilator-associated pneumonia (VAP) [23] . TREM-1 is a surface receptor on cells of the myeloid lineage. Activation of TREM-1 leads to the production of pro-inflammatory cytokines [9, 31, 32] . Binding of its ligand is possibly linked to the activation of several transcription complexes that synergize with NF-B in order to elicit transcription of genes of pro-inflammatory cytokines [8] . sTREM-1 is the soluble counterpart of TREM-1 and it is probably shed in the systemic circulation from cell membranes of neutrophils and monocytes [7, 33, 34] . The physiologic role of sTREM-1 remains under question despite data support a probable anti-inflammatory role [31, 35] . TREM-1 has been studied in patients with pneumonia, especially VAP [5, 21, [36] [37] [38] . Few data are available on the diagnostic role of TREM-1 and of sTREM-1 in patients with lung infiltrates. Our data are in agreement with observations from the study by Phua [36] . Their proposed sTREM-1 cut-off point was 163 ng/ml, which is different than the one we found. This may be result from the different method of assaying sTREM-1 the used being Western blotting. The results of our study are in contrast to those of another study [37] that did not disclose any difference in nTREM-1 expression between patients with and without a bacterial lung infection probably due to the small number of patients included in that former study. El Sohl et al [38] reported elevated alveolar levels of sTREM-1 in pulmonary aspiration syndromes, but not in serum. However, serial plasma sTREM-1 levels were not obtained and the possibility that plasma levels might rise on subsequent days cannot be excluded. Two recent studies [39, 40] evaluated the diagnostic role of CPIS and of sTREM-1 in BAL fluid from patients with bilateral lung infiltrates in the intensive care unit (ICU). These studies reported controversial results. However authors did not measure sTREM-1 in serum on consecutive days. The reported results of the present study are the first to our knowledge that evaluate the diagnostic value of TREM-1 among patients with lung infiltrates to discriminate CAP. They also disclose a relationship between levels of circulating sTREM-1 and causative pathogens. More precisely, infections caused by Streptococcus pneumoniae, Sthaphylococcus aureus and Haemophilus influenzae were accompanied by greater levels of sTREM-1 and by greater expression of TREM-1 on neutrophils than infections caused by other pathogens. Although it may be hypothesized that Gram-positive cocci and H. influenzae are strong inducers of TREM-1 expression, it should be emphasized that TREM-1 is one PRR, the exact agonist of which remains to be found [29, 31] . A former study of our group [29] and another by Gibot et al [22] in heterogeneous populations of patients with severe sepsis of diverse aetiology investigated the role of early assessment of sTREM-1 as a determinant of final outcome. Results revealed that concentrations greater than 180 pg/ml are accompanied by survival benefit. The exactly opposing finding is reported here. This discrepancy may be explained by the enrolment of more homogeneous populations of patients, compared to these former studies [22, 29] , all suffering with CAP. Our study presents two main limitations: a) no documented cases of CAP by Legionella pneumophila, Mycoplasma pneumoniae, protozoa or parasites were enrolled in group A; b) mortality in the CAP patient group was high probably due to the existence of severe co-morbid conditions. In conclusion, the presented results indicate that serum sTREM-1 and expression of TREM-1 on neutrophils and monocytes may serve as markers of CAP in patients with pulmonary infiltrates. Concentrations of sTREM-1 in serum are particularly increased in CAP caused by Gram-positive cocci and Haemophilus species. The real clinical value of sTREM-1 assay comes when TREM-1 levels are low, allowing the clinician to withhold empiric antibiotics until culture results are available, and thus eliminating unnecessary antibiotic exposure to the patient. And finally, early serum levels of sTREM-1 greater than 180 pg/ml in CAP are associated with unfavourable prognosis. and Evangelos J. Giamarellos-Bourboulis Ass. Prof. MD have no conflicts of interest to disclose related to this study. Evangelos J. Giamarellos-Bourboulis Prof. MD has received reimbursement for attending the 29 th International Symposium on Intensive Care and Emergency Medicine where participated as a speaker and unrestricted educational grants from ABBOTT Hellas SA; Wyeth Hellas SA; Sanofi-Aventis Hellas SA.

Nodeomics: Pathogen Detection in Vertebrate Lymph Nodes Using Meta-Transcriptomics

The ongoing emergence of human infections originating from wildlife highlights the need for better knowledge of the microbial community in wildlife species where traditional diagnostic approaches are limited. Here we evaluate the microbial biota in healthy mule deer (Odocoileus hemionus) by analyses of lymph node meta-transcriptomes. cDNA libraries from five individuals and two pools of samples were prepared from retropharyngeal lymph node RNA enriched for polyadenylated RNA and sequenced using Roche-454 Life Sciences technology. Protein-coding and 16S ribosomal RNA (rRNA) sequences were taxonomically profiled using protein and rRNA specific databases. Representatives of all bacterial phyla were detected in the seven libraries based on protein-coding transcripts indicating that viable microbiota were present in lymph nodes. Residents of skin and rumen, and those ubiquitous in mule deer habitat dominated classifiable bacterial species. Based on detection of both rRNA and protein-coding transcripts, we identified two new proteobacterial species; a Helicobacter closely related to Helicobacter cetorum in the Helicobacter pylori/Helicobacter acinonychis complex and an Acinetobacter related to Acinetobacter schindleri. Among viruses, a novel gamma retrovirus and other members of the Poxviridae and Retroviridae were identified. We additionally evaluated bacterial diversity by amplicon sequencing the hypervariable V6 region of 16S rRNA and demonstrate that overall taxonomic diversity is higher with the meta-transcriptomic approach. These data provide the most complete picture to date of the microbial diversity within a wildlife host. Our research advances the use of meta-transcriptomics to study microbiota in wildlife tissues, which will facilitate detection of novel organisms with pathogenic potential to human and animals.

Information about the commensal and pathogenic microbial communities associated with host species, including humans, is limited. The endemic microbial community of a healthy host is important to characterize because its perturbation can be a cause of disease [1, 2] . Pathogenic microbes often escape detection if the clinical consequences of infection are similar to known pathogens or if they infect non-domestic species [3] . The maintenance of unknown pathogens in wildlife species is particularly problematic because many emerging human and livestock infections arise from contact with wild animals [4] [5] [6] [7] . With the advent of meta-genomics methods, the entire community of microorganisms that exist in a given environment can potentially be identified. Pyrosequencing and other high throughput sequencing approaches have been applied to determine the microbial population in environmental samples such as soil and seawater [8] [9] [10] [11] and more recently to investigate the community of microbes on human mucosal surfaces [12] [13] [14] [15] , both of which are rich in microorganisms. Next generation sequencing methods have also been successfully applied to identify the microbial agents of several new diseases [16] [17] [18] [19] [20] . Recently, RNA based meta-transcriptomic studies [21] [22] [23] , which profile both protein-coding transcripts and ribosomal RNA (rRNA), have been used to study both functional and structural features of environmental microbial communities. The key question behind this study was whether viable microorganisms could be detected within healthy mammalian lymphoid organs by employing massively parallel sequencing coupled with computational techniques able to detect transcripts of microorganisms among the abundant transcripts of the mule deer host. Lymph nodes are the specific replication sites for certain pathogenic viruses and bacteria [24] [25] [26] [27] [28] [29] . Moreover, although the blood and the lymph systems are considered to be essentially free of viable microorganisms in healthy individuals, the transient and often asymptomatic presence of both bacteria and viruses have been detected in the circulation [30, 31] . Phagocytic cells engulf these microbes and migrate to lymph nodes. Thus, lymph nodes should concentrate the commensal, endemic, and potential pathogenic microbial communities of a host species. We evaluated the microbial community in retropharyngeal lymph nodes of mule deer to assess microbial exposure via the oral or respiratory route. Because ungulates browse and receive small punctures from sharp forage, we reasoned that healthy animals would potentially be exposed to microorganisms from their environment or to resident oral and rumen microorganisms that would be cleared in draining nodes. We used mule deer to highlight the utility of this approach in a wildlife host, but the method is broadly applicable to any host species. Our studies document for the first time that there is a community of viable microorganisms in retropharyngeal lymph nodes of healthy wild ungulates. Furthermore, our findings demonstrate the applicability of meta-transcriptomic techniques for the detection of novel bacteria and viruses in internal organs. The microbial community of mule deer lymph nodes Detection of protein-coding and ribosomal RNA transcripts provides strong support for the presence of viable and replicating microorganisms. Therefore, we enriched the total RNA obtained from lymph nodes for poly(A) + RNA to prepare cDNA libraries and subjected them to pyrosequencing on a Roche GS FLX sequencer (Roche-454 Life Sciences). Properties of sequencing runs are given in Table S1 . All reads were compared against the nonredundant NCBI protein database. The composite metatranscriptomic species profile for five individual and two pools of 4 or 8 mule deer samples, determined using the software MEGAN [32] , is depicted in Fig. 1A . On average, 51% of total transcript-tags could be assigned to known taxa with a bit score cutoff of 50 (see Table S2 ). Of the assigned tags, 99.3% were of eukaryotic origin, predominantly matching to Bos taurus and other close relatives of mule deer that are represented in the protein database. Approximately 0.3% of the assigned tags were to bacteria. Proteobacteria represented 60% of all bacterial hits; Enterobacteriaceae in the Gammaproteobacteria were the most commonly identified within this group. Firmicutes and Actinobacteria represented 22% and 5% of the identified bacterial taxa. Table S3 lists all bacterial genera detected in the seven data sets. Transcripts assigned to Archaea, family Halobacteriaceae, were identified in both pooled samples but none of the individual libraries. Only 37 transcripts were assigned to viruses. Twenty-nine of these matched to the Retroviridae and Poxviridae while the remaining were to phages, insect viruses, and a single assignment to herpesvirus. These results suggest that representatives of many bacterial phyla, archaea, and two major virus families are transcriptionally active in mule deer retropharyngeal lymph nodes. Meta-genomics studies evaluating microbial rich communities were pioneered based on genomic DNA sequences [8] [9] [10] 13] . Thus, we compared genomic libraries prepared from retropharyngeal lymph node tissue of MD 72360, MD 80228, MD 84709, and MD 84730 with our data from transcript libraries derived from those animals (Fig. 1 ). Many sequences from the genomic DNA libraries were to non-coding regions and could not be used for taxonomic profiling (Fig 1B, Table S2 ). Based on proteincoding sequences, only four bacterial genera were identified in the comprehensive MEGAN analysis of the four genomic data sets. Xylella and Burkholderia were identified in MD 72360, Acidovorax was found in MD 84709, and Bartonella was found in both MD 84709 and MD 84730. Bartonella and Xylella, as well as a member of the beta retroviruses (found in MD 80228 and MD 84709), were identified only in the genomic DNA data, suggesting that they might not represent actively replicating organisms. These findings indicate that meta-transcriptomics may be the preferred method for detecting the viable endemic microbial community in the tissues of healthy animals. The most commonly detected microorganisms in the transcriptome libraries comprised intestinal and skin-dwelling bacteria and soil and freshwater bacteria. Ruminococcus, which is part of the commensal intestinal microbial community of ungulates, was detected in all seven libraries ( Fig. 1A and Table S3 ). Other bacteria found in at least three of the seven data sets were Propionibacterium, a commensal bacterium of skin and the gastrointestinal tract, and the environmental soil or water inhabitants Magnetospirillum, Streptomyces and Pseudomonas. Members of the latter genus are able to colonize a wide range of niches and are also potential pathogens. Other animal and human pathogenic genera detected in at least three different libraries were Burkholderia, Streptococcus, Flavobacteria, and members of the Enterobacteriaceae (Escherichia, Providencia). The overall bacterial diversity and the number of unique transcripts assigned to each bacterial taxon varied among the samples. Notably, Helicobacter was only detected in the library constructed from MD 257 but there were 12 unique transcript-tags assigned to this genus. More commonly, bacterial taxon identification was based on a single tag. Many of the single transcript-tags came from MD 80228, which had the highest bacterial diversity profile of all libraries analyzed, and from MD 84730. Bacterial genera detected solely in either one or both of these two samples include Acinetobacter, Legionella, Enterobacter, Salmonella, Yersinia, Vibrio, Listeria, Mannheimia, and members of the Corynebacterineae, all of which contain known pathogens. In addition, both specimens depicted by far the highest numbers of reads taxonomically assigned to the family Enterobacteriaceae. The lowest diversity of bacterial genera was found in the MD OCT-pool, which was derived from eight different mule deer. Pooling RNA from several animals potentially increases the representation of transcripts common to all animals but might decrease the ability to detect transcripts that are unique to one animal. Consistent with this, the MD Bonner-pool, which was derived from four animals, provided a broader spectrum of bacterial genera than the MD OCT-pool. Thus, pooling samples did not improve our ability to detect microbial diversity in lymph node samples. In contrast, viruses were detected in both pooled samples, although the total number of transcript-tags was low. Of the individual libraries, only MD 257 had evidence of viral transcripts (Fig. 1A) . The majority of viral transcripts were from a cervid poxvirus [33] , and a novel gamma retrovirus. The computational analysis described above identified putative microorganisms in mule deer tissue based on detection of proteinencoding transcriptional activity. Although the cDNA used in our analyses was derived from total RNA enriched for polyadenylated RNA, it retained a considerable amount of the abundant ribosomal RNA (rRNA). These sequences contribute to the 'no hits' category in Figure 1 and Table S2 . Bacterial rRNA derived from the same dataset can, therefore, be used to provide additional support for species identification. By classifying the rRNA-tags from each library using the RDP rRNA classifier tool [34, 35] (http://rdp.cme.msu.edu/) we increased the number of bacterial genera identified (see Fig. 2 for MD 257, Fig. S1 for MD 80228 and MD OCT-pool, and Table S4 ). Abiotrophia, which is a component of the normal oral and intestinal microbial commu-nity, was detected in six of the seven libraries; environmental bacteria such as Thermoanaerobacter, which is frequently found in hot springs, were detected in four of the seven samples. Other genera that were identified based on rRNA in at least two of the libraries were Actinomyces, Campylobacter and Mycoplasma. Of particular importance, rRNA-tags supported the presence of Helicobacter in the MD 257 library (Fig. 2) , of Acinetobacter, Escherichia, Pseudomonas, Salmonella, Shigella and Variovorax in the MD 80228 library, and of Shigella in the MD Bonner-pool library ( Fig. 1 and Fig. S1 , Tables S3 and S4). The support for Helicobacter in the MD 257 library was particularly compelling because there were 12 unique transcripttags and one rRNA-tag to this genus. We evaluated the phylogenetic relatedness of the mule deer Helicobacter with other Helicobacter based on four of the protein-coding transcripts and on the single 16S rRNA sequence. All analyses demonstrated that the Helicobacter detected in the mule deer lymph node is a unique organism that affiliates with the H. pylori cluster ( Fig. 3A and 3B, and Fig. S2 ). Because 16S rRNA sequence data is available for more species, we were able to further demonstrate that the closest Figure 1 . MEGAN comparison of the taxonomic profiles of (A) cDNA transcript-tags from 454 sequencing five individual lymph node samples and two lymph node sample pools and (B) genomic DNA-tags from four individual lymph node samples. Depicted are assignments with bit score cutoffs $50. Circle sizes are scaled logarithmically. Not assigned: sequencing-tags matching to sequences in the NCBI database that are not assigned to taxa; no hits: sequencing-tags not matching to any sequences in the NCBI database. doi:10.1371/journal.pone.0013432.g001 Figure 2 . MEGAN comparison of taxonomic profiles of MD 257 cDNA transcript-tags analyzed against the protein database (red) and the ribosomal database (blue), and of V6 amplicon 16S rRNA-tags analyzed against the ribosomal database (green). Bit score cutoff for the protein database comparison was set at 50, and confidence cutoffs for the ribosomal database comparisons were set at 80%. doi:10.1371/journal.pone.0013432.g002 relative to mule deer Helicobacter is a newly described H. cetorum isolated from different dolphin species (Lagenorhynchus acutus, Lagenorhynchus obliquidens, and Tursiops truncatus) and a beluga whale (Delphinapterus leucas) [36] (Fig. 3A) . We also evaluated the phylogenetic placement of Acinetobacter detected in MD 80228 based on both 16S rRNA and rpo-b sequences. The number of rpo-b sequences for Acinetobacter in the database is limited. However, we demonstrated that the MD 80228 transcript-tag clustered with those of Acinetobacter (Fig. 3D ) [37] . Moreover, based on 16S rRNA, we determined that the Acinetobacter species identified in the MD 80228 cDNA library was distinct from all known Acinetobacter and was most closely affiliated with Acinetobacter schindleri (Fig. 3C) . The low representation of viral sequences was not unexpected because viruses causing acute infections should be difficult to detect in healthy animals. Retroviruses integrate into the host genome as part of their replication cycle, thus transcription of viral genes can be persistent in infected animals. Overall four transcript-tags were assigned to gamma retroviruses of the family Retroviridae. Based on the transcript-tag from the MD Bonner-pool and an upstream region that is conserved in gamma retroviruses, PCR fragments were amplified and sequenced from MD 191 cDNA, which was used in the MD Bonner-pool, and from genomic DNA of MD 80228. These sequences were compared to other gamma retrovirus sequences using maximum likelihood methods (Fig. 4) . The mule deer gamma retrovirus forms a distinct clade within the gamma retroviruses, which has many well-described members of primate, murine, and feline origin. A newly described gamma retrovirus from killer whale (Orcinus orca) [38] is the closest relative of this mule deer retrovirus. The killer whale virus was described as an endogenous retrovirus based on its finding in various tissues and individuals. However, our detection of transcripts to this virus in only three of the libraries and the sequence variation in the PCR fragment between genomic (MD 80228) and transcript-derived (MD 191) mule deer samples suggest that both endogenous and exogenous gamma retroviruses might be present. As an alternative approach to identifying bacterial microorganisms present in lymph node tissue, we utilized amplicon DNA library sequencing technology. The hypervariable region V6 of the 16S rRNA gene was used because it has been reported to differentiate between many bacterial species [39] . Amplicon libraries of V6 were generated from the 454 cDNA libraries of MD 257, MD 80228, and MD OCT-pool and subjected to multiplex pyrosequencing on a Roche GS FLX sequencer (for properties of amplicon sequencing runs, see Table S1 ). The V6 amplicon rRNA tags were evaluated using the RDP classifier tool (Table S5) . The assigned bacterial genera cluster in the Gamma-and Betaproteobacteria, the Actinobacteria and in the order Bacilli. A comparison of the three methods used to detect bacteria in mule deer lymph node samples is shown for MD 257 in Figure 2 and for MD 80228 and MD OCT-pool in Figure S1 . Acinetobacter, Burkholderia, Corynebacterium, Escherichia, Providencia, Salmonella, Pseudomonas, Ralstonia, Staphylococcus, Streptococcus and Variovorax were identified by both amplicon and cDNA sequencing in MD 257, MD 80228, and/or MD OCT-pool (Tables S3 and S5) . Although the overall taxonomic diversity in the V6 rRNA amplicon libraries was lower than that detected in the cDNA transcript libraries, the diversity within bacterial classes was higher. Newly identified genera comprised predominantly environmental soil, sediment and water inhabitants (e.g. Aeromicrobium and Bdellovibrio), and the potential pathogens Stenotrophomonas, Rhodococcus, Rothia, and Gardnerella [40] [41] [42] [43] . These findings indicate that the V6 rRNA amplicon sequencing technology is a valuable tool in complementing information about the bacterial community in host tissues. Microbiome profiles of environmental samples and animals have mostly been based on the analysis of genomic DNA [8] [9] [10] 13, 44, 45] . Further, studies on the microbiomes of humans or animals have been restricted to habitats known to harbor large collections of microorganisms, in particular skin, oral cavity or gut [12] [13] [14] [15] [46] [47] [48] . In this study, we sought evidence for viable microorganisms in lymph nodes, an organ hitherto believed to be largely amicrobic in the absence of overt disease [26] [27] [28] [29] . Our data demonstrate that transcriptional activity of a variety of bacteria and a limited number of archaea and viruses, including novel organisms, can be confirmed in healthy animals using a meta-transcriptomic approach. In our study, we faced the computational challenge of detecting a rare microbial community in a dominant pool of host genetic material. We utilized transcript-based libraries because there is an amplification of protein-coding sequences during transcription, which increased our detection ability and provided support that the identified microorganisms were viable. Further, the database for protein-coding regions is more extensive than that for noncoding regions for non-reference organisms. Thus, focusing on transcripts should facilitate classification of novel organisms and those without complete genome coverage. Indeed, our study demonstrates that at the moderate sequencing depth employed, there were more assignable sequencing tags to protein-coding regions utilizing cDNA compared to genomic DNA, which consequently increased our ability to detect microbial taxa. In addition, transcriptome sequencing yields bacterial ribosomal RNA, which is highly expressed in metabolically active microorganisms and is well documented as a taxonomic tool for bacteria. Because single protein-coding or rRNA transcript-tags from a putative microorganism were frequently encountered, our confidence in taxonomic assignment increased by employing bioinformatics methods to classify organisms based on both types of transcripts. Amplicon 16S rRNA sequencing increased the sensitivity to detect members of some bacterial classes. However, primer specific methods do not provide as comprehensive a perspective on the microbiota due to a possible amplification bias towards more abundant taxa or those exhibiting higher primer specificity. Therefore, neither our metatranscriptomic nor amplicon sequencing approaches should be considered quantitative. We note that in samples that are highly enriched for actively replicating microbial organisms, such as environmental samples or gastrointestinal tract specimens, cDNA-based approaches can yield an abundance of small RNA produced by complex microbial communities, which can facilitate studies on microbial ecology but be less useful for identification of individual microbes [49] . In addition established metagenomics or metatranscriptomic [11, 50, 51] approaches that utilize sample fractionation methods for microbial enrichment will likely provide a more comprehensive profile of the community structure. These methods were not applicable to our samples, which included phagocytized microorganisms and viable microbes that were not robustly proliferating. Nevertheless, as deeper sequencing of cDNA libraries using newer high-throughput sequencing methods becomes more accessible, it could complement the Roche-454 pyrosequencing data, potentially covering the entire viable microbial community. Our study confirms that there are viable microorganisms in intact lymph nodes of apparently healthy mule deer. In the analyzed samples, we identified members of all bacterial phyla, as well as archaea, a DNA virus and a Retrovirus. The bacteria were representative of organisms that are commensal to mule deer and to their external environment. For example, we detected the common rumen and intestine dwellers, Ruminococcus and Abiotrophia, based on transcript-and rRNA-tags, respectively, in most libraries, indicating that commensal gut and mucosal microorganisms may routinely be sampled in secondary lymphoid tissue, presumably from transient bacteremia. Streptomyces was the most common soil dwelling bacteria identified. Of interest, Legionella, which is found near hot springs, was identified only in an individual mule deer from the Yellowstone region. The finding of a considerable number of archaeal transcripts in MD OCT-pool and MD Bonner-pool libraries implies that members of this domain of life are likely present in mule deer habitats or resident in mule deer gastrointestinal tracts, as has been recently documented in humans [52] . Correspondingly, environmental bacteria identified in healthy deer lymph nodes may reflect the animal's habitat. Few viruses were identified with our analysis methods. This could represent the paucity of viruses in healthy animals. However, viral detection may be more difficult than bacterial identification using this technology in part due to extensive sequence diversity among viruses in the same family. For example, we were only able to detect the gamma retrovirus because a transcript was present which was homologous to a conserved region of the viral env gene, and the cervid poxvirus was detected because sequence data for this virus was present in the database. Other persistent viruses, such as herpes viruses (for which we detected a single transcript), would be expected to be present in some animals. However, detection of latent herpes virus infection may be difficult because protein-coding transcript levels are low and latent viruses express non-coding RNA [53] . In addition, viral detection can be compromised if viral sequence tags were misassigned to the host organism because of homology of viral and host genes. Thus, many virus tags might be found among the host transcripts or in the not-assigned or no-hits groups of the MEGAN analysis, which together comprise nearly half of the total sequenced transcript-tags of our data. In addition to our finding of a novel gamma retrovirus, we also identified new species of Helicobacter and Acinetobacter. Phylogenetic evaluation of Helicobacter transcripts and 16S rRNA from the MD 257 cDNA library placed this new organism in the Helicobacter pylori/Helicobacter acinonychis/Helicobacter cetorum complex. All members of this complex have been associated with gastritis and peptic ulcer disease in humans and animals [36, [54] [55] [56] . Our detection of this bacterium in only one animal suggests that this Helicobacter is not a mule deer commensal. Of interest in this respect is the high incidence of H. pylori infections and gastric ulcers in American Indian populations from the same geographical area in central Montana [57] . Acinetobacter and Pseudomonas were identified in MD 80228 libraries based on all detection methods used (cDNA transcripts for protein-coding and rRNA, and amplicon rRNA). Phylogenetic evaluation of Acinetobacter transcripts and 16S rRNA from the MD 80228 cDNA library placed the respective reads in close relationship to Acinetobacter schindleri. Acinetobacter species are important environmental organisms, however they also are notable pathogens. In particular, Acinetobacter schindleri infections appear to be increasing in prevalence in hospitalized patients [37, 58] . Therefore, both of the newly identified bacteria are potential mule deer pathogens. In conclusion, our study demonstrates that endemic microbiota can be detected in lymph nodes of healthy animals using metatranscriptomic approaches. These results suggest that metatranscriptomic analyses of secondary lymphoid organs could be valuable in monitoring endemic infections in wildlife or livestock as well as in detecting novel infectious organisms with the potential for causing emerging zoonotic or epizootic infectious diseases. Further, these studies have the potential to cast new light on the diversity of life within and among individuals. Retropharyngeal lymph nodes were obtained from a total of seventeen individual Montana mule deer that were presented by hunters to check stations approximately 5 hr (range 2-11 hr) of being shot. Because our samples were obtained from legally killed animals, the study is exempt from Montana State University guidelines governing animal experimentation. Lymph nodes were dissected from animals with sterile scalpel and forceps, and rinsed in 70% ethanol. After dissection from the animal, the lymph nodes were either frozen directly or stored in RNAlater (Applied Biosystems, Ambion, CA) until further processing. Lymph node tissue was taken from mule deer in several geographical regions. The Bonner pool consisted of tissue from four mule deer (167, 191, 196, 200) Preparation of genomic DNA, total RNA, poly(A) + RNA and cDNA Lymph node tissue cores were dissected into small pieces and further disrupted, lysed and homogenized using a TissueLyser with steel beads (Qiagen, Germany). Genomic DNA was isolated from lymph nodes of four individual Mule deer (MD 72360, MD 80228, MD 84709, and MD 84730) using either the Genomic DNA Buffer Set with 20/G Genomic-tips (Qiagen, Germany) or the AllPrep DNA/RNA Mini Kit (Qiagen, Germany). Total RNA was isolated using the RNAqueous-Midi Kit (Applied Biosystems, Ambion, CA). For the Bonner-and OCT-pools, equal quantities of total RNA from lymph nodes of four or eight individual Mule Deer, respectively, were combined. Poly(A) + RNA was enriched from total RNA using the MicroPoly(A) Purist Kit (Applied Biosystems, Ambion, CA). Poly(A) + RNA (0.9-5.0 mg each) was used for cDNA synthesis (Just cDNA Double-Stranded cDNA Synthesis Kit, Stratagene, CA) after elimination of residual contaminating genomic DNA using the Turbo DNA-free Kit (Applied Biosystems, Ambion, CA). In one case we explored an alternative empirical approach to enrich for rare microbial transcripts, using total RNA of the MD OCT-pool. Reverse transcription and amplification of cDNA was done as described by Cheung and coworkers [59] and included a normalization step, which effectively decreased overexpressed reads. The data resulting from this approach are included in the MD OCT-pool data. Up to 5.0 mg of cDNA or genomic DNA was subjected directly to preparation of 454-DNA libraries and subsequently to pyrosequencing without any prior PCR or cloning steps. Library preparation and pyrosequencing were performed as described previously [60] on a Roche GS20 sequencer FLX (Roche Applied Sciences/454 Life Sciences, Branford, CT), producing sets of RNA-tags or DNA-tags, respectively. The runs were performed on either quarter or half plates, resulting in read numbers between 10,673 and 176,878 and base numbers in the range of 1,411,420 to 41,066,808. The MD OCT-pool cDNA library was run twice due to low read and base numbers of the first run, and the transcript-tags of these two runs and of the run following the normalization approach (see above) were combined for all subsequent data analysis. Sequences are deposited to the Sequence Read Archive (in progress). The data of individual 454 runs (and the compilation of normal and normalized MD OCT-pool data) was compared against the NCBI non-redundant protein database (BLASTX-nr) with an evalue of 1e -4 to identify transcript RNA-derived tags. To filter repetitive elements, RepeatMasker (http://www.repeatmasker. org) was used to scan the mule deer sequences, with the latest version of Repbase 13.04 [61] . The output files were analyzed with the program MEGAN [32] version 3.7.2. The 16S ribosomal RNA content of the cDNA pyrosequencing reads was analyzed by comparison to the ribosomal database of the Ribosomal Database Project (RDP) version 10 (http://rdp.cme. msu.edu/) [34] . The selected output reads were classified by the RDP Classifier tool (Naïve Bayesian rRNA Classifier Version 2.0) using the Taxonomic Outline of the Bacteria and Archaea, release 7.8, for the setup of the taxonomical hierarchy [35] . The output files were analyzed with MEGAN version 3.7.2 [32] . For the MD OCTpool, the combined data of three individual 454 runs was used. The cDNA from MD 191, which was used in the MD Bonner pool, and genomic DNA from MD 80228 were subjected to PCR using forward primer 5-ATGTGGGGGAGTTGATTCTTTT-TA and reverse primer 5-CTGCGCCTGAGTGGTCTACATA. PCR conditions were 40 cycles of 95uC for 30 sec, 56uC for 30 sec and 72uC for 90 sec. Fragments were gel isolated, cloned using the Stratagene PCR cloning kit (Stratagene, La Jolla, CA) and Sanger sequenced. Partial nucleotide sequences of 16S rRNA and rpo-b for Helicobacter and Acinetobacter and of flgK, GDP-D-mannose dehydratase and UDP-3-O- [3-hydroxymyristoyl] glucosamine Nacyltransferase for Helicobacter from cDNA sequencing, and of env gene for the retrovirus from a PCR product were aligned with the respective homologous sequences available in GenBank using the MEGA version 4 [56] software. The appropriate nucleotide substitution model for each data set was selected by the Akaike information criterion implemented in the Modeltest version 3.7 [62] , and maximum likelihood (ML) trees were reconstructed using PhyML version 2.4.4 [63] . Using the same program (PhyML) nodal supports were estimated with 100 bootstrap replicates. The trees were visualized in FigTree version 1.2.2 (http://tree.bio.ed.ac.uk/software/figtree/). Fusion-primers were designed including the sequences of the 454-Amplicon DNA library specific primers A and B, respectively, (GS FLX Amplicon DNA Library Preparation Method Manual, www.roche-applied-science.com), 4-base barcode sequences for identifying amplicon products derived from mule deer specimen MD 257, MD OCT-pool, and MD 80228 (TGCA, ACGT, and CGAT, respectively), and the ''universal'' V6-specific PCR primer sequences V6F: 59 TCGATGCAACGCGAAGAA 39 and V6R: 59 ACATTTCACAACACGAGCTGACGA 39 (designed to conserved regions flanking V6 based on comparison of 110 bacterial DNA sequences [39] ). The MD 257 template for amplicon generation was based on the total RNA fraction depleted of poly(A) + RNA (see ''Preparation of genomic DNA, total RNA, poly(A)+RNA and cDNA''). The supernatant was cleared of small RNA molecules using the MEGAclear Kit (Applied Biosystems, Ambion, CA) and depleted of host ribosomal RNA performing two cycles of the MICROBEnrich (Applied Biosystems, Ambion, CA) protocol. Subsequent depletion of bacterial ribosomal RNA yielded an RNA sample enriched for bacterial transcripts (MICROBExpress, Applied Biosystems, Ambion, CA), which was subjected to cDNA synthesis (Just cDNA Double-Stranded cDNA Synthesis Kit, Stratagene, CA) after elimination of residual contaminating genomic DNA using the Turbo DNA-free Kit (Applied Biosystems, Ambion, CA). Either cDNA derived from RNA enriched for non-polyadenylated bacterial mRNA (MD 257) or cDNA sequencing library samples derived from reverse transcribed poly(A) + RNA (for MD OCT-pool and MD 80228) were used as templates for the generation of 16S rRNA V6 hypervariable region-specific amplicons using the FastStart High Fidelity PCR System (Roche, Switzerland). PCR conditions were 50 cycles of 94uC for 30 sec, 55uC for 45 sec and 72uC for 45 sec. The yielded amplicon products were purified using AMPure, and the resulting individual amplicon DNA libraries were clonally amplified by multiplex emulsion PCR followed by sequencing using the GS FLX pyrosequencing platform. The sequencing output data were computationally divided into subsets according to the barcodes (and the corresponding mule deer sample) and the primers A or B. Figure S1 Comparative MEGAN analysis of (A) MD 80228 and (B) MD OCT-pool transcript-tags analyzed by comparison to the protein database (red) and the ribosomal database (blue), and of amplicon 16S rRNA-tags compared to the ribosomal database (green). Bit score cutoff for the protein database comparison was set at 50, and confidence cutoffs for the ribosomal database comparisons were set at 80% and 80%, respectively.

In Vitro and In Vivo Studies Identify Important Features of Dengue Virus pr-E Protein Interactions

Flaviviruses bud into the endoplasmic reticulum and are transported through the secretory pathway, where the mildly acidic environment triggers particle rearrangement and allows furin processing of the prM protein to pr and M. The peripheral pr peptide remains bound to virus at low pH and inhibits virus-membrane interaction. Upon exocytosis, the release of pr at neutral pH completes virus maturation to an infectious particle. Together this evidence suggests that pr may shield the flavivirus fusion protein E from the low pH environment of the exocytic pathway. Here we developed an in vitro system to reconstitute the interaction of dengue virus (DENV) pr with soluble truncated E proteins. At low pH recombinant pr bound to both monomeric and dimeric forms of E and blocked their membrane insertion. Exogenous pr interacted with mature infectious DENV and specifically inhibited virus fusion and infection. Alanine substitution of E H244, a highly conserved histidine residue in the pr-E interface, blocked pr-E interaction and reduced release of DENV virus-like particles. Folding, membrane insertion and trimerization of the H244A mutant E protein were preserved, and particle release could be partially rescued by neutralization of the low pH of the secretory pathway. Thus, pr acts to silence flavivirus fusion activity during virus secretion, and this function can be separated from the chaperone activity of prM. The sequence conservation of key residues involved in the flavivirus pr-E interaction suggests that this protein-protein interface may be a useful target for broad-spectrum inhibitors.

The emergence and resurgence of human viral pathogens can be traced to a complex variety of causes including increased urbanization, human contact with animal reservoirs, a decrease in effective public health systems, and the spread of insect vectors that disseminate some viral infections [1, 2, 3] . Flaviviruses are a genus in the Flaviviridae family and include important emerging and resurgent human pathogens such as dengue virus (DENV), West Nile virus (WNV), tick-borne encephalitis virus (TBEV) and yellow fever virus [2, 4] . Flaviviruses are transmitted by insects such as mosquitoes and ticks, and can cause severe human diseases characterized by encephalitis, meningitis, and hemorrhages [2, 3] . More than one third of the world's population lives in dengue fever endemic areas, and there are an estimated 50-100 million cases of dengue infection and 500,000 cases of the more lethal complication, dengue hemorrhagic fever, per year [5, 6, 7, 8] . There are currently no antiviral therapies for flaviviruses. DENV vaccine development is underway but is problematic due to the presence of four DENV serotypes and the potential for antibody-dependent enhancement of infection [2, 6, 9, 10] . Antiviral therapies could thus be an important alternative for DENV and for viruses such as WNV in which the cost and potential side effects of vaccination must be weighed against the relatively low number of human cases [2] . Flaviviruses are small, highly organized enveloped viruses with a spherical shape [4, 11] . They contain a positive-sense RNA genome packaged by the viral capsid protein. The nucleocapsid is surrounded by a lipid bilayer containing the viral membrane protein E. Flaviviruses infect cells by receptor engagement at the plasma membrane, endocytic uptake, and a membrane fusion reaction triggered by the low pH of the endosome compartment [12, 13] . The viral E protein binds the receptor and drives the fusion of the viral and endosome membranes to initiate virus infection. The pre-fusion structure of the E protein ectodomain (here referred to as E9) shows that E contains three domains composed primarily of b-sheets: a central domain I (DI) connecting on one side to the elongated domain II (DII) with the hydrophobic fusion loop at its tip, and connecting via a flexible linker on the other side to the immunoglobulin-like domain III (DIII) [14, 15, 16, 17, 18, 19] (Figs. 1A, S1). Although these regions are not present in the truncated E9 ectodomain, DIII connects to a stem domain and C-terminal membrane anchor (TM). The E protein in mature infectious flavivirus is organized in homodimers that lie tangential to the virus membrane [20] . Within each dimer the E proteins interact in a head to tail fashion, with the fusion loop of each E protein hidden in a hydrophobic pocket formed by DI and DIII of the dimeric E partner. The E protein mediates virus-membrane fusion by refolding to a hairpin-like E homotrimer with the fusion loops and TM domains at the same end [21, 22] . This reaction involves low pH-triggered dissociation of the homodimer, fusion loop insertion into the endosome membrane, formation of a core trimer composed of DI and DII, and the foldback of the DIII and stem regions towards the target membrane and their packing against the core trimer. The prefusion and postfusion conformations of the flavivirus E fusion protein are structurally and functionally similar to those of the E1 fusion protein from the alphavirus Semliki Forest virus (SFV) [23, 24, 25] , and these fusion proteins are often referred to as ''class II'' [26, 27, 28] . In addition to the ectodomains whose trimer structures are described above, truncated fusion proteins composed of domains I and II (DI/II) can reconstitute SFV and DENV core trimer formation on target membranes [29, 30] . Such core trimers act as specific targets for DIII binding, thus recapitulating the protein-protein interactions during class II trimerization and hairpin formation. Flaviviruses bud into the endoplasmic reticulum (ER) and are transported as virus particles through the secretory pathway and released by exocytosis [4] . Given the low pH that is present in the Golgi complex and trans-Golgi network (TGN) [31] , how do flaviviruses avoid inactivation during their transport? The particles are assembled in the ER as immature non-infectious viruses containing heterodimers of the precursor membrane protein (prM) and E protein [4, 26, 32] . Subsequent exposure to low pH in the secretory pathway triggers a dramatic rearrangement to E homodimers and makes the prM protein accessible to furin cleavage [33, 34] . Processing of prM by cellular furin results in mature infectious virus in which E homodimers are poised to mediate fusion [33] . Important recent studies describe the structure of pr peptide in complex with E, and indicate that processed pr remains associated with the virus at low pH and can inhibit virus-membrane interaction [34, 35, 36] . Thus, pr on the virus could protect E protein from low pH in the secretory pathway. The flavivirus prM/pr protein plays multiple roles in the virus life cycle (reviewed in [26] ). prM acts as a chaperone for E protein folding [37] and associates with the tip of E [34] . prM also appears to respond to low pH to permit E rearrangement on the virus surface and allow furin access for prM processing [34, 38] . Following cleavage, the pr peptide may prevent premature virus fusion through bridging interactions that stabilize the E homodimer and thereby prevent dissociation to E monomers, a key fusion intermediate [35, 36] . To better understand these multiple roles of prM/pr, separation of its chaperone and pH-protection functions and characterization of the pr-E interaction are needed. Here we developed a system to produce DENV pr peptide and reconstitute the pr-E interaction in vitro. At low pH pr bound to both monomeric and dimeric forms of E and blocked their membrane insertion and trimerization. Addition of exogenous pr to mature DENV particles inhibited virus fusion and infection. Mutation of a key histidine residue in the pr-E interface, E H244, reduced pr's binding and inhibitory activity, and reduced DENV secondary infection and particle production. The defect in particle Figure 1 . Expression and purification of DENV2 pr and truncated E proteins. A) Linear diagrams of the DENV2 prM-E proteins and the truncated DENV2 pr and E proteins used in this work (not to scale). Domain and construct boundaries are marked, with numbering based on the individual proteins in the DENV2 New Guinea C (NGC) strain. The sequences appended to the diagrams contain the Strep (ST) affinity tag(s) used for protein purification (underlined), joined in the case of two Strep tags by a flexible linker region (STST). Pr was expressed in 293T cells and contains prM residues 1-86 plus N-terminal GS residues from the vector and the STST tag. The DI/II and E9 proteins were expressed in S2 cells and contain E residues 1-291 and 1-395, respectively, plus ST or STST tags. DIII was expressed in E. coli and contains E residues 289-430, comprising the linker, DIII, helix 1 and conserved sequence (LDIIIH1CS). The names in parentheses are the detailed nomenclature from [30] . B) 4 mg samples of purified pr peptide were incubated with DTT, Endo H, or PNGase F as indicated, analyzed by SDS-PAGE and stained with Coomassie blue. The positions of marker proteins are shown on the right with their molecular masses listed in kilodaltons. Asterisks indicate the positions of the added glycosidases. C) 4 mg samples of purified truncated E proteins were reduced with DTT as indicated, analyzed by SDS-PAGE and stained with Coomassie blue. Marker proteins are shown on the right with their molecular masses listed in kilodaltons. doi:10.1371/journal.ppat.1001157.g001 Enveloped viruses infect cells by fusing their membrane with that of the host cell. Dengue virus (DENV) is an important human pathogen whose membrane fusion is triggered by low pH during virus entry into the cell. However, newly synthesized DENV must also transit through a low pH environment during virus exit. DENV is believed to escape premature fusion in the exit pathway via the small viral protein pr, which is processed and associates with virus after biosynthesis, and is released from the virus particle in the neutral pH extracellular environment. Here we have reconstituted the interaction of pr with the DENV fusion protein E using soluble protein components. The interaction has a low pH optimum and inhibits membrane insertion of the fusion protein. The recombinant pr peptide can ''add back'' to fully infectious mature DENV and block virus fusion and infection. We found that mutation of a critical conserved histidine on the fusion protein inhibits the interaction of E and pr, and makes the virus susceptible to low pH-induced inactivation during exit. This work characterizes the mechanism of pr protection, and suggests that the conserved multifunctional pr-E interaction may be an important target for antiviral strategies. production could be partially rescued by neutralization of exocytic low pH, indicating the important role of pr in protecting DENV from premature fusion during transport to the plasma membrane. A number of truncated E proteins have been successfully produced by co-expression with prM (e.g., references [30, 39] ), while the pr-E structural studies were based on a secreted hybrid protein containing truncated prM linked to truncated E [34] . Previous studies indicated that full-length TBEV prM could fold correctly when expressed in the absence of E protein [37] , suggesting that production of pr peptide alone might be possible. We generated a construct based on residues 1-86 of DENV2 prM, truncating pr just before the start of the furin cleavage recognition site at residue 87 (Fig. 1A) . This sequence was linked to a mammalian signal peptide at the N-terminus and to an affinity tag at the C-terminus, and expressed in 293T cells. The protein was isolated in a highly purified form by affinity chromatography and gel filtration (Fig. 1B) , and was recognized by mAb prM-6.1 against prM [40] (data not shown). The pr peptide migrated at a position of ,17 kDa in reducing SDS-PAGE, in keeping with its predicted size of 13 kDa plus the presence of carbohydrate due to the glycosylation site at position 69. This carbohydrate was removed by Peptide N-glycosidase F (PNGase F) to give a peptide of the predicted size. The protein was largely resistant to Endoglyosidase H (Endo H) digestion, indicating maturation of the carbohydrate chain as the protein transited through the Golgi complex. A mobility shift was observed upon reduction of pr, in keeping with the presence of 3 disulfide bonds in the structure of pr [34] . We also produced and purified a dimeric ectodomain form of DENV2 E protein containing all three domains (E9), a monomeric form containing E domains I and II (DI/II), and E domain III (DIII) (Fig. 1A and 1C ), all as previously described in detail [30, 41] . As a first test of in vitro pr-E binding, we coupled pr to sepharose beads and tested its ability to pull-down truncated E protein containing only domains I and II. This form of E protein is monomeric and the tip of DII is thus accessible even at neutral pH. Previous studies showed that this and other DENV DI/II proteins are active in membrane insertion and trimerization at both neutral and low pH [30] . We observed efficient pull-down of DI/II protein by pr-sepharose ( Fig. 2A) , but in spite of the accessibility of the pr binding site on DI/II at neutral pH, pulldown was low pH-dependent. The pull-down of DI/II protein by pr was specific, as it was blocked by inclusion of mAb 4G2 against the E fusion loop at the DII tip, and did not occur with BSAsepharose beads. These data suggested that the recombinant pr peptide could bind to the tip of DI/II in a low pH-dependent reaction. For more detailed studies of pr-E binding, we performed surface plasmon resonance (SPR) assays using our various forms of recombinant E protein with immobilized pr peptide. Compared to the pull-down assay, SPR can detect low levels of protein-protein interactions as binding is detected in real time and does not require removal of unbound E. The E9 protein is a dimer at neutral pH and dissociates to monomers at low pH [30] . When SPR was performed with E9 protein buffered at pH 8.0 there was very low binding (low signal response) (Fig. 2B ). As the buffer pH was Figure 2 . Pr peptide binds DENV E proteins in a pH-dependent manner. A) Pull-down of DI/II protein by pr. DI/II was incubated with sepharose beads conjugated with pr peptide or BSA at the indicated pH for 1 h at room temperature. As indicated, reactions contained a 2:1 molar excess of mAb 4G2 to the E fusion loop or mAb to the ST tag (con.). Input lanes show an aliquot representing 20% of the reaction prior to pull-down. (Panels B-D) SPR analysis of pr-E binding. Pr peptide was immobilized on a CM5 sensor chip, and DENV2 E9 (B), DI/II (C) or SFV DI/II proteins (D) were flowed over the chip at concentrations of 1.2 mM in buffers of the indicated pH for 300 s, followed by injection of protein-free buffer at the same pH. Data are a representative example of two independent experiments. doi:10.1371/journal.ppat.1001157.g002 decreased, the signal gradually increased, with maximal response observed at ,pH 6.25 and no further increase at pH 6.0. A rapid decrease in signal was observed when the samples were shifted to protein-free buffer, indicating rapid dissociation of the pr-E interaction. Similar results were obtained using monomeric DI/ II, with the lowest binding at pH 8.0, highest binding at pH 6.25, and a slight decrease at pH 6.0 (Fig. 2C) . Thus, the dimeric E9 and monomeric E DI/II proteins bound pr peptide with similar pHdependence. Binding to pr was specific, as little interaction was observed using the structurally similar E1 DI/II protein of SFV (Fig. 2D ). In addition, binding of DENV E DI/II protein to pr was inhibited by preincubation with mAb 4G2 against the fusion loop (molar ratio 1:1) (data not shown). Determination of the affinity of pr-E binding was not performed as the data did not fit to a simple Langmuir model of 1:1 binding, presumably because of E protein aggregation at low pH. Previous studies showed that retention of endogenous pr peptide on the furin-processed DENV particle inhibits virus interaction with liposomes at low pH [35] . Structural considerations suggested that this inhibition occurs primarily by blocking low pH-triggered dissociation of the E dimer, a required first step in the fusion reaction. To test this mechanism, we evaluated the effect of pr on the membrane interactions of dimeric and monomeric forms of E protein. The E9 dimer was preincubated with pr peptide or an unrelated protein with the same affinity tag for 5 min at pH 8.0, and then treated at pH 5.75 in the presence of target liposomes. Membrane-associated proteins were separated by liposome floatation on sucrose gradients. There was no liposome cofloatation when E9 protein was incubated with liposomes at neutral pH (Fig. 3A) . About 70% of the total E9 floated with liposomes in the top part of the sucrose gradient after treatment at pH 5.75 in the presence (Fig. 3A , top panel) or absence (data not shown) of a control protein. In contrast, when E9 was preincubated with pr peptide (pr:E9 molar ratio 12:1) and treated with low pH, only ,2% of E9-ST floated with the liposomes (Fig. 3A , middle panel). Inhibition by pr was not observed when it was added after E9 was treated at low pH in the presence of liposomes for 30 min (Fig. 3A , bottom panel), and thus pr needed to be present during the membrane insertion step. Inhibition was concentrationdependent, with 22% E9 co-floatation at a pr:E9 molar ratio of 3:1, 8% at 6:1, and 0.4% for 24:1 (data not shown; see also We then tested the effect of pr on the DENV E DI/II protein. This protein is monomeric and its stable membrane interaction requires DIII to ''clamp'' the core trimer [30] . As shown in Fig. 3B , ,25% of DI/II co-floated with liposomes at low pH in the present of DIII, while no co-floatation was detected when BSA was Figure 3 . Pr peptide inhibits E protein-membrane interaction. A) E9-liposome co-floatation assay. E9 protein was mixed with pr peptide or an ST-tagged control protein (Seap) at a final concentration of 50 mg E9 protein and 200 mg pr/Seap protein/ml (molar ratio 12 pr/1E). Liposomes were added at a final concentration of 1 mM, and the samples were incubated at the indicated pH for a total of 60 min at 28uC. Where indicated, E9 protein plus liposomes were incubated for 30 min, pr peptide added to a final concentration of 200 mg/ml, and the incubation continued for an additional 30 min. The liposome-bound proteins were then separated by floatation on sucrose gradients at the indicated pH. Aliquots of the top, middle and bottom of the gradients were analyzed by SDS-PAGE and western blotting for E protein. B) DI/II-liposome co-floatation assay. 40 mg/ml DI/II plus DIII or BSA (200 mg protein/ ml) were incubated with liposomes plus 200 mg pr peptide/ml as indicated and assayed for liposome co-floatation as in panel 3A. C) SFV DI/II-liposome co-floatation assay. SFV DI/II protein (40 mg/ml) was mixed with BSA or pr peptide (160 mg/ml). Liposomes were added at a final concentration of 1 mM, and the samples were incubated at the indicated pH for a total of 30 min at 28uC. Liposome co-flotation was assayed as in panel 3A. D-E) Loss of pr inhibition of E protein in the pH range of the late endocytic pathway. D) pH dependence of pr inhibition. E9 protein was mixed with liposomes in the presence or absence of pr peptide (molar ratio ,12 pr/1E), treated at the indicated pH as in The structurally related alphavirus protein SFV E1 DI/II is monomeric and efficiently interacts with membranes at low pH (80% cofloatation, Fig. 3C , middle panel). No inhibition occurred when pr peptide was added prior to liposome addition (Fig. 3C , bottom panel), in keeping with the lack of pr-SFV DI/II binding in the SPR experiments discussed above. Thus, pr peptide specifically inhibits target membrane interaction of both monomeric and dimeric forms of the DENV E protein. E9 protein efficiently inserted into membranes over a wide range of pH values from 6.25-4.5 ( Fig. 3D-E) . However, pr's inhibition of E membrane insertion was less efficient in the pH range (pH 5.0) present in the late endocytic pathway (Fig. 3D-E) . This loss of pr inhibition at more acidic pH may be relevant to recent studies of infection by immature DENV [42] , as mentioned in the discussion section below. All of the results above were obtained with soluble forms of the E protein. In order to test the ability of exogenous pr peptide to interact with and inhibit intact DENV, we took advantage of a previously described assay that monitors low pH-triggered fusion of DENV with cells [41] . In this fusion-infection assay, virus is prebound to target cells on ice, and then treated at 37uC for 1 min at low pH to trigger virus fusion with the plasma membrane. This fusion reaction is then quantitated by detecting the infected cells by immunofluorescence. We tested the effect of pr peptide during this 1 min low pH treatment using DENV1 WP and DENV2 NGC. The sequence of E DI/II is 68% identical between these two serotypes. Both serotypes showed efficient fusion and infection after treatment at pH 6.0, with about a 10-fold increase compared to samples treated at pH 7.9 (Fig. 4 ). The addition of pr peptide during the 1 min low pH treatment strongly inhibited DENV fusion and infection. Inhibition was dose-dependent, with 45-49% inhibition at 6 mM pr and 81-85-% inhibition at 30 mM pr. In contrast, pr did not inhibit low pH-triggered fusion by the alphavirus SIN (Fig. 4) . Thus, exogenous DENV2 pr peptide can specifically interact with mature DENV1 and DENV2 to block virus fusion and infection. We did not observe inhibition when DENV was preincubated with 30 mM pr at pH 7.0 and then added to target cells in a standard infection assay, suggesting that under these conditions an inhibitory concentration of pr was not present during low pH-triggered fusion reaction in the endosome. This result also indicates that the presence of pr did not affect virus-cell binding. Although the interaction of pr with DENV can clearly prevent virus-membrane interaction and fusion (this study and [35] ), the importance of pr in protecting DENV during exocytic transport has not been defined. The binding interface between prM and E contains three complementary electrostatic patches containing 11 residues [34] (see also Fig. S1 ). Sequence analysis shows that these 11 residues (Fig. 5A , numbered residues) are highly conserved among the 4 DENV serotypes, and that D63 and D65 of pr, and the complementary H244 on E protein are conserved among all reported flavivirus sequences [34] . Optimal pr-E binding in vitro occurred at ,pH 6.25 (Fig. 2) , suggesting that protonation of H244 could be involved in this pH-dependence. To test this we substituted alanine for H244 in the DI/II protein. DI/II H244A was produced in highly purified form with electrophoretic mobility similar to that of the wild type (WT) protein in reducing and nonreducing SDS-PAGE (Fig. 1C) . We first tested the effect of the H244A mutation on pr-E binding. In agreement with our earlier results, WT DI/II protein was efficiently pulled-down by pr-sepharose (Fig. 5B ). Pull-down was low pH-dependent and blocked by mAb 4G2 against the E fusion loop at the DII tip. In contrast, almost no H244A DI/II protein was pulled-down by pr-sepharose at either low pH or neural pH (Fig. 5B ). SPR analysis of WT DI/II protein showed most efficient binding at pH 6.0, and binding was blocked by preincubating the DI/II protein with mAb 4G2 (molar ratio 1:1) before dilution into SPR buffer (Fig. 5C, upper panel) . Equivalent concentrations of H244A DI/II protein showed greatly reduced binding to pr compared to that of WT protein (Fig. 5C , lower panel). Although H244A binding was decreased, the residual binding was still blocked by mAb 4G2 and had an acidic pH optimum. This suggests that binding also involves other residues in the pr-E interface, such as the complementary residues identified in the structural studies and shown in Fig. 5A . We then asked if the H244A DI/II protein was still active in binding to target liposomes. WT or mutant DI/II proteins were mixed with liposomes at low pH in the presence of DIII protein to stabilize the core trimer. Both proteins efficiently bound liposomes in a DIII-dependent reaction (Fig. 6) , indicating that the mutant protein retains its ability to insert into target membranes and form a core trimer. In agreement with the results in Fig. 3C , floatation of the WT protein was blocked by inclusion of pr during the membrane insertion step (Fig. 6) . In contrast, the efficiency of floatation of the H244A mutant protein was 43% in the absence of pr and 47% in the presence of pr. Thus, the H244A mutation did not inhibit E-membrane interaction but made that interaction insensitive to the presence of pr. Since the E H244A mutation disrupts E protein's interaction with pr, we used this mutation to address the importance of pr in protecting DENV during transport through the exocytic pathway. We introduced the E H244A mutation into the infectious clone of DENV1 WP. WT and mutant viral RNAs were prepared by in vitro transcription and were electroporated into BHK cells. After culture for 3 d at 37uC, both WT and mutant RNA-electroporated cells expressed abundant E protein as detected by immunofluorescence microscopy (Fig. 7) . Parallel cultures were incubated for 6 d and progeny virus in the culture media was detected by infectious center assays on indicator BHK cells. WTinfected cells produced infectious progeny virus with a titer of ,1.5610 5 IC/ml. However, two independent infectious clones of the H244A mutant produced no detectable progeny virus, even though the viral RNAs mediated efficient primary infection as shown in Fig. 7 . This agrees with previous studies indicating lethal effects of an H244A mutation on DENV2 [43] . The absence of secondary infection by the H244A DENV1 mutant could be due to decreased virus particle production and/or production of particles that are non-infectious. Efficient DENV particle production is dependent on E protein folding, particle budding into the ER, and subsequent particle egress through the secretory pathway. To investigate these issues, we took advantage of the ability of the flavivirus prM and E proteins to assemble into virus-like particles (VLP) in the absence of other viral components or virus infection [44, 45, 46 ]. The VLP system avoids complications arising from selection of revertants of deleterious virus mutations such as H244A. Flavivirus VLP bud into the ER in the immature prM form, undergo furin maturation during transport through the secretory pathway, and display similar low pH- [34] , potential key residues in pr-E interaction are indicated by their numbers in the DENV2 NGC proteins. B) H244A mutation inhibits pr-E binding in pull-down assay. WT or H244A mutant forms of DI/II were assayed for binding to pr-sepharose beads as in Fig. 2A . C) H244A mutation inhibits pr-E binding in SPR assay. WT or H244A mutant forms of DI/II were assayed for binding to pr at various pH values using SPR as in Fig. 2C , shifting to buffer alone at 300 s. Where indicated, mAb 4G2 (molar ratio 1:1) was pre-incubated 15 min at room temperature with DI/II proteins at pH 6.0 prior to assay. Data are a representative example of two independent experiments. doi:10.1371/journal.ppat.1001157.g005 dependent fusion activity as infectious virions [44, 47] . The VLP system has been used extensively to follow the process of flavivirus particle production and the role of prM in this process [37, 44, 45, 48] . We established stable HEK 293 cells that inducibly express the DENV1 WT or H244A prM-E proteins. After 36 h induction with tetracycline, both WT and H244A cells show abundant intracellular expression of the DENV1 E protein as detected by immunofluorescence, while the parent cell line is negative for E expression (Fig. 8A) . To evaluate whether WT and H244A E proteins were correctly folded, cells were induced for 36 h, lysed, and immunoprecipitated with a rabbit polyclonal antibody to E DIII, and with two conformation-specific mAbs. mAb 4E11 recognizes a discontinuous epitope on DENV E DIII and requires proper DIII disulfide bond formation for recognition [49, 50] . mAb 4G2 recognizes the fusion loop at the tip of flavivirus E DII and its epitope is sensitive to reduction [51] . Expression studies have shown that the 4G2 epitope is not formed if the E protein is expressed in the absence of prM [52] , indicating that this epitope is particularly useful for diagnostic tests of prM's chaperone interaction with E (see also reference [37] ). As shown in Fig. 8B , lysates from cells induced to express prM plus WT or H244A E proteins showed strong reactivity with all three antibodies. Quantitation of multiple experiments confirmed that WT and H244A E proteins were comparably recognized by the 4E11 and 4G2 mAbs. Thus, by these criteria H244A E protein interacts with prM protein and is correctly folded. This result also agrees with our finding that truncated H244A E protein expressed with prM in the S2 cell system was fully active in low pH-dependent membrane binding and trimerization, suggesting correct folding (Fig. 6) . . DENV E H244A mutation inhibits release of virus-like particles via a low pH-dependent mechanism. A) WT and H244A mutant E proteins are comparably expressed. Stable cells inducibly expressing the WT or H244A mutant forms of prM-E were treated with tetracycline for 36 h at 37uC. E protein expression was detected by immunofluorescence and the nuclei were stained with DAPI. Fluorescence images are shown at the same magnification and exposure time. Bar represents 30 mm. B) WT and H244A mutant E proteins are comparably immunoprecipitated by conformation-specific mAbs. Stable cells inducibly expressing the WT or H244A mutant forms of prM-E were treated with tetracycline for 36 h at 37uC. E proteins in the cell lysates were immunoprecipitated by Sango, a rabbit polyclonal antibody to DIII, and by the mouse mAbs 4G2 and 4E11, as indicated at the top of the panel. Samples were then analyzed by SDS-PAGE and western blot using mouse anti-DENV2 Ab for the Sango samples and Sango for the mAbs samples. Asterisks indicate the positions of the IgG and IgG heavy chain, which cross-react in the western blot. Equivalent sample input was evaluated by western blot for b-actin (lower panel). C) Effect of low pH on WT and H244A VLP production. WT and H244A mutant cells were incubated with tetracycline for 2 h and then in this medium plus 20mM NH 4 Cl where indicated for a total of 36h. VLP released in the culture media were pelleted by ultracentrifugation, and E proteins in the cell lysates were immunoprecipitated using mAb 4G2. VLP and lysate samples were analyzed by SDS-PAGE and western blot using Sango. 5-fold more culture media from the H244A cells than the WT cells were loaded. Data are representative examples of two or more independent experiments. doi:10.1371/journal.ppat.1001157.g008 We then used the inducible cells to examine VLP production. Expression was induced for 36 h. The cells were then lysed and the E proteins immunoprecipitated, and the VLP in the culture media were pelleted by ultracentrifugation. Analysis by western blotting showed strong E protein expression in both WT and H244A cells, and no expression in the parent cells (Fig. 8C) . The WT cells released E protein in VLP, but VLP release from cells expressing the H244A mutant E protein was greatly reduced (Fig. 8C, -media samples) . This result is in keeping with the hypothesis that the H244A cells assemble VLP in the neutral pH environment of the ER but that VLP release is inhibited by the lack of pr protection from the low pH of the secretory pathway. To test this idea, we induced WT and H244A prM-E expression and cultured the cells in the presence of 20 mM NH 4 Cl to neutralize the acidic pH in the Golgi and TGN compartments (Fig. 8C , +NH 4 Cl lanes). The cellular expression level of either E protein was not significantly affected by NH 4 Cl treatment, and WT VLP production was similar in NH 4 Cl-treated cells and untreated cells. However, production of VLP containing the H244A mutant E protein was increased 4-7 fold in NH 4 Cl-treated cells. While H244A VLP production was still significantly decreased compared to that of WT, it was selectively rescued by NH 4 Cl treatment. During translation of the flavivirus polyprotein, prM is the first protein translocated into the ER lumen, where it acts as a chaperone during the folding of the subsequently translocated E protein [4, 37, 44] . In addition to this important role of prM during E protein synthesis, a variety of data suggest that the interaction of pr peptide with the viral E protein protects flaviviruses from low pH during their transport through the exocytic pathway [34, 35, 36] . Here we showed that a recombinant pr peptide was efficiently folded, glycosylated, and secreted from 293T cells in the absence of its normal prM context and furin processing. Recombinant pr bound to soluble E proteins at low pH, inhibited E-membrane insertion, and interacted with mature dengue virus to block fusion and infection. Alanine substitution of the conserved E H244 within the pr-E interface disrupted pr-E binding in vitro and blocked secondary virus infection. VLP production was inhibited by the H244A mutation and partially rescued by pH neutralization with NH 4 Cl. Together our data demonstrate the critical role of pr in protecting DENV from exocytic low pH. The in vitro interaction of pr with various truncated forms of E protein was strongly pH-dependent, with a pH optimum of ,6.25. In situ measurements indicate that the pH of the TGN is ,6 [53] , while the pH optimum of DENV2 NGC fusion is ,6.2 [41] . The low pH of the TGN is critical for the rearrangement of immature DENV to allow furin cleavage, but once the virus is processed it becomes fusion-active in this same pH range. Thus the pH dependence of the pr-E interaction appears optimized to protect DENV during its continued transit through the secretory pathway. Pr's inhibition of E membrane insertion was less efficient at a pH value (pH 5.0) similar to that in the late endocytic pathway (Fig. 3D-E) . This loss of pr inhibition at more acidic pH could help to explain the recent finding that infection by immature DENV is enhanced by antibodies to prM [42] . The antibodybound immature virus is likely to be endocytosed and processed by cellular furin in the endocytic pathway [54] . The lower pH conditions of the late endocytic pathway could then cause the loss of pr inhibition and allow virus fusion. The structure of furin-cleaved DENV at pH 6.0 shows that pr is bound to the virion through interactions with the DII tip of one E protein and DI on the neighboring E monomer [35, 36] . This suggested that pr might primarily block virus-membrane interaction by preventing dissociation of E dimers, a required first step in the fusion pathway [55] . Our results show efficient binding of pr to the dimeric form of the DENV E protein, but also to the monomeric DI/II form. We do not know if the E9 protein dimer is stabilized by pr interaction or if the dimer dissociates prior to interaction with pr, and experiments to address these points were inconclusive (data not shown). The similar pH dependence of pr binding to monomeric and dimeric E proteins suggests that pr may bind the same site in both cases. mAb 4G2 against the fusion loop inhibited pr interaction with E DI/II, confirming that pr was binding to the DII tip rather than to other sites on expressed E proteins. In keeping with its binding site in the vicinity of the fusion loop, pr peptide blocked the membrane insertion and liposome co-floatation of E9 and DI/II proteins. Prior studies showed that a monomeric DI/II protein with a single Strep affinity tag stably inserts into liposomes at either neutral or low pH [30] , and pr blocked this insertion even at pH 8.0 where its interaction with DI/II was suboptimal (data not shown). Thus, while the pr-E interaction is strongly low pH-dependent, its functional inhibition of membrane insertion can still be observed at neutral pH in the presence of excess pr. Several other studies have addressed the role of E H244 in the flavivirus lifecycle. Experiments in TBEV evaluated particle production and membrane fusion activity using a VLP system [56] . Mutation of H248 (TBE numbering) to A or I blocks VLP secretion, in agreement with our results. However, an H248N mutant efficiently produces VLP, and these particles show WT levels of fusion activity. WNV E H246A or Q mutations inhibit release of infectious reporter virus particles from cells, as do a number of other substitutions at this position [57] . Replacement of H246 with aromatic residues such as phenylalanine allows both particle release and infectivity. An H244A mutation in DENV2 NGC inhibits infectious virus production [43] . E H244 and its interacting partners D63 and D65 on pr are conserved within the flaviviruses, and thus these data from several flaviviruses plus our DENV results support an important role for the E 244 position. However, a histidine residue at this position does not seem to be strictly required for particle production, suggesting that substitutions such as 244F and 244N can support the interaction of E with pr. In contrast to the block in production of H244A VLP, the H244A DI/II protein was efficiently secreted from cells. Mutant protein secretion was somewhat reduced, with the final yield of DI/II H244A about half that of the WT protein in two separate preparations (data not shown), suggesting some effects of nonoptimal pr interaction. However, unlike the E protein in virus or VLP, the truncated DI/II protein lacks the TM region and does not mediate membrane fusion, and thus may be relatively independent of the pH-protection function of pr. The purified WT and mutant DI/II proteins were able to bind liposomes and form core trimers that were stabilized by DIII (Fig. 6) . Thus, the mutant protein is correctly folded and active in membrane insertion. Studies with conformation-specific mAbs also provided evidence for the correct folding of H244A E protein (Fig. 8B) . Together, these results suggest that the H244A E protein is still able to access the chaperone functions of prM, while its decreased pr binding indicates that it can no longer utilize the pH protection functions of pr. These data are consistent with the idea that, similar to WT E, the mutant protein is assembled with prM into VLP in the ER. The membrane insertion and trimerization activity of H244A suggest that the full-length mutant protein would be fusion-active on such VLP once they are transported from the neutral pH of the ER to the low pH of the Golgi and TGN [31] . Thus, the decreased release of H244A VLP and its partial rescue by neutralization of the exocytic pathway support a critical role for pr in protecting DENV from exocytic low pH, and suggest that virus/VLP fuses in the TGN in the absence of pr-E interaction. Rescue of H244A VLP production by NH 4 Cl was clearly incomplete. This may be due to complex aspects of both virus and cell, such as direct effects of the H244A mutation on particle assembly in the ER, or difficulties in blocking fusion of a virus with the relatively high pH threshold of DENV. Several strategies have been used to block flavivirus and alphavirus fusion reactions and thus inhibit virus infection. SFV and DENV fusion are specifically blocked by exogenous DIII, which binds to the core trimer and prevents the foldback of endogenous DIII and hairpin formation [41] . A later stage in DENV fusion is targeted by a stem-derived peptide, which binds to the ectodomain trimer in which DIII has folded back but stem packing has not yet occurred [58] . These virus protein-protein interactions can be reconstituted in vitro [29, 30, 58] , opening the possibility of using them as screens for small molecule inhibitors of virus fusion and infection. The in vitro reconstitution of the pr-E interaction using soluble components could also act as a screen for small molecule inhibitors of this important flavivirus protein-protein interaction. Such inhibitors could act at multiple points in the virus lifecycle. During virus protein biosynthesis, an inhibitor could block the chaperone interaction of prM with E, leading to misfolding of E and its elimination by the ER quality control pathway. An inhibitor of pr interaction could make E protein susceptible to premature fusion in the TGN and could thus block virus production similar to the H244A mutation. It is also possible that small molecule inhibitors of pr-E binding could interact directly with the DII tip on mature virus particles, perhaps stabilizing the dimer and/or blocking membrane insertion of the fusion loop, thereby blocking virus fusion. Thus the in vitro system we describe here has the potential to identify molecules that could aid in the study of the flavivirus lifecycle and that could act to inhibit specific steps. Previous studies showed that after cleavage endogenous pr is retained on the virus particle if the virus is maintained at acidic pH [35] . Under these conditions, the virus-pr complex does not bind target membranes, while virus from which pr is first released at neutral pH efficiently binds membranes upon shift to acid pH. Thus, the bound endogenous pr inhibits virus-membrane interaction and presumably blocks virus fusion [35] . Our results demonstrated that even after maturation to fully infectious DENV particles, exogenous pr could add back to the virus and inhibit low pH-triggered virus fusion and infection. The flavivirus membrane fusion reaction is very rapid, occurring within seconds of low pH treatment [47] . Recombinant DENV2 pr peptide inhibited fusion by both DENV1 and DENV2, suggestive of a fairly broad spectrum inhibition in agreement with the strong sequence conservation at the pr-E interface [34] . The structure of the flavivirus E protein in its pre-fusion and post-fusion conformations defines the dramatic conformational changes between these two states. Many questions about the intermediates that connect the pre-and post-fusion conformations remain. In particular, it will be important to define the membrane protein rearrangements in the context of the highly organized flavivirus particle. For example, a neutralizing E mAb that blocks virus fusion was used to trap a West Nile virus fusion intermediate [59] . It will be interesting to evaluate if exogenous pr peptide could also be used as a novel probe to capture intermediates in the flavivirus fusion pathway. BHK-21 cells and C6/36 mosquito cells were cultured as described previously [60] . 293T cells and T-REx TM -293 cells were cultured as previously described using tetracycline-deficient fetal calf serum for the latter cells [61] . The DENV2 New Guinea C (NGC) strain and the DENV1 Western Pacific (WP) strain were propagated in C6/36 cells in DMEM containing 2% heatinactivated fetal calf serum and 10 mM Hepes, pH 8.0, as previously described [41, 62] . Sindbis virus expressing green fluorescent protein was obtained as an infectious clone (a kind gift from Dr. Hans Heidner) and propagated in BHK cells [63] . 4G2 is a mouse monoclonal antibody (mAb) that recognizes the fusion loop of flavivirus E proteins [51, 64] . mAb prM-6.1 recognizes a linear epitope on prM, and was a kind gift of Drs. Chunya Puttikhunt and Nopporn Sittisombut [40] . 4E11 is a mouse mAb that recognizes DIII of DENV E protein and neutralizes all 4 serotypes of dengue virus [49, 50] , and was a kind gift of Dr. Fernando Arenzana-Seisdedos (Institute Pasteur, Paris). The anti-DIII polyclonal antibody Sango was raised by immunization of a rabbit with purified DENV2 DIII protein [30] . Western blot detection of truncated E proteins used 4G2 or Sango antibodies. A mAb to b-actin was obtained from Sigma and used to confirm equivalent loading of cell lysate samples. Immunofluorescence detection of DENV-infected cells used the antibody to DIII or mouse polyclonal anti-DENV2 hyperimmune ascitic fluid (obtained from Robert B. Tesh, University of Texas Medical Branch), with Alexa Fluor 488 or rhodamine-conjugated secondary antibodies (Molecular Probes). The sequence encoding residues 1-86 of pr was amplified by PCR of an expression plasmid for DENV2 NGC prM-E DI/II [30] . The PCR product was ligated into the pPUR vector (Clontech), with the 21-residue TPA signal peptide [65] fused at the N-terminus and a tandem Strep tag at the C terminus (Fig. 1) . The plasmid, referred to as pPUR-TPA-pr-STST, was transfected into 293T cells using polyethylenimine (PEI, Polysciences). For optimal protein production, 3.5610 6 cells were plated per 10 cm dish and cultured for 24 h in 10 ml of complete medium. 7.5 mg plasmid in 1 ml DME was mixed with 30 mg PEI, incubated 10 min, then added drop wise to the cell culture medium. After 12 h, the medium was changed to 10 ml DME plus 2% serum. The culture medium was collected after 48h and again after 72h. Pr was purified by affinity chromatography on a Strep-Tactin column from IBA BioTAGnology and by gel filtration using a Sephadex G75 column [30] . Final yields were ,2 mg purified protein/1 liter culture supernatant. Truncated DENV E proteins (Fig. 1) were obtained by inducible expression in Drosophila S2 cells and purified by affinity chromatography as previously described in detail [29, 30] . The H244A mutation was introduced into the DI/II protein by in vitro mutagenesis, and S2 cell expression and purification were performed as above. DENV2 NGC DIII (Fig. 1) was previously referred to as LDIIIH1CS [30] , and contains domain III, the linker between domain I and domain III, and the H1 and CS regions of the stem domain. DIII was expressed in E.coli and refolded as previously described [41] . SFV E1 DI/II protein was produced as previously described [29] . All purified proteins were stored in TAN buffer (20 mM Triethanolamine[TEA], pH 8.0; 130 mM NaCl) at 280uC. SDS-PAGE analysis was performed using 10-12% acrylamide gels with a Bis-Tris buffer system (Invitrogen). Western blots were performed with Alexa Fluor 688-conjugated secondary antibodies (Molecular Probes), and were quantitated using an Odyssey Infrared Imaging system and Odyssey InCell Western software (LI-COR Biosciences) [30] . Standard curves with purified E proteins confirmed the linearity of this analysis (data not shown). Pr or BSA was coupled to NHS-activated sepharose 4 fast flow (GE Healthcare) as described in the manual. In brief, sepharose was washed with 1mM HCl, and incubated with 660 mg pr or BSA/ml in 0.2 M NaHCO 3 , 0.5 M NaCl, pH 8.3 at room temperature for 1.5 hr. The reaction was quenched with 0.1M Tris-HCl pH 8.5 for 30 min and free protein removed by washing with PBS. About 1mg of protein was coupled to 1ml beads. For the pull-down assay, 3 mg DI/II protein was pre-incubated where indicated with 24 mg 4G2 (molar ratio 1:2) or control mAb for 10 min at room temperature, and then incubated for 1 h on a rocker at room temperature with 10 ml of pr-or BSA-sepharose in a buffer containing 20 mM MES, 20 mM TEA, 130 mM NaCl, 0.2% Tween 20 at pH 8.0 or 6.25. The beads were then washed twice with the corresponding buffer and the bound DI/II was analyzed by SDS-PAGE and western blot. SPR studies were performed on a BIAcore 2000 instrument (GE Healthcare). Purified recombinant pr was immobilized on a CM5 biosensor chip by primary amine coupling as described in the manual. In brief, pr peptide was diluted to 10 mg/ml in 10 mM sodium acetate pH 4.7 and pre-concentrated on the chip surface. The chip was then activated by a mixture of 1-ethyl-3-(3dimethylaminopropyl)carbodiimide and N-hydroxysuccinimide, followed by quenching with 1M ethanolamine at pH 8.5. Under these conditions, pr was immobilized to a final density of 600 or 1000 response unit (RU). A control cell was mock-coupled with protein-free solutions. To test interaction, truncated E proteins were diluted to 1.2 mM in a MES/TEA buffer (20 mM MES, 20 mM TEA, 130 mM NaCl) at a pH range of 6.0 to 8.0, and flowed over the chip for 300 s at 0.3 ml/min, followed by buffer alone at the same flow rate. After each round, the chip was regenerated by washing with 50 mM NaOH in 1 M NaCl. The pr chip showed undiminished E binding activity for at least 50 rounds. Liposomes were prepared by freeze-thaw and extrusion through 200 nm polycarbonate filters [66] , and were stored at 4uC in TAN buffer under N 2 and used within 2 weeks of preparation. Liposomes were composed of a 1:1:1:3 molar ratio of 1-palmitoyl-2-oleoylsn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-snglycero-3-phosphoethanolamine (POPE), sphingomyelin (bovine brain) (Avanti Polar Lipids; Alabaster, AL), and cholesterol (Steraloids, Inc.; Wilton, NH), plus trace amounts of 3 H-cholesterol (Amersham; Arlington Heights, IL). Protein-membrane interaction was monitored using a liposome co-floatation assay [29, 30] . E9 or DI/II proteins at a final concentration of 50 mg/ml were incubated in TAN buffer (pH 8.0) for 5 min at 28uC in the presence of 200 mg pr peptide/ml as indicated. Liposomes were then added to a final concentration of 1mM lipid and the samples were adjusted to pH 5.75 by the addition of 0.3 M MES or maintained at pH 8.0, and the incubation continued at 28uC for 30-60 min. The samples were then adjusted to 20% sucrose and loaded on top of a 300 ml cushion of 40% sucrose, then overlaid with 1.2ml 15% sucrose and 200 ml 5% sucrose. All sucrose solutions were at the same pH as the samples, and were wt/wt in TAN buffer at pH 8.0 or in MES buffer (50 mM MES, 100 mM NaCl) at pH 5.5. Gradients were centrifuged for 3 hr at 54,000 rpm at 4uC in a TLS55 rotor, and fractioned into the top 700 ml, middle 400 ml and bottom 1 ml. The 3 H-cholesterol marker was quantitated by scintillation counting. 200 ml of each fraction were precipitated with 10% trichloroacetic acid and analyzed by SDS-PAGE and western blotting [29] . Purified human secreted placental alkaline phosphatase with a ST affinity tag (Seap) was used as a control protein [67] , and was a kind gift from Yves Durocher, Biotechnology Research Institute, Montreal. The fusion-infection assay was performed essentially as described previously [41] . In brief, BHK cells grown on 96-well plates were washed twice with ice cold binding medium (RPMI without bicarbonate, 0.2% BSA, 10 mM Hepes, and 20 mM NH 4 Cl, pH 7.9). Virus stocks were diluted in binding medium and incubated with cells on ice for 3 h with gentle shaking. Cells were washed twice with binding medium to remove unbound virus and pulsed for 1 min at 37uC in 100 ml RPMI without bicarbonate, containing 0.2% BSA, 10 mM Hepes and 30 mM sodium succinate at pH 6.0 or 7.9, containing the indicated concentration of pr peptide. Infected cells were incubated in MEM plus 2% FCS and 50 mM NH 4 Cl for 4 h at 37uC, and then at 37uC for 2 d in the presence of 20 mM NH 4 Cl. The number of infected cells was quantitated by immunofluorescence using mouse polyclonal anti-DENV2 antibody. Infection observed at pH 7.9 represents virus that is endocytosed and fuses during 1 min at this pH. The DENV1-WP infectious clone (reference [68] , a kind gift from Dr. Barry Falgout) was digested with KpnI and a 3.3kb fragment including the E sequence was sub-cloned into the pGEM3Z vector to generate pGDENV1 3.3. pGDENV1 3.3 was used as a template to generate the E H244A mutation, using circular mutagenesis as previously described [69] . A 2.6kb BstB1/ XhoI fragment containing the H244A mutation was sub-cloned into the DENV1-WP infectious clone to obtain DENV1-E H244A. The mutation was confirmed by restriction analysis and sequencing of the complete prM-E region. Two independent infectious clones were used to confirm the phenotype. The WT and the mutant infectious clones were linearized by Sac II digestion and used as templates for in vitro transcription [70] . RNAs were electroporated into BHK cells and cells were cultured overnight at 37uC followed by 6 d at 28uC in MEM containing 2% FBS and 10 mM HEPES, pH 8.0. Progeny virus in the medium was quantitated by infectious center assay on indicator BHK cells, using mouse polyclonal anti-DENV2 antibody. To detect primary infection, aliquots of the electroporated cells were plated on coverslips, cultured 3 d at 37uC, and processed for immunofluorescence microscopy as above. WT and E H244A mutant DENV1 prM-E sequences were PCR-amplified from the pGDENV1 3.3 subclones described above, and cloned into pcDNA4/TO (Invitrogen). These constructs were transfected into T-REx TM -293cells using Lipofectamine 2000 (Invitrogen) and selected in T-REx HEK medium containing 125 mg/ml Zeocin, all as previous described [61] . To test E protein folding and expression, 1610 6 WT and mutant E expressing cells were seeded in 10 cm plates, cultured for 24h, and then E protein expression was induced by culture for 36 h in 1.5 mg/ml tetracycline in DME medium with 10% FCS at 37uC. Cells were lysed in RIPA buffer (50 mM Tris-HCl pH 7.4, 150mM NaCl, 1% NP40, 0.5% Na-deoxycholate, 0.1% SDS, 1mM PMSF, 16 Roche complete protease inhibitor cocktail) on ice for 1 hr. The cell lysates were cleared by centrifugation for 30 min at 10,0006g and protein concentrations were quantitated and normalized. E proteins were immunoprecipitated from cell lysate samples (500 mg total cellular protein) using 20 mg purified mAb 4G2 or mAb 4E11 and 20 ml protein-G sepharose, or 30 ml Sango antibody and 20 ml protein-A sepharose. 4E11 and 4G2 immunoprecipitated samples were blotted with Sango. Sango immunoprecipitated samples were blotted with mouse anti DENV2 serum. For VLP secretion studies, 2-3610 6 cells were seeded in 10 cm plates, cultured for 24h, and then induced by culture for 36 h in 1.5 mg/ml tetracycline in DME medium with 10% FCS at 37uC. The culture media were centrifuged at 10,0006g for 30 min to remove cell debris. VLPs were then pelleted through a 0.5 ml sucrose cushion by centrifugation at 54,000 rpm for 2 h at 4uC using a TLS55 rotor. To test the effect of neutralizing the pH of acidic cellular compartments, cells were seeded and induced as above. After 2 h of induction the media were changed to DME medium containing 20 mM HEPES pH 8.0, 2% FCS, and 1.5 mg/ml tetracycline plus 20 mM NH 4 Cl as indicated, and the incubation continued for a total of 36 h. E proteins in the cell lysates were immunoprecipitated using mAb 4G2. VLP and lysate samples were then analyzed by SDS-PAGE and western blot using Sango. Figure S1 Open-book view of pr-E interface. Pr peptide is shown in cyan. DI, DII and DIII of E9 protein are colored red, yellow and blue, and the fusion loop at the DII tip is labeled. The important charged residues in the pr-E interface are numbered and shown as stick drawings in blue (positive) or red (negative). In this structure from DV2 16681, E9 residue 71 is a Glu, while the corresponding residue in NGC E9 protein is an Asp. Figure prepared from Protein Data Bank accession number 3C5X [34] using PyMOL. Found at: doi:10.1371/journal.ppat.1001157.s001 (0.39 MB PDF)

Alternative splicing of CD200 is regulated by an exonic splicing enhancer and SF2/ASF

CD200, a type I membrane glycoprotein, plays an important role in prevention of inflammatory disorders, graft rejection, autoimmune diseases and spontaneous fetal loss. It also regulates tumor immunity. A truncated CD200 (CD200(tr)) resulting from alternative splicing has been identified and characterized as a functional antagonist to full-length CD200. Thus, it is important to explore the mechanism(s) controlling alternative splicing of CD200. In this study, we identified an exonic splicing enhancer (ESE) located in exon 2, which is a putative binding site for a splicing regulatory protein SF2/ASF. Deletion or mutation of the ESE site decreased expression of the full-length CD200. Direct binding of SF2/ASF to the ESE site was confirmed by RNA electrophoretic mobility shift assay (EMSA). Knockdown of expression of SF2/ASF resulted in the same splicing pattern as seen after deletion or mutation of the ESE, whereas overexpression of SF2/ASF increased expression of the full-length CD200. In vivo studies showed that viral infection reversed the alternative splicing pattern of CD200 with increased expression of SF2/ASF and the full-length CD200. Taken together, our data suggest for the first time that SF2/ASF regulates the function of CD200 by controlling CD200 alternative splicing, through direct binding to an ESE located in exon 2 of CD200.

CD200 is a type 1 membrane glycoprotein, delivering immunoregulatory signals through binding to its receptors (CD200Rs) (1) (2) (3) (4) . It is present on neurons, B cells, activated T cells, thymocytes, dendritic cells and endothelium in mice, rats and human (5, 6) . A large and growing body of studies demonstrates that expression level of CD200 regulates graft survival (7) (8) (9) , susceptibility to autoimmune diseases (10) (11) (12) , fetal loss (13) , inflammation/infection (14) and tumor immunity (15) (16) (17) (18) . Alternative splicing is a major mechanism for regulating biological systems, producing multiple messenger RNA (mRNA) and protein isoforms. Some of these isoforms have distinct or even opposing functions (19) . Many genes in the immune system have been found to be alternatively spliced (20) (21) (22) and a growing number of human diseases are associated with aberrant splicing of the genes (23) (24) (25) . However, few studies to date have identified the mechanisms that regulate alternative splicing in the immune system. While CD200 exists as a single copy gene, data from Borriello et al. (26) , confirmed by our experiments (27) , have reported that a splice variant of CD200 exists. Although exon 2 deletion of CD200 caused by alternative splicing results in a frame shift and premature translational termination, we noted the existence of a downstream ATG start codon in a perfect Kozak context (27) . When the first start codon is followed shortly by a terminator codon and creates a small open reading frame (ORF; 5 0 -mini-cistron), the 40S ribosomal subunit remains bound to the mRNA, resumes scanning, and potentially reinitiates at the next ATG codon downstream (28) . It is known that the NH2-terminal region of CD200 is important for its biological interaction with CD200Rs (29, 30) , and translation from the second ATG start codon would produce a truncated form of CD200 (CD200 tr ) lacking the NH2-terminal 43 amino acids which includes regions important for the interaction with CD200Rs. Indeed, our previous studies have shown that expressed CD200 tr is a functional antagonist to CD200 (27) . Exons often contain specific short oligonucleotide sequences that affect their ability to be spliced. Exonic splicing enhancers (ESEs) within exons promote splicing of the corresponding exons and subsequent exon inclusion mediated by splicing regulatory proteins. The best-studied family of splicing regulatory proteins are Serine/ Arginine-rich proteins (SR proteins), which include the proteins SF2/ASF, SC35, SRp20, SRp30c and many others (31, 32) . It has become clear that many exons *To whom correspondence should be addressed. Tel: 4163404800 (ext. 6759); Fax: 4169289466; Email: zhiqi.chen@utoronto.ca contain ESE elements that bind to specific members of the SR family (25) , leading to exon inclusion. Since CD200 is involved in many diseases and its splice variant CD200 tr is an antagonist to CD200, identification of the mechanism controlling the relative expression levels of CD200 versus CD200 tr may provide insight into novel strategies for treatment of clinical disorders. In the present study, we have explored the mechanism controlling CD200 alternative splicing and show that SF2/ASF regulates CD200 alternative splicing through its direct binding to an ESE site in exon 2 of this gene. The level of SF2/ASF determines the alternative splicing patterns in different tissues or cells. Interestingly, in a mouse model of viral infection, we detected for the first time that the normal splicing pattern of CD200 was reversed in the lung tissue of A/J mice infected with mouse hepatitis virus strain I (MHV-1), following an increase in expression of SF2/ASF in this MHV-1 susceptible mouse strain. All human cell lines were obtained from American Type Culture Collection. Human B cell lines Daudi, Raji and TEM were maintained in RPMI 1640 (Invitrogen) supplemented with 10% fetal bovine serum (FBS). The human neuronal cell lines SK-N and HCN-1A were cultured with 10% FBS in a-MEM media (Invitrogen). Total RNAs from different human tissues were purchased from Clontech. A human BAC clone containing the whole human CD200 gene and a pcDNA3.2 expression vector containing SF2/ASF were obtained from The Center for Applied Genomics (Hospital for Sick Children, Toronto). Taq DNA polymerase, T4 DNA ligase and all restriction endonucleases were purchased from New England Biolabs. Random Primers, Superscript Reverse Transcriptase II, Elongase Enzyme, pcDNA3.0 expression vector and all competent cells were purchased from Invitrogen. EndoFree Plasmid purification Maxi Kit and QIAEX II Gel Extraction Kit were ordered from QIAGEN. A purified SF2/ASF recombinant protein was kindly provided by Dr. Blencowe (University of Toronto). Anti-human and mouse SF2/ ASF antibody was obtained from Santa Cruz Biotechnology. Anti-human and mouse b-actin antibody was purchased from BD Biosciences. Small-interfering RNA (siRNA) including SF2/ASF siRNA and a 'scrambled' siRNA were synthesized by Eurogentec. RNA oligonucleotides were synthesized by DNA and RNA Synthesis Center at Hospital for Sick Children (Toronto). All the primers used for polymerase chain reactions (PCRs), real-time PCRs and mutations were synthesized by Invitrogen. Female A/J and C57BL/6J mice, 6-8 weeks of age were purchased from Jackson laboratories. The mice were maintained in microisolator cages, housed in the animal facility at The Toronto Hospital Research Institute, University of Toronto, and fed standard lab chow diet and water ad libitum. All protocols were approved by the animal Welfare Committee. Parental virus Mouse Hepatitis Virus strain 1 (MHV1) was ordered from the American Type Culture Collection. As previously described (33) , MHV1 infection was carried out in a viral isolation room. A/J and C57BL/6J mice were anesthetized by intraperitoneal injection with 0.2 ml 10% pentobarbital diluted in normal saline. Mice were left untreated or received 5000 plaque forming unit (PFU) of MHV1 intranasally. Mice were sacrificed 12, 36 and 96 h postinfection and lung tissue was collected. Total RNA was isolated from human B cell lines (Daudi, Raji, TEM), human neuronal cell lines (SK-N, HCN-1A) and mouse lung tissue using TRIzol reagent. Five micrograms of total RNA from human tissues (brain, heart, skeletal muscle, colon, liver, thymus, kidney, intestine, lung, placenta and spleen), or human B cell lines (Daudi, Raji, TEM) and human neuronal cell lines (SK-N, HCN-1A), or mouse lung tissue was treated with DNase I and reverse transcribed in the presence of 250 ng of Random Primers, 1Â PCR Buffer, 10 mM dNTPs and 200 U of SuperScript II reverse transcriptase (RT; Invitrogen) in a final reaction volume of 20 ml. Reactions were carried out at 25 C for 10 min, 42 C for 50 min, followed by a 15-min step at 70 C to denature the enzyme. For regular PCR, 2 ml of first strand complementary DNA (cDNA) was amplified in a 50-ml reactions in the presence of 1Â PCR buffer, 1.5 mM MgCl 2 , 2.5 mM of dNTPs, 5 U of Taq DNA Polymerase (New England Biolab). A first cycle of 5 min at 94 C was followed by 30 cycles of 30 s at 94 C, 30 s at a different annealing temperature (based on different primer pairs), and 1 min at 72 C. The final extension step was at 72 C for 15 min. For real-time PCR, first strain cDNA was diluted 1:20 and quantified using an ABI 7900HT Sequence Detection System (Applied Biosystems). The sequences of the primers used for regular and real-time PCR were indicated in Table 1 . The endogenous human CD200 primer pairs for regular PCR were also used to construct an amplicon-containing plasmid (endogenous) for a standard curve. An exogenous amplicon-containing plasmid (exogenous) for a standard curve was constructed using the primers shown in Table 1 . Samples were tested in triplicate using 4 ml of first strand cDNA in a 20 ml total volume with 1Â universal master mix (Applied Biosystems). The results were normalized to that of the housekeeping gene GAPDH and HPRT. The copy number of transcripts was determined by comparison with a calibration curve of known amounts of amplicon-containing plasmid. Control reactions were performed for the specificity of the real-time PCR primers. A DNA fragment, containing either exon 1, exon 2 and exon 3 or only exon 1 and exon 3, was gel purified and subcloned into pcDNA 3.0 between NotI and XhoI sites. The CD200-bearing plasmids were then linearized by XhoI. In vitro transcription was carried out using TranscriptAid T7 High Yield Transcription Kit (Fermantas Inc.) following the manufacturer's instruction. Transcribed RNA was treated with DNaseI to remove template DNA and purified by phenol:choloroform extraction and ethanol precipitation. First strand cDNA was then synthesized and real time PCR was performed. The primer pairs used for real-time PCR are shown in Figure 3A and Table 1 . A human BAC clone containing the whole human CD200 gene was used as a template for long-distance PCR to obtain a region bearing exon 1, intron 1, exon 2, intron 2 and exon 3 of the human CD200. Two mixtures were prepared: mix 1 (20 ml) contained 0.1 mg of DNA template, 0.5 mM dNTP mix and 0.5 mm of sense and antisense primers; mix 2 (30 ml) included elongase enzyme mix and 1Â long-distance PCR buffer A and B provided by the manufacturer (the ratio of buffer A and B is 1:4). The sense primer started with the NotI cleavage site and the antisense primer with the SalI site. The sequences of the primers were shown in Table 1 . Mix 1 and mix 2 were combined on ice and subject to PCR under the following condition: 94 C for 1 min followed by three cycles at 94 C for 30 s, 59 C for 30 s, 69 C for 20 min, and then 29 cycles of 94 C for 30 s, 69 C for 20 min. The final extension was 69 C for 15 min. The 12-kb CD200 fragment was displayed on 0.7% TAE-agarose gel and purified using QIAEX II Agarose Gel Extraction Kit following the manufacturer's instruction. For more efficient elution of the large size DNA, the final incubation time was extended to 30 min at 60 C. The gel-purified DNA fragment was verified by restriction enzyme digestion with BamHI, BglII, EcoRI and HindIII, respectively, and DNA sequencing. For ligation to pcDNA 3.0 expression vector, the CD200 fragment was digested with NotI and SalI. Meanwhile, pcDNA 3.0 expression vector was digested with NotI and XhoI. Afterwards, pcDNA 3.0 vector was further dephosphorylated to remove the 5 0 phosphoryl group, preventing the vector from selfligation. The enzyme-treated CD200 fragment and pcDNA 3.0 were ligated, at a molar ratio of 3:1, using Primers for real-time PCR (the location of the numbered primers was shown in Figure 3A ) Endogenous human full-length CD200 (1) sense (exon 2) 5 0 -CAGCCTGGTTTGGGTCATG-3 0 (2) antisense (exon 3) 5 0 -GCAGAGAGCATTTTAAGGAAGCA-3 0 Endogenous human truncated CD200 Ligation products containing the alternative splicing construct were transformed into DH 10b Escherichia coli cells by electroporation using a Cell-Porator Electroporation System (Life Technologies) at 401 V, 330 mF capacitance, low and 4 k (for Booster). The cells were plated onto LB/ampicillin plates and incubated at 37 C overnight. Twenty isolated clones were randomly picked. Only one clone showed a DNA supercoil band with much larger size than that of the vector clone on the gel. This clone was further characterized by the combination of restriction enzyme digestion and sequence analysis. An ESE site was identified in exon 2 of the human CD200 using computational methods RESCUE-ESE (34) and ESEfinder (35) . To mutate the ESE site, site-directed mutagenesis was employed using QuickChange II XL site-directed mutagenesis kit from Stratagene. Two mutagenic primers were synthesized, in which the ESE site was replaced by a BsiWI site or deleted, and purified by polyacrylamide gel electrophoresis (PAGE). The sequences of the primers used are shown in Table 1 (the mutated region was underlined). The mutagenesis reaction was carried out in 50 ml total volume with 40 ng of template DNA, 125 ng of each primer and 2.5 U PfuUltra high-fidelity (HF) DNA polymerase and 3 ml of QuickSolution reagent provided by Stratagene. The cycling conditions included a 1-min initial denaturation at 95 C, 18 cycles with 50 s denaturation at 95 C, 50 s annealing at 58 C and 40 min extension at 68 C, and a final extension of 7 min at 68 C. The product was then subjected to digestion with 10 U of DpnI for 2 h at 37 C, selectively removing the parental, methylated, and nonmutated strands. Four microliters of DpnI-treated DNA was then transformed into XL10-Gold Ultracompetent cells. Cells were plated and incubated for selection of ampicillin-resistant clones. Ten isolated ampicillin-resistant clones were picked at random and their mutated or deleted regions were characterized by DNA sequencing. The B cell line Daudi was washed and resuspended in 1Â Hanks Balanced Salt Solution (HBSS) to a cell density of 2 Â 10 7 cells/ml. The neuronal cell line SK-N was trypsinized and resuspended in 1Â phosphate-buffered saline (PBS) with 2% FBS at a density of 10 7 cells/ml. Thee-hundred microliters of the Daudi cells or 500 ml of the SK-N cells were transfected with 10 mg of the alternative splicing minigene construct, the minigene construct with the ESE site deleted or mutated, the minigene construct plus SF2/ASF expression vector, or the ESE deleted construct plus SF2/ASF expression vector. Electroporation was performed with square waves of 700 V, 99 ms pulse length for four pulses for Daudi and square waves of 200 V, 70 ms pulse length for one pulse for SK-N using T820 ElectroSquarePorator (BTX). Both Daudi and SK-N cells were cultured in 5 ml of pre-warmed complete medium for 48 h before harvesting. The RNA oligonucleotides used for gel mobility shift assay were as follows: CD200 exon 2 with the wild-type ESE, 5 0 -GUGAUCAG GAUGCCCUUCUC-3 0 ; CD200 exon 2 with the mutated ESE, 5 0 -GUGACGUAC GUGCCCUUCUC-3 0 ; The RNA gel mobility shift assay was carried out as previously described (36) . The RNA oligonucleotides were 5 0 -end labeled with g-32 P-ATP (Perkin Elmer) using KinaseMax kit from Applied Biosystems following the manufacturer's instruction. Unincorporated nucleotides were removed by using G-25 Sephadex Columns. Fifteen femtomoles of radiolabeled RNA oligonucleotides were mixed with 4 pmol of SF2/ASF recombinant protein in a 20-ml binding reaction containing 2 mg yeast tRNA (Applied Biosystems). For competition, 100Â cold CD200 exon 2 oligonucleotide was added to the reaction containing the radiolabeled CD200 exon 2 oligonucleotide and SF2/ASF. After incubation for 20 min on ice, the RNA-protein complexes were separated from free RNA by electrophoresis on a 5% native polyacrylamide gel, run at 170 V for 2 h in 0.5% TBE buffer. The gel was then dried and autoradiographed at À80 C with intensifying screen. SF2/ASF siRNA was designed based on the information described by Cartegni et al. (37) . A 'scrambled' siRNA, which has no match with any mRNA of the human database, was used as a control. The siRNAs were synthesized by Eurogentec with the following sequences: 7.5 Â 10 5 Daudi cells or 5 Â 10 5 SK-N cells were seeded into 12-well plates 24 h before transfection. Twoand-a-half micrograms of siRNA was transfected to Daudi or SK-N cells using Lipofectamine 2000 (Invitrogen) to examine endogenous expression pattern of CD200 following silencing SF2/ASF. Two-and-a-half micrograms of siRNA, together with 10 mg of the alternative splicing construct DNA, was transfected to Daudi or SK-N cells by electroporation to detect exogenous expression pattern of CD200 following silencing SF2/ASF. The cells were harvested 48 h posttransfection. Total RNA and protein were then extracted. Nuclear extracts from Daudi and SK-N cells were isolated using NE-PER Nuclear and Cytoplasmic Extraction Reagents (38) from Pierce Biotechnology following the manufacturer's instruction. Western blotting was performed using 20 mg of nuclear extracts. After separation on a 10% SDS-PAGE gel, the proteins were transferred to a nitrocellulose membrane and probed with anti-human SF2/ASF antibody [1:200 dilution, goat polyclonal immunoglobulin G (IgG; Santa Cruz Biotechnology] followed by washing in 2% milk-PBS Tween. The membrane was then incubated with Donkey anti-goat IgG (1:5000 dilution; horseradish peroxidase-conjugated (BD Biosciences) and followed by washing again. Substrates, luminal and enhancer were added onto the membrane and incubated for 1 min. The membrane was exposed to Kodak XAR-5 film with intensifying screens for 5 min. Anti-human b-actin antibody (1:6000 dilution, goat monoclonal IgG; BD Biosciences) was used as loading controls. The exposure time for b-actin was 20 s. Statistical significance was calculated with one-way analysis of variance (ANOVA) followed by Tukey tests. P-values 0.05 were considered significant and shown in the figures. The existence of discrete CD200 splice variants is cell and tissue specific Human CD200 splice variants were examined in human tissues, B cells and neuronal cells. Total RNAs from different human tissues or human B cell and neuronal cell lines were used for RT-PCR using a sense primer located in exon 1 of human CD200 and an antisense primer in exon 3. As shown in Figure 1A and B, two transcripts were detected in all the human tissues, B cell lines (Daudi, Raji and TEM) and neuronal cell lines (SK-N and HCN-1A). The larger transcript was by far the dominant one seen in the brain and neuronal cell lines. Accordingly, for subsequent experiments, the B cell line Daudi and neuronal cell line SK-N were used as representatives of the two different splicing pattern of CD200. The only tissue not expressing CD200 was human skeletal muscle. The two transcripts were purified from the agarose gel and sequenced. It was confirmed that the larger one represented an exon 2 inclusion, whereas the smaller one represented an exon 2 exclusion (CD200 tr ). Since alternatively spliced exons often contain ESEs for binding of splicing regulators that determine the fate of the exon (exon inclusion or exclusion), we wondered whether ESEs for binding of splicing regulatory proteins existed in exon 2 of CD200. For this purpose, both RESCUE-ESE (34) and ESEfinder (35) were used to search for ESEs in the exon 2 of CD200. Only one ESE was identified in exon 2 by both RESCUE-ESE and ESEfinder. The ESE existed in exon 2 of CD200 in human, mouse and rat, with the sequence TCAGGA ( Figure 2A ). The identified ESE represents a known binding site for a splicing regulatory protein SF2/ASF, a member of the SR protein family (35) . Exogenous expression of CD200/CD200 tr shared the similar pattern with the corresponding endogenous one To gain insight into the role of the ESE in exon 2 of CD200, we generated an alternative splicing minigene construct containing the genomic region from exon 1 to exon 3 of the human CD200 ( Figure 2B) . A 12-kb fragment bearing this genomic region was characterized by sequencing and restriction enzyme digestion, and ligated to a pcDNA 3.0 expression vector. The construct was transfected independently to human B cell line Daudi and neuronal cell line SK-N. After 48 h, RNA was extracted from each cell population for detection of the exogenous expression of splicing pattern of CD200. RNA was also isolated from nontransfected Daudi and SK-N cells for detection of the endogenously expressed splicing pattern. To measure quantitatively the expression levels of the two splice variants, real-time RT-PCR was performed using the primer pairs located in different regions ( Figure 3A ). The specificity of the primers for amplification of full-length and truncated CD200 was examined. As shown in Figure 3B , the primer pair used for full-length CD200 did not amplify the template from CD200 RNA lacking exon 2 (truncated form), whereas the primer pairs for truncated CD200 were not able to amplify the template from CD200 RNA containing exon 2 (full-length form). Each primer pair generated only a single product (Supplementary Figure 1A) and the standard curves generated from each primer pair are parallel with slopes between À3.1 and À3.6 (Supplementary Figure 1B) . The exogenous expression of CD200/ CD200 tr had a similar pattern to the corresponding endogenous one in Daudi cells or SK-N cells ( Figure 3D and E). To examine further whether the ESE in exon 2 of CD200 determined the fate of the exon (inclusion or exclusion), site-directed mutagenesis was performed to mutate the ESE element in the alternative splicing construct, replacing the ESE (TCCTGA) with a restriction enzyme BsiWI site (CGTACG) ( Figure 3C ) or to delete the ESE. After characterizing the mutation or deletion construct by sequencing, the splicing construct was transfected to Daudi and SK-N cells. Total RNA was extracted from cells 48 h after transfection and real-time RT-PCR was carried out. As shown in Figures 3C and D, and 4A and B, expression of the full-length transcript (exon 2 inclusion) was reduced in both Daudi and SK-N cells after mutation or deletion of the ESE in exon 2. These data suggest that the ESE in exon 2 of CD200 promotes exon 2 inclusion. A splicing regulatory protein SF2/ASF directly binds to the ESE and determines the fate of exon 2 of CD200 Since the ESE described above is known to contain a putative binding site for SF2/ASF, we investigated whether SF2/ASF binds to the ESE. An RNA-EMSA was performed. As shown in Figure 5 , an RNA-protein complex was detected after the SF2/ASF recombinant protein with ÁRS domain was mixed with a radiolabeled RNA oligonucleotide containing the ESE site. This protein/RNA interaction is specific since SF2/ASF did not bind to a radiolabeled RNA oligonucleotide containing mutated ESE site and the above binding was eliminated by competing 100Â unlabelled oligonucleotide containing the same ESE ( Figure 5 ). Moreover, this binding was not competed by the same level of cold oligonucleotide with the ESE site mutated (data not shown). As previously described, the full-length CD200 was expressed predominantly in brain and neuronal cells. One explanation of this observation is that the expression of SF2/ASF is higher in neuronal cells and brain. To test this hypothesis, we assessed SF2/ASF levels in Daudi and SK-N cells by Western blotting. As shown in Figure 6A , the natural level of SF2/ASF was clearly higher in SK-N cells than in Daudi cells. To gain further insight into the role of SF2/ASF in controlling alternative splicing of CD200, an siRNA against SF2/ASF was employed to knock down SF2/ ASF in Daudi and SK-N cells. A scramble siRNA was used as a negative control. After Figure 3 . The pattern of expression of exogenous full-length CD200 or truncated CD200 in different cells parallels that of the endogenous molecules and mutation of the ESE in exon 2 abolishes exon 2 inclusion. (A) The location of the primers used for real-time RT-PCR. Primers 1 and 2 were used for endogenous expression of full-length CD200; primers 3 and 4 were used for endogenous expression of truncated CD200; primers 5 and 6 were used for exogenous expression of full-length CD200; primers 7 and 8 were used for exogenous expression of truncated CD200; primers 9 and 10 were used for the constitutive expression of V region of CD200. (B) The specificity of the primers used for full-length or truncated CD200. CD200 RNA containing exon 2 or lacking exon 2 from in vitro transcription was reverse transcribed and used for real-time PCR using the primer pairs labeled in the figure. (C) Mutation of the ESE in exon 2 was confirmed by DNA sequencing. (D) Endogenous and exogenous expression of the full-length and truncated CD200 in Daudi cells, and exogenous expression of two isoforms after mutation of the ESE. (E) Endogenous and exogenous expression of the full-length and truncated CD200 in SK-N cells, and exogenous expression of two isoforms after mutation of the ESE. The data represent the mean ± SE (three independent experiments, triplicate determinations). Broken lines reflect exogenous expression of the full-length CD200 decreased after mutation of the ESE relative to that of wild type (P < 0.01 in Daudi; P < 0.05 in SK-N). Continuous lines reflect exogenous expression of the truncated CD200 increased after mutation of the ESE relative to that of wild type in SK-N cells (P < 0.05). and total RNAs extracted for real-time RT-PCR, along with nuclear proteins for western blot. As shown in Figure 6A , SF2/ASF expression was eliminated after treatment with 2.5 mg of siRNA. b-Actin was used as a loading control. Real-time RT-PCR was then performed using RNA samples treated with siRNA. As shown in Figure 6B and C, the endogenous expression of full-length CD200 (exon 2 inclusion) was reduced in both Daudi and SK-N cells, compared with Mock (no siRNA) or scramble siRNA-treated cells. The same pattern was observed for the exogenous expression of CD200 in Daudi ( Figure 6D or SK-N cells ( Figure 6E ) following silencing SF2/ASF. Consistent with the observation resulting from the ESE mutation or deletion, the expression pattern of full-length versus truncated CD200 was reversed in SK-N cells after knockdown of SF2/ASF. To investigate further the function of SF2/ASF in exon 2 inclusion or exclusion, we performed overexpression analysis by transfection of SF2/ASF expression vector to Daudi or SK-N cells and examined the fate of exon 2. As shown in Figure 4A -C, overexpression of SF2/ ASF induced exon 2 inclusion but this function was abolished in the absence of the ESE in exon 2, indicating that SF2/ASF regulates CD200 isoforms only via the ESE. These results support the hypothesis that the splicing regulatory protein SF2/ASF, acting through binding to the ESE in exon 2 of CD200, plays an important role in controlling alternative splicing of CD200, and regulates Figure 4 . ESE deletion or overexpression of SF2/ASF affects exogenous expression patterns of CD200 isoforms. Ten micrograms of the wild-type minigene construct, the minigene construct with the ESE site deleted, the minigene construct plus SF2/ASF expression vector, or the minigene construct with the ESE site deleted plus SF2/ASF expression vector was transfected into Daudi or SK-N cells by electroporation. After 48 h, cells were collected and total RNA was isolated for real-time RT-PCR.The expression levels of the full-length and truncated CD200 as well as total CD200 in Daudi (A) and SK-N (B) were normalized to the housekeeping genes GAPDH and HPRT. The data shown are expression levels of full-length or truncated CD200 relative to total CD200. The data represent the mean ± SE (three independent experiments, triplicate determinations). the relative ratio of expression of full length to truncated CD200. The alternative splicing pattern is altered in vivo in A/J mice infected with MHV-1 Previous studies have shown that several viruses express a viral protein which mimics human CD200 and down-regulates host immunity to the virus following interaction with a human CD200 receptor on host cells (39) (40) (41) . Whether viral infection itself affects the expression of CD200 in host is an issue which remains to be explored. Intranasal infection of A/J mice with the coronavirus murine hepatitis virus strain 1 (MHV-1) has been described to induce pulmonary pathology with features reminiscent of severe acute respiratory syndrome (SARS) (33) . To examine the correlation between the viral (MHV-1) infection and expression of CD200 in host we collected lung tissues from MHV-1 susceptible A/J mice and MHV-1-resistant C57BL/6J mice after infection. RT-PCR was performed using a sense primer located in exon 1 and an antisense primer present in exon 3. Interestingly, we observed a reversal of the normal CD200 splicing pattern in lung tissues of A/J mice postinfection ( Figure 7A ). Real-time RT-PCR provided a more accurate result of this phenomenon. We documented that the full-length CD200 was increased after viral infection and was 2-fold higher at 36 h postinfection compared with that before infection ( Figure 7B ). All the susceptible A/J mice were dead at 96 h postinfection. In contrast, the relative ratio of full-length to truncated CD200 did not change in infected C57BL/6J mice ( Figure 7C and D) . Thus, the pattern of alternative splicing of CD200 was correlated with susceptibility of these strains to viral infection. Since the above studies have shown that SF2/ASF regulates alternative splicing of CD200, we wondered whether expression of SF2/ASF increased in A/J mice post infection. We performed western blotting using anti-SF2/ASF antibody. As shown in Figure 7E , no obvious difference of SF2/ASF level was seen between A/J and C57BL/6J mice before viral infection. Increased expression of SF2/ASF was detected in lungs of A/J mice 12 h postinfection, whereas no increase of SF2/ASF in C57BL/6J mice even 36 h postinfection, suggesting that the role of virus on host CD200 expression is mediated by SF2/ASF. The studies reported here show that the relative expression of two isoforms (CD200 and CD200 tr ) is tissue and cell specific and the alternative slicing patterns are different between the pattern in the lymphoid tissues and that of neuronal tissues. The relative expression of the two isoforms of CD200 is of interest, given our recent evidence that the truncated form (CD200 tr ) can antagonize the functional suppression induced by full-length CD200 (27) . Although Borriello et al. (26) reported no change in the alternative splicing pattern of murine CD200 in lymphoid tissue after stimulation by Con A or LPS in vivo, in our in vivo studies of mouse lung tissues before/after infection of MHV-1 virus we observed that, unlike in the natural condition, following viral infection the expression of total CD200 increased in lung of both MHV-1 susceptible A/J mice and MHV-1-resistant C57BL/6 mice. However, the splicing pattern of CD200 is reversed only in A/J mice, with the full-length transcript, capable of inducing immunosuppresion, becoming the predominant one. In contrast, for C57BL/6J, an MHV-1-resistant mouse strain, no change in the splicing pattern of CD200 was seen in the lung. This result importantly demonstrates that only the splicing pattern, but not the total transcription level, of CD200 determines the murine immune response to MHV-1 and is consistent with the hypothesis that the shift in the balance of expression of CD200/CD200 tr to decrease expression of the truncated product allowing CD200 to function in its immunosuppressive role, possibly contributing to the increased susceptibility to MHV-1 in the A/J mice. Further studies showed an increased expression of SF2/ ASF in A/J mice postinfection and the increase in SF2/ ASF occurred prior to increased full-length CD200, strongly suggesting that the regulation of alternative splicing of CD200 is mediated by SF2/ASF. It remains to be determined what viral proteins of MHV-1 have this effect and how the proteins regulate expression of SF2/ASF. Our studies suggest that viruses escape elimination by the host's immune system not only through producing viral proteins which mimic CD200 but also by inducing host CD200 expression and reducing expression of the antagonist CD200 tr . Posttranscriptional regulation, including mRNA stability, plays an important role for gene expression (42) . Whether the increase of full-length CD200 in A/J mice is also due to differential mRNA stability cannot be ruled out. In this report, we searched ESEs in the human and murine exon 2 sequence using two ESE-detecting algorithms RESCUE-ESE and ESE finder (35, 43) . Only one ESE, which is a putative binding site for SF2/ASF, was detected by both RESCUE-ESE and ESEfinder 2.0. No ESE was identified in the whole exon 2 when using higher stringent ESEfinder 3.0. Thus, we focused on this ESE for the rest of the experiments. Since an ESE can promote exon inclusion, mutation or deletion of the ESE would lead to less full-length but more truncated CD200. Our results showed that after mutating the ESE in exon 2, expression of full-length CD200 was reduced in both Daudi and SK-N cells. This expression pattern is the reverse of that seen for endogenous CD200 expression in SK-N cells, in which the predominant expression is of full-length CD200. To exclude the possibility that the mutation created a new exonic splicing silencer (ESS) which led to decreased full-length, and increased truncated CD200, we deleted the ESE and examined the changes in CD200:CD200 tr . Our result showed that deletion of the ESE promoted exon 2 exclusion, the same result as we obtained from mutation analysis, indicating that mutation of the ESE does not create an ESS. Identification of a putative ESE for SF2/ASF binding does not provide direct evidence that SF2/ASF recognizes and binds to the ESE. To examine whether the identified ESE in exon 2 is bound by SF2/ASF, we performed RNA-EMSA using RNA radiolabeled oligonucleotides bearing the ESE in exon 2 and a recombinant SF2/ASF with ÁRS domain to reduce nonspecific binding. The result showed a binding of SF2/ASF to the ESE and the binding is specific because either mutated ESE or 100Â cold oligonucleotides abolished the binding. Knockdown of SF2/ASF decreased expression of full-length CD200 in both Daudi and SK-N cells. Consistent with data seen following mutation or deletion of the ESE, the expression pattern of CD200 was again reversed in SK-N cells. The western blot performed confirmed the efficiency of knockdown of SF2/ASF. In contrast, overexpression of SF2/ASF increased expression of full-length CD200 but only in the presence of the ESE in exon 2, highlighting the critical role of the ESE in the mechanism of alternative splicing of CD200. Ubiquitously expressed splicing factors, among them is SF2/ASF, are thought to control tissue specific alternative splicing through their different expression levels in different tissues (44) . Our result showed that the natural level of SF2/ASF was higher in the neuronal cell line SK-N than in B cell line Daudi. This may help explain why endogenous full-length CD200 (exon 2 inclusion) is expressed at much higher level than that of truncated CD200 (exon 2 exclusion) in SK-N. A recent report has described a higher expression level of SF2/ASF in many tumors, including lung, thyroid, kidney, colon, small intestine and melanoma, relative to their respective normal controls. One mechanism to explain this observation is that SF2/ASF abolished the tumor suppressor activity of BIN1, a tumor suppressor gene, by inclusion of exon 12A which interferes with MYC binding (45) . In contrast to its roles in transplantation, autoimmune diseases and inflammation, CD200 enhances the growth of malignant tumors and it has been suggested that a novel approach to anticancer therapy might include blockade of CD200 (15, 16, (46) (47) (48) (49) . Since CD200 tr is an antagonist to CD200 (27) , our data are consistent with the hypothesis that increased CD200 tr expression and decreased expression of full-length CD200 by blockade of SF2/ASF may also be of potential benefit for cancer treatment. In conclusion, we have identified an alternative splicing pattern for expressed human CD200 in different cells and tissues, and compared this with the pattern observed in vivo following viral infection. Our data suggest that regulation of expression of alternative splicing transcripts may be important in controlling susceptibility to viral infection. An ESE in exon 2 of CD200 is a binding site for a splicing regulatory protein, SF2/ASF, which we have shown to control the alternative splicing pattern of CD200. A drug-mediated manipulation of alternative splicing has recently been reported which includes modulation of SF2/ASF (25) . It would be of interest to know if this drug treatment alters the expression ratio of CD200 to CD200 tr and thereby produces change in immune function. Supplementary Data are available at NAR Online. Funding for open access charge: The Heart and Stroke Foundation (NA6164 to R.M.G.).

Efficacy of Oseltamivir-Zanamivir Combination Compared to Each Monotherapy for Seasonal Influenza: A Randomized Placebo-Controlled Trial

BACKGROUND: Neuraminidase inhibitors are thought to be efficacious in reducing the time to alleviation of symptoms in outpatients with seasonal influenza. The objective of this study was to compare the short-term virological efficacy of oseltamivir-zanamivir combination versus each monotherapy plus placebo. METHODS AND FINDINGS: We conducted a randomized placebo-controlled trial with 145 general practitioners throughout France during the 2008–2009 seasonal influenza epidemic. Patients, general practitioners, and outcome assessors were all blinded to treatment assignment. Adult outpatients presenting influenza-like illness for less than 36 hours and a positive influenza A rapid test diagnosis were randomized to oseltamivir 75 mg orally twice daily plus zanamivir 10 mg by inhalation twice daily (OZ), oseltamivir plus inhaled placebo (O), or zanamivir plus oral placebo (Z). Treatment efficacy was assessed virologically according to the proportion of patients with nasal influenza reverse transcription (RT)-PCR below 200 copies genome equivalent (cgeq)/µl at day 2 (primary outcome), and clinically to the time to alleviation of symptoms until day 14. Overall 541 patients (of the 900 planned) were included (OZ, n = 192; O, n = 176; Z, n = 173), 49% male, mean age 39 years. In the intention-to-treat analysis conducted in the 447 patients with RT-PCR-confirmed influenza A, 46%, 59%, and 34% in OZ (n = 157), O (n = 141), and Z (n = 149) arms had RT-PCR<200 cgeq/µl (−13.0%, 95% confidence interval [CI] −23.1 to −2.9, p = 0.025; +12.3%, 95% CI 2.39–22.2, p = 0.028 for OZ/O and OZ/Z comparisons). Mean day 0 to day 2 viral load decrease was 2.14, 2.49, and 1.68 log(10) cgeq/µl (p = 0.060, p = 0.016 for OZ/O and OZ/Z). Median time to alleviation of symptoms was 4.0, 3.0, and 4.0 days (+1.0, 95% CI 0.0–4.0, p = 0.018; +0.0, 95% CI −3.0 to 3.0, p = 0.960 for OZ/O and OZ/Z). Four severe adverse events were observed. Nausea and/or vomiting tended to be more frequent in the combination arm (OZ, n = 13; O, n = 4; and Z, n = 5 patients, respectively). CONCLUSIONS: In adults with seasonal influenza A mainly H3N2 virus infection, the oseltamivir-zanamivir combination appeared less effective than oseltamivir monotherapy, and not significantly more effective than zanamivir monotherapy. Despite the theoretical potential for the reduction of the emergence of antiviral resistance, the lower effectiveness of this combination calls for caution in its use in clinical practice. TRIAL REGISTRATION: www.ClinicalTrials.gov NCT00799760 Please see later in the article for the Editors' Summary

Neuraminidase inhibitors (oseltamivir [O] , zanamivir [Z] ) are thought to be efficacious as compared to placebo in outpatients with uncomplicated seasonal influenza [1] [2] [3] [4] [5] [6] , both clinically in terms of reduction in duration of symptoms, as well as in terms of a reduction in viral shedding. In 2008, they were considered an important strategy to limit the impact of an influenza pandemic both individually, by reducing morbidity and mortality, and collectively, by slowing spread of the virus to allow time for vaccine production, the cornerstone of influenza control [2] [3] [4] 7] . It was hypothesized that the widespread use of a single antiviral might result in the emergence of resistant strains whose subsequent spread could dramatically reduce the effectiveness of antiviral therapy. The combination of two antiviral agents, if well tolerated, and if producing at least additive antiviral activity, theoretically offers several advantages: reducing disease severity, viral shedding, and viral excretion period, thereby also lowering the attack rate and risk of selection of resistant viruses, specifically in individuals with prolonged viral shedding, such as immunocompromised patients [8, 9] . Indeed, mathematical modelling showed a reduction in risk of emergence of resistant strains during early phases of a pandemic, associated with use of two antivirals as compared to single antiviral therapy [9] . Finally, another theoretical advantage of combining two drugs would be to ensure optimal treatment of all circulating influenza virus types, subtypes, or variants, as susceptibility of influenza viruses has been shown to vary, and seasonal H1N1 viruses naturally resistant to oseltamivir, which remain susceptible to zanamivir, emerged in 2008 [10] . Among antivirals active against influenza virus, the combination of neuraminidase inhibitors is attractive, because both compounds are licensed for seasonal influenza, they are delivered to the respiratory tract by distinct means (directly through a diskhaler for zanamivir, after gastrointestinal absorption and hepatic metabolism for oseltamivir), and key mutations associated with resistance are different for each drug. However, negative interactions cannot not be ruled out owing to the possible competition between these two drugs, which target the same binding pocket in the neuraminidase. In 2006, in the context of pandemic planning, we designed a double-placebo randomized controlled trial in patients presenting with seasonal influenza-like illness to compare the oseltamivirzanamivir combination to each of the monotherapies plus placebo. The trial was conducted in France, during the winter of [2008] [2009] . Because of the emergence of the pandemic 2009 (H1N1) virus in humans, in April 2009 in North America, and its subsequent worldwide spread, the independent data-monitoring committee requested that we terminate the trial early and analyze the results earlier than planned, given the possible impact of the results on antiviral treatment management during the pandemic [11] . From January 7th to March 15th 2009 (period of the winter 2008-2009 influenza epidemic in France), we enrolled throughout France adults 18 y old and older who consulted their general practitioner within 36 h of influenza symptoms onset (following the first influenza symptoms reported by the patient), with a temperature greater than or equal to 38uC (reported or observed by the practitioner), one or more respiratory symptoms (cough, sore throat), one or more general symptoms (headache, dizziness, myalgia, sweats and or chills, fatigue), and a positive nasal rapid test for influenza A (Clearview Exact Influenza A & B) performed by the practitioner. Enrolment of women required a negative urine pregnancy test. Exclusion criteria were vaccination against influenza during the 2008-2009 season, recent exacerbations of chronic obstructive pulmonary disease (COPD), asthma or severe chronic disease, previous history of depression, and prior inclusion in this trial. Prior to inclusion, patients gave informed written consent. The protocol was approved by the Ethics Committee of Ile de France 1 (Texts S1 and S2). At enrolment (day 0), a nasal swab for virological analysis was performed using a standard operating procedure (sample kit plus instructional video). Patients were allocated to treatment by a randomization list, with an arm ratio of 1:1:1, balanced by practitioner. A computer random number generator was used to select random permuted blocks of size 3. This randomisation code was given to the central hospital pharmacy that prepared blinded treatment units in conformity with good manufacturing practices (GMP). Each general practitioner received six treatment units and was told to distribute them by order of inclusion of his patients in the trial. Allocation was concealed through the similarity of all the containers and the impossibility for the GP to identify the treatment arm when opening the container. The three treatments were (1) oseltamivir capsule for oral use plus inhaled zanamivir, (2) oseltamivir plus inhaled placebo, (3) zanamivir plus oral placebo. Oseltamivir dosage was 75 mg orally twice daily; zanamivir dosage was 10 mg by oral inhalation using the commercialized GlaxoSmithKline Diskhaler, twice daily. Active drugs and placebo were kindly provided by Roche and Glaxo-SmithKline laboratories. A visiting nurse performed a nasal swab for virological analysis on day 2. Patients returned to their general practitioner at day 7 for a follow-up examination, and were contacted by phone on day 14. Patients, general practitioners assigning the patients, and outcome assessors (practitioners, virologists, patients), were blinded to treatment assignment throughout the study and statisticians until the end of the analysis. Nasal swabs placed into a transport medium (Virocult, Elitech) were transported at 4uC by special courier to the nearest National Influenza Centre (NIC) (Hospices Civils de Lyon, Lyon, or Pasteur Institute, Paris, France). Upon arrival, the swab samples were eluted into 2 ml of transport medium, processed for real-time reverse transcription (RT)-PCR analyses and inoculated onto MDCK cells for virus isolation and subsequent subtyping using a standard hemagglutination inhibition assay. For RT-PCR analyses, RNA extraction from 200 ml of specimen was performed using the QIAmp virus RNA mini kit (Qiagen) with RNA elution into a final volume of 60 ml. All real-time RT-PCR assays were performed in a final volume of 15 mL with 5 mL RNA, 0. mM of each primer, 0.2 mM probe, and 0.8 ml enzyme mix (Super-ScriptIII platinum one-step quantitative RT-PCR system, Invitrogen). Type A influenza virus RNA was detected by a real-time RT-PCR targeting the conserved matrix gene using GRAM/7Fw (59-CTTCTAACCGAGGTCGAAACGTA-39) and GRAM/ 161Rv (59-GGTGACAGGATTG GTCTTGTCTTTA-39) primers and GRAM probe/52/+ (59[Fam]-TCAGGCC CCTCAA-AGCCGAG-[BHQ-1]39) probe. The quality of the specimens was assessed by real-time RT-PCR targeting the GAPDH cellular gene [12] . Amplification was performed on a LightCycler 480 (Roche Diagnostics) (NIC, Pasteur Institute, Paris) or an ABI 7500 (Applied Biosystems) (NIC, Lyon). Cycling conditions are available upon request. Quantified synthetic RNA transcripts corresponding to the M and GAPDH genes were used as controls in parallel [13] . To take into account the variability in the quantity of cells collected by nasal swab, we calculated a normalized influenza viral load for each specimen; this normalized viral load was defined as the ratio of the M RT-PCR and GAPDH RT-PCR multiplied by the average GAPDH RT-PCR at day 0 to express results in copies of genome equivalent/ml (cgeq/ml). The virological response was defined as a normalized viral load below 200 cgeq/ml at day 2. This threshold was determined according to results on specimens from patients with a positive influenza A rapid test from winter 2008-2009 included in the French influenza surveillance network (GROG), and analysed by the 2NICs, and because it resulted in 5% false positive and 5% false negative results with respect to virus isolation (Table S1 ). It was validated by the independent datamonitoring committee prior to any analysis. Sensitivity of the qRT-PCR was assessed using serial dilutions of quantified synthetic transcripts corresponding to the target genes (Text S3). To assess comparability of the data between the two NICs, specimens were exchanged showing excellent concordance. The threshold of the qRT-PCR used was set well above the limit of detection. Oral temperature was recorded and severity of seven symptoms (nasal stuffiness, sore throat, cough, muscle aches, tiredness or fatigue, headache, and feverishness) was rated by the patient twice daily (morning and evening) up to day 5 and then once daily on a four-point scale (0, none; 1, mild; 2, moderate; 3, severe) [2] [3] [4] 14] . The time to resolution of illness was defined as time from study drug initiation to time of symptom alleviation. Symptom alleviation was defined as the first 24-h period during which the above seven symptoms were absent or only mild as previously described [3, 4] . Influenza-related clinical events were defined as incidence of a secondary complication (such as pneumonitis or otitis) independently of any antibiotic initiation, and/or occurrence of exacerbation of a preexisting chronic disease. Patients reported treatment compliance using a self-administered questionnaire; full compliance during the day 0 to day 2 period was considered when 100% of planned drug intakes had been completed. The primary efficacy endpoint was the proportion of patients with RT-PCR,200 cgeq/ml on day 2 of treatment. Given the viral shedding kinetics in patients with seasonal influenza receiving neuraminidase inhibitors, the day 2 virological endpoint was considered to be best suited to measurement of virological effects [2, 4] . Other endpoints were (1) the decrease of log 10 viral load between days 0 and 2 in the patients with confirmed influenza A on day 0 and available samples both at days 0 and 2; (2) the time to resolution of illness; (3) the number of patients with alleviation of symptoms at the end of treatment (day 5); (4) the symptoms score at the end of treatment; (5) the incidence of secondary complications of influenza such as otitis, bronchitis, sinusitis, pneumonia, and the use of antibiotics; (6) the occurrence of adverse events in all participants having received at least one dose. According to the protocol, the intention-to-treat (ITT) analysis was performed on two populations: (1) all enrolled patients (primary objective), and (2) enrolled patients with an influenza A virus infection confirmed by RT-PCR on day 0 (influenza Ainfected population). Sample size evaluation assumed that virological response was obtained in 70% of patients in the oseltamivir-zanamivir arm, compared to 55% in each of monotherapy arms on the basis of the extrapolation of the results of previous trials [2, 4] . Samples of 300 subjects per arm had 90% power to detect this difference, with a two-sided test and a type I error of 0.025 because of the two tests (factoring 20% lack of follow-up). Proportions of response at day 2 were compared between the combination therapy arm (OZ) and each monotherapy arm (O or Z) separately using two tests with a type I error of 0.025 because of the two tests. Patients without a day 2 sample were considered treatment failures. Mean decreases of log 10 viral load were compared using a t-test in patients who had both day 0 and day 2 samples assuming a value of 0.5 cgeq/ml when RT-PCR was negative. For clinical endpoints nonparametric tests were used. Times to resolution of illness and symptoms score at the end of treatment were compared using Wilcoxon tests. If time to symptom alleviation was missing it was imputed to be 14 d, i.e., the end of the trial; 95% confidence intervals (95% CIs) of median differences were estimated by bootstrap. Probability of symptoms alleviation versus day of treatment was estimated using the Kaplan Meier method and was compared between groups with the log-rank test. Proportions of clinical events and of patients with alleviation of symptoms were compared using Fisher's exact tests. As an exploratory analysis, 95% CIs for differences of response between the two monotherapies were also estimated. All analyses were performed using SAS software, version 9.1 (SAS Institute). Out of the 900 patients initially planned, a total of 541 patients were enrolled by 145 general practitioners. They were randomly assigned to oseltamivir plus zanamivir (OZ, n = 192), oseltamivir plus inhaled placebo (O, n = 176), zanamivir plus oral placebo (Z, n = 173) ( Figure 1 ). Mean age was 39 y (standard deviation [SD] = 13), 49% were male, 14% had preexisting chronic diseases, mean fever was 38.2uC (SD = 0.8). The mean duration of illness before enrolment was 25 h (SD = 10). Other characteristics of patients appeared well balanced in the three arms ( Table 1 ). The rate of fully compliant patients was not significantly different among the three arms (84% for the OZ arm, 88% for the O arm, and 85% for the Z arm). Out of the 541 enrolled patients, 447 (83%) had a RT-PCR laboratory confirmation of influenza A virus infection on the day 0 specimen, with a mean viral load of 4.38 log 10 cgeq/ml (interquartile range [IQR] 3.75-5.30). All the day 0 specimens were GAPDH RT-PCR positive with a mean value of 3.88 log 10 copies/ml. Primary endpoint. In the ITT analysis, considering the 541 enrolled patients with positive influenza A rapid test, the proportion of patients with a RT-PCR,200 cgeq/ml on day 2 of treatment was 52.6% in the oseltamivir-zanamivir arm, 62.5% in the oseltamivir monotherapy arm (p = 0. Table 2 ). The same trends were observed in the 382 fully compliant influenza A-infected patients (Table S3) , and in the 395 patients with H3N2 infection with proportions of 42.4% in the oseltamivir-zanamivir arm, 58.6% in the oseltamivir monotherapy arm, and 30.3% in the zanamivir monotherapy arm. Other virological endpoints. In the 414 influenza RT-PCR confirmed patients with both day 0 and day 2 available specimens, the day 2 to day 0 decrease was 2.14 log 10 cgeq/ml in the oseltamivir-zanamivir arm, 2.49 log 10 The median time to resolution of illness in the 541 enrolled patients was 3. Sum of the severity of the seven day 0 influenza symptoms (feverishness, nasal stuffiness, sore throat, cough, muscle aches, tiredness-fatigue, and headache) using a four-point scale [2, 14] . b The score is expressed as a percentage of the maximal score of 21. (Tables 2 and 3 ). Other clinical outcomes showed similar trends (Tables 2, 3 , and S3). Four serious adverse events occurred during the study, one of which was considered unrelated to study drugs (acute bacterial pneumonia at day 3 in a patient receiving oseltamivir-zanamivir combination). Two adverse events also occurred in patients receiving the oseltamivir-zanamivir combination: severe headaches leading to interruption of therapy and facial oedema following the first administration, disappearing within 24 h postdrug interruption. The remaining patient experienced repeated vomiting after oseltamivir monotherapy drug administration. All four patients completely recovered. Other nonserious adverse events reported in more than 1% of the total population were in the OZ, O, and Z arms, respectively, nausea and/or vomiting (in 13, 4, and 5 patients), diarrhoea (in 2, 1, and 5 patients), and rash (in 1, 2, and 2 patients). This large publicly funded clinical trial examined the effect of combination neuraminidase inhibitor antiviral therapy in influenza, as compared to each monotherapy plus placebo. It showed that, during the prepandemic winter of 2009 with a predominance of H3N2 viruses (more than 85%) in France, the oseltamivir-zanamivir combination seemed less effective than oseltamivir monotherapy, and not significantly more effective than zanamivir monotherapy in adults with seasonal influenza A virus infection. Analysis of the different antiviral regimens' efficacy was based on a primary virological endpoint, which we hypothesized could be a sensitive and a more specific indicator than a primary clinical endpoint. Clinical endpoints, used as primary endpoints in previous studies, were used as secondary endpoints in the present study [1, 5] . Clinical endpoints, which are based on a global assessment of both general (mainly immunologically linked) and respiratory (mainly virologically linked) symptoms, are probably not the best way to monitor the virological effect of treatment, because clinical symptoms are not exclusive to influenza. We thus considered that a difference in viral shedding rate would be the best indicator of the virological effects of combined therapy, and consequently a valuable surrogate. Our initial hypothesis was that the combination of two antivirals may reduce the rate of resistant virus emergence (for a naturally susceptible pandemic virus and a nonimmune population). In addition, we hypothesized that for cases of infection with susceptible seasonal influenza viruses, this could not be easily shown, owing first to the rarity of this phenomenon in adults, and second, to the necessity of monitoring virus excretion for several days, whereas for cases of influenza due to H1N1 viruses, which are naturally resistant to oseltamivir, the question was not relevant. Given the viral shedding kinetics in patients with seasonal influenza receiving neuraminidase inhibitors, the day 2 virological endpoint was considered to be best suited to quantify virological effects. The 200 cgeq/l threshold was chosen, as it was the best compromise in terms of specificity and sensitivity as compared to standard culture. Of note, the same trends were observed when a 100 cgeq/l or a 1,000 cgeq/ml cut-off was used to define virological success (Table S2 ). Furthermore, the study was designed to be statistically two-sided to take into account the possibility that the combination would perform worse than either drug alone because of the theoretical concern of antagonism at the receptor level. The oseltamivir-zanamivir combination seemed, both virologically and clinically, significantly less effective than the oseltamivir monotherapy. This result seems robust because (1) it was found using a double-blind placebo methodology, (2) there was overall concordance both among virological endpoints, and between virological and clinical endpoints, (3) it was confirmed over the three different subgroups of subjects included in the global population (541 enrolled patients, 447 influenza A-infected patients, 382 influenza A-infected and fully compliant patients). This lower clinical and virological response to the combination may suggest a negative effect of zanamivir on oseltamivir, as in the absence of interactions the effect of the combination should at least be additive [15] . A negative interaction at the level of binding at the catalytic pocket of the neuraminidase is an explanation that should be further investigated in vitro for both seasonal H3N2 and H1N1 viruses. Recent in vitro data showing the lack of synergy between oseltamivir and zanamivir, and some antagonism at higher concentrations of zanamivir on pandemic H1N1 2009 virus, are in agreement with this hypothesis [16] . Furthermore, contrary to oseltamivir, which upon digestive absorption needs to be metabolized, thus delaying arrival of the active drug at the infection site (t max = 4 h), inhaled zanamivir is delivered directly to the primary site of influenza virus replication. The hypothesis that zanamivir is more likely to occupy the catalytic pocket first, thus preventing the action of oseltamivir, must be tested. According to this hypothesis, the combination would be largely reduced to a zanamivir monotherapy. Whereas the results of the primary virological endpoints indicated a superiority of the oseltamivir-zanamivir combination to zanamivir monotherapy, clinical results were not significantly different, suggesting that oseltamivir adds essentially nothing to zanamivir monotherapy. This view is concordant with the above hypothesis of the predominant catalytic site occupation by zanamivir when the combination is administered. As an exploratory analysis, oseltamivir showed a significantly higher clinical and virological efficacy as compared to zanamivir. This finding could be the consequence of a suboptimal treatment regimen in the zanamivir arm, since the IC50 values for the A(H3N2) viruses of the 2008-2009 season were 2-to 3-fold higher for zanamivir as compared to oseltamivir, but remained within the range for susceptible strains (GROG surveillance; NICs, unpublished data). The virological result is confirmed by the longer time Table 3 . Virological and clinical response according to treatment arms in the 447 influenza A-infected patients between day 0 and day 2 (ITT analysis). to alleviation of the influenza symptoms in patients receiving zanamivir. As the present study was conducted before the 2009 H1N1 pandemic during an influenza season where A(H3N2) viruses predominated, the impact of prepandemic seasonal A(H1N1) oseltamivir-resistant viruses on the results is expected to be negligible. We observed the same trends after excluding patients infected with seasonal H1N1 from our population. Of note, A(H3N2) viruses are to date sensitive to both drugs. It remains to be determined to what extent the present results can be extrapolated to susceptible viruses of other subtypes, e.g., H1N1, and in particular to the pandemic H1N1 2009 virus, which displays significant differences in the catalytic pocket of the neuraminidase [17] , or to a mixed viral season with H3N2 and H1N1 cocirculating viruses. We must acknowledge several limitations to our study. First, this preliminary analysis was conducted on a partial set of data after enrolment of 541 patients instead of the 900 initially planned. However, it is highly unlikely that the lower response of the combination as compared to oseltamivir would have been reversed if all originally planned 900 patients had been enrolled. Second, as previous randomized clinical trials had shown the superiority of each monotherapy as compared to placebo in terms of time to symptom alleviation and viral shedding, it was decided, on the basis of ethical reasons, that the study would not comprise a double placebo arm. Third, the proportion of patients with unavailable viral swab on day 2 was higher in the combination arm. As the missing value equals failure, this may have biased the results in the combination arm towards reduced performance. Indeed, in the analysis of the 414 patients with available day 0 and day 2 nasal swabs, the same trends were observed. Fourth, the virological response was assessed only in one site (nose) and at one time (day 2), which prevents extrapolation of the results to the entire virological response over time and throughout the respiratory tract. However, clinical endpoints completed the picture, giving information on the overall response. Fifth, as mentioned above, day 2 sampling was chosen to show the virological effect. However, this is probably not the best moment to look for resistance emergence induced by drug selective pressure, as it has been shown to occur later in the course of treatment [18] [19] [20] . Nevertheless, we looked for neuraminidase inhibitor resistance using a standard fluorimetric test in the 65 patients with day 2 positive viral culture; none of them carried a resistant virus, except for one patient infected with an H1N1 virus resistant to oseltamivir but susceptible to zanamivir, as were all H1N1 viruses circulating during the study period. However, the absence of resistance at day 2 does not rule out any further (postday 2) resistance selection. Finally, this trial was conducted in adult outpatients, which prevents any extrapolation of the results to adults with severe presentation necessitating hospitalisation, and to children, who usually have more prolonged viral shedding. We chose the outpatient adult population because it seemed to be the most homogenous and the easiest in which to test our hypothesis. Despite the theoretical potential for the reduction of the emergence of antiviral resistance, the lower efficiency of the oseltamivir-zanamivir combination found in this study calls for caution in its use in clinical practice. Thus, also considering the superiority of oseltamivir monotherapy over zanamivir monotherapy observed in this trial, oseltamivir should be the recommended primary anti-influenza treatment during influenza seasons with predominant H3N2 viruses naturally susceptible to oseltamivir. These results would need to be confirmed for the 2009 H1N1 pandemic virus and in the coming years, for future circulating influenza viruses. Table S1 Proportion of false positive and false negative results for various viral load thresholds compared to viral isolation in the sample of GROG patients.

A comparison of methods for purification and concentration of norovirus GII-4 capsid virus-like particles

Noroviruses (NoVs) are one of the leading causes of acute gastroenteritis worldwide. NoV GII-4 VP1 protein was expressed in a recombinant baculovirus system using Sf9 insect cells. Several methods for purification and concentration of virus-like particles (VLPs) were evaluated. Electron microscopy (EM) and histo-blood group antigen (HBGA) binding assays showed that repeated sucrose gradient purification followed by ultrafiltration resulted in intact VLPs with excellent binding to H type 3 antigens. VLPs were stable for at least 12 months at 4°C, and up to 7 days at ambient temperature. These findings indicate that this method yielded stable and high-quality VLPs.

Abstract Noroviruses (NoVs) are one of the leading causes of acute gastroenteritis worldwide. NoV GII-4 VP1 protein was expressed in a recombinant baculovirus system using Sf9 insect cells. Several methods for purification and concentration of virus-like particles (VLPs) were evaluated. Electron microscopy (EM) and histo-blood group antigen (HBGA) binding assays showed that repeated sucrose gradient purification followed by ultrafiltration resulted in intact VLPs with excellent binding to H type 3 antigens. VLPs were stable for at least 12 months at 4°C, and up to 7 days at ambient temperature. These findings indicate that this method yielded stable and high-quality VLPs. Noroviruses (NoVs) cause sporadic acute nonbacterial gastroenteritis in all age groups [12, 13] . NoVs are divided into five genogroups, GI to GV, of which GI and GII strains mostly affect humans [21] . Recently, NoV genotype GII-4 has been responsible for the majority of sporadic gastroenteritis cases and outbreaks [13] . The NoV genome consists of a single-stranded RNA of about 7.6 kb, organized into open reading frames (ORF 1-3). ORF-1 codes for the RNA-dependent RNA polymerase, and ORF-2 and -3 encode the two structural proteins VP1 and VP2 [10] . Expression of the capsid VP1 gene by recombinant baculoviruses leads to self-assembly into empty virus-like particles (VLPs) that are morphologically and antigenically similar to native NoV [9] . NoV VLPs are widely used as antigens in diagnostic serological assays and as candidate vaccines against NoVs [2, 7] . NoV VLPs are highly stable and resistant to variable conditions, particularly to low pH [1, 9] . There are limitations in NoV VLP production in terms of inadequate yield and quality of the VLPs [1, 3, 9, 20] . Both sucrose and CsCl gradients ultracentrifugation have been used for purification of NoV VLPs [1, 7, 9, 17] , even though studies on rotavirus-like particles demonstrated a low yield and impurities resulting from CsCl gradient purification [16] . In the present study, we compared commonly used methods for NoV GII-4 VLP purification [1, 14, 17] and concentration [6, 19] , considering the purity, yield, morphological integrity, antigenicity and functionality of the purified VLPs. The steps for cloning the NoV GII-4 (GenBank sequence database accession number AF080551) full-length capsid gene are described elsewhere [11] . VLPs were produced in Sf9 insect cells infected with the recombinant baculovirus according to the manufacturer's instructions (Invitrogen, Carlsbad, CA). Baculovirus titers expressed as the multiplicity of infection (MOI) of the P2 stocks were determined using a BacPak Rapid Titer Kit (Clontech Laboratories, Mountain View, CA). At day 6, infected cell culture (200 ml) was clarified by centrifugation at 30009g for 30 min at 4°C. VLPs in the supernatant were concentrated by ultracentrifugation (L8-60M ultracentrifuge, Beckman SW-32.1 Ti rotor) at 100,0009g for 2 h at 4°C, and pellets were resuspended in 0.2 M Tris-HCl, pH 7.3. VLPs were loaded onto a 10-60% discontinuous sucrose gradient and ultracentrifugated at 100,0009g for 1 h at 4°C as described before [17] . Fractions were collected by bottom puncture. The fractions containing VLPs were pooled. An additional discontinuous sucrose gradient (35-60%) ultracentrifugation was performed. Sucrose was removed by overnight dialysis against 1 liter of PBS. VLPs were concentrated by dialysis against polyethylene glycol (PEG; 50%) [6] or by ultrafiltration [19] . VLPs were concentrated using an Amicon Ultra 30 kDa centrifuge filter device (Millipore Corporation, Billerica, Germany). VLPs were stored at 4°C in PBS. Alternatively, a less time-consuming sucrose density gradient purification method was employed [14] . Clarified supernatants were pelleted twice by ultracentrifugation. Pellets were resuspended in 0.2 M Tris-HCl, pH 7.3, and placed on a discontinuous sucrose density gradient (10-60%) for ultracentrifugation at 100,0009g for 16 h at 4°C. The VLP band, which was visible at the 35% sucrose layer, was collected. Sucrose was removed by dialysis against 1 liter of PBS, and VLPs were concentrated by ultracentrifugation at 100,0009g for 2 h at 4°C. In addition, clarified supernatants were concentrated, and the pellets were resuspended in sterile water. VLPs were sedimented by ultracentrifugation through cesium chloride (CsCl) (0.4 g/ml) at 116,0009g for 18 h at 4°C as described earlier by others [1] . CsCl was removed by dialysis against PBS, and VLPs were concentrated using an Amicon Ultrafilter. The total protein content of the purified VLP preparation was determined using the Pierce BCA Protein Assay (Thermo Science, Rockford, USA). Endotoxin levels in the VLP preparations were quantified using the Limulus amebocyte lysate (LAL) assay (Lonza, Walkersville, MD, USA). The level of endotoxin was\0.1 EU/10 lg of protein, which is below the international standard of B30 EU/20 lg of protein [15] . All samples were analyzed for protein expression by sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE) and western blot. The presence of NoV VLPs was verified by electron microscopy (EM). VLP preparations were negatively stained with 3% uranyl acetate (UA), pH 4.6. The VLPs were examined using an FEI Tecnai F12 electron microscope operating at 120 kV. Binding of GII-4 VLPs to carbohydrate receptors was examined by the histo-blood group antigen (HBGA) binding assay as described by others [18] , with slight modifications. Briefly, VLPs were coated at 50 ng/well, and synthetic biotinylated H type 3 and Lewis b histo-blood group carbohydrates (Lectinity Holdings Inc. Moscow, Russia) were used in serial threefold dilutions, starting from 6 lg/ml. Wells lacking the synthetic carbohydrates were used as a negative control. The conditions for Sf9 cell infections were optimized in order to obtain a high yield of the VLPs. For recombinant baculovirus P2 stock, an MOI of 1 was found to be optimal, and VLPs were harvested from the supernatant after 6 days of infection. VLPs were further purified by several different procedures, as described in Fig. 1a . The best yield (2-3 mg/200 ml) was obtained after the purification procedure described in Fig. 1a, panel B. In comparison, the method described in Fig. 1a , panel C, yielded a ten times lower amount of NoV capsid VLPs. The purity of the VLPs obtained by each method was determined by 12% SDS-PAGE and staining of the proteins with Page-Blue (Fig. 1b) . 1 lg of the total protein was loaded into each lane. Each method resulted in equally pure protein bands corresponding to the size of the NoV capsid. Residual PEG present in the VLPs, concentrated by PEG dialysis, may have interfered with the protein concentration determination, and Fig. 1 Purification, concentration and characterization of NoV VLPs. a Several purification procedures (A-D) were used, and the purity, morphology, antigenicity and yield of the VLPs were compared. b NoV capsid protein analysis by 12% SDS-PAGE. c Western blot analysis using a human convalescent serum against norovirus GII-4. d EM of purified NoV VLPs observed at a magnification of 930,000. Bar 100 nm. Samples from purification procedures A, B, C, and D correspond to lanes and panels A, B, C, and D, respectively. M protein weight marker this could explain the lower capsid protein concentration observed (Fig. 1b, lane A) . NoV capsid identity and antigenicity of the VLPs were assayed by western blot using human NoV-specific convalescent sera (Fig. 1c) . The results show pure NoV capsid proteins without degradation and with similar antigenicity for each of the purification methods. Next, we determined the morphological integrity and homogeneity of the VLPs by EM (Fig. 1d) . The VLPs obtained by purification with sucrose density gradients followed by dialysis and either ultrafiltration (Fig. 1d, panel B) or ultracentrifugation (Fig. 1d, panel C) were approximately 38 nm in size, with the classical appearance of NoV capsid VLPs. By contrast, CsCl purification (Fig. 1d, panel D) and concentration of the VLPs by dialysis against PEG (Fig. 1d , panel A) resulted in VLPs of heterogeneous size, which appeared broken and aggregated. The VLPs purified and concentrated by the method schematically presented in Fig. 1a, panel B , were of the best quality and were subjected to further analysis. To determine the optimal storage conditions and stability of the NoV capsid VLPs, VLPs were treated under different conditions, and the samples were analyzed by SDS-PAGE to test their protein integrity (Fig. 2a) and by EM to examine their morphology (Fig. 2b) . VLPs were stable for at least 12 months at 4°C in PBS, pH 7.4 ( Fig. 2a and b, panel 1) , and at room temperature (23°C) up to 7 days ( Fig. 2a and b, panel 3) . The VLPs withstood the sterile filtration conditions when 0.22-lm filters (Durapore, Millipore, Ireland) were used. Next, we intentionally disrupted the VLP morphology by heat treatment at 60°C for 1 h (Fig. 2b, panel 4) [1] without degrading the capsid protein (Fig. 2a, lane 4) . An HBGA binding assay [8, 18] was used to test the functionality of the purified VLPs as well as to determine the significance of the conformational binding sites on the VLP, which would presumable require intact VLPs. A comparison of the binding of VLPs purified by sucrose gradient centrifugation and by CsCl sedimentation, as well as heat-treated VLPs (60°C, 1 h), to synthetic biotinylated H type 3 carbohydrate is shown in Fig. 3 . The binding was clearly dependent on the morphology and preserved structure of the VLPs, with the lowest level of binding observed with the heat-treated VLPs with disrupted conformational binding sites. Lewis b antigen was used as a control in the assay, and this did not bind to any of the VLPs (Fig. 3) . NoV VLPs have been used extensively to study protein interactions [8] , and virus assembly [17] , and have been used as a tool in diagnostic serological assays [7] . Clinical trials have been performed with the NoV VLPs used as a vaccine [2] . For all these applications, high-quality VLPs would be preferable. In this study, different methods of VLP purification and concentration were used, and the purity, integrity, morphology, antigenicity and functionality of the GII-4 NoV VLP preparations were examined. VLPs purified by each method (Fig. 1a , methods A-D) had a similar appearance and migration pattern on the SDS-PAGE gel. A western blot with a human convalescent serum from an individual infected with GII-4 confirmed the identity of the capsids and the lack of degradation products. The reason for the lower protein yield after purification procedure C might have been that a narrow visible band of VLPs was collected from 35% sucrose, causing some of the protein to be omitted, thus affecting the yield. However, a yield of up to 2-3 mg of VLPs, which was obtained by the best purification method described in the present study, is remarkably high when compared to other reports [3, 4] . EM analysis of the morphological integrity and homogeneity of the VLPs showed that VLPs obtained by purification with sucrose density gradients followed by ultrafiltration or ultracentrifugation (procedures B and C, Carbohydrate concentration (µg/ml) Sucrose VLPs+H type 3 Sucrose VLPs+Lewis b Heat treated VLPs+H type 3 Heat treated VLPs+Lewis b CsCl VLPs+H type 3 CsCl VLPs+Lewis b Fig. 3 Binding of NoV VLPs to synthetic ABH histo-blood group antigens. Sucrose-purified VLPs (Sucrose VLPs, purified according to procedure B in Fig. 1a) , heat-denatured NoV VLPs (60°C, 1 h) (Heattreated VLPs) and CsCl-purified VLPs (CsCl VLPs) were tested for binding to H type 3 and Lewis b carbohydrates at the indicated concentrations. The pH value in the binding assay was 7.4 respectively) were homogenous and intact, and approximately 38 nm in size. However, CsCl-purified VLPs appeared heterogeneous in size, with a few broken particles, although comparable to the morphology seen by others [7] . CsCl purification is known to introduce several impurities at the end of the process and cause aggregation of the VLPs during storage [5] . The poorest morphology was seen after concentration of the VLPs by dialysis against PEG, which resulted in aggregation. In addition, residual PEG, which leaks through the dialysis membrane, might interfere with further applications of the VLPs [19] . Our data clearly demonstrate that the purification process affects the integrity of the native quaternary structure of NoV VLPs and, subsequently, the receptor-binding functionality of the VLPs. Although the majority of the VLPs purified by the CsCl method seemed intact in the EM image, the difference from sucrose-density-gradient-purified VLPs in HBGA binding is striking. This result is supported by the recent finding that CsCl has a negative impact on the functionality of VLPs [5] . We also demonstrated that even heat-disrupted VLPs have binding capability, but intact homogenous VLPs have a significantly greater binding intensity. Standardization of the purification method for NoV VLPs used in diagnostic serological assays and blocking assays [8] would greatly strengthen the results obtained. To the best of our knowledge, this is the first time that VLPs purified by conventional purification and concentration methods were compared in terms of yield, purity, morphological integrity, antigenicity and functionality. The results show that NoV GII-4 VLPs purified twice by sucrose density gradient centrifugation followed by ultrafiltration maintain their icosahedral capsid structure and their capacity to bind HBGAs.

Urine Peptidomic and Targeted Plasma Protein Analyses in the Diagnosis and Monitoring of Systemic Juvenile Idiopathic Arthritis

PURPOSE: Systemic juvenile idiopathic arthritis is a chronic pediatric disease. The initial clinical presentation can mimic other pediatric inflammatory conditions, which often leads to significant delays in diagnosis and appropriate therapy. SJIA biomarker development is an unmet diagnostic/prognostic need to prevent disease complications. EXPERIMENTAL DESIGN: We profiled the urine peptidome to analyze a set of 102 urine samples, from patients with SJIA, Kawasaki disease (KD), febrile illnesses (FI), and healthy controls. A set of 91 plasma samples, from SJIA flare and quiescent patients, were profiled using a customized antibody array against 43 proteins known to be involved in inflammatory and protein catabolic processes. RESULTS: We identified a 17-urine-peptide biomarker panel that could effectively discriminate SJIA patients at active, quiescent, and remission disease states, and patients with active SJIA from confounding conditions including KD and FI. Targeted sequencing of these peptides revealed that they fall into several tight clusters from seven different proteins, suggesting disease-specific proteolytic activities. The antibody array plasma profiling identified an SJIA plasma flare signature consisting of tissue inhibitor of metalloproteinase-1 (TIMP1), interleukin (IL)-18, regulated upon activation, normal T cell expressed and secreted (RANTES), P-Selectin, MMP9, and L-Selectin. CONCLUSIONS AND CLINICAL RELEVANCE: The urine peptidomic and plasma protein analyses have the potential to improve SJIA care and suggest that SJIA urine peptide biomarkers may be an outcome of inflammation-driven effects on catabolic pathways operating at multiple sites. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s12014-010-9058-8) contains supplementary material, which is available to authorized users.

Clinical Relevance Sensitive and specific diagnostic biomarkers for systemic onset juvenile idiopathic arthritis (SJIA) would allow its differentiation from other febrile illnesses, such as Kawasaki disease (KD) or acute infections (febrile illness (FI)) and facilitate prompt initiation of appropriate treatment at disease onset. Early treatment may reduce the risk of long-term complications and subsequent disabilities. In addition, biomarkers that distinguish intercurrent SJIA flare from infection in patients with known SJIA would be clinically useful, as would markers that predict impending disease flare or responder status to particular therapies, or provide an early indication of a treatment response. Finally, biomarkers may provide clues to unanswered questions concerning SJIA pathogenesis. Our comparative analysis of SJIA, KD, and FI urine peptidomes identified a small subset of the urine peptidome that effectively discriminates SJIA patients in the active, quiescent, and remission disease states, and also discriminates patients with active SJIA from confounding conditions including KD and FI. Urine peptide diagnostic and prognostic biomarkers could be of clinical use, especially for serial sampling of pediatric SJIA patients. Targeted sequencing revealed that these peptide markers fall into several tight clusters indicating SJIA-specific proteolytic events. Plasma cataloging analysis of the normal plasma peptidome shows that at least some of these nested peptide markers originate in the circulation. A customized antibody array was used to compare the plasma abundance of proteins known to be involved in inflammatory and protein catabolic processes, revealing a SJIA flare signature. Taken altogether, the urine peptidomic and plasma protein and peptide analyses suggest a testable model that SJIA urine peptide biomarkers may be an outcome of inflammation-driven effects on catabolic pathways operating at multiple sites. Electronic supplementary material The online version of this article (doi:10.1007/s12014-010-9058-8) contains supplementary material, which is available to authorized users. biomarkers may be an outcome of inflammation-driven effects on catabolic pathways operating at multiple sites. Systemic onset juvenile idiopathic arthritis is a chronic inflammatory disease of childhood characterized by a combination of systemic features [fever, rash, serositis (e.g., pericarditis, pleuritis)] and arthritis. Current diagnosis of SJIA is based solely on clinical findings [1] and requires arthritis, daily fever for at least 2 weeks, and at least one of the following: evanescent erythematous rash, generalized lymph node enlargement, hepatomegaly and/or splenomegaly, or serositis. This makes early diagnosis of SJIA challenging, as its clinical manifestations are similar to other diseases, including malignancy, infection, Kawasaki disease (KD), and other autoimmune or inflammatory disorders. Long-term disease outcome in SJIA is variable. About 50% of patients experience a single episode that resolves. However, the other half experience either polycyclic or non-remitting disease. Sensitive and specific diagnostic biomarkers for SJIA would allow its differentiation from other febrile illnesses, such as KD or acute infections (febrile illness (FI)), and facilitate prompt initiation of appropriate treatment at disease onset. Early treatment may reduce the risk of long-term complications and subsequent disabilities. In addition, biomarkers that distinguish intercurrent SJIA flare from infection in patients with known SJIA would be clinically useful, as would markers that predict impending disease flare or responder status to particular therapies, or provide an early indication of a treatment response. Finally, biomarkers may provide clues to unanswered questions concerning SJIA pathogenesis. There have been several previous biomarker discovery efforts in SJIA. Initial studies, including ours [2] , attempted to identify early clinical variables that predict long-term outcomes, such as joint damage or functional disability at ≥2 years after disease onset [3] [4] [5] [6] . Studies of serum found elevated cytokines, chemokines, and acute-phase reactants in active SJIA [7] [8] [9] [10] . More recently, transcriptional profiling of peripheral blood mononuclear cells from SJIA subjects with active disease revealed a signature of active SJIA that normalized in association with clinical response to treatment [11, 12] . A single SELDI-based analysis of plasma identified serum amyloid A as a plasma biomarker of disease activity [13] . However, all these efforts fall short of robust diagnostic and prognostic biomarkers with practical clinical utility. We sought to explore urine as a source of biomarkers. Such markers would permit frequent tests, which would be of use, especially in children, for a chronic pediatric disease with a polycyclic course. A normal adult human excretes 30-130 mg of protein and 22 mg of peptides per day in urine [14, 15] . Urine proteomic analysis has identified more than 1,500 proteins including a large proportion of membrane proteins [16] . Urine peptidomic analysis revealed over 100,000 different peptides [17] . Our own in-depth 2D mass spectra (MS)/MSMS analysis led to the identification of 11,988 different urine peptide sequences from 8,519 unique protein precursors in normal human urine [18] . Recent reviews have indicated that analysis of the urinary proteome/peptidome can be highly informative for both urogenital and systemic diseases and used for disease classification [18, 19] . Naturally processed urine peptides have certain advantages over proteins as biomarkers. The roughly equal mass of protein and peptide in urine translates into at least a tenfold greater molar abundance of peptides. While the urine proteome contains a number of abundant proteins that obscure the lower abundance proteins more likely to be biomarkers, this problem does not complicate analysis of peptides in urine. A one-dimensional HPLC separation is sufficient for the analysis of greater than 25,000 urine distinct peptides. Among the emerging quantitative proteomics technologies, isobaric tags for relative and absolute quantification (iTRAQ) allows concurrent protein sequence identification and relative quantification of those peptides with known protein sequences in up to eight different biological samples in a single experiment [20] . However, due to its limited throughput and current cost, iTRAQ is not feasible for simultaneous comparison of the large number of disease subjects needed to achieve the discovery of differential features with sufficient statistical power. As an alternative, a label-free liquid chromatography-mass spectrometry (LC-MS)-based approach has been applied as a quantitative biomarker discovery method. The label-free LC-MS approach can compare and quantify peptides with precision and accuracy comparable to those based on isotope labeling [21] . Although LC/ESI mass spectrometry is typically used in label-free quantitative proteomics, matrix-assisted laser desorption/ionization-timeof-flight (MALDI-TOF) mass spectrometry is increasingly being used and demonstrates low average coefficients of variation for all peptide signals across the entire intensity range in all technical replicates [22, 23] . Using the label-free LC/MALDI-TOF profiling approach, we previously discovered candidate urine peptide biomarkers of renal transplant rejection [19, 24] . Subsequent urine peptide biomarker validation [19] by multiple reaction monitoring (MRM) [25, 26] showed significant correlation between the urine peptide measurements obtained from label-free MALDI-TOF and from MRM using stable isotope-labeled synthetic marker analogues to derive absolute quantification. The label-free LC-MALDI-TOF approach involves the comparison of urine peptidomes of different samples, and thus, multiple LC-MS spectra. However, comparing multiple LC-MS spectra in a label-free analysis is computationally intensive, demanding robust detection of LC-MS peaks, alignment of all LC-MS peaks, and determination of the common peak indices across all assayed samples. The output of data processing is essentially a P X N table in which each of the indexed P peptides has been quantified across the N studied samples. This table, reduced from LC-MS spectra of all samples, can be subjected to downstream statistical learning including transformation, normalization, and unsupervised/supervised analyses suited to the experimental design to mine for a differential subset of the P peptides, which will then be subjected to MSMS protein sequence identification and future quantitative prospective MRM [25, 26] or antibodybased validation. We identified naturally occurring urine peptides with specificity for active systemic SJIA compared with other sources of fever. We hypothesized that SJIA flare is associated with increased levels of circulating mediators of inflammation that activate catabolic pathways leading to the generation of novel peptide biomarkers that are found in urine. We tested this hypothesis through global LC-MS analysis of urine and plasma peptides as well as targeted analysis of plasma proteins using antibody arrays. The following reagents were used for the proteomics sample analysis: nanopure or Milli-Q quality water (~18 megohm cm or better); Amicon Ultra centrifugal filtration tubes were obtained from Millipore (Bedford, MA, USA) ammonium bicarbonate, ammonium formate, and formic acid were obtained from Fluka (St. Louis, MO, USA); Tris-HCl, urea, thiourea, DTT, iodoacetamide, calcium chloride, and TFA were obtained from Sigma-Aldrich (St. Louis, MO, USA); HPLC-grade methanol (MeOH) and HPLC-grade ACN were purchased from Fisher Scientific (Fair Lawn, NJ, USA); 2,2,2-trifluoroethanol was obtained from Aldrich Chemical (Milwaukee, WI, USA); and sequencing grade-modified trypsin was purchased from Promega (Madison, WI, USA). Sodium tetraborate, glycine, and picrylsuofonic acid were obtained from Sigma-Aldrich (St. Louis, MO, USA). Informed consent was obtained from the parents of all patients and assent from all patients >6 years of age. This study was approved by the human subject protection programs at UCSD, UCSF, and Stanford University. Urine samples were obtained from two new onset SJIA disease (ND), 18 active systemic disease plus arthritis (SAF), nine SJIAwith active arthritis (AF), 18 quiescent SJIA on medication (QOM), nine SJIA in remission off medication (RD), and ten healthy control (HC) from Stanford University Medical Center and UCSF. In addition, urine samples were obtained from 23 KD and 23 age-similar FI control patients evaluated for fever at Rady Children's Hospital San Diego. All KD patients had fever and ≥4 of the five principal clinical criteria for KD (rash, conjunctival injection, cervical lymphadenopathy, changes in the oral mucosa, and changes in the extremities) or three criteria plus coronary artery abnormalities documented by echocardiography [27] All FI control, patients had naso-or oro-pharyngeal and stool viral cultures. Urine sample patient demographics are described in Tables 1 (SJIA) and 2 (KD and FI). Plasma samples included 25 SJIA flare (F), 14 SJIA (Q) for the training analysis, and 41 SJIA F and 11 Q for the "bootstrapping" testing analysis. Instead of in silico bootstrapping simulation, samples belonging to different visits of the same patient and even the same samples were assayed, i.e., "bootstrapped" experimentally, for testing. For the bootstrapping testing, a total of 52 SJIA samples were analyzed by the antibody array, where 41 samples were from 19 patients at the time of SJIA flare, and 11 samples were from eight patients at the time of SJIA quiescence. Plasma sample patient demographics are described in Table 4 . Urine samples (5-10 mL) were collected in sterile tubes and held at 4°C for up to 48 h before centrifugation (2,000×g× 20 min at room temperature) and freezing of the supernatant at −70°C. Urine processing, preparation of peptides, extraction, and fractionation are as previously described [18, 19, 24] . Urinary samples were processed by centrifugal filtration at 3,000×g for 20 min at 10°C through Amicon Ultra centrifugal filtration devices (10 kDa cutoff; Millipore, Bedford, MA) preequilibrated with 10 ml Milli-Q water. The filtrate (urine peptidome) containing the low MW naturally occurring peptides was processed with Waters Oasis HLB Extraction Cartridges (Waters Corporation, Milford, MA) and extracted with ethyl acetate. The resulting urine peptide samples were quantified by the 2,4,6-trinitrobenzenesulfonic acid (TNBS) assay, as previously described [28] . Three nanomoles peptides were injected on a 100 μm×15 cm C18 reverse-phase column (Michrom) and eluted with a gradient of 5% to 55% acetonitrile over 50 min using a Michrom MS4 HPLC. Twenty-second fractions were collected onto MALDI targets with a Probot fraction collector (LC Packings). A total of 100 fractions were collected and analyzed on 4700 MALDI-TOF/TOF (Applied Biosystems) in MS mode. One microliter of matrix solution containing 4.8 mg/ml α-cyano-4-hydroxycinnamic acid (Agilent Technologies, Palo Alto, CA, USA) and 30 fmol/μl glu-fibrinopeptide (Sigma-Aldrich, St. Louis, MO, USA) was automatically deposited by the Probot on each spot. The plasma peptidome preparation protocol was adapted from that of the urine peptidome analysis. The plasma samples were centrifuged at 3,000×g for 20 min at 10°C through Amicon Ultra centrifugal filtration devices (10 kDa cutoff; Millipore, Bedford, MA) pre-equilibrated with 10 ml Milli-Q water. The retentate (plasma proteome) was washed twice, brought to the final volume of 400 μl with 20 mM Tris-HCl (pH 7.5), and quantitated by the BCA protein assay (Pierce, Rockford, IL). The filtrate (plasma peptidome) containing the low MW naturally occurring peptides was processed with Waters Oasis HLB Extraction Cartridges (Waters Corporation, Milford, MA), and extracted with ethyl acetate. The resulting plasma peptide samples were quantified by the TNBS assay, as previously described [28] . Three nanomoles of peptides were fractionated by two-dimensional chromatography-a SCX column as the first and a RP column as the second dimension, and then subjected to extensive MSMS sequence identification involving a Thermo Finnigan LTQ-FTICR spectrometer. The ABI 4700 oracle database MS spectra were exported as raw data points via ABI 4700 Explorer software version 2.0 for subsequent data analyses. The m/z ranges were from 800 to 4,000 with peak density of maximum 30 peaks per 200 Da, minimal S/N ratio of 5, minimal area of 10, minimal intensity of 150, and 200 maximum peaks per spot. We previously had developed an informatics platform [18] which contains an integrated set of algorithms, statistical methods, and computer applications, to allow for MS data processing and statistical analysis of LC-MS-based urine peptide profiling. The MS peaks were located in the raw spectra of the MALDI data by an algorithm that identifies sites (mass-to-charge ratio, m/z values) whose intensities are higher than the estimated average background and the~100 surrounding sites, with peak widths~0.5% of the corresponding m/z value. To align peaks from the set of spectra of the assayed samples, we applied linkage hierarchical clustering to the collection of all peaks from the individual spectra. The clustering, computed on a 24-node LINUX cluster, is two dimensional, using both the distance along the m/z axis and the HPLC fractionation time, with the concept that tight clusters represent the same biological peak that have been slightly shifted in different spectra. We then extracted the centroid (mean position) of each cluster, to represent the "consensus" position as the peak index (bin) across all spectra. The normalization of the MALDI-TOF signal intensity for each peptide feature was performed at two steps: (1) within each LC fraction (MALDI plate spot), all peptide peak signal intensities were normalized to the externally spiked reference peptide (30 fmol/μl glu-fibrinopeptide) at each MALDI plate spot; (2) each clustered peptide, with unique m/ z and LC fraction time, was normalized to the total signal intensity of all the clustered peptides within the same sample. For urine peptidome analysis, we used the approach of ion mapping [29, 30] , whereby biomarker candidate MS peaks were selected on the basis of discriminant analysis, and then targeted for MS/MS sequencing analysis to obtain protein identification. Extensive MALDI-TOF/TOF and LTQ Orbitrap MS/MS analyses coupled with database searches [29, 30] were performed to sequence and identify these peptide biomarkers. The identity of a subset of peptides detected was determined by searching MS/MS spectra against the Swiss-Prot database (June 10, 2008) restricted to human entries (15,720 sequences) using the Mascot (version 1.9.05) search engine. Searches were restricted to 50 and 100 ppm for parent and fragment ions, respectively. No enzyme restriction was selected. Since we were focusing on the naturally occurring peptides, hits were considered significant when they were above the statistical significant threshold (as returned by Mascot). Selected MS/MS spectra were also searched by SEQUEST (BioWorks™ rev.3.3.1 SP1) against the International Protein Index human database version 3.5.7 restricted to human entries (76,541 sequences). mMASS, an open source mass spectrometry tool (http://mmass.biographics.cz/), was used for manual review of the protein identification and MS/MS ion pattern analysis for additional validation. Customized antibody arrays, consisting of pairs of capture and detection antibodies against 43 proteins, were utilized (SDF-1), IGF1, IFNG, IGFBP3, IGFBP4, IGFBP6, IL-1A, IL-1B, IL-1R1, IL-2, IL-4, IL-5 . Antibody array fabrication, processing, data extraction and analysis were performed as previous described [31] . Patient demographic data were analyzed using "Epidemiological calculator" (R epicalc package, version 2.10.0.0). The binned LC-MALDI MS peak data obtained for all urine peptidome samples were analyzed for discovery of discriminant biomarkers using algorithms [32] of nearest shrunken centroid (NSC) for biomarker feature selection, tenfold cross-validation analyses, and Gaussian linear discriminant analysis (LDA) for classification analyses. To control the number of false significant features found during NSC mining, we permuted the data set 500 times to calculate GFDR [33] . To quantify the difference between classes for the identified peptide biomarkers, Student's t test and Mann-Whitney U tests were used for hypothesis testing, and local false discovery rate (FDR) [34] tool was used to correct multiple hypothesis testing. In order to test whether the selected discriminated features could serve as a diagnostic biomarker panel, a logistic regression model was used to find a linear combination of the biomarkers that minimizes the total classification error. In order to avoid bias in data sets, we utilized a bootstrapping technique to bootstrap 500 times to evaluate the impact of the data construction on overall classification performance of the biomarker panel. For each of the bootstrapping sets, we used the LDA-derived prediction scores for each sample to construct receiver operating characteristic (ROC) curves [35, 36] . To summarize the results, the vertical average of the 500 ROC curves was plotted, and the boxes and whiskers were used to describe the vertical spread around the average. We collected 56 intraday urine samples from pediatric SJIA patients at two sites, Stanford and UCSF (Tables 1, 2, and 3). These included patients with ND (n=2), SAF (n=18), AF (n=9), SJIA quiescence (inactive disease on medication; QOM, n=18), SJIA remission (inactive disease off all medications) (RD, n=9). For comparison, samples from subjects with KD (n=23), and acute FI (n=23) ( Table 2; collected at UCSD), and healthy, age-matched controls (HC, n=10, collected at Stanford) were collected. We also collected 66 plasma samples from pediatric SJIA patients (Tables 4 and 5 ). These included patients with SAF (n=25) and (n=14). If available, plasma samples from multiple visits, considered as experimentally "bootstrapped" samples, of the same SJIA patient at different disease states were also collected for confirmatory analyses using bootstrapping. Thirteen patients provided both urine and plasma samples. As expected, based on known differences in demographics [37] , there were differences in the age and gender distribution of our SJIA and the KD and FI urine subjects. Except for ND patients (median age, 3 years; range, 1-5 years), the SJIA patients (SAF: median age, 12.5 years; range, 3-17 years; AF: median age, 13 years; range, 11-16 years; QOM: median age, 13 years; range, 5-17 years; RD: median age, 14 years; range, 6-21 years) are older than KD (median age, 3 years; range, 1-10 years) and FI (median age, 2 years; range, 1-10 years) patients. Except for ND patients (100% male), there are fewer male SJIA patients (SAF, 33% male; AF, 33% male; QOM, 39% male; RD, 22% male) than KD (82% male) and FI (61% male) patients. As expected, active Table 5) . Mass spectrometry-based urine peptidomics analyses suffer from two major sources of variance [18] : analytical issues including mass spectrometric ion suppression; and biological issues including dilution of urine by different hydration states of the urine donors. To standardize amount of urine peptides for comparative analysis, we have quantified each extracted urine peptidome by the TNBS assay [28] and 3 nmol of peptides were subjected to the downstream LC-MALDI-TOF profiling analysis. The initial step in our biomarker discovery effort was to collect urine peptide spectra by LC-MALDI-TOF profiling from the 56 urine samples. The MS spectrum of each HPLC fraction was analyzed by "MASS-Conductor" software (Ling, unpublished) , which extracts peaks from raw MALDI spectra, enables common peak alignment, generates consensus representative peaks across all spectra via two-dimensional hierarchical clustering of both mass/ charge and the HPLC fractions, and normalizes peak signal measurements. To discover an SJIA systemic flare signature, the urine spectra of the subjects with systemically active disease, SJIA ND (n=2), and SAF (n=18) patients, were compared simultaneously to the non-systemic group of SJIA AF (n=9) and the QOM (n=18) patients and the other systemic inflammation groups: KD (n=23) and FI (n=23). The data mining process included selection of the discriminative urine "Bootstrapping" differing from in silico bootstrapping (re-sampling) simulation, samples belonging to different visits of the same patient and even the same samples were assayed, i.e. "bootstrapped" experimentally peptides, supervised classification, bootstrapping, and ROC analysis, as outlined in Fig. 1 . Classifier discovery and feature selection by the NSC algorithm [32] were performed using all the features in the data set. NSC algorithm iteratively shrinks the standardized class mean of the abundance for each peptide. Eventually all urine peptides were ranked by the difference between the shrunken class means. Tenfold internal cross-validation analysis and LDA led to the discovery of a biomarker panel of 17-urine-peptide biomarkers effectively differentiating SJIA flare (ND and SAF) from contrasting group of AF, QOM, RD, KD, and FI samples. Extensive MALDI-TOF/ TOF and LTQ Orbitrap MS/MS analysis coupled with database searches [29, 30] were then performed to identify these peptide biomarkers. As shown in Tables 6 and 7, the 17-peptide biomarkers were found to be degradation products of eight different proteins: alpha1 antitrypsin (A1AT, two peptides having overlapping sequences), collagen type I alpha 1 (COL1A1; five peptides and three of them having overlapping sequences), collagen type I alpha 2 (COL1A2; one peptide), collagen type III alpha 1 (COL3A1; one peptide), collagen type IX alpha 2 (COL9A2; one peptide), fibrinogen alpha (FGA; two peptides having overlapping sequences), fibrinogen beta (FGB; two peptides having overlapping sequences), and uromodulin (UMOD; three peptides having overlapping sequences). Sequence alignment of these peptide biomarkers revealed tight sequence clusters for A1AT-, COL1A1-, FGA-, FGB-, and UMODderived biomarkers. The Mann-Whitney U tests (Table 6) were performed to evaluate the significance of discriminations "Bootstrapping" differing from in silico bootstrapping (re-sampling) simulation, samples belonging to different visits of the same patient and even the same samples were assayed, i.e. "bootstrapped" experimentally (Table 7) Fig. 2 Evaluation of the 17-urine-peptide biomarker panel as a classifier of SJIA versus systemic inflammation from Kawasaki disease or acute febrile illness. a A logistic regression model was used to find a panelbased algorithm that minimizes the total classification error discriminating SJIA systemic disease from inflammation due to KD/FI. The maximum estimated probabilities for each of the wrongly classified samples, are labeled with arrows. b A modified 2×2 contingency table shows the percentage of classifications that agreed with clinical diagnosis. c The discriminant analysis-derived prediction scores for each sample were used to construct a receiver operating characteristic (ROC) curve; 500 testing data sets, generated by bootstrapping, from the SJIA systemic flare, KD, and FI data were used to derive estimates of standard errors and confidence intervals for our ROC analysis. The plotted ROC curve is the vertical average of the 500 bootstrapping runs, and the box and whisker plots show the vertical spread around the average. d Distribution of the standardized ROC AUC values of the 500 falsely discovered panels upon the 500 class-label permutated data set of the cohort of SJIA F and KD/FI urine peptidomes. Examining all the 500 falsely discovered biomarker panel ROC AUC values, the number of falsely discovered same-size panels that have ROC AUC values greater than that of the original urine biomarker panel (represented by the red vertical line) dividing the total number of the "falsely discovered" biomarker panels led to the estimation of false discovery rate FDR Seventeen-Urine-Peptide Biomarker Panel Effectively Discriminates SJIA Flare from KD and FI In order to test whether the 17-peptide biomarkers could collectively serve as a diagnostic biomarker panel, a logistic regression model was then used to find a linear combination of the 17-peptide biomarkers that minimizes the total classification error discriminating SJIA systemic ND and SAF patients from KD and FI patients. Figure 2a plots the linear discriminant probabilities of the peptide biomarker panel. Samples had good separation between the highest and next highest probability for the classification. Seventeen of the 20 SJIA flare and all 46 non-SJIA (KD and FI) patients were correctly classified. The maximum estimated probabilities for each of the wrongly classified samples, are labeled with arrows. A modified 2×2 contingency table (Fig. 2b) To evaluate the performance of our peptide panel for separating SJIA flare from KD and FI, we used the discriminant analysis-derived prediction scores for each sample and constructed ROC curves [35, 36] . In addition, we utilized bootstrapping, a re-sampling technique to construct multiple-testing data sets to further evaluate the classification performance of the 17-urine-peptide biomarker panel. Figure 2b summarizes the 500 bootstrapping runs of the SJIA systemic flare, KD, and FI samples to derive the estimates of standard errors and confidence intervals for our ROC analysis. The plotted ROC (Fig. 2c) curve is the vertical average of the 500 bootstrapping runs, and the boxes and whiskers plot the vertical spread around the average. The ROC analysis yielded an averaged area under the curve (AUC) value of 0.999, indicating high performance. We next sought to determine whether the panel of the 17urine-peptide biomarkers could serve as a flare signature to discriminate SJIA flare samples from samples of patients at QOM and RD. A logistic regression model was used to find a linear combination of the 17-urine-peptide biomarkers to minimize the total classification error, classifying patients of SJIA systemic flare from QOM and RD. Figure 3a plots the linear discriminant probabilities of the peptide biomarker panel. Samples had good separation between the highest and next highest probability for the classification. Eighteen of 20 SJIA flare (90%) and all 27 SJIA quiescent or remission patients were correctly classified. The maximum estimated probabilities for each of the wrongly classified samples are labeled with arrows. A modified 2×2 contingency table (Fig. 3b) shows the percentage of classifications that agreed with clinical diagnosis. Overall, the 17-urinepeptide biomarker panel classified the SJIA flare samples with 90% positive agreement with the clinical diagnosis and quiescent or remission samples with 100% agreement with the clinical diagnosis (P=4.16×10 −11 ). Figure 3c summarizes the 500 bootstrapping runs of the SJIA systemic flare, quiescent and remission samples to derive the estimates of standard errors and confidence intervals for our ROC analysis. The plotted ROC (Fig. 3c) curve is the vertical average of the 500 bootstrapping runs, and the boxes and whiskers plot the vertical spread around the average. The ROC analysis yielded an AUC value of 0.998. In the ROC analyses of the 17-urine-peptide biomarker panel for discriminating SJIA F versus QOM/RD or SJIA F versus KD/FI, bootstrapping (a re-sampling technique) was used to avoid bias due to the presence of outliers in our assayed samples. In both cases shown in the ROC plots, the ROC analyses (Figs. 2 and 3c) yielded a significant AUC, indicating the ROC curve was not affected significantly by the bootstrapping process and demonstrating the robustness of our 17-urine-peptide biomarker panel in discriminative analyses. As observed in other high throughput analyses, e.g., microarray expression profiling, where the number of profiled features greatly exceeds that of the assayed samples, concurrent analysis of MALDI-TOF spectral peaks to evaluate null hypotheses for differential urine peptide biomarkers leads to the multiple-testing problem. To address the multiple-hypothesis testing problem, we estimated the FDR in concurrent statistical tests of peptide panels, of the same size as our biomarker panel; multiple permutated "random" training data sets were constructed. The class labels of our training samples in either cohorts of SJIA F and QOM/RD, or cohorts of SJIA F and KD/FI, were permutated 500 times such that each time every sample would be randomly assigned a new class label (SJIA F or QOM/RD in F and QOM/RD discrimination; SJIA F or KD/FI in SJIA F and KD/FI discrimination). For each of the 500 simulated "training" sets, NSC algorithm was applied to rank all the MALDI-TOF spectral peak features based upon their ability to discriminate the binary classes: SJIA F versus QOM/RD; and SJIA F versus KD/ FI, respectively. The NSC-selected top 17-peak features were then designated as the "panel" for LDA analysis. ROC analysis subsequently was used to calculate the AUC for this "falsely discovered panel". The AUC values of the 500 falsely discovered panels were standardized, and the density distribution was plotted in Figs. 2 and 3d. FDR was calculated as the number of AUC values greater than that of our 17-urine-peptide panel divided by the total number of AUC values of the "falsely" discovered panels. As shown in Figs. 2 and 3d , the FDRs of our urine peptide biomarker panel are estimated as 0.2% in SJIA F versus QOM/RD discrimination, and 0.2% in SJIA F versus KD/ FI discrimination respectively. These results support the notion that the discovery of our peptide biomarker panel is Fig. 3 Evaluation of the 17-peptide biomarker panel as a classifier of active SJIA versus inactive (quiescent or remitted) SJIA. a A logistic regression model was used to find a panel-based algorithm that minimizes the total classification error discriminating active systemic SJIA from inactive SJIA. The maximum estimated probabilities for each of the wrongly classified samples, are labeled with arrows. b A modified 2×2 contingency table shows the percentage of classifications that agreed with clinical diagnosis. c The discriminant analysis-derived prediction scores for each sample were used to construct a receiver operating characteristic (ROC) curve; 500 testing data sets, generated by bootstrapping, from the SJIA systemic flare, and inactive SJIA data were used to derive estimates of standard errors and confidence intervals for our ROC analysis. The plotted ROC curve is the vertical average of the 500 bootstrapping runs, and the box and whisker plots show the vertical spread around the average. d Distribution of the standardized ROC AUC values of the 500 falsely discovered panels upon the 500 classlabel permutated data set of the cohort of SJIA F and QOM/RD urine peptidomes. Examining all the 500 falsely discovered biomarker panel ROC AUC values, the number of falsely discovered same-size panels that have ROC AUC values greater than that of the original urine biomarker panel (represented by the red vertical line) dividing the total number of the "falsely discovered" biomarker panels led to the estimation of false discovery rate FDR unlikely to be the outcome of chance in multiple hypothesis testing. Direct Sequencing and Cataloging of Naturally Occurring, Normal Plasma Peptides Revealed Nested COL1A1, FGA, and FGB Peptides That Are Related to SJIA Urine Peptide Biomarkers We reasoned that, at least, some of the SJIA urine peptide biomarkers, such as FGA peptide biomarkers, are likely filtered from the circulation into urine. To explore this, we fractionated normal plasma by two-dimensional chroma-tography to extract the naturally occurring peptides for MSMS peptide sequencing. Similar to other plasma peptide direct sequencing efforts [38] , we observed FGA peptide clusters. We found seven FGA peptide clusters in plasma (Electronic Supplementary Table 1) , and our urine peptide biomarkers FGA (20-38) 1,536.69, FGA (605-628) 2,560.2, and FGA (605-629) 2,659.24 were observed in plasma FGA peptide clusters I and VII. We did not detect FGA (605-621) 1,826.80, although this peptide was found in a published plasma peptidome sequencing effort [38] . Urinary peptide FGA (607-622) Table 8 Identification of peptides found in normal plasma that are related to SJIA urine peptide biomarkers previously published one [38] . The urinary FGA peptides found in active SJIA samples (Table 8 ) lack one or two Cterminal residues compared with a related plasma peptide and thus appear to derive from exopeptidase activity. The urinary COL1A1 peptide that is most differentially expressed in active SJIA (Table 8) extends beyond the C terminus of a closely related peptide found in normal plasma, suggesting it may be generated by inhibition of normal protease activity during SJIA. A urinary FGB peptide found in SJIA (at similar levels in KD, but not comparison groups with less systemic inflammation) is identical to a peptide found in normal plasma, suggesting that the precursor protein is increased during inflammation. Our data indicate that at least some SJIA urine peptide biomarkers likely originate in circulation and are filtered into the urine. Identification of TIMP1, IL18, RANTES, P-Selectin, MMP9, and L-Selectin as SJIA Plasma Flare Biomarkers Fibrinogen is degraded by MMP9 [39] [40] [41] , and the fibrinogen degradation fragments have been shown to be biologically important molecules with numerous proinflammatory actions [42] . We reasoned that the generation of the SJIA urine biomarker peptides, including those derived from fibrinogen, may be an outcome of the actions of inflammatory mediators on protease expression and regulation, generating a disease-specific degradation pattern of source proteins, such as fibrinogen. To explore this hypothesis, we utilized an antibody array, consisting of pairs of capture and detection antibodies against 43 proteins of chemokines and cytokines, protein catabolism regulators, and cell surface molecules involved in leukocyte adhesion, to profile and compare the F and Q plasma samples (demographics shown in Tables 4 and 5 ). Our training data set derived from plasma samples from 25 patients at the time of SJIA systemic flare and 14 patients at the time of quiescence. Classifier discovery and feature selection by a nearest shrunken centroid (NSC) algorithm [32] was performed with all the 43 proteins. Ten fold internal cross-validation analysis led to the discovery of a candidate flare signature consisting of six proteins: TIMP1, IL-18, RANTES, P-Selectin, MMP9, and L-Selectin (Fig. 4a) . We used the NSC algorithm to derive shrunken class means of biomarker protein abundance and gauged the relative quantity of each plasma protein in the SAF and QOM samples to assess the relative resolving power of each biomarker (Fig. 4a) . To validate the antibody array observations, TIMP1 and MMP9 concentrations in SJIA plasma were also determined using enzyme immunometric assay kits from RayBiotech, Inc (Norcross, GA; data not shown). All of the SJIA flare biomarker proteins were found at higher levels in plasma at SJIA flare state. The LDA classification results were used to calculate the percentage of classification that agreed with clinical diagnosis, as shown in a modified 2×2 contingency table (Fig. 4b, left panel) . The six-protein biomarker panel classified the SJIA flare samples with 92% positive agreement and the non-flare samples with 71.4% agreement (Fig. 4c, left panel) To assess the performance of the peptide biomarker panel in the classification of "unknown" samples, we carried out an experimental bootstrapping approach. Instead of in silico bootstrapping simulation, samples belonging to different visits of the same patient and even the same samples were assayed, i.e., "bootstrapped" experimentally, for testing. For the bootstrapping testing, a total of 52 SJIA samples (demographics shown in Tables 4 and 5) were analyzed by the antibody array where 41 samples were from 19 patients at the time of SJIA flare, and 11 samples were from eight patients at the time of SJIA quiescence. Figure 4b plots the linear discriminant probabilities of the peptide biomarker panel for the training (left) and bootstrapping data (right); in both cases, samples had good separation between the highest and next highest probability for the classification. Our six-biomarker panel classified blindly the bootstrapping samples with 87.8% agreement with the clinical diagnosis for the flare samples and 81.8% agreement for the quiescent samples (Fig. 4c , right panel) (P=2.4×10 −5 for the bootstrapping test). Based upon the discriminant analysis-derived prediction scores for each sample, we constructed ROC curves [35, 36] to evaluate the performance of our plasma protein panel for distinguishing flare from quiescence samples. Figure 4d summarizes the 500 bootstrapping runs of the assayed SJIA flare and quiescent samples to derive the estimates of standard errors and confidence intervals for our ROC analysis. The ROC analysis yielded an AUC values of 0.922 for the training Fig. 4 Identification of six plasma proteins as a SJIA plasma flare panel. a All of the six plasma biomarker proteins are of higher abundance in SJIA flare. Relative abundance: the nearest shrunken centroid values [32] have been utilized to represented the relative abundance of biomarkers in either SJIA F or Q patient class. b A logistic regression model was used to find a panel-based algorithm that minimizes the total classification error discriminating SJIA F from Q. The maximum estimated probabilities for each of the wrongly classified samples, are labeled with arrows. c A modified 2×2 contingency table shows the percentage of classifications that agreed with clinical diagnosis. d The discriminant analysis-derived prediction scores for each sample were used to construct a receiver operating characteristic (ROC) curve; 500 testing data sets, generated by in silico bootstrapping, from the SJIA F and Q, both the training and the experimentally bootstrapped, data were used to derive estimates of standard errors and confidence intervals for our ROC analysis. The plotted ROC curve is the vertical average of the 500 bootstrapping runs, and the box and whisker plots show the vertical spread around the average Fig. 4d left panel) and 0.907 for the bootstrapping testing ( Fig. 4d right panel) , respectively. Urine based proteomic profiling is a novel approach that may lead to the discovery of non-invasive biomarkers for diagnosing patients with different diseases, with the aim to ultimately improve clinical outcomes [18] . Given new and emerging analytical technologies and data mining algorithms, the urine peptidome has become a rich resource for the discovery of naturally occurring peptide biomarkers. For pediatric diseases, urine is expected to become one of the most useful body fluids in clinical proteomics for diagnosis and risk-stratification. Mass spectrometry-based urinary protein and peptide profiling has led to the discovery of highly informative biomarkers for both urogenital and non-urogenital diseases [43] . At the current time, urine proteomics have been applied primarily to diseases affecting the kidney and urinary tract. Our focus on changes in urine that reflect systemic inflammation is novel and potentially of broad use [18] . One of our long-term goals is to use urine biomarkers to develop clinical tests that are non-invasive and feasible for frequent sampling and determination. With this in mind, urine peptidomes from SJIA patients were profiled to identify naturally processed urine peptide biomarkers and 17-urine peptides emerged as a candidate SJIA flare panel. This panel was found to be robust using statistical analyses. Nonetheless, the panel requires validation using a new sample set of sufficient size, guided by power analysis. The panel discovered in this study appears capable of discriminating patients with active SJIA from those with quiescent or remitted disease. Similar to other molecular changes, such as plasma protein profiles [44] , this urine peptide panel may detect incipient SJIA disease activity prior to clinical evidence of disease. In order to offer a significant clinical advantage to justify routine monitoring of urine biomarkers, a test would have to be more sensitive than the history and physical exam at predicting impending flare, and would need to predict those disease flares which do not self-resolve and therefore require escalation of medical therapy. Serial evaluation of urine samples from SJIA subjects using MRM analysis will be performed to test this hypothesis. The urine peptide panel also identifies subjects with active SJIA when compared with those with KD and FI. Our ability to discriminate between SJIA patients and other acute systemic inflammatory conditions is promising for future development of diagnostic tools. However, the SJIA patients in this study, except for the two new onset patients, are older than KD and FI patients. Collagens, bone growth and other connective tissue production may differ substantially. Therefore, development of diagnostic markers discriminating new onset SJIA from confounding acute inflammatory diseases, e.g., KD and FI, requires agematched subjects. To continue the discovery efforts and to validate the current biomarker panel, we plan to assemble a larger cohort of new onset SJIA. Another potential utility of the SJIA urine biomarker panel is to distinguish SJIA flare from infection in a patient with known SJIA, as these scenarios require different therapeutic responses. Urine samples from cohorts of both SJIA flare and SJIA patients with known infection will be assembled to validate the urine peptide biomarker panel revealed from this study. The parent proteins of the urine peptide biomarkers can be found in the circulation or kidney. For example, A1AT, FGA, and FGB are all acute-phase plasma proteins, which are synthesized by hepatocytes and megakaryocytes [45] and are found at high levels in the circulation in the setting of acute or chronic inflammation. Increased fibrinogen and fibrin deposition within joints are prominent indicators of active SJIA flare [46] and arthritis [47] . We hypothesize that the SJIA urine peptide biomarkers may result from changes in the concentrations of inflammatory mediators and protein catabolism regulators, altering the levels of peptides that are ultimately filtered into urine or generated in the urinary tract from local protease activity. The consequence is the generation of a disease-specific molecular phenotype in SJIA urine. In support of this model, we and others [38] have found urine FGA peptide biomarkers in plasma. This would suggest that the FGA urine peptide biomarkers are likely to be present in circulation. Future prospective studies are needed to determine whether the plasma A1AT, FGA, and FGB peptides have diagnostic or prognostic value in SJIA disease management. However, UMOD protein is not derived from blood, but is produced by the thick ascending limb of the loop of Henle in kidney. Our plasma analysis failed to find any UMOD peptides, suggesting that UMOD peptide biomarkers are coming from kidney. The differential abundance of UMOD urine peptide biomarkers in SJIA suggests that SJIA is likely to have an impact on normal kidney function. Renal disorders in SJIA patients are not well characterized. One 9-year-old SJIA patient was characterized by an aggressive disease course and developed renal amyloidosis just 2 years after the disease onset [48] . A variety of renal disorders can occur in patients with rheumatoid arthritis (RA), which may due to the underlying disease. The most common disorders associated with RA are membranous nephropathy, secondary amyloidosis, a focal, mesangial proliferative glomerulonephritis, rheumatoid vasculitis, and analgesic nephropathy [49, 50] . It is unclear whether SJIA directly affects renal function or indirectly causes renal inflammation. It would have been of interest to assess changes in total protein or peptide excretion in our tested SJIA and contrasting KD/FI samples. Amounts of inflammatory proteins excreted in urine can change dramatically, and febrile states have often been associated with increased protein excretion [51, 52] . Future characterization of the SJIA and KD/FI urine proteomes can help determine whether disease-related changes in total of protein excretion explain the changes in urine peptide profiles we observe in SJIA. Disease-specific alterations of gene transcription in the affected tissue and change in the balance of proteolytic and anti-proteolytic activities in urine, as we have proposed previously [18] , may also contribute to the altered pattern of urine peptides in SJIA. To further explore the possible underlying mechanisms related to urine peptide generation, we used an antibody array with 43 proteins known to be involved in the inflammation of SJIA, including certain proteases and their regulators. Plasma profiling of SJIA flare and quiescent samples using this antibody array identified a biomarker panel of TIMP1, IL-18, RANTES, P-Selectin, MMP9, and L-Selectin, all of which are present at higher abundance in SJIA flare than in quiescence. Given that fibrinogen is a substrate of MMP9 [39] [40] [41] , it is possible that up regulation of MMP9 and TIMP1 in circulation may be directly associated to the generation of FGA peptide biomarkers that ultimately are enriched in urine during active SJIA. It has been shown that treatment of active rheumatoid arthritis with golimumab (human monoclonal antibody to TNF-α) plus methotrexate significantly decreases serum IL-18, E-selectin, TIMP1 and MMP9 levels [53] . IL-18 also has been reported to be a candidate for a key cytokine in the pathogenesis of SJIA [54] . Notably, IL-18 synthesis is increased in SJIA, but not in KD [55] , indicating that there are differences in the inflammatory milieu in these (sometimes clinically similar) diseases. Such differences may explain the differences in expression of the FGA peptide biomarkers in urine between SJIA flare and Kawasaki disease. The observation of P/L-Selectins as part of the plasma biomarker panel suggests that P/L-Selectin-mediated leukocyte migration might be important in SJIA pathogenesis, possibly by mediating the recruitment and/or trafficking of specific leukocyte subtypes into inflammatory foci. Previously analysis [56] of rheumatoid arthritis (RH)-specific collagen breakdown products indicates that RH-specific fragments are formed locally in synovial fluids during diseases process and then released into the circulation. It is likely that the SJIA urine peptide biomarkers, in a similar formation mechanism as RH-specific collagen degradation products, originate due to local inflammation, and then are released in the circulation, which are ultimately enriched and ended in urine. Together, urine peptidomics and targeted plasma profiling revealed a urine biomarker panel of 17-urine peptides and a plasma biomarker panel of six plasma proteins as SJIA flare signatures. Shown in Fig. 5 , our integrated analyses suggest that the differential abundance of urine peptides in SJIA urine may be an outcome of both the pathophysiological changes initiated by IL-18 and RANTES and P/L-Selectinmediated inflammatory responses and the function of leukocyte-derived TIMP1/MMP9; the latter would influence protein catabolism in SJIA. The inflammatory cytokines may also directly affect levels of expression of substrate proteins and influence levels of expression of peptide derivatives of these proteins. Evaluation of urine peptide profiles in future prospective studies will test the robustness and diagnostic/prognostic values of these urine peptide biomarkers and may provide new insights into SJIA pathogenesis.

Translation Elongation Factor 1A Facilitates the Assembly of the Tombusvirus Replicase and Stimulates Minus-Strand Synthesis

Replication of plus-strand RNA viruses depends on host factors that are recruited into viral replicase complexes. Previous studies showed that eukaryotic translation elongation factor (eEF1A) is one of the resident host proteins in the highly purified tombusvirus replicase complex. Using a random library of eEF1A mutants, we identified one mutant that decreased and three mutants that increased Tomato bushy stunt virus (TBSV) replication in a yeast model host. Additional in vitro assays with whole cell extracts prepared from yeast strains expressing the eEF1A mutants demonstrated several functions for eEF1A in TBSV replication: facilitating the recruitment of the viral RNA template into the replicase complex; the assembly of the viral replicase complex; and enhancement of the minus-strand synthesis by promoting the initiation step. These roles for eEF1A are separate from its canonical role in host and viral protein translation, emphasizing critical functions for this abundant cellular protein during TBSV replication.

Genome-wide screens for host factors affecting RNA virus infections have led to the identification of several hundreds host proteins in recent years [1, 2, 3, 4, 5, 6, 7] . These works demonstrated complex interactions between the host and plus-stranded (+)RNA viruses, the largest group among viruses. (+)RNA viruses contain relatively small genomes and greatly depend on the resources of the infected hosts in many steps during the infection process. These viruses recruit numerous host proteins to facilitate their replication and spread [8, 9, 10] . Many host RNA-binding proteins have been implicated in replication of (+)RNA viruses, including ribosomal proteins, translation factors and RNA-modifying enzymes [8, 9, 10, 11, 12, 13, 14] . In spite of the extensive effort, the actual function of host factors in (+)RNA virus replication is known only for a small number of host factors [8, 10, 15, 16, 17] . Tomato bushy stunt virus (TBSV) and other tombusviruses are model plant RNA viruses with 4.8 kb genomic (g)RNA coding for two replication proteins, termed p33 and p92 pol , and three proteins involved in cell-to-cell movement, encapsidation, and suppression of gene silencing [18, 19] . Yeast (Saccharomyces cerevisiae) expressing p33 and p92 pol replication proteins can efficiently replicate a short TBSV-derived replicon (rep)RNA [20, 21] . The tombusviral repRNA plays several functions, including serving as a template for replication and as a platform for the assembly of the viral replicase complex [19, 22, 23] . The viral RNA also participates in RNA recombination [6, 18, 24] , which likely plays a major role in virus evolution. One of the major advantages of studying TBSV replication is the availability of genomic and proteomic datasets on virus-host interactions [4, 5, 6, 7, 10, 15, 25, 26, 27] . For example, systematic genome-wide screens of yeast genes have revealed that TBSV repRNA replication is affected by over 100 different host genes [5, 7] . Additional genome-wide screens with TBSV also identified ,30 host genes affecting TBSV RNA recombination [4, 6, 28] . The identified host genes code for proteins involved in various cellular processes, such as translation, RNA metabolism, protein modifications and intracellular transport or membrane modifications [3, 5, 7] . Additional global approaches based on the yeast proteome microarray (protein array) have led to the identification of over 100 host proteins that interact with viral RNA or the viral replication proteins [25, 26] . Also, proteomics approaches with the highly purified tombusvirus replicase has determined at least seven proteins in the complex, including the viral p33 and p92 pol , the heat shock protein 70 chaperones (Hsp70, Ssa1/2p in yeast), glyceraldehyde-3-phosphate dehydrogenase (GAPDH, encoded by TDH2 and TDH3 in yeast), pyruvate decarboxylase (Pdc1p), Cdc34p ubiquitin conjugating enzyme [14, 26, 27] and eukaryotic translation elongation factor 1A (eEF1A) [25] . The functions of GAPDH and Hsp70 have been studied in some detail [14, 29, 30, 31] , but the roles of the other host proteins, such as eEF1A, in the replicase complex are currently undefined. eEF1A is a highly abundant cellular protein with a role in delivering aminoacyl-tRNA to the elongating ribosome in a GTPdependent manner. Many additional functions have been ascribed to eEF1A including quality control of newly produced proteins, ubiquitin-dependent protein degradation, and organization of the actin cytoskeleton [32, 33] . Although eEF1A has been shown to be part of replicase complexes of several RNA viruses [16, 34, 35, 36] , studies on determining its functions in virus replication are hindered by several major difficulties. These include (i) genetic redundancy: yeast has two eEF1A genes (TEF1 and TEF2), whereas animals and plants have 2-7 genes and several isoforms of eEF1A. (ii) eEF1A provides essential functions for cell viability and mutations could have pleiotropic effects on protein translation, actin bundling and apoptosis. (iii) eEF1A is a very abundant protein that constitutes 1-5% of total cellular proteins, making it difficult to completely remove eEF1A from biochemical assays using cell extracts. (iv) eEF1A is also required for the translation of viral proteins in infected cells, making it difficult to separate its effect on translation versus replication, processes that are interdependent. The first evidence that translation elongation factors, such as EF-Tu and EF-Ts, play a role in (+)RNA virus replication was obtained with bacteriophage Qbeta [34] . The eukaryotic homolog of EF-Tu, eEF1A was found to bind to many viral RNAs, including the 39-UTR of Turnip yellow mosaic virus (TYMV) [37] , West Nile virus (WNV), Dengue virus, Tobacco mosaic virus (TMV) and Turnip mosaic virus (+)RNA [35, 38, 39, 40] . In addition, eEF1A has also been shown to interact with various viral replication proteins or the replicases, such as the NS5A replication protein of Bovine viral diarrhea virus (BVDV) [41] , NS4A of hepatitis C virus (HCV) [42] , the TMV replicase [43] , and the Gag polyprotein of HIV-1 [44] . It is also part of the replicase complex of vesicular stomatitis virus, a negative-stranded RNA virus [45] . The actual biochemical functions provided by eEF1A for (+)RNA virus replication are currently poorly understood. In case of WNV, eEF1A is co-localized with the WNV replicase in the infected cells and mutations in the WNV (+)RNA within the mapped eEF1A binding site have led to decreased minus-strand synthesis [46] . On the contrary, eEF1A was shown to enhance translation but repressed minus-strand synthesis of TYMV in vitro [37, 47, 48] . Overall, eEF1A likely plays a role in the replication of many RNA viruses. The interactions of eEF1A with viral RNAs and viral replication proteins and its high abundance in cells might facilitate recruitment of eEF1A into virus replication. eEF1A has been shown to interact with the components of the tombusvirus replicase, including the 39-UTR of the repRNA, as well as the p33 and p92 pol replication proteins [25] . eEF1A is also known to interact with the yeast Tdh2p (GAPDH) [49] , which is also a component of the tombusvirus replicase. Overall, the multiple interactions of eEF1A with various components of the tombusvirus replicase could be important for eEF1A to regulate yet unknown functions of the viral replicase complex. In this paper, we characterized the functions of eEF1A in TBSV replication based on identification of functional eEF1A mutants in yeast as well as using in vitro approaches. The obtained data support the model that eEF1A plays several roles during TBSV replication, including facilitating the assembly of the viral replicase complex. Moreover, using in vitro replication assays, we demonstrate that eEF1A enhances minus-strand synthesis via stimulating the initiation step of the viral RNA-dependent RNA polymerase. Since eEF1A is also associated with several other viral replication proteins or binds to viral RNAs, it is possible that the uncovered functions of eEF1A might be utilized by other RNA viruses during their replication as well. To determine the functions of eEF1A during tombusvirus replication, we generated ,6,000 yeast strains expressing eEF1A with random mutations (see Fig. S1A ) and tested the level of TBSV repRNA accumulation in a high-throughput assay [50] . In this assay, we used yeast strains, in which the two wt eEF1A genes (TEF1/TEF2) were deleted from the chromosome, while the wt or mutated eEF1A was expressed from plasmids. Importantly, a given eEF1A mutant is the only source of eEF1A in the yeast cells used. Using the high-throughput assay, we identified one yeast strain (N21) expressing an eEF1A mutant that supported reduced TBSV repRNA replication, while the other three strains with eEF1A mutants (named C42, C53 and C62) showed increased level of repRNA accumulation ( Fig. 1A and S1B-G). Interestingly, the eEF1A mutants supporting increased steady-state level of repRNA accumulation did not increase the relative level of p33 and p92 pol replication proteins (Fig. 1A , bottom panel; S1D-E). Thus, these eEF1A mutants likely affect TBSV replication directly. Accordingly, affinity purification of the solubilized tombusvirus replicase complex from yeast cells, followed by in vitro replicase activity assay revealed that the replicase from C42, C53 and C62 mutant eEF1A-expressing yeast strains had ,2-fold increased activities when compared with wt eEF1A-expressing yeast strain (Fig. 1B , lanes 1-6 versus 7-8). The amounts of replication protein p33 and the co-purified eEF1A were comparable in the purified replicase samples (Fig. 1B, bottom panel) , indicating that the differences in replicase activities in the mutants are likely due to enhanced replicase functions, and not due to altered proteins levels in the replicase complexes. Testing the ability of C42, C53 and C62 mutant eEF1As to bind to the viral RNA or to the p33 and p92 pol replication proteins in vitro ( Fig. S2B -C) did not reveal significant differences between the mutants and the WT. This further supports that these eEF1A mutations likely increase the function of the viral replicase without altering the protein and RNA components in the replicase. Placing the identified mutations in the three novel gain-offunction mutants of eEF1A (V 301 D, L 374 V/N 377 K, and F 413 L; Fig. 1C , indicated with yellow balls), which exhibited increased tombusvirus replication, over the known structure of eEF1A [51] revealed a cluster on one face of eEF1A (namely, the actin bundling domain III), away from the domains known to bind to tRNA and translation factor eEF1Ba. On the other hand, the new reduced function mutant (A 76 V, Fig. 1C , indicated with green balls) and the previously identified T 22 S [25] , which exhibited Plus-stranded RNA viruses are important pathogens of plants, animals and humans. They replicate in the infected cells by assembling viral replicase complexes consisting of viral-and host-coded proteins. In this paper, we show that the eukaryotic translation elongation factor (eEF1A), which is one of the resident host proteins in the highly purified tombusvirus replicase complex, is important for Tomato bushy stunt virus (TBSV) replication in a yeast model host. Based on a random library of eEF1A mutants, we identified eEF1A mutants that either decreased or increased TBSV replication. In vitro studies revealed that eEF1A facilitated the recruitment of the viral RNA template for replication and the assembly of the viral replicase complex, as well as eEF1A enhanced viral RNA synthesis in vitro. Altogether, this study demonstrates that eEF1A has several functions during TBSV replication. eEF1A mutants enhance TBSV RNA replication in a cellfree extract Since eEF1A is part of the tombusvirus replicase complex [25] , it is possible that C42, C53 and C62 eEF1A mutants might affect the assembly/activity of the tombusvirus replicase. To test this idea, we prepared cell-free extracts (CFE) from yeast strains expressing selected eEF1A mutants in the absence of the wt copy of eEF1A. These yeast extracts contained comparable amount of total proteins as well as the amounts of eEF1A, ALP, PGK and Hsp70 (Ssa) yeast proteins were comparable ( Fig. 2A) . The advantage of the CFE extracts is that they can then be programmed with the TBSV (+)repRNA in the presence of purified recombinant p33 and p92 pol obtained from E. coli that leads to the in vitro assembly of the viral replicase, followed by a single cycle of complete TBSV replication, resulting in both (2)stranded repRNA and (+)-stranded progeny [31, 52] . Therefore, this assay can uncouple the translation of the viral proteins from viral replication, which are interdependent during (+)RNA virus infections. Using CFEs from yeast expressing one of the three mutant eEF1As resulted in ,3-fold increased TBSV repRNA accumulation when compared with the extract obtained from yeast expressing the wt copy of eEF1A ( Fig. 2A , lanes 2-4 versus 5). These data suggest that the viral replicase complex containing the mutant eEF1A can support in vitro TBSV repRNA replication more efficiently than the replicase with the wt eEF1A. In contrast, CFE from N21 yeast supported TBSV repRNA replication to similar extent as the CFE containing wt eEF1A ( Fig. 2A , lanes 1 versus 5), indicating that N21 eEF1A mutant can perform the same functions as the wt eEF1A in vitro, when the same amount of p33 and p92 pol was provided. To test if the increased TBSV repRNA replication in vitro was due to enhanced (+) or (2)-strand synthesis, we analyzed the replication products under non-denaturing versus denaturing conditions (Fig. 2B) . These experiments showed that the amount of dsRNA [representing the 32 P-labeled (2)RNA product hybridized with the (+)RNA] increased ,3-fold in case of C42, C53 and C62 mutants (lanes 3-8, Fig. 2B ) in comparison with the wt (lanes 9-10). The dsRNA nature of these products was confirmed by the ssRNA-specific S1 nuclease digestion assay (Fig. 2C ). On the other hand, the ratio of dsRNA and ssRNA did not change in the various CFEs containing the eEF1A mutants or the wt (Fig. 2B) . These results are consistent with the model that the replicase complex carrying the eEF1A mutants increased mostly the level of (2)RNA production, which then led to proportionately higher level of (+)RNA progeny. Cell-fractionation assay, followed by the cell-free TBSV replication assay demonstrated that the soluble fraction from the C42, C53 and C62 mutant yeasts stimulated the in vitro replication of TBSV repRNA by ,3-fold, while the membrane fraction when derived from C42, C53 and C62 mutant yeasts had a lesser effect (Fig. 2D , lanes 12-14 versus 7-9). These data are in agreement with the expected mostly cytosolic distribution of eEF1A, albeit eEF1A is also present in the membrane fraction in a smaller amount (Fig. 2D , bottom panel). To test directly if eEF1A could stimulate RNA synthesis by the viral RdRp, we chose the E. coli-expressed recombinant p88 pol RdRp protein of Turnip crinkle virus (TCV), which is unlike the E. coli-expressed TBSV or CNV p92 pol RdRp, does not need the yeast cell-free extract to be functional in vitro [53, 54] . The template specificity of the recombinant TCV RdRp with TBSV RNAs is similar to the closely-related tombusvirus replicase obtained from yeast or infected plants [21, 54, 55, 56] . However, the recombinant TCV RdRp preparation lacks co-purified eEF1A, unlike the yeast or plant-derived tombusvirus replicase preparations, facilitating studies on the role of eEF1A on the template activity of a viral RdRp. When we added the highly purified wt eEF1A to the RdRp assay containing TCV RdRp protein and a TBSV derived (+)RNA template, which is used by the TCV RdRp in vitro to produce the complementary (2)RNA product (Fig. 3A , lanes 3-4) [55] , we observed a ,6-fold increase in (2)RNA synthesis by the TCV RdRp (lanes 11-12), while, as expected, we did not detect new (+)RNA progeny (not shown). This suggests that eEF1A can greatly stimulate TCV RdRp activity in vitro, confirming a direct role for eEF1A in (2)RNA synthesis by a viral RdRp. Since it is known that eEF1A can bind to the 39-UTR of TBSV (+)RNA as well as to the tombusvirus replication proteins [25] and to the TCV RdRp ( Fig. S2A) , we wanted to test if the above stimulating activity of eEF1A in the in vitro RdRp assay was due to binding of eEF1A to the (+)RNA template and/or to the TCV RdRp protein. Pre-incubation of the purified wt eEF1A with the TCV RdRp prior to the RdRp assay led to a ,5-fold increase in in vitro (2)RNA synthesis (Fig. 3A , lanes 9-10), while pre-incubation of the purified eEF1A with the TBSV (+)RNA template prior to the RdRp assay led only to a ,2-fold increase in (2)RNA products (lanes 7-8). Also, preincubation of the TCV RdRp with the (+)RNA template prior to the RdRp assay containing purified eEF1A led only to a ,2-fold increase in (2)RNA synthesis (lanes 5-6), suggesting that eEF1A can stimulate (2)RNA synthesis less efficiently after the formation of the (+)RNA-RdRp complex. Overall, data shown in Fig. 3 imply that eEF1A stimulates (2)RNA synthesis most efficiently when it forms a complex with the viral RdRp prior to binding of the template RNA to the eEF1A-RdRp complex. To test if eEF1A stimulates the rate of initiation of (2)RNA synthesis, we analyzed the amount of abortive RNA products, which are generated during de novo initiation of RNA synthesis by the TCV RdRp [57] . We found that the amount of the 5-11 nt long abortive RNA products increased by 3.5-fold in the presence of purified eEF1A in the TCV RdRp assay (Fig. 3B , lanes 3-4 versus 1-2). We also tested the RdRp activity in the presence of Top panel: Replication of the TBSV repRNA was measured by Northern blotting 24 h after initiation of TBSV replication. The accumulation level of repRNA was normalized based on the rRNA (middle panel, the 18S ribosomal RNA levels were estimated by Northern blotting). Bottom two panels: Accumulation of p33/p92 pol and eEF1A was estimated by Western blotting using anti-His and anti-eEF1A antibody, respectively. Note that * marks an SDS-resistant p33 homodimer band. (B) An in vitro replicase assay to test the relative activity of the tombusvirus replicase obtained from yeast expressing various mutants of eEF1A. Top panel: We tested the in vitro replicase activity using comparable amounts of affinity-purified replicase with added DI-72 RI(2) RNA template. Bottom panels: Western blot analysis showing p33 viral replication protein and the co-purified eEF1A in the above purified replicase preparations. (C) Critical eEF1A residues for tombusvirus replication. Three novel mutants of eEF1A were identified, which exhibited increased tombusvirus replication (V301D, L374V/N377K, and F413L; yellow balls) while the new A76V and the previously identified T22S exhibited decreased tombusvirus replication (green balls). The structure of eEF1A was generated using Jmol with PDB coordinates 1IJE. doi:10.1371/journal.ppat.1001175.g001 eEF1A using a (+)RNA template with a mutation opening the closed structure in the promoter region that leads to increased template activity [58] . The mutated template also showed 2-fold increased abortive RNA products in the RdRp assay with eEF1A ( Fig. 3B , lanes 5 versus 6). These data strongly support the model that eEF1A stimulates the de novo initiation step in the RdRp assay. The denaturing PAGE analysis of the 32 P-labeled repRNA products obtained is shown. The full-length repRNA is pointed at by an arrow. Panels below show Western blot analysis of the whole cell extracts for the indicated yeast proteins based on specific antibodies. Bottom panel shows the coomassie-blue stained SDS-PAGE gel to visualize total protein levels in the whole cell extracts. (B) Detection of single-and double-stranded RNA products produced in the cell-free TBSV replicase assay. Odd numbered lanes represent replicase products, which were not heat treated (thus both ssRNA and dsRNA products are present), while the even numbered lanes show the heat-treated replicase products (mostly ssRNA is present). The amount of dsRNA and the ratio of ssRNA/dsRNA in the samples are shown. Note that, in the nondenatured samples, the dsRNA product represents the annealed (2)RNA and the (+)RNA, while the ssRNA products represents the newly made (+)RNA products. (C) Denaturing PAGE analysis of the TBSV replicase products obtained in the cell-free replicase assay after S1 nuclease treatment, which cleaves the ssRNA, but not the dsRNA product. (D) The denaturing PAGE analysis of the 32 P-labeled repRNA products obtained in the in vitro reconstitution assay is shown. The membrane fraction of the whole cell extracts prepared from eEF1A mutant strains were mixed with the supernatant fraction of CFE prepared from WT eEF1A (lanes 6-10) or the supernatant fraction of CFE from the mutant strains were added to the membrane fraction from the wt strain (lanes [11] [12] [13] [14] [15] . The reconstituted extracts were programmed with purified recombinant TBSV p33/p92 pol and (+)repRNA. Bottom panel: Western blot analysis shows the amount of endogenous eEF1A in various fractions (see above) prepared from yeast expressing various mutants of eEF1A. doi:10.1371/journal.ppat.1001175.g002 To test if eEF1A stimulates the rate of RNA synthesis in the absence of de novo initiation, we analyzed the amount of 39terminal extension (39-TEX) RNA products, which are generated from an internal primer by the TCV RdRp (Fig. 3D ) [55] . Addition of purified eEF1A did not increase the amount of 39TEX products (lanes 2, 4, 6, Fig. 3D ), suggesting that the elongation step The TBSV (+)RNA template was the short 39 end region (SL1/SL2/SL3), which contain the promoter region (SL1) for initiation and the replication silencer element (within SL3) that down-regulates initiation. The second template was SL1m with a point mutation within the promoter sequence, which is being used more efficiently by the TCV RdRp in vitro. Note that eEF1A has been shown to bind to the replication silencer element. The RdRp assay had two steps: first, the shown components were incubated at room temperature to facilitate their interaction, followed 5 min latter the addition of the shown component and the ribonucleotides to start RNA synthesis. The RdRp activity in samples containing the template RNA and the RdRp were chosen as 100% (lanes 3-4). (B) Detection of abortive RdRp products in the in vitro assay. 15% PAGE/UREA gel was used to resolve the 4-10 nt long products produced during initiation followed by rapid termination. Note that abortive RNA products are characteristic products for RNA polymerases that initiate de novo (in the absence of a traditional primer). (C) Lack of stimulation of 39-terminal extension by eEF1A in vitro. The template RNAs (shown schematically) contain a common artificial hairpin structure at the 39 end that facilitates 39-TEX by the TCV RdRp. The black bar represents 3 different sequences in the three constructs, derived from RIV(+)(includes SL1/SL2/SL3 sequences), RIII(2) and RIII(+) of DI-72 RNA, respectively. The gel image shows the results of 39-TEX in the presence of 0 or 1 mg eEF1A as shown in a TCV RdRp assay. doi:10.1371/journal.ppat.1001175.g003 during complementary RNA synthesis is not affected by eEF1A. Altogether, the obtained in vitro TCV RdRp data suggest that eEF1A can mostly stimulate the initiation step during de novo viral (2)RNA synthesis. To further test the function of eEF1A in TBSV replication, we used chemical inhibitors of eEF1A, including Didemnin B (DB) and Gamendazole (GM). DB inhibits the activity of GTP-bound eEF1A during translation by binding to a pocket in eEF1A involved in the interaction with the aminoacylated tRNA and the nucleotide exchange factor eEF1Balpha [59, 60] . GM has been shown to inhibit the actin bundling function, while it does not inhibit protein translation or GTP binding functions of eEF1A [61] . We found that both DB and GM efficiently inhibited TBSV repRNA replication in the in vitro assay with CFE, which contains the endogenous eEF1A (Fig. 4A ). Time-course experiments revealed that the inhibition by DB was the most effective when the inhibitor was added at the beginning or during the first 10-15 min of the assay (Fig. 4B , lanes 2-5), while GM inhibited the cell-free replication of TBSV repRNA when added not only at the beginning, but up to 40 min after the start of the assay (lanes [12] [13] [14] [15] [16] [17] . It is known that the recruitment of the viral RNA and replication proteins as well as the assembly of the viral replicase complex take place during the first 40-60 min in the cell-free assay [31] . Since DB could inhibit translation, we also tested the effect of another translation inhibitor, namely cycloheximide, which did not affect TBSV repRNA replication in our assay (Fig. S3) . These data suggest that the inhibition by DB and GM is unlikely through decreased translation in the replication assay. Therefore, the above data are consistent with the model that DB and GM interfere with the assembly of the viral replicase complex in the CFE. Also, GM seems to be a more potent inhibitor of TBSV replication than DB. To further test if DB and GM can interfere with the assembly of the tombusvirus replicase complex, we performed a two-step in vitro assembly/replication assay, also based on CFE containing endogenous eEF1A [31] . In this assay, first, we only provide ATP and GTP in addition to the replication proteins, the (+)repRNA and CFE, which can support the assembly of the replicase, but cannot perform RNA synthesis due to the lack of CTP and UTP [31] . After 1 hr incubation, once the replicase assembly had taken place, we collected the membrane fraction of the CFE by centrifugation and removed the supernatant containing the unbound p33, p92, repRNA as well as the cytosolic fraction of the CFE. Then we added all four rNTPs (including 32 P-labeled UTP) to the membrane fraction of the CFE to allow for RNA synthesis by the pre-assembled replicase complex (second step, Fig. 4C ) [31] . Interestingly, adding either DB or GM during the first step resulted in robust inhibition of TBSV repRNA synthesis during the second step of the assay (Fig. 4C, lanes 2-3 versus 1) , whereas providing the same amount of DB and GM at the beginning of the second step did not result in inhibition of repRNA replication (lanes 4-6) . These data support a model that DB and GM could inhibit the assembly of the tombusvirus replicase complex, but not the RNA synthesis by the already assembled replicase. Similarly, DB and GM failed to inhibit TBSV RNA synthesis in an in vitro assay with a highly purified RdRp from yeast (Fig. S4A) . Since the assembly of the tombusvirus replicase also depends on events prior to the replicase assembly step, such as template RNA binding by the viral replication proteins/host proteins (such as eEF1A), and template recruitment to intracellular membranes [21, 52] , we also tested the effect of DB and GM on these processes as well based on purified recombinant eEF1A. We found that GM strongly interfered with the binding of eEF1A to the viral RNA in an EMSA assay (Fig. 4D, lanes 3-6 versus 2) , whereas DB did not affect the binding under the assay conditions (lanes 9-12). Since DB binds only weakly to eEF1A in solution, but it binds much more effectively to eEF1A in the presence of GTP and the ribosome [62] , we also performed in vitro co-purification experiments. First, 35 S-labeled eEF1A was produced in an in vitro translation system (containing ribosome and GTP) and, second, biotin-labeled viral (+)repRNA was added. After short incubation in the absence or presence of various amount of DB, we performed affinity-purification of the viral RNA. Phosphoimaging revealed that eEF1A was co-purified with the viral RNA and the amount of protein co-purified with the viral (+)repRNA was inhibited by increasing amount of DB in the assay (Fig. 4E , compare lane 1 with [2] [3] [4] [5] . This demonstrated that DB inhibits the binding of eEF1A to the viral repRNA. Moreover, both DB and GM interfered with the recruitment of the viral template RNA to the membrane of the CFE containing endogenous eEF1A (Fig. 4F, lanes 5-8 versus 3-4) . On the other hand, DB and GM do not seem to affect the interaction between eEF1A and p33 or p92 replication proteins in vitro (Fig. S4B) . Altogether, these data suggest that inhibition of eEF1A function by DB and GM could block several steps during the assembly of the tombusvirus replicase complex, including template binding by eEF1A and viral RNA recruitment into replication. (+)RNA virus replicases contain viral-and host-coded components, which likely provide many yet undefined functions to facilitate robust virus replication in infected cells. Translation factors, such as eEF1A, are among the most common host factors recruited for (+)RNA virus replication. eEF1A is an integral component of several viral replicases, including the highly purified tombusvirus replicase complex. Since eEF1A is an essential G protein involved in translation elongation, it is difficult to obtain evidence for its direct involvement in virus replication in living cells. Indeed, down-regulation of eEF1A in cells has led not only to decreased TBSV repRNA accumulation, but also reduced p33 levels [25] . However, using a small set of functional eEF1A mutants defective in various functions revealed that eEF1A is involved in stabilization of p33 replication protein in yeast [25] . Based on the previous successful strategy of analyzing eEF1A mutants, here we generated ,6,000 random mutants covering the entire eEF1A sequence and found four mutants, which greatly affected TBSV repRNA accumulation in yeast (Fig. 1A) . Among these mutants, C42, C53 and C62 increased TBSV repRNA replication. Importantly, this effect by the eEF1A mutants was not due to changing the translation efficiency of p33/p92 pol , but likely via directly altering viral replication and affecting the activity of the viral replicase. On the other hand, N21 mutant of eEF1A resulted in decreased TBSV RNA accumulation and also led to reduction in the level of p33 replication protein. This is reminiscent of the previously characterized GDP-binding mutant T 22 S [25], which supported greatly reduced level of viral RNA replication and p33 accumulation due to shortened half-life of p33. Overall, N21 mutant further supports that one of the functions of eEF1A in TBSV replication is to stabilize the p33 replication protein in yeast. In addition to this genetic evidence on the relevance of eEF1A in TBSV replication in yeast, we also obtained additional supporting data by showing that chemical inhibitors of eEF1A, such as DB and GM, strongly inhibited replication of TBSV eEF1A Role in the Assembly of Virus Replicase repRNA in the cell-free replication assay (Fig. 4A ). Since we used the same amount of purified recombinant p33/p92 pol in this in vitro assay (i.e., translation in the CFE is not needed for production of p33/p92 pol ), the role of eEF1A in TBSV replication must be separate from its role in protein translation. Altogether, these data strongly established that eEF1A is directly involved in TBSV replication, independent of the role of eEF1A in protein translation. Fig. 2 . DB (150 mM) and GM (100 mM) were added at various time points and the replicase assay was stopped after 3 hours for each treatment, followed by RNA analysis in a denaturing PAGE gel. The replicase activity in the samples containing DMSO added at the 0 time point was chosen as 100%. (C) A step-wise approach was used to separate the possible effect of DB and GM during either the assembly of the TBSV replicase or RNA synthesis steps. In step 1, the purified recombinant TBSV p33, p92 pol and (+)repRNA were added to the whole cell extract in the presence of ATP and GTP, which only supports the assembly of the TBSV replicase, but prevents RNA synthesis. This was followed by removal of the extra amount of p33, p92 pol and repRNA, which were not bound to the membranes of cell-free extract, and then by the standard replicase assay in a buffer containing 32 P-UTP and ATP, CTP and GTP (step ''RNA synthesis''). The denaturing PAGE analysis of the 32 P-labeled repRNA products obtained is shown. Note that DB (150 mM) and GM (100 mM) were added to the assay either at the beginning (prior to replicase assembly) or after the replicase assembly. See further details in panel B. (D) The effect of DB and GM on binding between the purified eEF1A and 32 P-labeled template RNA (SL1/SL2/SL3) based on EMSA. The bound and unbound RNAs are pointed at by arrowheads. GM and DB were applied in the following amounts: 0, 5, 50, 250, and 1000 mM. Note that the amount of unbound RNA in the absence of eEF1A (lane 1) was chosen as 100%. (E) The inhibitory effect of DB on co-purification of eEF1A with the viral repRNA. WT 35 S-labeled eEF1A was produced in a translation assay using rabbit reticulocyte lysate, followed by incubation with biotin-labeled DI-72(+) repRNA in the presence of 0, 50, 150, 500 and 1000 mM DB. Then the repRNA was captured with streptavidin-coated magnetic beads, followed by elution of the co-purified proteins from the beads. SDS-PAGE analysis shows the amount of co-purified 35 S-labeled eEF1A. (F) The inhibitory effect of DB and GM on the template recruitment step in vitro. Purified recombinant p33/p92 and 32 P-labeled DI-72 (+)repRNA (indicated as W, lanes 1 and 3-8) or C 99 -G mutant (+)repRNA (indicated as M, lane 2) were added to a whole cell extract (CFE) in the presence of DMSO (control), 100 mM GM or 150 mM DB, followed by centrifugation/washing to remove the 32 P-labeled repRNA that is not bound to the membrane. Then the membrane-bound RNA was analyzed in a denaturing PAGE gel. Note that the recruitment deficient C 99 -G mutant repRNA bound to the membrane nonspecifically (,20% level). doi:10.1371/journal.ppat.1001175.g004 eEF1A selectively enhances minus-strand synthesis during TBSV replication The identified eEF1A mutants were also useful to dissect the functions of eEF1A in TBSV replication. Based on a cell-free TBSV replication assay in CFE prepared from yeast expressing the C42, C53 or C62 mutants, we found that the minus-strand synthesis was enhanced by ,3-fold, while the rate of plus-strand synthesis was proportionate with (2)RNA synthesis, resulting in ,10-fold more (+) than (2)RNA products for wt and each mutant. We confirmed a direct role for eEF1A in RNA synthesis in vitro by using a highly purified eEF1A and the recombinant TCV RdRp, which is closely homologous with the TBSV p92 pol . Interestingly, it seems that eEF1A stimulates the RdRp activity directly, since pre-incubation of eEF1A and the RdRp prior to the RdRp assay led to the highest level of stimulation of (2)RNA synthesis (Fig. 3A) . On the other hand, pre-incubation of eEF1A with the TBSV-derived template RNA led only to ,2-fold increase in RNA synthesis in vitro (Fig. 3A) . Analyzing the amount of short abortive RdRp products, which are produced through initiation followed quickly by abortive termination [57] , in the in vitro assays revealed that eEF1A strongly enhanced the initiation of minus-strand synthesis (Fig. 3B) . Although the actual mechanism of stimulation of RdRp activity by eEF1A is currently unknown, we propose that eEF1A might facilitate the proper and efficient binding of the RdRp to the 39 terminal sequence of the viral RNA prior to initiation of (2)-strand synthesis (Fig. 5) . Accordingly, eEF1A was shown to bind to the so-called replication silencer sequence (RSE) in the 39-UTR, which is required for the assembly of the viral replicase complex [22, 58] . The binding of eEF1A-RdRp complex to the RSE might assist in placing the RdRp over the 39-terminal promoter sequence, thus facilitating the initiation of (2)RNA synthesis starting from the 39-terminal cytosine. Similar function of eEF1A in stimulation of (2)RNA synthesis has been proposed for WNV, based on mutations in the viral RNA within the eEF1A binding sequence that reduced the binding affinity of RNA to eEF1A and inhibited (2)RNA synthesis in infected cells [46] . eEF1A stimulates the assembly of the viral replicase complex during TBSV replication Recent intensive work revealed that the assembly of the viral replicase complex is a regulated process involving viral-and host factors, cellular membranes and the viral (+)RNA [8, 10, 19, 63, 64, 65] . The assembly of the viral replicase also depends on steps occurring prior to the actual assembly process, such as selection of the viral template RNA and the recruitment of (+)RNA/protein factors to the sites of assembly. Although our current understanding is rather poor about the factors involved and their functions during replicase assembly, rapid progress is being made in this area due to the development of a new cell-free assay based on yeast CFE [30, 31] . The yeast CFE is capable of assembling the tombusvirus replicase complex in vitro in 40-60 min in the presence of recombinant p33/p92 pol and the viral (+)repRNA [31] , allowing for studies on direct roles of various factors. We find that inhibition of eEF1A activity by either DB or GM also inhibited the assembly of the tombusviral replicase complex based on time-course experiments (Fig. 4B ) as well as a direct replicase assembly assay (Fig. 4C) . On the contrary, the replicase activity was not inhibited by these compounds after the assembly took place (Fig. 4B-C) . It is possible that after the formation of the eEF1A-RdRp-repRNA complex DB or GM are not effective in inhibiting the stimulatory effect of eEF1A on the RNA synthesis by the viral RdRp. Additional in vitro experiments with purified tombusvirus replicase preparations confirmed the lack of inhibition of RNA synthesis by DB or GM (Fig. S4A ) on pre-assembled virus replicases. The inhibition of the tombusvirus replicase complex by DB or GM might come from the ability of these compounds to inhibit the template RNA recruitment step (Fig. 4F ). If the recruitment of the viral (+)RNA is inhibited, then the assembly of the viral replicase cannot take place in yeast or in vitro [21, 22, 31] . A target for GM and DB could be the inhibition of binding between eEF1A and the viral (+)RNA (Fig. 4D, E) . Since the actual steps during the replicase assembly process are not yet known, it is possible that eEF1A might play additional roles in the assembly of the viral replicase complex. The presented data are also in agreement with the function of eEF1A as a chaperone of the viral RdRp. Binding between the eEF1A and RdRp might alter the structure of the RdRp that favors de novo initiation for RNA synthesis. Indeed, the chaperone activity of eEF1A and its bacterial homolog EF-Tu has been shown before [66, 67] . Moreover, the EF-Tu-EF-Ts complex is thought to function in the Qbeta replicase complex as a chaperone for maintaining the active conformation of the RdRp protein [68] . Overall, the current work demonstrates two major functions for eEF1A in TBSV replication (Fig. 5) : (i) stimulation of the assembly of the viral replicase complex, likely by facilitating the recruitment of the viral RNA template into the replicase; and (ii) enhancement of the minus-strand synthesis by promoting the initiation step. These roles for eEF1A are separate from its canonical role in host and viral protein translation. Saccharomyces cerevisiae strain BY4741 (MATa his3D1 leu2D0 met15D0 ura3D0) was obtained from Open Biosystems (Huntsville, AL, USA). Plasmid-borne TEF1/2 TKY strains (MATa ura3-52 leu2-3, 112 trp1-D1 lys2-20 met2-1 his4-713 tef1::LEU2 tef2D pTEF2 URA3) were published before [69, 70, 71, 72] . The plasmid pESCHIS4-ADH-His33/CUP1-DI-72 expressing Cucumber necrosis virus (CNV) p33 and the TBSV replicon RNA, called DI-72, was described earlier [25] . The LYS2-based plasmid pRS317-Tet-His92, expressing CNV p92 under the control of Tetracycline-regulatable (Tet) promoter was constructed as follows: the Tet promoter sequence was obtained from pCM189-His92/Tet [73] by digestion with EcoRI and BamHI, and CNV p92 coding sequences from pGAD-His92 [7] digested with BamHI and PstI, followed by ligation into pRS317 vector treated with EcoRI and PstI. To generate mutations within TEF1 coding sequence by random mutagenesis, we constructed the TRP1-based plasmid pRS314-pTEF1-TEF1, which expressed TEF1 under the control of its native promoter. The TEF1 promoter sequence, the TEF1 coding region and the Cyc1 terminator sequences were amplified by PCR with the following primer pairs, #2764 (CCGCGAGCTCATAGCTTCAAAAT-GTTTCTAC)/#2765 (CCGCGGATCCGTAATTAAAACT-TAGATTAGATTGC), #2768 (CCGCGGATCCAAAATGGG-TAAAGAGAAGTCTC)/#1877 (CCGCCTCGAGTTATTTC-TTAGCAGCCTTTTGAGCAGC), and #2769 CCGCCTCG-AGGAGGGCCGCATCATGTAA/#2770 (CCGCGGTACCA-GCTTGCAAATTAAAGCCTTC), respectively. This was followed by cloning the PCR products into pRS314 digested with SacI and KpnI. The mutagenic PCR conditions were as follows: 50 mM KCl, 10 mM Tris (pH 8.3 at 25uC), 7 mM MgCl 2 , 0.3 mM MnCl 2 , 1 mM dCTP and dTTP each, 0.2 mM dGTP and dATP each, 0.2 mM of each primer, 20 pM of template DNA and 10 units of Taq polymerase in a 10 ml reaction volume in 10 aliquots. The PCR was performed for 30 cycles at 94uC for 1 min, 50uC for 1 min, and 72uC for 1 min in a conventional thermal cycler. Three overlapping ,300-500 bp N-, central-and C-terminal segments of the TEF1 gene were amplified separately by PCR using primer pairs: #2767 (GTT-TCAGTTTCATTTTTCTTGTTC)/#2788 (GAGTCCATCT-TGTTGACAG), #2787 (CATCAAGAACATGATTACTGGT-AC)/#2790 (GACGTTACCTCTTCTGATTTC) and #2789 (CGGTGTCATCAAGCCAGGT/#2771, (TTCGGTTAGAGC-GGATGTGG), respectively. Yeast strain TKY102 was co-transformed with constructs pESCHIS4-ADH-His33/CUP1-DI-72 and pRS317-Tet-His92 to induce TBSV repRNA replication according to standard Lithium acetate-PEG protocol [74] . The transformed yeast cultures were grown in a Synthetic Complete (SC) media with 2% glucose lacking leucine, histidine, lysine and uracil (SC-ULHK 2 ) by shaking at 29uC overnight. To completely suppress TBSV replication before induction, 1 mg/ml Doxycycline was added to the media to inhibit the expression of p92. The plasmid pool carrying the randomly mutated TEF1 gene was introduced into the yeast cells already transformed with the two virus expression plasmids by in vivo gap repair mechanism via homologous recombination (Fig. S1A) [75] . Briefly, pRS314-pTEF1-TEF1 was digested with enzymes to truncate the TEF1 coding sequence, and then the digested plasmid was recovered. The gapped plasmid (5-10 mg) was transformed together with overlapping PCR (20 mg) products carrying the TEF1 mutations created by random mutagenic PCR (see above). The transformed yeast cells were selected on SC media lacking uracil, tryptophan, leucine, histidine and lysine. The colonies were further streaked onto SC media plate lacking tryptophan, leucine, histidine and lysine (SC-TLHK 2 ) with 0.1% (w/v final) 5-Fluoroorotic Acid (5-FOA) media to select against the URA3-based wild-type TEF1 plasmid (Fig. S1A) . This selection was repeated once and the loss of URA3 plasmid was confirmed by the inability of the yeast strains to grow on uracil-minus media. The yeast cells carrying the randomly mutated TEF1 were grown at 29uC for 24 h in SC-TLHK 2 media with 50 mM CuSO 4 to induce virus replication. Total RNA extraction from yeast cells and Northern blotting and Western blotting were done as previously described [7, 25] . Whole cell yeast extract capable of supporting TBSV replication in vitro was prepared as described [31] . The in vitro TBSV replication assays were performed in 20-ml total volume containing 2 ml of whole cell extract, 0.5 mg DI-72 (+)repRNA transcript, 400 ng purified MBP-p33, 100 ng purified MBP-p92 pol (both recombinant proteins were purified from E. coli), 30 mM HEPES-KOH, pH 7.4, 150 mM potassium acetate, 5 mM magnesium acetate, 0.13 M sorbitol, 0.4 ml actinomycin D (5 mg/ml), 2 ml of 150 mM creatine phosphate, 0.2 ml of 10 mg/ml creatine kinase, 0.2 ml of RNase inhibitor, 0.2 ml of 1 M dithiothreitol (DTT), 2 ml of 10 mM ATP, CTP, and GTP and 0.25 mM UTP and 0.1 ml of [ 32 P]UTP [31] . The reaction mixture was incubated at 25uC for 3 h. The reaction was terminated by adding 100 ml stop buffer (1% sodium dodecyl sulfate [SDS] and 0.05 M EDTA, pH 8.0), followed by phenol-chloroform extraction, isopropanol-ammonium acetate precipitation, and a washing step with 70% ethanol as described [52] . The newly synthesized 32 P-labeled RNA products were separated by electrophoresis in a 5% polyacrylamide gel (PAGE) containing 0.56Tris-borate-EDTA (TBE) buffer with 8 M urea. To detect the double-stranded RNA (dsRNA) in the cell-free replication assay, the 32 P-labeled RNA samples were divided into two aliquotes: one half was loaded onto the gel without heat treatment in the presence of 25% formamide, while the other half was heat denatured at 85uC for 5 min in the presence of 50% formamide [31] . S1 nuclease digestion to remove single-stranded 32 P-labeled RNA was performed at 37uC for 30 min in a buffer containing 5 mM sodium acetate (pH 4.5 at 25uC), 0.28 M NaCl, 4.5mM ZnSO4 and 40 U S1 nuclease (Boehringer). Fractionation of the whole cell extract was done according to [52] . The total extract was centrifuged at 21,0006 g at 4uC for 10 min to separate the ''soluble'' (supernatant) and ''membrane'' (pellet) fraction. The pellet was re-suspended and washed with buffer A (30 mM HEPES-KOH pH 7.4, 150 mM potassium acetate, and 5 mM magnesium acetate) followed by centrifugation at 21,0006 g at 4uC for 10 min and re-suspension of the pellet in buffer A. In vitro TBSV replication in the fractions was performed as described [31] . Expression and purification of the recombinant TBSV p33 and p92 and TCV p88C replication proteins from E. coli were carried out as described earlier with modifications [54] . Briefly, the expression plasmids were transformed separately into E. coli strain BL21 Rosetta (DE3). Protein expression was induced using isopropyl b-D-thiogalactopyranoside (IPTG) for 8 h at 16uC, then the cells were collected by centrifugation (5,000 rpm for 5 min). The recombinant TCV p88C protein was purified on an amylose resin column (NEB), as described [54] . The cells were suspended and sonicated in MBP column buffer containing 20 mM Tris-Cl pH 8.0, 150 mM NaCl, 1 mM EDTA, 10 mM b-mercaptoethanol and 1 mM phenylmethylsulfonyl fluoride (PMSF). The sonicated extract was then centrifuged at 27,000 g for 10 min, followed by incubation with amylose resin (NEB) for 1 h at 4uC. After washing the resin 3 times with the column buffer and once with a low salt column buffer (25 mM NaCl), the proteins were eluted with a low salt column buffer containing 0.18% (V/W) maltose and 6% (V/V) glycerol and stored at 280uC. MBP-p33 and MBP-p92 pol were purified as above, except 30 mM HEPES-KOH pH 7.4 was used instead of 20 mM Tris-Cl pH 8.0. eEF1A was purified from yeast as described [76] and stored in aliquots at the vapor temperature of liquid nitrogen. Protein fractions used for the replication assays were 95% pure, as determined by SDS-PAGE. Yeast strains (WT, C42, C53, C62) were transformed with plasmids pESCHIS4-ADH-HF33/CUP1-DI-72 expressing 6XHis-and Flag-tagged CNV p33 and the TBSV DI-72 repRNA, and pRS317-Tet-His92, expressing CNV p92 under the control of Tet promoter [25] . Co-purification was done according to a previously described procedure with the following modification [25] . Gel mobility shift assay (EMSA) and co-purification of eEF1A-repRNA EMSA was performed in a 10 ml-reaction containing 20 mM HEPES [pH 7.6], 50 mM KCl, 2 mM MgCl 2 , 1 mM DTT, 0.1 mM EDTA, 10% [vol/vol] glycerol, 10 U of RNase inhibitor, 10 nM 32 P-labeled DI-72 (+) RNA probe and 0.5 mg purified eEF1A protein [76] . Reactions were incubated at room temperature for 20 min and then resolved by 4% nondenaturing polyacrylamide gel as described previously [25] . For in vitro eEF1A-repRNA co-purification, DI-72(+) repRNA was biotin-labeled in standard T7 transcription reaction in the presence of 20 mM Biotin-16-UTP (Roche). After the T7 transcription, the unincorporated biotin-UTP was removed on a Bio-Rad mini gel filtration column. The biotinylated RNA was immobilized on a column containing Streptavidin MagneSphere Paramagnetic Particles (SA-PMPs). Briefly, a 30-ml suspension of SA-PMPs (Promega) was washed three times with 1 ml of water and re-suspended in 16 Phosphate Buffered Saline (PBS). Biotinylated DI-72(+) RNA (5 mg) was then added to the suspension of SA-PMPs, followed by 30 min incubation at 4uC with gentle rotation. The SA-PMPs were collected on the side of the tube in a magnetic stand and washed 3 times with 16 PBS buffer. eEF1A was translated in vitro and labeled with 35 In vitro TCV p88C RdRp assay The TCV RdRp reactions were carried out as previously described for 2 h at 25uC [54] . Briefly, the RdRp reactions were performed in a 20 ml reaction containing 50 mM Tris-HCl (pH 8.2), 10 mM MgCl 2 , 10 mM DTT, 1.0 mM each ATP, CTP, and GTP, 0.01 mM UTP plus 0.1 ml of [ 32 P]UTP, 7 pmol template RNA, 2 pmol affinity-purified MBP-p88C. 20 pmol eEF1A was added to the reaction at the beginning or as indicated in the text and Fig. 3 legend. The 32 P-labeled RNA products were analyzed by electrophoresis in a 5% or 15% PAGE/8 M urea gel [57] . The 86-nt 39 noncoding region of TBSV genomic RNA was used as the template in the RdRp assay [25, 54] . The use of eEF1A inhibitors in the in vitro replicase assembly assay Purified Didemnin B (NSC 325319) was kindly provided by the Natural Products Branch, NCI (Bethesda, MD, USA), while Gamendazole was a generous gift from Dr. Tash (University of Kansas Medical Center). Both chemicals were dissolved in DMSO (the final concentration was 20 mM). The concentrations of chemical and time point of the addition of the chemicals to the in vitro reaction are indicated in the text. The cell-free TBSV replicase assay and the in vitro TBSV replicase assembly assay were performed according to [31] . Briefly, the purified recombinant TBSV p33, p92 pol and (+) repRNA were added to the cell-free reaction in the presence of 1.0 mM ATP and GTP in step 1. After incubation at 25uC for 1 h, the in vitro reactions were centrifuged 21,0006 g at 4uC for 10 min. The supernatant containing extra p33, p92 pol and repRNA, which were not bound to the membranes in the cell-free extract, was discarded, while the membrane pellet was re-suspended in a standard in vitro replicase assay buffer containing [ 32 P]-UTP and ATP, CTP, and GTP, and incubated at 25uC for 3 h [31] . The TBSV viral RNA gets recruited to the membrane from the soluble fraction with the help of TBSV replication proteins and host factors present in the yeast CFE. The in vitro RNA recruitment reaction was performed according to [31] , except that 32 P-labeled DI-72 (+)repRNA were used and rCTP, rUTP, 32 P-labeled UTP, and Actinomycin D were omitted from the reaction. As a negative control, a recruitment-deficient repRNA, termed C 99 -G mutant, was used (Fig. 4F , lane 2) [23] . This mutant RNA is not recognized by p33/p92 replication proteins and it does not replicate in plants, in yeast or in the CFE in vitro [23, 31, 52, 77] . The RNA recruitment assay results in the assembly of the functional viral replicase, when wt repRNA is used, and nonfunctional replicase when the C 99 -G mutant is used in the assay (J. Pogany and P.D. Nagy, not shown) [31] . Inhibitors DB and GM were added at final concentration of 150 and 100 mM, respectively. After two hours of incubation at room temperature, 1 ml of reaction buffer was added to the in vitro assay, followed by incubation on ice for 10 min. Samples were centrifuged at 35,0006 g for 1h, and the pellet was washed with 1 ml reaction buffer, followed by centrifugation at 35,0006 g for 10 min. The membrane-bound repRNA was extracted from the pellet by adding 0.1 ml stop buffer and 0.1 ml phenol/chloroform and vortexing, followed by isopropanol/ammonium acetate precipitation [52] . The RNA samples were analyzed by denaturing PAGE and phophoimaging as described [52] . Figure S1 Schematic presentation of the random mutagenesis strategy used to obtain 6,000 eEF1A mutants. (A) The yeast strain (tef1Dtef2D carried a plasmid that expressed one of the random eEF1A (TEF1) mutants from the native promoter. Each yeast strain also carried pESCHIS4-ADH-His33/CUP1-DI-72 and pRS317-Tet-His92 to induce TBSV repRNA replication as described in the M&M section. (B) Additional experiments on the effect of eEF1A mutations on TBSV repRNA accumulation in yeast. The yeast strain expressed only one form of eEF1A, as indicated. Top panel: Replication of the TBSV repRNA was measured by Northern blotting 24 h after initiation of TBSV replication. (C) The accumulation level of repRNA was normalized based on the rRNA (the 18S ribosomal RNA levels were estimated by Northern blotting). Panels (D), (E) and (F) show the accumulation of p92 pol , p33 and eEF1A, respectively, estimated by Western blotting using anti-His and anti-eEF1A antibody. (G) SDS-PAGE analysis of total protein extract from the above yeast strains, after Coomassie blue-staining. Found at: doi:10.1371/journal.ppat.1001175.s001 (0.15 MB PDF) Figure S2 Binding of eEF1A to TBSV and TCV replication proteins in vitro. (A) MBP-tagged TCV p88C (lacking the p28overlapping domain from the N-terminus), MBP-TBSV p92, MBP-TBSV p92C (lacking the p33-overlapping domain from the N-terminus) and MBP-TBSV p33 or MBP (1 mg each) were separately immobilized on amylose beads, followed by incubation with a cytosolic extract prepared from yeast. The bound host proteins were eluted from the beads and were analyzed by 10% SDS-PAGE and detected via Western blotting using anti-eEF1A antibody (Top panel). The affinity-purified recombinant MBP-TCV p88C, MBP-TBSV p92, MBP-TBSV p92C, MBP-TBSV p33 and MBP were analyzed by 10% SDS-PAGE and Coomassie blue-staining (Bottom panel). (B) The effect of eEF1A mutations on binding to the viral p33 and p92 proteins in vitro. MBP-tagged p92, p33 or MBP were separately immobilized on amylose beads, followed by incubation with a cytosolic extract prepared from yeast expressing wt or mutated eEF1A. The bound eEF1A was eluted from the beads and were analyzed by 10% SDS-PAGE and detected via Western blotting using anti-eEF1A antibody (Top panel). The affinity-purified recombinant MBP-TBSV p92, MBP-TBSV p33 and MBP were analyzed by 10% SDS-PAGE and Coomassie blue-staining (Bottom panel). (C) The effect of eEF1A mutations on binding to the viral repRNA. CFE containing WT or mutated eEF1A was incubated with biotin-labeled DI-72(+) repRNA. Then the repRNA was captured with streptavidincoated magnetic beads, followed by elution of the co-purified proteins from the beads. Western blot analysis shows the amount of co-purified eEF1A using anti-eEF1A antibody. Found at: doi:10.1371/journal.ppat.1001175.s002 (0.12 MB PDF) Figure S3 Lack of inhibition of TBSV repRNA replication by Cyclohexamide in a cell-free TBSV replicase assay. The cell-free TBSV replicase assay was performed as described in Fig. 4 . Cyclohexamide was added in the following amounts: 0, 2, 10, 50, 100 mg/ml. The membrane-bound tombusvirus replicase in a yeast lysate was solubilized with Triton X-100/SB3-10 detergent, followed by purification on a FLAG-affinity column as described. The activity of the affinity-purified TBSV replicase was tested on the same amount of DI-72(2) RNA added to each sample. DB (panel on the left) and GM (panel on the right) were added in the following amounts: 0, 100, 150, 200, 250 mM for DB and 0, 25, 50, 100, 200 mM for GM. Denaturing PAGE analysis of the 32 P-labeled RNA products obtained with the purified tombusvirus replicase is shown. Note that this replicase preparation is only capable of complementary RNA synthesis on the added template RNA, but incapable of supporting a full cycle of replication. (B) The effect of GM and DB on binding to the viral p33 and p92 proteins in vitro. MBP-tagged p92 and p33 were separately immobilized on amylose beads, followed by incubation in the presence of 0 or 100 mM GM or 150 mM DB with a cytosolic extract prepared from yeast expressing wt eEF1A. The bound eEF1A was eluted from the beads and were analyzed by 10% SDS-PAGE and detected via Western blotting using anti-eEF1A antibody (Top panel). The affinity-purified recombinant MBP-TBSV p92, MBP-TBSV p33 and MBP were analyzed by 10% SDS-PAGE and Coomassie blue-staining (Bottom panel). Found at: doi:10.1371/journal.ppat.1001175.s004 (0.09 MB PDF)

The calculation of information and organismal complexity

BACKGROUND: It is difficult to measure precisely the phenotypic complexity of living organisms. Here we propose a method to calculate the minimal amount of genomic information needed to construct organism (effective information) as a measure of organismal complexity, by using permutation and combination formulas and Shannon's information concept. RESULTS: The results demonstrate that the calculated information correlates quite well with the intuitive organismal phenotypic complexity defined by traditional taxonomy and evolutionary theory. From viruses to human beings, the effective information gradually increases, from thousands of bits to hundreds of millions of bits. The simpler the organism is, the less the information; the more complex the organism, the more the information. About 13% of human genome is estimated as effective information or functional sequence. CONCLUSIONS: The effective information can be used as a quantitative measure of phenotypic complexity of living organisms and also as an estimate of functional fraction of genome. REVIEWERS: This article was reviewed by Dr. Lavanya Kannan (nominated by Dr. Arcady Mushegian), Dr. Chao Chen, and Dr. ED Rietman (nominated by Dr. Marc Vidal).

Organismal complexity is difficult to define and to measure, especially quantitatively. When DNA was discovered to be the material basis of inheritance in all organisms, it was thought that the DNA content of an organism should correlate with its phenotypic complexity, but soon thereafter the C-value paradox was found. C-value refers to the amount of DNA contained within a haploid nucleus, and usually equals to genome size. Salamanders and lungfishes have the largest genomes of 120pg, while the C-value of humans is only 3.5pg [1] . C-values vary enormously among species. In animals they range more than 3,300-fold. Variation in C-values bears no relationship to the complexity of the organism. The discovery of non-coding DNA in the early 1970 s resolved the C-value paradox. Although it is still unclear why some species have a remarkably higher amount of non-coding sequences than others of the same level of complexity, it was believed that the number of genes contained in the genome, rather than the genome size, correlated with the complexity of the organism. However, the human genome project and other model organism genome projects revealed that there are only about 25,000 genes in the human genome [2] , while simple organism nematode have 19,500 genes [3] and rice even has more genes than humans, 46,000~55,000 [4] . Obviously, the number of genes bears no direct relationship to phenotypic complexity. This is called Gvalue paradox. As proteins are the ultimate bearers of organismal structure and function, it is now believed that the diversity of proteins as well as the interactions between the proteins correlates with the phenotypic complexity. The study of proteomics proceeds extensively after the genome projects. However, proteins are different from nucleic acids. Many proteins are subjected to a wide variety of chemical modifications after translation. A lot of these post-translational modifications, such as phosphorylation, ubiquitination, methylation, acetylation, glycosylation, oxidation, nitrosylation and protein splicing, are critical to the protein functions. The modified proteins display different physical and chemical properties and biological functions from the unmodified. If a modified protein can be seen as a new protein, then the size of the proteome would be much larger than the number of genes. The number of proteins in a living organism may never be accurately counted as the number of genes, because every modified protein can only be individually studied. That is technically much more difficult because different proteins have different properties. A proteomic study can become quite complex even if the object of the study is very restricted. Therefore, it seems difficult to calculate the phenotypic complexity of an organism at least at the present time. There are some studies about the problem of biological complexity. The traditional mathematical notion of complexity is Kolmogorov complexity, which can be thought of as the length of the shortest message in which the given sequence can be encoded [5] . The complexity is minimal for a homopolymer, and is maximal for a random sequence, in which case complexity is equal to the sequence length. The later case means the sequence cannot be simplified at all, so is the most complex. However, Kolmogorov complexity does not correspond to our intuitive notion of biological complexity. The other complexity is to calculate the Shannon's information entropy of a sequence [5] [6] [7] [8] . The complexity is the length of the sequence subtracting the Shannon's entropy, and is actually the information content of the sequence. This kind of complexity has biological meaning. The more complex means the sequence is more conserved and therefore carries more information. However, this complexity is only the complexity of sequences, which has nothing to do with the phenotypic complexity of organisms. There are three parts of information contained in an organism's genome: first, the information to construct the organism; secondly, the information to constitute DNA structures, including replicons, centromeres, telomeres, etc; and finally, the information for the mechanisms of evolution, which we do not know at the present time. Here we take the amount of minimal information needed to construct an organism as a measure of organism phenotypic complexity because apparently more complex organisms should need more minimal information to construct and simple organisms are supposed to need less information to construct. The information to constitute DNA structure is relatively simple and less important, and we know very little about the mechanisms of evolution, so for the purpose of this article we will not be considering these areas of information. The information needed to construct organism, to put it simply, is the information needed to express proteins in time, in space, and in quantity. We actually still know very little about this information up to now. The conserved gene coding sequences are only part of the information. To calculate this information, we need to construct organism mathematically using permutation and combination formulas based on the numbers of proteins and cell types. While some biologists may know how to calculate the information content of sequence, most are not familiar with how to calculate amount of information. According to Shannon's information concept, information is to decrease uncertainty. The more uncertainty information decreases, the more the information. Information is the difference between the entropy of known and unknown. How much uncertainty you need to decrease, how much information you need. If you know everything, you do not need information. The less you know, the more you need to know, the more information you need. The uncertainty can often be calculated as possibility. The certainty means only one possibility. The uncertainty means many possibilities. The more possibilities excluded, the more the information. Shannon's information entropy can be calculated using formula: where p are the probabilities of events. When the probabilities of all events are equal, H gets the maximal value. Let N be the number of events or possibilities, then The number of possibilities N can be calculated by using permutation and combination formulas. H is the entropy of unknown because you only know the probabilities or possibilities. When you know everything, H becomes 0 because there is only one possibility left with probability 1 and the probabilities of other possibilities are 0. The entropy of known is 0. In order to know, you need to reduce the entropy to 0, so you need information: So information I has the same value as entropy H, but they are different concepts. Entropy is a quantity to describe disorder. Information is a quantity to reduce disorder. Information can be calculated based on the entropy you need to reduce. It is easy to calculate information if the number of possibilities can be calculated. For example, in order to guess a random 8-digits telephone number, you need information ) . The probability for each event is same here. Sometimes the probability for each event may not be same, but you do not know the probabilities, so any probability distribution is possible and equal probability distribution is also possible. In order to know, you have to need information to reduce the entropy from the maximum to 0. You have to assume equal probability because in this way you need minimal information. For example, we need to encode a protein sequence with 10 amino acids (all the amino acids are independent). Although the actual amino acids distribution across the sequence may not be equal, we do not know the distribution information. We have to assume equal probability to write the sequence. In this way, we need minimal information. The formula I=log(N) can still be used. We need information to write a protein sequence with 10 amino acids. If we know the sequence is MMMMMMMMMM, we need information I=0 to write the sequence because it is already known. If we know the sequence is a DNA sequence, then the entropy of unknown is H = 10 log (4), so we need information I = ⋅ = 10 4 20 log( ) bits to write the sequence. If we know it is an English sequence but we do not know any English except 26 letters, then the entropy of unknown is H=10 log (26) . We need information to write the sequence. If we know English quite well, we will need much less information to write the sequence. For example, for the last character of the sentence "I love yo_", you may need no information to guess the underscore is "u" if you know English, but I need information log (26) bits to guess the character if I do not know English except 26 letters. Even though the distribution of each English letter is actually not equal at all, but I do not know the probabilities of each letter, I have to assume equal probability. I still need information log(26) bits for each character. Different persons need different information to know because they know differently. Information is the difference between the entropy of known and unknown. How much information you need depends on how much you have known, therefore depends on the entropy of unknown. The more you know, the less the entropy of unknown, and the less the information you need. For a sequence, we can also say the information of the sequence is the information we need to write the sequence. Another example is to calculate the regulatory information of viruses. If a virus has 10 protein coding genes and all the genes express one by one in sequence and only express once, how much should be the regulatory information? The probability for each gene expression is the same, so the formula I=log(N) can be used. For the first gene to express, you have 10 genes to choose. For the second gene, the number of genes to choose is 9. The information needed for the order of expression is: . 10 21 This information is regulatory information, which is actually composed of regulatory sequences or Transcription Factor (TF) binding sites. Because every base pair of DNA contains information log(4) = 2 bits, the length of all TF binding sites should be at least 21.8/2 = 10.9 bp. A virus may use longer sequence for the binding sites, but the minimum should be 11 bp. The concept of information content of sequence is a little different from the concept of information above. The calculation of information content also needs to use Shannon's information entropy. To calculate information content, you must know the probability distribution of each site of a sequence, which is based on the data of population genetics. The information content or complexity of the sequence can be calculated as [5, 7, 8] : where L is the length of the sequence, H is the entropy of known. For example, for a sequence AXT, the first site A is very conserved with probability 1 for A, 0 for other base pairs. The information content of the first site is: If the second site X is actually random, any bases are possible, the information content is: C 2 = 1-1/4·log(4)·4 = 1-1 = 0 or C 2 = log 2 (4)-1/4·log 2 (4)·4 = 2-2 = 0bit If the probabilities of the third site T are 1/2 for T, 1/4 for A, 1/4 for G, the information content of third site can be calculated as: The more conserved the sequence, the more the information content. The amount of information of the sequence is determined by how much you know to write the sequence. If you know it is a DNA sequence and the probability distribution of each site of the known sequence, you need information I = H unknown -H known = 6-3.5 = 2.5 bits to write the sequence AXT. The information of the sequence is 2.5 bits. For each genetic codon, which is composed of 3 base pairs, the amount of information is I=H unknown -H known =3 log 2 (4)-0 = 6 bits. For each amino acid, because many amino acids have more than one codon, the information content is different. For example, proline has 4 codons and the third base pair of the codons is wobbled, the information content of proline C=H max -H known =6-2 = 4 bits. In the same way, the information content of Arg is 3 bits, of Asn is 5 bits, and of Trp is 6 bits. Organisms probably know the importance of each amino acid and how to regulate gene expressions with codon bias, and evolved this codon system to regulate translation efficiently and reduce the harm of mutation of important amino acids. The redundance of information is R=I-C. The less the information content is, the more the redundance of information, the more the space to silence or neutralize mutation. So Arg is more important, more protected from mutation, and more expression regulated than Trp. Although some amino acids may need less information to be encoded (for example, 2 base pairs for the amino acids with information content less than 4 bits), organisms actually use 3 base pairs for every amino acid, so it cost organisms 6 bits genomic information to encode every amino acid. In another word, the amount of information of every amino acid in genome is 6 bits although there is information redundance in the 6 bits. Using fixed number of base pairs rather than different number to encode all amino acids manifests that organisms probably do not know the distribution of amino acids across all protein sequences and assume equal probability for all amino acids. In this way organisms can change protein sequences and amino acid distributions flexibly and efficiently under changing environments. Otherwise, organisms have to change the codon system when the distributions change. For evolution, it is impossible for a living organism to know the future protein sequences and the amino acid distributions across the sequences. In order to generate any unknown future sequences efficiently, living organisms need to use a mechanism that can encode any protein sequences efficiently. In fact, all organisms use a mechanism that can reduce the maximal entropy to 0, which means equal probability for all amino acids when encoding protein sequences. For the same reason, organisms may not know the distribution of bases across all nucleotide sequences, and may assume equal probability for all bases even though the actual probability distribution across some genomes may not be equal at all. For biological information, most cases are like this with maximum entropy needed to be reduced and equal probability need to be assumed by organisms, so formula I=log(N) can be used to calculate the information. Fixed number of base pairs of genetic code also simplifies the reading of DNA protein coding sequence and reduces the total information needed to encode protein. If the lengths of codons were different, "space" would have to be used to separate each codon in protein coding sequence, and the total mechanisms of translation would be more complex. Although proteins and other molecules may contain information beside DNA genetic information, most information that can pass to next generation is in DNA, so DNA directly controls enough information to construct organism. All the processes not directly controlled by DNA, including post-translational protein modifications and interactions, are the functions of proteins and will proceed automatically when the proteins are synthesized. All the information necessary for these functions is already contained in DNA sequences. Therefore, we do not need to take the post-translational processes into consideration when calculating genomic information needed to construct an organism. The term "effective information" can be used to describe the minimal amount of information needed to construct an organism. The effective information actually means the genome size except "junk DNA" or "functional fraction" of genome. There are different formulas to calculate the effective information of viruses, bacteria, and eukaryotes because they have different genomic information structures. Every organism has proteins, which need to be expressed in order, in time, and in quantity. Viruses have several to hundreds of proteins. For a complex virus with hundreds of proteins, it may need to separate expression of early and late proteins and maintain precise ratios of different protein products expressed simultaneously. Because all virus proteins are expressed only once, we only need to consider the order of virus protein expressions, while the quantities of protein expressions are controlled by the feedback through the affinity of regulatory factors to the binding sites. There are many possibilities for the order of virus protein expressions and the protein sequences. If each protein is expressed one by one, the number of possibilities N can be calculated by using permutation and combination formula: where x is the number of proteins, n 1 , n 2 , ... n x are the length of protein sequences (the unit is the number of amino acid residues). As the probabilities for all amino acids and proteins are equal as analyzed before, the effective information can be calculated: where n are lengths of virus proteins. The protein expressions may not be one by one in sequence. Some proteins may be expressed simultaneously. In this way, organisms may use less information to express proteins, depending on the specific expression routes. However, this way will be inflexible to change the routes under changing environments. Like encoding protein sequence, organisms may possibly use a mechanism that can cope with any routes and reduce the maximum entropy to 0. The maximum entropy is the case that all proteins are expressed one by one in sequence. In this way, organisms can change the expression routes easily without changing the mechanism of how the information system works. Genetic information is stored in the form of DNA. DNA genetic code has degeneracy. In order to better compare the value of I and the genome size, we will not calculate the actual minimal information but the DNA information. So the formula becomes: where 3 is the number of base pairs that make up a codon. The information I 1 forΣn 6 is of the protein coding sequences, because it is just the length of protein coding genes in bits. The information I 2 for log(x!) is of regulatory sequences. The x should be the number of protein linkage groups or operons. If all virus proteins are in one operon or linkage group or expressed simultaneously, no regulatory information is necessary. Because the number of protein linkage groups is usually unknown, we still use the number of proteins instead. The real I 2 in viruses is very possibly equal to x log(x) because in this way viruses can use fixed length of regulatory sequence log(x) for every gene or operon. This may simplify the mechanism of regulation. Although the information I 2 is calculated by assuming one by one protein expression, the result is that each gene or operon has a regulatory sequence, which is very possibly true. The amount of effective information of Avian infectious bronchitis virus is: I=76,512 bits. The genome size is: C=length of genome 2 = 55,216 bits. So the I value is larger than the C value. After checking the virus genome, we find there is an overlap between genes. It's the overlap that makes the I value too large. When calculating the information of the overlapping part of the genes for the second time, because these base pairs already cannot change any more and there are no possibilities to be excluded, the amount of information for the second time calculation is 0. For example, sequences AACCC and CCCGG have overlapping part CCC. The information of the first sequence is 5 2 = 10 bits. The information of the second sequence is 2 2 = 4 bits because the information of sequence CCC is already known and this part contains no new information. So the information of sequence AACCCGG is 14 bits rather than 20 bits. The actual value of I 1 should be exactly equal to the size of the protein coding area. For virus genomes with overlapping genes, we use the actual size of protein coding area instead of Σn 6 to calculate the effective information. The effective information of some viruses is as follows ( Table 1) . The effective information I of viruses range from thousands of bits to hundreds of thousands of bits. In general, single strand DNA viruses have the minimum amount of effective information, down to 3 thousands bits, while double strand DNA viruses have the maximum amount of effective information, up to hundreds of thousands of bits. Single strand RNA viruses and retro-transcribing viruses are between the above two. The amounts of calculated effective information of viruses are consistent to the complexity defined by the number of proteins or genome size. In fact, the effective information is roughly proportional to the number of proteins. For viruses, the advantage of using effective information is not obvious because viruses are simple and the number of protein coding genes or genome size can be used to determine the complexity of viruses. For higher multicellular organisms, due to G-value paradox, the effective information will be more useful. The calculation of effective information of bacteria is more complex than viruses' because the genes may express more than once in the growth of bacteria. There are regulations on protein sequency and quantity, but no regulation on the spatial deployment. The proteins can be thought to reach their positions in a cell automatically after synthesized. A grown bacterium has roughly fixed volume, mass, and the number of protein molecules. It needs to produce all the protein molecules in its growth. The production of proteins in the growth is like chain reactions controlled by complex regulatory cascades with feedbacks. Even though the production of proteins is complex, we can always think that a bacterium grows theoretically from producing the first protein molecule to the last molecule in sequence. In this way, we can calculate the effective information. The problem is that many molecules may be produced simultaneously in the process. This may need less information. However, like viruses, in order to change protein expressions flexibly, organisms may use an information mechanism that can produce protein molecules one by one in sequence to reduce the maximum entropy to 0, which means each protein or operon has at least one regulatory sequence. Even if the calculated information may be more than necessary, we can still use k value later to adjust the information to fit to the actual amount of effective information based on real regulatory information of organisms. The number of possibilities for the first molecule is: where x is the number of proteins, n 1 is the length of the first protein. If all the molecules are independent each other, the number of possibilities for all molecules by using permutation and combination formula is: where y is the number of all protein molecules in a grown bacterium, n 1 , n 2 , ... n x are the length of proteins. The regulatory information is calculated as encoding the order of all protein molecules in a cell, which includes timing and volume of expression. If you know the first 10 molecules produced are protein A, the next 100 molecules are protein B, the next 50 molecules are protein A again, the last 200 molecules are protein Z, you know the timing and volume of expression. Although the chance for each protein in an organism may not be equal, the distribution is actually unknown for the organism because the distribution always needs to change under changing environment. For similar reason for encoding protein sequences, in order to cope with changing environments, organisms have to use a mechanism that can generate any distribution, including equal distribution, to reduce the maximal entropy to 0. This kind of mechanism provides flexibility for organisms to change the quantities of certain proteins any time to adapt the environments. The effective information can be calculated: The effective information includes two parts: first, the information to encode the protein sequences (I 1 ), and secondly, the regulatory information (I 2 ). The formula to calculate I 1 is the same as virus's. Converting the information to DNA information, the calculation can be simplified as: where n is the average length of bacteria proteins, which is 308 for all bacteria [9] . I 1 should be adjusted to the exact size of the proteins coding area if there is overlapping of genes. For regulatory information I 2 , in fact, every protein molecule is not produced independently. A bacterium does not once synthesize only one protein molecule, but a batch of protein molecules. Only the first protein molecule in the batch is independent, the possibility is × kinds of protein. The other following protein molecules are not independent. The possibilities follow the previous protein, which can only be 1. If a bacterium synthesizes 100 protein molecules at a time, I 2 is: It can be expected that the number of protein molecules a bacterium once synthesizes may be proportional to the average quantity of the proteins y/x, which means the bigger the bacterium of the same structural complexity, the more the protein molecules it synthesizes at once. where k is the average number of times a protein is synthesized. The value of k can be estimated as 5 based on the genomic information structure of E. coli (about 3% of E. coli genome is estimated as regulatory information). This is equivalent to that E. coli synthesizes 67 same protein molecules on average at a time. When the bacteria are very small, y is very close to x. The quantity of a kind of protein molecules may be less than 5. In order to ensure at least one protein molecule synthesized at once, the formula needs to be calibrated as: This formula means that each protein has at least one regulatory sequence. The average number of regulatory sequences is k. The quantity of a certain protein is controlled by the times the protein is synthesized. The more the quantity, the more the times the protein is synthesized, the more the regulatory sequences. The average number of times is k. The quantity of protein synthesized each time is not directly determined by DNA information, but determined by the feedback through the affinity of regulatory factors to the binding sites. It is the number of regulatory binding sites upstream the protein coding gene sequence that determines the quantity of the protein. In fact, most bacterial proteins are synthesized within the unit of operons, i.e. all the proteins in an operon are linked. They are not independent and are synthesized together. So the x in the formula should be the number of the addition of all operons and independent proteins in the genome. For example, the number of operons in E. coli is 834 [10] , with the number of independent proteins, together amounts to about 2,584. The x should be 2,584 for E. coli. Because the number of operons in most bacteria genomes is still unavailable at the present time, we calculated I 2 based on the number of proteins. The complete formula to calculate the amounts of information of bacteria is: where y is the total number of protein molecules in a bacterium. The effective information of some bacteria is as follows ( Table 2) . The amounts of effective information of bacteria range from millions of bits to tens of millions of bits, just one order of magnitude higher than viruses'. The regulatory information, except very small bacteria and mycoplasma, accounts for 2~3% of the genome. The I values correlate well with the bacterial complexity defined by number of protein-coding genes, genome size, and volume. Bacteria's I values are higher than virus's. This is also consistent with our knowledge about complexity. The only thing that looks like an anomaly is that the I value of Haemophilus influenzae and Streptococcus pneumoniae exceeds the C value. Checking the average length of the proteins, we find that the actual protein average length of Haemophilus influenzae is only 235 (< 308) and the actual size of protein coding area is only 3.26e6 bits, while the calculated I 1 value is 4.26e6 bits, which already exceeds the C value. The corrected actual I value of Haemophilus influenzae is 3.39e6 bits, accounting for 91.6% of C value. Similarly, the average length of proteins of Streptococcus pneumoniae is 250, which is also much lower than the average value of bacteria. The corrected actual I value of Streptococcus pneumoniae is 3.77 e 6 bits, accounting for 92% of C value. Therefore, if precise I 1 value is needed, it should be directly calculated from the actual size of protein coding area, which is: While the calculation of effective information of unicellular eukaryotes is the same as bacteria's, the calculation of I of multi-cellular organisms is much more complex because multicellular organisms not only need to produce all proteins to build different cells, but also need to put all the cells in spatial structures to build the organisms. The number of possibilities for all the cells put together and the effective information can be calculated by using permutation and combination formula similarly, but the equation is very long and the explanation can be quite complex. It is better to separate the effective information directly to three parts and calculate separately: first, the information to encode all the proteins (I 1 ); secondly, the information to produce all the differentiated cells (I 2 ); and finally, the information to construct the spatial structures (I 3 ). The information to encode all the proteins is the size of the protein coding area in the genome, like bacteria's I 1 . There are also overlapping genes in eukaryotic genomes, and the overlapping genes account for considerable weight [11] . So these overlapping parts must be adjusted according to the actual size of protein coding area, otherwise these parts will mix with other parts of information and can cause confusing results. If the genome is not yet sequenced, this information can be calculated: where g is the number of protein coding genes, n is the average length of proteins of eukaryotes, which is 448 [9] . In this way, the calculated I 1 is usually larger than the actual size of protein coding area. To produce all the differentiated cells, proteins need to be chosen from the complete proteome. For one type of cells, the algorithm is similar to bacteria's, Let x be the size of complete proteome (number of functional proteins before post-translational modifications), let t be the average size of cellular proteome of differentiated cells, cn is the number of cell types, a 1 , a 2 , a 3 ,..., a cn are the diversities of differentiated cellular proteomes from t. Because many genes expressed in the differentiated cells are the same, t+a 1 +a 2 +a 3 +...+a cn =x, then the information to produce diverse differentiated cells is: a k x t a a a a k 2 1 12 where k 2 is a coefficient. The formula means choosing t+a 1 from × to construct the first type of cell, and choosing a 2 from x-t-a 1 to construct the second cell, and so on. The possibilities for every protein molecule in all cell types are t, and the numbers of free protein molecules are proportional to the diversities of cellular proteome. As the calculated value of Σlog(C) is actually quite small and can be negligible, the information to produce diverse differentiated cells is: For unicellular eukaryotes, the calculation of I 2 of is the same as bacteria's: where x is the number of proteins. k 1 is estimated as 30 based on the information structure of Saccharomyces cerevisiae (Baker's yeast) genome to make the amount of effective information account for about 80% of the genome. This is equivalent to that a yeast cell synthesizes average 1000 protein molecules at once. Because there are regulations between the cells of multicellular organisms, k 2 can be larger than k 1 , estimated as 110, which is equivalent to a cell synthesizes average 820 protein molecules at a time. The average number of genes expressed in specific tissues is about 5000. The proteome size of a specific tissue ranges from a few thousand to tens of thousands. Because a tissue contains diverse cells, the proteome size of one differentiated cell may be a few thousand. Because most proteome data are still unavailable and are confusing with the proteome containing post-translational modifications, transcriptome data from clustered EST can be used instead. The number of unique sequences of clustered ESTs from human breast tumors is 6501 [12] , so we estimated t=6500 as the average cellular proteome size of differentiated cells. With all the differentiated cells available, organism spatial structure can be constructed. Let z be the total number of cells in an organism. In the process of organismal development, cell divisions are controlled by complex regulatory signals. We can always think that cells are produced from the first one to the last one in sequence. When the last cell is produced, the development ends and the organism reach its adult weight. In this way, information can be calculated. Although some cells may be produced simultaneously, organisms may use an information mechanism to reduce the maximum entropy to 0, which is the case that all cells are produced one by one in sequence, in order to keep the flexibility to change the cell development routes easily. Even if the calculated information may be more than necessary, it can still be adjusted by k 3 value later to fit to the actual amount of information. Like cell division, there is only one appropriate position between two adjacent cells. In this way, the spatial structure can be easily constructed. Let us start from deploying the first cell. If all cells are independent, the number of possibilities to produce and deploy all cells can be calculated using permutation and combination formula: The information is: The chances for different cell types and different positions may not be equal, but they are unknown. Organisms have to use mechanisms that can generate all possible distributions of cells and positions, including equal distribution. This kind of mechanisms provides flexibility for organisms to change the quantities and positions of cells easily and efficiently to adapt the environments. When z is larger than 10 5 , log(z!) is very close to zlog (z) (> 90%). In order to simplify the calculation, log(z!) is replaced by zlog(z). When z is small, log(z!) is still used. The actual situation is that all the same cells are generated and deployed by many times. where k 3 is the average number of times one type of cells is generated. cn should be the number of cell types or the number of linkage groups. In fact, in the organs or tissues of multicellular organisms, many cells are linked, causing repetitive pattern in the organ or tissue. So the numbers of cell types cn in the formula should be replaced by the number of linkage groups. There is no regulatory information needed for genes or cells inside a linkage group. However because the numbers of linkage groups in organisms are still unknown, we calculated I 3 based on the number of cell types. The formula also means each type of cells has at least one regulatory sequence. The average number is k 3 . The quantities of each type of cells are controlled by the number of times this type of cells is generated. The more the quantity, the more the times this type of cells is generated. The quantity of cells generated each time is not directly determined by DNA information, but by the cellular feedback signals. It is the times of generating that determines the quantities of each type of cells. k 3 can be estimated as 2.5e4, which is equivalent to that human body averagely generates and deploys 4.2e7 cells at once, that is about 0.012g. The k 3 value is determined by the information structure of Schistosoma mansoni and Caenorhabditis elegans. Because C. elegans is very small and the I 3 of which is almost negligible, and Schistosoma mansoni is bigger than C. elegans but the phenotypic complexity should be a little bit less than C. elegans, therefore we adjusted the k 3 value to make the I value of Schistosoma mansoni close to, but a little bit lower than the value of C. elegans. So we take k 3 value as 2.5e4. When z is very small, z is close to cn. That may make organisms once generate less than one cell. To avoid this kind of error, the formula can be calibrated as: The final formula to calculate the effective information of Eukaryotes is: 110 6500 2 5 10 1 2 5 10 The x in the formula is the size of complete transcriptome. Because the size of proteome before post-translational modifications is still unknown, we use the size of complete transcriptome instead. The sizes of complete transcriptome come from clustered EST data. As the number of genes is quite different from the number of clustered EST, it is a problem how to use the data from EST databases. There are a few databases having EST fragments and mRNA assembled and clustered to reduce the redundancy for gene discovery, but different databases give different results. For example, the number of human UniGene clusters is about 120,000, while the number of unique sequences of human EST clusters in The Gene Index project (TGI) of Harvard University is 1,080,000 [13, 14] . The difference between the two databases is supposed to be that TGI separated alternative splicing sequences and tried to produce tentative consensus, while UniGene put all the overlapping sequences together in one cluster. However only knowing this does not help match the data from the two databases. We still did not know the actual size of human complete transcriptome. Even in one database, the data are often conflicting each other. For example, the number of human Uni-Gene clusters is about 120,000 [15] . UniGene means unique gene, and is supposed to cluster the transcribed ESTs and mRNA into unique genes. So the number of UniGene clusters should be equal to the number of genes. However, there are only about 25,000 genes in human genome, while the number of UniGene clusters that contain only one sequence is more than 40,000. There must be errors inside. Zhang et al analyzed the results of UniGene clusters [16] and pointed out that most narrowly expressed transcripts (NETs), whose expression is confined to a few tissues, resemble intergenic sequences, and most NETs are singleton clusters containing only one EST or mRNA sequence. So those singleton clusters seem unreliable. The sequences in these clusters may come from non-coding RNA, contamination of pre-mRNA, genomic DNA, non-canonical introns or foreign sources, or simple sequencing errors. Owing to the establishment of other specialized databases, we can resolve this problem, at least the size of the human transcriptome. The Alternative Splicing Prediction Data Base (ASPicDB) in Italy has predicted almost all human alternative-splicing transcripts and has them listed in detail [17] . They analyzed 18,193 human genes and found 319,745 transcripts, which on the whole can represent the size of complete transcriptome because their data correspond quite well with the data from TGI. In TGI's data, 730,000 of 1,080,000 human unique consensuses are singleton. The total number of human genes is about 25,000, among which 35~60% contain alternative splicing, so the number of singleton should be about 10,000~15,000. Obviously, most sequences in the singleton are the result of errors. The real singleton sequences perhaps are already included in the tentative consensuses (TC). So we discarded all unique singletons and only count the number of TCs and obtain the result of 328,301. As ASPicDB only analyzed 18,193 genes, while human beings have 25,000 genes. If other genes do not have alternative splicing, then the size of total transcriptome should be 326,552, which is quite close to the result of TGI. It is a pity that only human data in ASPicDB is fairly comprehensive, the data of other species are quite sparse and cannot be used in the same manner. Because TGI is not updating the data regularly and many data was released in 2006, which may be out of date, so we use UniGene data as supplement. In order to take advantages of both UniGene and TGI, we took the average value of the two databases as the complete transcriptome size x. We discarded all the error prone clusters in UniGene that contain only one sequence. Having the number of the remaining UniGene clusters multiplied by possible average number of alternative splicing, we could match the two databases. For example, the number of human UniGene clusters after the treatment is 82,718. After multiplied by the possible number of alternative splicing 4, we got 330,872, which is close to the result of TGI. The UniGene result of mouse is 56,365 4 = 225,460, which is close to TGI result 210,249. We supposed the average number of alternative splicing for mammals is 4, for birds is 3, for fishes, amphibians and chordates is 2, for other animals is 1.5, for plants is 2~2.5, to make the results from the two databases as consistent as possible. We also referred to other databases. The sizes of complete transcriptome of some eukaryotes are as follows ( Table 3 ). The species were chosen for two reasons: first, we chose the species with higher numbers of UniGene clusters among the close species because the data are still incomplete; secondly, the species should have TGI data or other data sources. Plant species were chosen only to illustrate that this method can apply to plant. In general, the results from the two databases are consistent. Some results of mammals obtained from TGI are lower than UniGene's. It's probably because some of TGI's data are too old, or because there are real differences in the average number of alternative splicing among mammals, but if so, it will be difficult to understand the huge difference between mouse and rat. At the present time, although the transcriptome data (clustered EST data) of many species are available, most of them are incomplete. The TGI result of Danio rerio is too high. We cannot explain why. Perhaps those clusters contain too much gene fragments. The data of UniGene and TGI are far from perfect because they cannot correspond to the number of genes. Only if every transcript corresponds to every gene, like ASPicDB, the data can be more reliable. The data of the number of cell types of eukaryotes can be calculated. We know the number of cell types of adult human body is 210 [18] ; sponges have 12 kinds of cell types [19] ; the simplest multicellular organism Trichoplax adhaerens has 4 types of cell [20] ; C. elegans has 27 types of cell [21] . Because the tanscriptome size of specific cells are relatively fixed, it can be anticipated that the larger the size of complete transcriptome of an organism, the more the number of cell types. Based on the data of Valentine's [22] , a linear relationship between the number of cell types cn and the size of complete transcriptome x can be drawn (Fig 1) . The number of cell types can be roughly calculated as: ( ) Table 3 The size of transcriptome of eukaryotes There is not yet evidence if the formula apply to plants, but the cn of plants are calculated using the formula in this paper just to illustrate rough range of effective information of plants. With all the data available, the effective information of eukaryotes can be calculated (Table 4 ). If the genomes of some organisms are not yet sequenced, the numbers of genes are estimated according to the close species (marked by * sign). Because eukaryotes also have the problem of gene overlapping, it is better to find the exact size of protein coding area to calculate I 1 . Sometimes the size of protein coding area is quite different from the result calculated from the number of genes. For unicellular organisms, if the transcriptome sizes x are not available, then x are the numbers of genes. After attentive observation of the I values, one can clearly see that the results demonstrate a definitive correlation between the amounts of effective information and the organismal phenotypic complexity defined by biological taxonomy and evolutionary theory. C. pombe has the lowest I value, while human beings have the highest. Nematode, insects, amphioxus, fish, frog, bird falls in between. The I values of eukaryotes range from tens of thousands to hundreds of thousands of bits, which are just one order of magnitude higher than prokaryotes'. These results are also consistent to our intuition about organismal complexity, whilst the number of genes is a poor index here. P. tetraurelia has quite high number of genes 39,642, but is actually a simple single cellular organism with low I value. So the effective information is a much better index of organismal phenotypic complexity. The I value of Danio rerio (Zebra fish) is too high because the x is too high (higher than chicken's). The effective information of Ciona intestinalis accounts for 69% of the genome (almost all the non-repetitive sequence), that means the genome is quite compact and there are less junk DNA, therefore it will be easier to study the genomic information structure of this organism. The x should be the number of transcripts that produce functional protein sequences. One transcript should correspond to one protein sequence, and vice versa. Every UniGene cluster should correspond to one gene and every TC in TGI should correspond to one protein sequence, and vice versa. ASPicDB uses a better algorithm because it makes all transcripts correspond with genes and proteins very well. The value of x is very important because the effective information is roughly proportional to x. The diversity of proteins before posttranslational modifications can reflect the complexity of organisms. The number of cell types was calculated with all neuron cells as one type. Some argued that neurons should be counted as different kinds of cells because they are functionally differentiated. It is difficult to count the number of neuron types at the present time. Perhaps in the future, a more objective and high throughput method can be found to count the number of neuron types. This may explain why the number of cell types reaches saturated at mammal level. Among the alternative splicing isoforms in organism transcriptome, how many are functional is still disputable. Some alternative splicing may cause premature termination codon (PTC); and the alternative isoforms with PTC can be potentially targeted for degradation by the nonsense mediated mRNA decay (NMD) surveillance machinery. According to ASPicDB, the number of transcripts of the human genome is about 320,000, among which only 30,000 may generate PTC+ isoforms. If this part is discarded, there will be no important effect on the calculation. As PTC related data of most organisms are still unknown, this part is not taken into consideration at this time. There is an implicit assumption in this paper: all prokaryotes have the same k value and so do eukaryotes. It needs to be verified for this assumption to hold. The exact values of k can be determined when the genomic information structures of model organisms are completely clear. For example, when the genomic information structures of C. elegans are completely known, the value of k 2 can be completely determined. When the genomic information structures of C. intestinalis are known, the value of k 3 can be more precisely determined. The information contained in I 2 and I 3 cannot be included in the sequences of regulatory proteins. I 2 is composed of regulatory sequences or regulatory factor binding sites. I 3 may be composed of regulatory noncoding RNA sequences. It is clear that the effective information increases along with the increase of organismal phenotypic complexity defined by taxonomy, evolution, and intuition. The simpler the organism, the lower the I value. The more complex the organism, the higher the I value. The effective information in viruses is between 10 3~1 0 5 bits, while in bacteria is between 10 6~1 0 7 bits, and in eukaryotes between 10 7~1 0 8 bits. For multicellular organisms, the effective information increases from 4.68e7 bits of placozoa to 8.38e8 bits of human beings. Worm, insects, amphioxus, fish, frog, bird falls in between. These results are consistent to other observations with the number of cell types [22] , and the number of miRNA families [23] . Aburomia et al calculated the morphological complexity of 21 extant higher-order chordate groups based on the presence or absence of 479 morphological characters [24] . Their results are consistent to ours. Therefore, the effective information can be used as a quantitative measure of organismal phenotypic complexity from the simplest viruses to the most complex human beings. While limited by the incomplete data presently available, some results may not be so accurate, but the approximate range will remain true. When data become more complete and accurate, more precise calculation can be conducted. Studies have reported increasing morphological complexity in multiple parallel lineages of the Crustacea [25] . When the phenotypic complexity of more organisms can be precisely calculated as effective information, it will be easier to study the evolution of organismal complexity. The results of effective information of mammals are also consistent to the recently published article regarding the amount of constrained sequence in genome shared between eutherian mammals [26] . The constrained sequence means the sequence under functional constraint. The total amount of constrained sequence in rodents is estimated as 260Mb (5.2e8 bits), which is close to 5.09e8~6.34e8 bits effective information of rat and mouse. 300Mb (6.0e8 bits) of human genome is estimated under functional constraint. This is also close to 8.38e8 bits effective information of human. For fruit fly, the amount of constrained sequence is estimated as 55.5~66.2Mb (1.1~1.32e8 bits), which is also close to 1.09e8 bits effective information. Therefore, the effective information can be used as an estimate of functional fraction of genome. The genomic information of viruses, bacteria, and some eukaryotes can be found at GenBank. where 128 and 18 is the mean molecular weight of amino acids and water respectively, 6.02e23 is the number of molecules of one mole. The data for x comes from the genome database of GenBank. The data for volumes of bacteria can be found at website: http://www.ionizers.org/Sizes-of-Bacteria.html. The volumes of bacteria can be calculated based on rod lengths and rod or coccus diameters. Given the average cell volume v of multicellular eukaryotes as 300 μm 3 , the total cell number z of an organism is: z = w/v = (w/3) 10 10 , where w is the weight of the organism. The data of UniGene can be found at website: http:// www.ncbi.nlm.nih.gov/sites/entrez?db=UniGene. On the homepage of UniGene, you can see the number of Uni-Gene clusters of diverse organisms. Click on the organism name and enter the page of statistic data, you can see "Final Number of Clusters" and "Histogram of cluster sizes". With the number of total sets subtracting the number of cluster sizes with only one sequence, you can obtain the number of UniGene clusters. TGI's data can be found at website http://compbio. dfci.harvard.edu/tgi/tgipage.html. Click on the organism names, you can enter the database of that organism. There are statistical data for every organism. Only the number of TC sequences is what we need. The singleton data can be ignored. Alternative splicing data can be found at http://t.caspur.it/ASPicDB/. There is statistic data of human alternative splicing on the homepage, including number of genes, transcripts, etc. Although there are also alternative splicing data of other species, they are not yet complete enough. You can also obtain the numbers of genes expressed in a specific tissue from this database. In the advanced search page, you can search genes with different specific tissue names in the search bar, and then you can get the number of genes. We had these numbers averaged and got about 5000. The database does not give the data of transcripts expressed in different tissues; otherwise the average size of transcriptome t can be obtained this way. The data of average adult weights of eukaryotes are calculated based on the body sizes, which can be found or estimated from various sources. All the calculation can be conducted by simple Perl scripts, which are available on request. Reviewer #1: Dr Lavanya Kannan (nominated by Dr. Arcady Mushegian) and Dr. Arcady Mushegian The paper presents a method to calculate the phenotypic complexity of organisms. The phenotypic complexity of an organism is a measure of uncertainty associated with the size of the genome sequence, and is mathematically defined as the information entropy of the system. The amount of effective genomic information needed to produce a gene/protein sequences from a random sequence is at least the information entropy. The approach uses permutation and combination formulas to model the information needed to encode proteins for simple organisms like viruses, bacteria and other single celled organisms; and also extends the method to compute the information needed to produce differentiated cells and to construct spatial structures formed by the cells in higher organisms. The approach is not without interest, but several questions need to be addressed. 1. The main question is whether the complexity estimates given by the computations in this manuscript are any better than simply the number of protein-coding genes. Examining the I values for various species, from viruses to higher eukaryotes, one gets an impression that I is roughly proportional to gene numbers. Is this the case or not? If yes, what is the advantage of using I, and if no, then the relationship between the two is worth discussing in some detail. 2. In the calculation of I =Σn 6 for viruses (which is also I 1 for all the higher organisms discussed), where Σn is the summation of the lengths of all protein sequences, the authors make the following note: I is the same as the size of the protein coding area in the DNA sequence. It would be helpful if this equality may be explained. This paragraph, as many others, suffers from simplistic explanation of biological phenomena. In p.4: \...there are no regulations of quantity of gene products. Viruses only need to produce their proteins one by one in order." -this is not true, most if not all viruses have elaborate mechanisms of, e.g., separating the expression of early and late proteins; of maintaining quite precise ratios of different protein products expressed simultaneously; etc. Many of these processes require action of virus-encoded signals and cellular proteins. But is this relevant for computing complexity? in p. 5:\In fact, mutations are not normally allowed for a real protein sequence" -not true, viruses are notorious for rapid evolution that is facilitated, in the case of RNA viruses, by a particularly high mutation rate (but again, is this information even needed for what authors are proposing?). Why does the quantity I hold for cases of overlapping genes? A simple example that exemplifies both the above facts would be beneficial for the readers. 3. In the calculation for eukaryotes: For genomes that are not sequenced, it is noted that I 1 =6n g. It is also mentioned that in the case of overlapping genes, this quantity should be adjusted to the size of the proteincoding genes. How can this be done for the genomes that are not sequenced? Can this be elaborated? Authors' response The comments are insightful. We have revised the manuscript based on the review, especially the calculation of effective information of viruses. raised if viruses are therefore more primitive than rickettsiae, bacteria, fungi, algae, plants or animals. Indeed, a single cell alga is more complex than bacteria. Can gastrula be more advanced than a parasite in the gut of a termite? Can an organism like a viroid more primitive than free living algae? To prevent such issues, the domain of study and the scale of complexity must be clearly defined. 4. The manuscript needs a technical editing. Just to mention some example problems here: the sentence "Set × is the size of complete proteome.." is not clear. I believe what authors intend to say is "Let × be the size of complete proteome..". I also notice that the word "once" has been misplaced or misused in some sentences. Note that "once synthesized" means differently from "synthesized at once" Authors' response The comments are constructive. We have revised the manuscript based on the comments. We focus on the methodology how to calculate effective information rather than evolutionary claims in this manuscript. Reviewer #3: Dr. ED Rietman (nominated by Dr. Marc Vidal) The authors are to be commended for taking on a challenging and important biological question. Their basic hypothesis is that one can use the standard information measures on DNA strings and induce similar information measures on numbers of proteins in all types of cells. The premise is that there is in increase in complexity over the course of biological evolution and that this increase in complexity comes about as a result of a reduction in entropy. The soundness of the hypothesis will not be commented upon, because there is so much doubtful with the basic premise. Start with a simple self-replicating autocatalytic set of molecular species. If mutation-based evolution can operate on this set, there will be an increase in the complexity of the molecular species in the set. This increase in complexity is driven primarily by chemical potential and reduction in free energy. The increase in the number of new molecular species capable of participating in the reaction set results in more ways to dissipate the free energy, and thus an increase, not a decrease, in entropy [28] [29] [30] . Similarly, in a microbiological ecosystem with competing microorganisms, mutation-based evolution will increase the microorganism-based species diversity and drive up the number of ways the free energy may be dispersed. Again this results in an increase in entropy. Besides misunderstanding fundamental thermodynamic issues, there are cellular biology errors. To support the calculations it is assumed that all proteins are produced at once. This conjecture can again be argued away, because chemical reaction networks cannot produce all molecular species at once. The chemical potential imbalance, at various points in the network, is the driving factor to produce other chemical species (Le Chatelier's principle). This same logic carries over into molecular systems biology and thus into cellular biology. There is an insufficient review of the pertinent literature. The paper is not a review, but still a few paragraphs of review of other approaches to addressing this important question would have put this new work in perspective. Authors' response It is true that mutation-based evolution is an entropy increase process, but evolution may not only be mutation-based. There may be other evolutionary mechanisms. Anyway, we have deleted those parts regarding entropy in this manuscript.

Environmental factors preceding illness onset differ in phenotypes of the juvenile idiopathic inflammatory myopathies

Objective. To assess whether certain environmental factors temporally associated with the onset of juvenile idiopathic inflammatory myopathies (JIIMs) differ between phenotypes. Methods. Physicians completed questionnaires regarding documented infections, medications, immunizations and an open-ended question about other noted exposures within 6 months before illness onset for 285 patients with probable or definite JIIM. Medical records were reviewed for 81% of the patients. Phenotypes were defined by standard clinical and laboratory measures. Results. Sixty per cent of JIIM patients had a reported exposure within 6 months before illness onset. Most patients (62%) had one recorded exposure, 26% had two and 12% had three to five exposures. Patients older than the median age at diagnosis, those with a longer delay to diagnosis and those with anti-signal recognition particle autoantibodies had a higher frequency of documented exposures [odds ratios (ORs) 95% CI 3.4, 31]. Infections were the most common exposure and represented 44% of the total number of reported exposures. Non-infectious exposures included medications (18%), immunizations (11%), stressful life events (11%) and unusual sun exposure (7%). Exposures varied by age at diagnosis, race, disease course and the presence of certain myositis autoantibodies. Conclusion. The JIIMs may be related to multiple exposures and these appear to vary among phenotypes.

The juvenile idiopathic inflammatory myopathies (JIIMs) are a heterogeneous group of acquired systemic autoimmune diseases characterized by symmetric proximal weakness, the presence of characteristic rashes and other systemic features. While the aetiology of these disorders remains unknown, many lines of evidence suggest that they result from the interaction of multiple genetic risk factors and environmental exposures [1] . The JIIMs, like other autoimmune disorders, appear to be comprised of a number of clinical and serological phenotypes, each of which defines more homogeneous subsets of patients in terms of demographic features, the presence of certain myositis-associated autoantibodies, immunogenetics and outcomes [2, 3] . For example, patients with anti-p155 autoantibodies form a phenotype characterized by the frequent presence of cutaneous involvement and characteristic photosensitive rashes of JDM and the HLA-DQA1*0301 allele, whereas patients with anti-synthetase autoantibodies frequently have moderate to severe weakness, arthritis, RP, mechanic's hands, fevers, interstitial lung disease and HLA DRB1*0301 [3] [4] [5] . Clinical features of illness also appear to differ by age, gender, race and even disease course phenotypes [6] [7] [8] . Such homogeneous phenotypes might share unique combinations of environmental and genetic risk factors that result in a discrete disorder [9] . Several genetic risk factors for the JIIM have been defined, including MHC Class II alleles [10, 11] , cytokine polymorphisms [12, 13] , the protein tyrosine phosphatase gene N22 [14] and Gm and KM allotypes [15] . Environmental risk factors in JIIMs are not as well understood, and most efforts have focused on the potential role of infections in their aetiologies. Studies of cohorts of patients with JDM indicate that respiratory and gastrointestinal infections may be temporally associated with the onset of JIIM [16, 17] . Prior studies of other autoimmune diseases suggest differences in environmental risk factors in different phenotypes [9] , but the relationship between environmental risk factors and phenotypes has not been examined in the JIIMs [16, 17] . We, therefore, undertook this study to examine whether environmental factors that are temporally associated with the clinical onset of JIIM differ in selected phenotypes, focusing on a large, well-characterized population with data on both infectious and non-infectious exposures. Four hundred and twenty-three patients with probable or definite JDM or juvenile PM (JPM) [18] were enrolled into the NIH Clinical Center or Food and Drug Administration's investigational review board-approved natural history protocols from September 1994 until July 2008; subjects' written consent/assent was obtained according to the Declaration of Helsinki. The study was approved by the NIDDK/NIAMS Institutional Review Board. Enrolled patients provided a blood sample for autoantibody testing and the treating physician completed a questionnaire that included clinical, demographic and laboratory data. For 285 of these patients, questions about factors temporally associated with illness onset were also completed, which is the basis of the present study. Informed consent/parent assent was consistent with the Declaration of Helsinki. Phenotypes were defined by age of illness onset, clinical features, disease course, race or autoantibodies. Disease course was classified as monocyclic if the patient achieved remission without evidence of active disease, based on clinical examination and laboratory testing, within 2 years of diagnosis; as polycyclic if the patient had recurrence of active disease after a definite remission; as chronic continuous if disease activity persisted for >2 years; and as undefined if follow-up was <2 years from the time of diagnosis [8] . Clinical, demographic and autoantibody characteristics of the study population are described in Table 1 . Only the autoantibody phenotypes defined as anti-aminoacyl-tRNA synthetase, anti-signal recognition particle (anti-SRP), anti-Mi2, anti-p155, anti-MJ, anti-U1 RNP and autoantibody negative were included in the analyses of environmental factors. The physician questionnaire contained three questions about environmental exposures that had been previously suggested to be possibly associated with the onset of JDM [16, 17, 19, 20] . These included whether the patient had any documented infections, received any immunizations or took any medications (including vitamins, minerals, herbal preparations and dietary supplements) within 6 months before illness onset. The questionnaire also included an additional open-ended question about other environmental exposures within 6 months before illness onset relating to other possible triggers of disease and to specify these and when they occurred. Stressful life events were categorized as major vs minor and as One hundred and twenty-one patients were tested by IP immunoblotting. a Other myositis autoantibodies, which were not examined in the environmental exposure analysis, included: anti-Ro (n = 15), anti-PM/Scl (n = 5), anti-Sm (n = 3), anti-La (n = 2), anti-U5 RNP (n = 1), anti-U3 RNP (n = 1), anti-Ku (n = 1) and anti-Th (n = 1). Some patients have more than one myositis autoantibody. One hundred and one patients were not tested for myositis autoantibodies. network, family, academic or unknown type, based on the Adolescent Perceived Event Scale of Compas et al. [21] (personal communication: B. Compas, Vanderbilt University). Illness onset was defined as the month and year when the first symptom related to myositis developed. A paediatric rheumatologist (L.G.R. or G.M.) reviewed available medical records for 81% of the patients in order to confirm the reported exposures, as well as the diagnostic and clinical material contained in the questionnaire. Patient sera were tested for myositis autoantibodies by validated methods [22, 23] . For anti-p155/140 and anti-MJ autoantibodies, serum samples were screened by immunoprecipitation (IP), and this was confirmed by IP blotting [5, 24] . Sera were considered positive if they blotted the antigen in immunoprecipitates prepared using reference serum (direct) or if reference serum blotted the antigen in immunoprecipitates prepared using patient serum (reverse). Since some IP-positive sera do not react by immunoblotting, reverse IP blotting was used for most sera [5] . Case-only analyses were conducted to describe the frequency of exposures overall and in relation to patient phenotype. Statistical analysis was performed using Sigma Stat Version 3.1 (Systat Software, Inc., Chicago, IL, USA), including 2 and Fisher's exact tests to determine differences in the proportion of patients with different environmental exposures. Odds ratios (ORs) and 95% CIs were calculated using GraphPad InStat version 3.06 (GraphPad Software, San Diego, CA, USA). P-values were adjusted for multiple testing using Holm's procedure [25] , using SAS System for Windows, version 9.1.3 and SAS Enterprise Guide Version 4.1 (SAS Institute, Cary, NC, USA). Sixty per cent of JIIM patients had one or more reported exposures within 6 months before illness onset ( Table 2 ). The total number of reported exposures was more frequent in white patients than in other racial groups (P = 0.008; OR 2.1; 95% CI 1.2, 3.5; Table 2 ). Although most patients (62%) had only one reported exposure, 26% had two exposures, and 12% had three to five recorded exposures in the 6 months before illness onset ( Table 2) . Of the 64 patients with more than one exposure, 50% had a combination of infection and medication, and 27% had a combination of infection and immunization. The combination of infection and immunization was more frequent in patients from non-white racial groups (P < 0.0001; OR 22.0; 95% CI 5.9, 81.7). Patients who were >7.5 years of age at diagnosis (the median age at diagnosis) more often had three to five reported exposures compared with younger patients (P = 0.027; OR 3.7; 95% CI 1.3,10.6; Table 2 ), and patients with anti-SRP autoantibody more often had three to five exposures compared with patients without a myositis autoantibody (P = 0.027; OR = 31.0; 95% CI 1.9, 507). Patients with a longer delay to diagnosis (>4 months, the median delay) were more likely to have three to five exposures than patients with a shorter delay (P = 0.034; OR 3.4; 95% CI 1. 2, 9.8) . There were no other significant differences in the frequency or number of noted exposures between clinical phenotypes [JDM vs juvenile PM (JPM)], nor by gender, disease course, delay to diagnosis or between other autoantibody phenotypes (data not shown). Infections were the most common type of exposure identified within 6 months before diagnosis, consisting of 45% of the total number of reported exposures, followed by medications (18%) and immunizations (11%) ( Table 2 ). Patients 47.5 years of age at onset, those with 44 months delay to diagnosis, those with a polycyclic illness course and those who were myositis autoantibody negative were more likely to have an infection in the 6 months before diagnosis than older patients, those with greater delay to diagnosis, those with a monocyclic or chronic continuous illness course or those who had anti-p155 or anti-SRP autoantibodies (ORs 1.8-4.3, Table 2 and data not shown). There were no other differences between documented types of exposure between clinical or other autoantibody phenotypes, nor by gender, disease course, race or delay to diagnosis. From the open-ended exposure question, stressful life events constituted 11% of the reported exposures. Patients >7.5 years of age reportedly experienced a stressful life event more frequently in the 6 months before diagnosis than younger patients (P = 0.003; OR 3.5; 95% CI 1.5, 7.9). Unusual sun exposures comprised 7% of the total exposures in the 6 months before diagnosis and occurred exclusively in patients with JDM, not JPM. Unusual sun exposures included those resulting in sunburn, as well as receiving more sun than usual or travel to a more sunny location. An unusual chemical exposure was recorded 3% of the time and included application of pesticides inside or around the home, painting the home, use of formaldehyde to clean the child's bed and application of a hair perming chemical. Seven (2%) reported exposures involving unusual animal contact within 6 months of illness onset, including a dog or cat scratch, exotic bird bite or multiple flea or mosquito bites. Four (2%) exposures involved weight-training exercise or physical trauma, and two (0.6%) exposures involved dietary supplement usage before illness onset, including creatine monokinase and Echinacea. These less-frequent exposures were present exclusively in patients with JDM, except that weight-training exercise was also noted in one JPM patient. Weight training, physical trauma and dietary supplements were seen exclusively in patients >7.5 years of age at diagnosis and in patients with a greater delay to diagnosis. White patients and patients who did not have an identified myositis autoantibody were more likely to have a documented infection within 6 months before illness onset than those from other racial groups or those with the anti-p155 autoantibody (P = 0.0008; OR 2.7; 95% CI 1.5, 4.8 and www.rheumatology.oxfordjournals.org None 114 (40) 62 (41) 52 (38) 69 (35) 45 (52) 15 (32) 31 (48) 49 (42) 20 (46) 19 (36) No. of patients exposed (22) 8 (25) 10 (30) 15 (22) 8 (34) 8 (24) 3-5 20 (12) 5 (6) (12) 4 (10) 2 (7) 1 (3) 12 (18) 2 (8) 2 (6) (18) 27 (17) 35 (20) 48 (18) 14 (20) 11 (19) 13 (22) 26 (18) 15 (33) zz Immunization 37 (11) 16 (10) 21 (12) 29 (11) 8 (12) 10 (18) 4 (7) 14 (10) 5 (11) 6 (8) Other 85 (25) 28 (18) P = 0.007; OR 3.9; 95% CI 1.5, 9.9, respectively; Table 3 ). The majority (68%) of patients with an infectious exposure had one documented infection, but 28% had two and 4% had three to five infections documented within 6 months of illness onset. Patients >7.5 years of age were more likely than younger patients to have two infections (P = 0.038; OR 2.7; 95% CI 1.1, 6.3; Table 3 ). Respiratory infections were the most common type of infection reported, followed by mucocutaneous and gastrointestinal infections (Table 3) . Pharyngitis was the most frequent specific infection and was more prevalent in patients >7.5 years of age than in younger patients (P = 0.017; OR 2.7; 95% CI 1.2, 6.0; Table 3) . A flu or febrile illness and otitis media were each seen in 13% of patients, and an upper respiratory infection in 10% of patients within 6 months before illness onset. There were no other differences noted in the infection site or type of infection among clinical or autoantibody phenotypes, nor by gender, race, disease course or delay to diagnosis. Patients 47.5 years of age were more likely to have one drug exposure (P = 0.004; OR 15.4; 95% CI 1.8, 135), whereas those >7.5 years of age were more likely to have two drug exposures (P = 0.008; OR 18.9; 95% CI 1.0, 358) in the 6 months before illness onset (Table 4) . Of interest, >25% of the medication usage documented included drugs that were potentially photosensitizing or myopathic [26] [27] [28] [29] [30] [31] . There were no other differences noted in medication usage between clinical or autoantibody phenotypes, nor were there any differences by gender, race, disease course or delay to diagnosis. There was no difference in the proportion of patients who received an immunization in the 6 months before illness onset or in the number of immunizations received, between clinical or autoantibody phenotypes, nor by age, gender, race, disease course or delay to diagnosis. Patients with a polycyclic illness course were more likely than patients with a monocyclic illness course to have received an immunization or to have received a measles-mumps-rubella (MMR) vaccine in the 6 months before illness onset (21 vs 6%; P = 0.023; OR 4.1; 95% CI 1.2, 13.9 and 50 vs 6%; P = 0.035; OR = 17.0; 95% CI 1.3, 223, respectively). Given the time period under study, it was not surprising that patients >7.5 years of age at diagnosis were more likely to have received a hepatitis B vaccine than younger patients (47 vs 12%; P = 0.002; OR 6.2; 95% CI 2.0, 19.7), whereas patients 47.5 years of age were more likely to have received a diphtheria-(pertussis)-tetanus vaccine (22 vs 6%; P = 0.033; OR 8.2; 95% CI 0.98, 69.8). Nine per cent (n = 26) of patients had at least one stressful life event in the 6 months before illness onset, with 72% of these being major stressors and the remainder being minor stressors. The majority of these patients (65%) had one stressor, but 31% had two recorded stressors and 4% had three stressors. The categorization of stressors included network (50%), family (25%), academic (19%) and unknown types (6%). Patients >7.5 years of age had a stressful life event more frequently than younger children (P = 0.003; OR 3.5; 95% CI 1.5, 7.9). There were no differences in the proportion of patients with a reported stressor or in the number or type of stressor in 6 months before illness onset between clinical or autoantibody phenotype, nor by gender, race, disease course or delay to diagnosis. The availability of a large, well-characterized population enabled us to examine the relationship between environmental exposures before illness onset and phenotypes in JIIM. We confirmed a number of exposures that had also been seen in prior studies of JDM, particularly the temporal association of respiratory infections preceding illness onset [16, 17] . We identified for the first time that a number of other non-infectious exposures occurred within 6 months of the first signs of illness, including medications, many of which are potentially myopathic or photosensitizing, immunizations, stressful life events and sun exposure. The main novel findings of this study were differences in some exposures by age at diagnosis, delay to diagnosis, race, disease course and autoantibody phenotypes. For example, children younger than the median age at the time of diagnosis had a higher frequency of documented infections, whereas older children had a higher frequency of stressful life events in the months before illness onset. Patients without a myositis autoantibody had a higher frequency of infections in the 6 months before illness onset than was seen in patients with anti-p155 or anti-SRP autoantibodies, whereas patients with anti-SRP autoantibodies had a greater number of documented exposures than patients without a myositis autoantibody. These findings suggest that environmental exposures may differ by phenotype, and that they could be useful in understanding pathogeneses [1] . We found that an infectious illness, particularly a respiratory infection, frequently occurs within several months before juvenile myositis onset, supporting the findings of other studies of exposures temporally associated with the onset of JDM. In one study, a prospective registry of patients within 6 months of illness onset in which data were based on a parent environmental interview and medical record review, respiratory infections were identified within 3 months of illness onset in 57% of patients [16] . The other, a retrospective cohort with review of medical records by infectious disease specialists, identified infections within 3 months before Conventions as per Table 2 . Bold values represent P 4 0.05 after Holm's adjustment for multiple comparisons (using family-wise error rates of 5%). a Based on total number of drug exposures, because some patients had >1 documented drug exposure. Potentially myopathic drugs included penicillin (n = 2) and ranitidine (n = 1) [28] [29] [30] [31] . Potentially photosensitizing drugs included loratadine (n = 1), diphenhydramine (n = 1) and sertaline (n = 1) [26, 27] . Drugs classified as potentially both myopathic and photosensitizing included ibuprofen (n = 5), trimethroprim/sulphamethoxazole (n = 3), isoniazid (n = 2) and erythromycin-sulfisoxazole (n = 1). Drugs not known to be either myopathic or photosensitizing included amoxicillin (n = 4), cefaclor (n = 3), pseudoephedrine (n = 2), cetirizine (n = 1), albuterol inhaler (n = 1), flecainide (n = 1), bromopheniramine maleate (n = 1), nedocromil (n = 1), oxybutynin (n = 1), permethrin (n = 1), pyrethrine (n = 1), cantharidine (n = 1), cefadroxil (n = 1), acetaminophen (n = 1), streptomycin (n = 1), erythromycin (n = 1) and nystatin (n = 1). Drugs whose classification is unknown included unknown antibiotic (n = 10), birth control (n = 3) and anaesthetic (n = 1). *P = 0.004; OR 15.4; 95% CI 1.8, 135. y P = 0.008 ; OR 18.9; 95% CI 1.0, 358. the first symptoms of JDM in 33-50% of patients, and respiratory infections accounted for 80% of the infections [17] . The lack of control comparator groups in all of these studies, however, does not enable one to conclude that these exposures differ from a healthy population, nor that they are associated with the onset of illness. While infections, particularly upper respiratory infections, are reported frequently in school age children [32] , a prospective matched cohort of new-onset JDM patients reported a higher frequency of antecedent illness in the JDM patients compared with friend controls from the same geographical region [19] . We identified for the first time that a number of other non-infectious exposures also occurred within 6 months of the first signs of illness, including medications, many of which are potentially myopathic or photosensitizing, immunizations, stressful life events and sun exposure. Pachman et al. [16] noted medication use in >60% of patients, including medications for symptoms of early illness or antibiotics to treat associated infections. A listing of medications taken by patients in the present study and in others includes similar medications (Table 4) , and we noted that many of the medications could be potentially myopathic or phototoxic [26, 27, 29, 30] . Drug-induced myositis has been well described with a number of different medications, including D-penicillamine, lipid-lowering agents, L-tryptophan and IFN-a [33, 34] . Myopathic or phototoxic drugs, however, could lead to the first symptoms of myositis. Other environmental factors reported here, including ultraviolet light exposure, emotional stress and heavy weight lifting, have been reported as possible risk factors for adult DM or PM in case-controlled studies [35] [36] [37] [38] . Almost 40% of the patients in this study had two or more reported exposures within 6 months before illness onset, rather than a single documented exposure. This is consistent with the concept that, just as systemic autoimmune diseases are polygenic [39] , they might also be polyenvironmental, meaning that patients may have more than one exposure before developing the disease. These exposures may also be dependent on gene-gene, environment-environment and gene-environment interactions. In diseases such as cancer, multiple infectious and non-infectious environmental factors have been associated with specific malignancies, and these environmental exposures have been shown to affect the development of disease in different ways, including altering mutagenesis, promotion and direct carcinogenesis [40] . Synergistic interactions between some of these environmental factors, including viral and non-infectious exposures, have also been seen in certain malignancies [41, 42] . It is possible, though, that there was a confounding between exposures, such as an infectious illness and the use of antibiotics, as noted by Pachman et al. [16] . Our data suggest that further investigation of the interaction between environmental exposures may be useful. It is important to emphasize that the temporal association of environmental exposures with illness onset does not imply causality. For example, certain exposures, such as trauma or weight training, could have occurred after the onset of illness as a consequence of the first unrecognized symptoms of disease, such as fatigue or muscle weakness. Rather, exposures with temporal relationships to disease onset, as were seen in this hypothesisgenerating study, constitute a first step for determining which factors may trigger the onset of illness and warrant further investigation. Additional support for a relationship between these exposures and disease pathogenesis could be provided by dechallenge data, which did not exist in this cohort-based study, from laboratory investigations and from case-controlled epidemiological studies [43] . A case-control study by Pachman et al. [19] did not find any significant differences in pesticide use, psychological stress or exposure to animals in 80 JDM patients within 6 months of illness onset compared with 63 age-matched geographically similar healthy controls with similar school or daycare experiences, nor was parvovirus found to be an aetiological factor in recent-onset JDM patients compared with age-, genderand race-matched controls [44] . However, both of those studies may not been adequately powered to detect differences between the cases and controls. Also, the extent of matching of controls may have obscured differences with JDM patients. For example, in the parvovirus study [44] , the controls were age, race and gender matched to patients, but they were not geographically matched, whereas in the study of Pachman et al. [19] , the healthy controls, frequently age-matched classmates and neighbours, may have been geographically overmatched, but they were not gender or race matched. An appropriately powered prospective case-controlled study is needed to confirm the observations from this and other previous reports. There are a number of potential limitations in this study. A primary limitation is the absence of a control group. Thus, the frequencies of exposures observed in juvenile myositis patients overall may not differ from healthy control populations and these exposures may not be associated with the onset of illness. In addition, there could be under-or over-reporting of potential exposures, including a selection bias in the patients who had the environmental component of the questionnaire completed. We also found more exposures, including infectious illnesses, in white patients compared with patients in other racial groups. This could potentially be the result of differences in access to health care, resulting in better documentation of such exposures. The somewhat arbitrary period of 6 months before the onset of illness for identification of environmental factors might not be relevant to the initiation of myositis for all exposures. Certain exposures could require a longer period to induce their effects, as has been reported in malignancies, silicosis and other disorders, while for other exposures a shorter time frame might be more relevant [37, 38] . Also, exposures other than infections, drugs and vaccines were reported in an open-ended manner, and patients were not required to be directly interviewed to obtain information about environmental exposures. We attempted to www.rheumatology.oxfordjournals.org overcome these possible biases by conducting a formal review of most of the medical records of the study subjects. However, the medical records might also have selection bias by reporting only some of the significant environmental exposures. Certain exposures, such as exposures in the home and use of certain chemicals, are likely not captured uniformly in the medical record by the treating physician. Nonetheless, the fact that our data on infections before illness onset are similar to those of other large cohorts suggests that the quality of the data and reporting are reliable [16, 17] . Finally, while some of the ORs in our study are large, the CIs may be wide and estimates could be inflated due to relatively small numbers of patients in some groups. In summary, we have identified a number of environmental exposures, including infectious and non-infectious agents that occurred within 6 months before illness onset, varied by phenotype and may be important in the pathogenesis of JIIM. These findings suggest that a search for a single environmental factor that causes or triggers a single disease as currently defined, such as JIIM, may be unproductive, as patients could have several environmental exposures and these could vary with the disease phenotype that develops. These exposures require confirmation in case-controlled studies to identify whether they are associated with illness onset and whether they play any role in aetiology, yet they suggest focused areas of further research to better understand the environmental factors associated with the onset of JIIM phenotypes and their possible interrelationships with genetic risk factors. Rheumatology key messages . Environmental exposures before the onset of juvenile myositis include infections, medications, vaccinations, sun exposure and stressful life events. . Exposures vary by disease phenotype, defined by age of illness onset, race and autoantibody status.

Network Analysis of Global Influenza Spread

Although vaccines pose the best means of preventing influenza infection, strain selection and optimal implementation remain difficult due to antigenic drift and a lack of understanding global spread. Detecting viral movement by sequence analysis is complicated by skewed geographic and seasonal distributions in viral isolates. We propose a probabilistic method that accounts for sampling bias through spatiotemporal clustering and modeling regional and seasonal transmission as a binomial process. Analysis of H3N2 not only confirmed East-Southeast Asia as a source of new seasonal variants, but also increased the resolution of observed transmission to a country level. H1N1 data revealed similar viral spread from the tropics. Network analysis suggested China and Hong Kong as the origins of new seasonal H3N2 strains and the United States as a region where increased vaccination would maximally disrupt global spread of the virus. These techniques provide a promising methodology for the analysis of any seasonal virus, as well as for the continued surveillance of influenza.

Influenza, a negative-sense RNA orthomyxovirus, is one of the few diseases that is truly global in scale. It is responsible for approximately three to five million cases of severe acute respiratory illness and 250,000 to 500,000 deaths each year throughout the world [1] . In 2009, the swift isolation of swineorigin H1N1 strain (S-OIV) from all continents within several weeks of onset reinforced the idea that influenza is a highly infectious agent circulating worldwide [2, 3] . Although vaccination remains one of the most powerful ways of combating influenza, choosing a representative strain for vaccine composition poses a challenging problem. Due to the virus's high evolutionary rate, significant resources must be spent to update vaccines each year in order to match the dominant epitope of the season. Even with annual strain selection, major antigenic reassortment can obviate otherwise promising vaccine candidates, as occurred with the 'Fujian/411/2002'-like H3N2 strain in 2003 [4, 5] . To prevent such vaccine failures, a solid understanding of the global spread of influenza must inform the design process. If reservoirs for new viral strains can be identified, surveillance in these areas can better optimize prediction of seasonal variants in seeded regions. Previous papers investigating the global circulation of H3N2, the major seasonal influenza subtype prior to pandemic H1N1, focused on transmission within and between climate zones. Important motivating factors for such analysis include increased aerosol transmission in cold and dry conditions, as well as increased indoor crowding and decreased host immunity in cold and wet conditions [6, 7] . In the temperate zones, influenza exhibits distinct seasonality with flu-related cases spiking in the winter. However, several papers have confirmed the presence of viral diversity even between these epidemic peaks [8, 9, 10] , suggesting two possible scenarios during the inter-epidemic period: either viral infections locally persist at a low level only to reemerge as the dominant strains of the epidemic season, or an outside source introduces new genetic diversity into temperate populations each year. Although a degree of local persistence may occur, phylogenetic analysis supports the latter scenario, with few direct links between strains of the same region but successive seasons [8, 9, 10] . For a given temperate zone, these conclusions suggest the tropics or the opposite temperate zone as plausible external seeding regions. At first blush, northern-southern temperate oscillations seem credible. Each year, northern and southern temperate climates have alternating seasonal influenza epidemics, lasting from November to April, and May to September respectively [11] . A possible mechanism of viral spread could involve transmission from the seasonal peak of one temperate zone into the season ebb of the other. On the other hand, specific epidemiological characteristics suggest a tropical origin for influenza. For example, although both climates share a similar yearly burden of mortality from influenza, the tropics do not possess the same consistent seasonal peaks during the winter months [9, 12, 13] . With a constant, low-level circulation of viruses year-round, the tropics represent an ideal epicenter for the extended transmission of new viruses to the rest of the world [14, 15, 16] . Several papers tracking H3N2 across continents have asserted that this tropical reservoir of influenza strains lies within East-Southeast Asia [12, 14, 17] . Russell, et al. analyzed H3N2 data to identify regions of the world that are antigenically and genetically leading or trailing. They found that newly emerging strains appeared in E-SE Asia roughly 6-9 months earlier than in other parts of the world, while South America experienced delayed transmission of roughly 6-9 months following other parts of the world [8] . However, such studies have been limited by several drawbacks. Most papers focus on H3N2 as a single entity, when in reality, it co-circulates with several other subtypes, the most important of which is seasonal H1N1 [11] . Although they possess different surface antigens, H3N2 and H1N1 share enough genetic similarity to display cross-immunity. As a result, seasonal H1N1 may demonstrate transmission patterns distinct from H3N2's [18, 19] . Such codependence between different subtypes is exemplified by the pandemic years of 1957 and 1968, when H2N2 replaced preexisting H1N1 and H3N2 replaced preexisting H2N2, respectively [20, 21] . Similarly, the antigenically different pandemic H1N1 strain of 2009 has largely overtaken previously circulating H1N1 and H3N2 [22] . During the years our dataset took place, evidence that H3N2 and H1N1 rarely co-dominate in a season further supports the idea of codependent dynamics [7] . A second shortcoming stems from biases in the number of sequences from different regions and different seasons [8] . Most isolates of H3N2 and H1N1 were sampled from North America, whereas Africa and South America have been largely neglected [23] . Many sequences were obtained within the last 15 years, making reliable tracking over long periods of time problematic. On the level of climate zones, the number of temperate isolates far outstrips the tropics. Although hemagglutinin (HA), the HA1 domain, and neuraminidase (NA) have the most globally representative distributions of sequences, even these remain skewed ( Figure S1 , Figure S2 ). In this paper, we present a novel probabilistic model for tracking the spread of influenza that employs two strategies to eliminate regional and seasonal data bias. The first involves clustering isolates of high sequence similarity by region and season. Since we would expect highly similar sequences from the same time and location to be related, we considered seeding events between clusters to be of greater significance. Consideration of clusters rather than individual sequences nullifies the overrepresentation of a high number of isolates from a single region and season ( Figure 1 ). As a second strategy for eliminating bias, we determined statistical significance of inter-cluster seeding events by modeling transmission as a binomial distribution with prior probabilities based on the proportion of sequences isolated before a given time point. To illustrate our methodology, Figure 2 depicts the 2003-2004 flu season, which was marked by failure to predict the dominant, tropically-derived Fujian/411/2002-like H3N2 strain. We identified a strong seeding pattern from the tropics to all three climate zones, supporting the effectiveness of our methodology. We applied this model to the H3N2 and H1N1 coding regions of HA and NA, the most antigenic proteins of the eight viral segments. Clustering H3N2 sequences confirmed previous findings that this strain originates in the tropics, specifically E-SE Asia, and seeds South America by way of North America last. Clustering H1N1 NA also revealed a similar pattern of circulation beginning in the tropics. However, similar H1N1 analysis by continent and country was not possible due to the absence of a larger number of countries in the dataset. Applying the same methodology to the H3N2 HA1 domain increased the geographic diversity enough to enable reconstruction of the global influenza network prior to the 2009 pandemic strain at a country level. Our results suggest a possible flu seeding hierarchy beginning in China and spreading throughout a highly interconnected E-SE Asian subnetwork. From there, viruses transmit to an Oceanic subnetwork dominated by interchange between Australia and New Zealand. Both subnetworks seed into the USA, which in turn seeds many countries, particularly in South America. Expanding upon the sink-source hypothesis of global influenza dynamics proposed by Rambaut, et al. [15] , we applied techniques of graph theory to identify important source and sink regions in the global flu network. These techniques better describe the dynamic nature of influenza movement across the globe, as well as suggest different vaccination strategies to disrupt maximally viral flow around the world. Spatiotemporally clustering the complete H3N2 and H1N1 coding sequences for HA and NA allowed the determination of multiple statistically significant seeding seasons between 1988 and 2009. For our initial analysis, we clustered sequences into three climate zones-northern temperate, tropical, and southern temperate. To determine seasonal boundaries, we defined the northern temperate season to last from 1 st July to the 30 th June of the following year and the southern temperate season to last from 1 st January to the 31 st December of the same year [11] . Although the tropics do not have a well-defined seasonal pattern, we determined a consensus tropical flu season from 1 st October to 30 th September of the next year (Text S1, Table S1 ). Results for H3N2 showed that the overwhelming majority of statistically significant seeding seasons came from the tropics, confirming previous findings ( Figure 3A , Figure S3A ). Clustering H3N2 by the six major continents rendered an even more detailed picture. For HA, Asia was the primary seeder of Asia, North America, and Oceania. Prominent transmission from North America to Europe and South America was also observed ( Figure S3B ). Interestingly, this hierarchical seeding structure reflects the findings of Russell, et al., which identified Asia and South America as antigenically advanced and lagging continents respectively [8] . This network of hierarchical seeding can be visualized as a directed graph plotted against the world map ( Figure 4A ). Analysis As evidenced by several historic vaccine failures, the design and implementation of the influenza vaccine remains an imperfect science. The virus's rapid rate of evolution makes the selection of representative strains for vaccine composition a difficult process. From a global health viewpoint, how to optimally implement a limited stockpile of vaccines is another fundamental question that remains unanswered. An understanding of how influenza spreads around the world would greatly aid the design and implementation process, but regional and seasonal bias in collected virus samples hampers epidemiologic analysis. Here, we show that it is possible to counter this data bias through probabilistic modeling and represent the global viral spread as a network of seeding events between different regions of the world. On a local scale, our technique can output the most likely origins of a virus circulating in a given location. On a global scale, we can pinpoint regions of the world that would maximally disrupt viral transmission with an increase in vaccine implementation. We demonstrate our method on seasonal H3N2 and H1N1 and foresee similar application to other seasonal viruses, including swine-origin H1N1, once more seasonal data is collected. of NA produced similar findings with the exception of North America being its own primary seeder ( Figure 3B ). No complete HA and NA isolates existed in the NCBI Influenza Virus Resource database [24] for Africa. The complete dataset of HA and NA represented only 17 and 21 countries respectively. Despite the sparse number of countries for analysis, both HA ( Figure S3C ) and NA ( Figure 3C ) consistently identified Hong Kong (considered a country by NCBI sequence annotation) as the primary external seeder of USA and New Zealand among others, and New Zealand as the primary external seeder of Australia. Due to fewer available sequences, clustering H1N1 did not yield as many significant seeding events as H3N2; however, our tests suggest that H1N1 adopts a similar seeding pattern with the tropics as a source. Of the two segments, NA sequences display a broader geographical profile than HA. In particular, our HA dataset for H1N1 contained no sequences from Hong Kong and only 1 (0.091%) China sequence, while NA contained 9 (0.69%) Hong Kong and 3 (0.23%) China sequences. Consequently, we considered NA to be more suitable for comparison between H3N2 and H1N1 and HA to be a background signal to assess the effect of Hong Kong and China on global influenza transmission. Even so, the number of these H1N1 Hong Kong and China sequences remained vastly disproportionate to the 361 (7.42%) Hong Kong and 133 (2.73%) China sequences of H3N2. Clustering H1N1 NA by climate zone supported the theory of global viral spread from the tropics ( Figure 5B ). Unlike H3N2, H1N1 analysis by continent and country was inconclusive due to low (typically fewer than 3 seeding events), homogeneous counts. Although inconclusive, the fact that a tropical signal could be detected at all from such few tropical countries, including Hong Kong and China, suggests that H1N1 adopts a similar seeding pattern out of the tropics. Due to insufficient sampling, however, a more detailed transmission pattern could not be discerned. Although using the complete HA and NA coding genomes facilitated differentiation of isolates by Hamming distance, the absence of data from certain countries limited the information gained from clustering at this geographic detail, a problem that has plagued previous studies [8] . To increase the amount of data from different geographical regions, we clustered H3N2 sequences of the HA1 epitope, expanding the number of isolates in the dataset from 2,251 to 4,864, and the number of countries from 17 to 81. A necessary consequence of expanding geographic coverage was an increase in the number of non-unique solutions (Text S1). Importantly, clustering HA1 by climate and continent was corroborated by findings from the complete HA and NA sequences, lending credence to the validity of the dataset. Due to the inclusion of isolates from Africa, which was hitherto not present in our datasets, H3N2 HA1 analysis also revealed Europe and North America tied for being the primary seeders of Africa. Country clustering of the HA1 data produced a highly detailed global network of influenza variants. USA, Hong Kong, Australia, and China were identified as the four most prominent seeding countries in that order ( Figure 3D , Table S2 ). From the data, an inferred seeding hierarchy would begin with China at the epicenter of an E-SE Asian influenza subnetwork. Our analysis supports China as the most predictive seeder of many Asian countries, including Hong Kong. Both China and Hong Kong then serve as a launching pad for the dispersal of new seasonal variants to the rest of the world [14, 17] , in particular USA and an Oceanic subnetwork dominated by interchange between Australia and New Zealand. Viruses from USA, the largest seeder of the entire world, then spread to a number of South American, European, and African countries. Interestingly, Australia and Hong Kong are equally probable seeders of the USA ( Figure 3D ). Detailed transmission events are enumerated in Table S2 . An inset of the Asian subnetwork is depicted in Figure 4C , a demonstration of this study's high geographic resolution. After clustering, there were a total of 10 observed seeding events into the northern temperate zone: 1 from the north, 4 from the tropics, and 5 from the south. Up until that year, the skewed regional distribution of HA sequences included 541 (48.3%) northern temperate, 240 (21.4%) tropical, and 339 (30.2%) southern temperate isolates. Multiplying these percentages with the 10 observed seeding events yielded expected counts of 4.8, 2.1, and 3.0. Therefore, the number of seeding events from the north was less than expected, and from the tropics and the south, more than expected. Corresponding binomial p-values-0.986, 0.043, and 0.049, respectively-indicated that there were two statistically significant events, the most significant of which was transmission from the tropics into the northern temperate zone. Similar analysis for transmission into the tropics and the southern temperate showed that only the tropical zone was a significant seeder. doi:10.1371/journal.pcbi.1001005.g002 As can be seen with the world map plots ( Figure 4A,B) , a natural representation of the global influenza network is a directed graph with each node representing a clustered region (climate, continent, and country) and each edge representing a seeding event with a weight equal to the number of significant seeding seasons. To quantify observed patterns, we employed principles of graph theory to measure the importance of nodes using four different metrics. By counting the number of indegrees and outdegrees of each node for H3N2, we identified that the tropics and the northern temperate zone ( Figure S4A ), specifically Asia and North America ( Figure Figure S5A ), transmit and receive the most seeding events to and from the rest of the world, respectively. In a similar manner, we identified USA, Hong Kong, Australia, and China as the greatest seeders, and USA, Japan, Australia, and Hong Kong as the most seeded ( Figure 6A) . In this analysis, we differentiated between internal (self-seeding) and external (seeding between nodes) transmission events. Importantly, we can accurately detect internal events in temperate countries since their flu seasons are discrete. On the other hand, the specificity for internal events in the tropics is much lower due to unpronounced seasonal peaks. To minimize the number of local false positives, we demarcated seasons within the tropics on a per country basis. We found that for all climate zones except the tropics ( Figure S4A ) and all continents except Asia ( Figure S5A ), the number of internal seeding events paled in comparison to the proportion of external seeding events,. The more numerous internal events in the tropics and Asia indicate a high level of circulation between tropical countries and between Asian countries. This pattern is supported by the highly interconnected E-SE Asian subnetwork depicted in Figure 4C . The small proportion of internal events for countries supports the notion that local persistence often plays only a minor role in influenza transmission [8, 9, 10] (Figure 6A ). Beyond the absolute number of seeding events, a region's influence on global viral spread is also dependent on the topological structure of the graph itself. As an analogy, consider the influenza network as a system of connected train stations each representing a single region seeding influenza. In such systems, trains begin and end their routes at terminal stations. Similarly, influenza commuters begin their journeys at terminal sources and end at terminal sinks in each season. These start and end terminals can represent regions where new influenza variants respectively originate and ultimately spread to. To quantify the terminal characteristic, we calculated the outdegree minus the indegree of each node, which we term ''degree flow.'' Positive degree flow indicates terminal sources, while negative indicates terminal sinks. Countries were also ranked by calculating the proportion of nodes in a 1,000 randomized networks with a greater, or lesser, degree flow (Text S1). For analysis by climate zone, the tropics was identified as the only terminal source, suggesting that flu spreads from the tropical belt outward to both temperate zones ( Figure S4B ). As for continental clustering, Asia was the only terminal source, indicating that global circulation begins in Asia and ends in terminal sink continents, of which North America was the most prominent ( Figure S5B) . On a country level, Hong Kong and China were the greatest terminal sources, corroborating our observations ( Figure 3D) . Australia was also a conspicuous terminal source, especially within the Oceanic subnetwork where it seeded the greatest terminal sink, New Zealand. Several South American countries, including Chile and Argentina, figure as terminal sinks too, correlating with such countries as antigenically delayed [8] (Figure 6B ). Trains also stop at waypoint stations, which can be the junction of a large number of routes. Correspondingly, certain regions act as waypoint sources: important intermediate launch pads to other destinations. Others act as waypoint sinks: important points of convergence for multiple routes. Eigenvector centrality can gauge this property on the principle that connections to high-scoring nodes contribute more to the score of the node in question than equivalent connections to low-scoring nodes. We used a method akin to PageRank, Google's method of assigning importance to web pages [25] . Using this method, the northern temperate zone was the most important waypoint source and sink ( Figure S4C) . Similarly, the predominantly northern temperate continents of North America and Europe were identified as prominent waypoint sources and sinks. Asia, however, was the greatest waypoint source but a poor waypoint sink, correlating with its role as a greater terminal source than North America or Europe ( Figure S5C ). Interestingly, USA was both the greatest waypoint source and sink ( Figure 6C ). H1N1 NA clustering by climate zone produced results similar to that of H3N2 NA. The tropics consistently scored highest by seeding outdegree, positive degree flow, and PageRank source. In addition, the tropics possessed a large amount of internal seeding events. These results emphasize that similar to H3N2, H1N1 circulates within the tropics across seasons only to spread eventually to the temperate zones. Betweenness measures the number of shortest paths between any two vertices in a network that lie on a given node. In the context of influenza, increasing vaccinations in regions of high betweenness would hypothetically have the greatest effect on diminishing the spread of infection worldwide. This novel strategy contrasts with previous studies simulating containment only at the source of influenza [26, 27] . For H3N2, this criteria highlighted Europe and North America as promising candidates for vaccination programs ( Figure S5D ). Clustering by country revealed USA, Japan, and Australia as sites in the influenza network vulnerable to disruption ( Figure 6D ). Using statistical and network theory analysis, we analyzed H3N2 and H1N1 sequence data to determine the global spread of influenza. Our novel method employs two main strategies to eliminate geographic and seasonal bias: 1) Spatiotemporal clustering of sequence data to count seeding events between clusters and 2) Use of binomial prior probabilities based on the regional proportion of viral isolates to screen for significant seeding events. Applying these techniques to coding HA and NA segments of H3N2 by climate zone and continent revealed a seeding pattern stemming from the tropics, particularly Asia. HA1 analysis produced a more detailed picture: each year, a wave of seasonal flu originates in China to feed an E-SE Asian subnetwork. From there, China and Hong Kong seed two major subnetworks, each dominated by Australia and USA. Similar clustering of H1N1 NA sequences by climate zone reproduced tropical transmission to the rest of the world. However, due to inadequate geographic coverage, clustering H1N1 by continent and country proved inconclusive with few significant seeding events detected. One explanation for these results is that important seeding countries, such as China and Hong Kong, were too underrepresented in the dataset. Alternatively, global patterns may be weaker for H1N1 due to crossreactivity between the two strains [18, 19] , a conclusion reflected by the smaller number of seeding events for the strain. In our analysis, the total number of seeding seasons for each region did not necessarily correspond to the total number of isolates from each region, indicating that our methodology counters data bias. However, certain confounders may affect results. First, selection bias in sampling remarkable variants, such as patients suffering severe rather than mild or non-symptomatic influenza, would poorly represent flu in the general population. Moreover, many sequences had to be excluded from our dataset due to poor annotation and lack of date information. Finally, although our probabilistic methodology accepts regional and temporal variability, it has low sensitivity for detecting anything but particularly significant seeding events for regions with very few sequences. This issue becomes important in analyses with regions that have no sequences whatsoever, as with near-absent sequences from Hong Kong and China for H1N1 HA. The persistence of such bias highlights the continuing need to sequence viruses in underrepresented areas, especially the tropics. Each year, the current influenza vaccine is formulated separately for the Northern and Southern Hemisphere; one can surmise that two viral strains may not be enough to represent the entire pool of influenza strains around the world. Although there are many other economic and political concerns to consider, our methodology suggests several ways of guiding vaccine strain selection based on biological and epidemiological principles. Graph theory metrics-terminal and waypoint sinks and sources, as well as degree and betweenness centralities-pinpoint potential regions in which increased vaccinations could stem the transmission of influenza globally as well as locally. Increased analytical resolution could optimize vaccine design by choosing the dominant antigenic strain of a country's most predictive seeder. Vaccines could be catered to each country, rather than each hemisphere. At the very least, our analysis advises strain selection from the tropics, from which seasonal strains are dispersed each year. On the other hand, local strain selection within a country should prove comparatively ineffective, as few viruses persist in the inter-epidemic period to seed the following flu season. Our analysis of terminal sources resonates with an old hypothesis that in southern China, zoonotic infection from liveanimals markets [28] selling in particular duck-a natural host of influenza [29] -combined with a dense population for sustained viral circulation, could be the main ingredients for the creation of new seasonal influenza variants. In support, two major acute respiratory infections-SARS [30] and H5N1/97 [31, 32] -have been definitively traced back to southern China, with Hong Kong serving as an important sentinel post for the rest of the world. Other influenza pandemics, 1968 H3N2 (Hong Kong) [28] and even as early as 1889 pandemic influenza [33] , have suspected origins in southern China. It would be interesting to dissect the factors that govern waypoint sources and sinks. For example, air travel and other transportation may play a major role in the dispersal of virus worldwide [8, 19, 34, 35] . Many important hubs of the global flu network, including USA, Australia, Hong Kong, and China, have several of the world's busiest airports [36] . Understanding the reasons for these seeding patterns may offer other strategies for arresting the movement of flu. The advent of 2009 pandemic S-OIV has largely depleted the number of seasonal H3N2 and H1N1 infections, most likely via cross-reactivity between novel and seasonal strains [22] . Consequently, the conclusions of this paper may not necessarily apply to current dynamics of seasonal H3N2 and H1N1. However, the fact that H1N1 shares a tropic-centric movement pattern with H3N2 despite cross-reactivity suggests that these patterns may still persist even in the presence of the cross-reactive S-OIV. Moreover, this paper demonstrates that when more sequence data is deposited in NCBI, a similar methodology can be applied to predict global circulation of S-OIV as well. All sequence data used in this study was publicly available from the National Center for Biotechnology Information database (NCBI) [37] . For each segment, only protein coding regions were considered. Furthermore, we only used sequences with full date (year, month and day) and location information to build hierarchies. Geographical coordinates of each isolate were obtained using geolocation information from Google Maps. Sequences were then aligned using the ClustalW v. 1.83 multiple sequence alignment package using default parameters for H3N2 and H1N1, respectively. For each segment, sequences were aligned and those that were poorly aligned compared to the rest of the dataset were removed until all sequences aligned with a Hamming distance no greater than 0.15. Given estimated mutation rates of 6.7610 23 nucleotide substitutions per site per year [12, 19] , Hamming distances over the 20-year span of our dataset are expected to be no more than 0.15 of the sequence length. Outlying sequences were most likely incorrectly sequenced and were discarded from analysis. Our methodology aimed to minimize data bias from geospatial and temporal variability in sequences from NCBI. First, we determined the most parsimonious evolutionary paths traversed by the flu virus. To this end, we sorted sequences from earliest to most recent viral isolates. Working backwards from newest to oldest, we calculated the sequence similarity of each virus to all earlier isolates regardless of geography. We defined a virus's most likely ancestor to be the sequence with minimum Hamming distance. From this data we built evolutionary paths for each virus. Related sequences were clustered (grouped) together by common geography and season to simplify the paths. For example, a chain of related viruses in the same region and season would be collapsed into a single umbrella node representing all of them. Our analysis was then based on looking at the transitions between clusters rather than individual viruses. We counted these ''seeding events,'' where the closest ancestor of a given cluster of sequences is from a different region or season [8] (Figure 1 ). When tallying seeding events, non-unique solutions were not considered where a given viral isolate possessed multiple closest ancestors from different geographical zones or seasons (Text S1, Figure S6 ). The observed frequencies of seeding events between clusters were compared to expected frequencies based on the prior probability of randomly choosing a sequence from a given geographical zone in the past. Using the binomial distribution with the proportion of prior NCBI sequences as a binomial probability, a p-value was calculated for observing more seeding events than expected. The best predictor of a seeding region for each season had the greatest ratio of observed to expected seeding events with a p-value smaller than 0.05 ( Figure 2 ). Figure S4 Rankings of significant seeding and seeded climate zones for H3N2 and H1N1 using different graph theory metrics. (A) The indegree and outdegree of a node represent the total number of seeding events into and out of a region, respectively. Local seeding events depicted in gray play little role in overall seeding except in the tropics. (B) Degree flow measures the difference between seeding events out of and into a node and determines whether it is a terminal sink or source. (C) PageRank uses an algorithm similar to that employed by Google to categorize nodes based on the number and quality of links pointing to that node. Found at: doi:10.1371/journal.pcbi.1001005.s004 (0.88 MB EPS) Figure S5 Rankings of significant seeding and seeded continents for H3N2 using different graph theory metrics. (A) The indegree and outdegree of a node represent the total number of seeding events into and out of a region, respectively. Local seeding events depicted in gray play little role in overall seeding except in Asia. (B) Degree flow measures the difference between seeding events out of and into a node and determines whether it is a terminal sink or source. (C) PageRank uses an algorithm similar to that employed by Google to categorize nodes based on the number and quality of links pointing to that node. (D) Betweenness measures the number of shortest paths in a network passing through a given node. Text S1 Detailed description of the methodology, including evaluation of clustering, determining flu seasons, timing of observed seeding events, and network randomization. Detailed description of the methodology, including evaluation of clustering, determining flu seasons, timing of observed seeding events, and network randomization. Found at: doi:10.1371/journal.pcbi.1001005.s009 (0.02 MB DOCX)

Willingness to accept H1N1 pandemic influenza vaccine: A cross-sectional study of Hong Kong community nurses

BACKGROUND: The 2009 pandemic of influenza A (H1N1) infection has alerted many governments to make preparedness plan to control the spread of influenza A (H1N1) infection. Vaccination for influenza is one of the most important primary preventative measures to reduce the disease burden. Our study aims to assess the willingness of nurses who work for the community nursing service (CNS) in Hong Kong on their acceptance of influenza A (H1N1) influenza vaccination. METHODS: 401 questionnaires were posted from June 24, 2009 to June 30, 2009 to community nurses with 67% response rate. Results of the 267 respondents on their willingness to accept influenza A (H1N1) vaccine were analyzed. RESULTS: Twenty-seven percent of respondents were willing to accept influenza vaccination if vaccines were available. Having been vaccinated for seasonable influenza in the previous 12 months were significantly independently associated with their willingness to accept influenza A (H1N1) vaccination (OR = 4.03; 95% CI: 2.03-7.98). CONCLUSIONS: Similar to previous findings conducted in hospital healthcare workers and nurses, we confirmed that the willingness of community nurses to accept influenza A (H1N1) vaccination is low. Future studies that evaluate interventions to address nurses' specific concerns or interventions that aim to raise the awareness among nurses on the importance of influenza A (H1N1) vaccination to protect vulnerable patient populations is needed.

The 2009 pandemic of influenza A (H1N1) infection has alerted many governments to make preparedness plan to control the spread of influenza A (H1N1) infection. With evidence on the effectiveness of vaccination in the control and prevention of seasonal influenza [1, 2] , vaccination for pandemic influenza is one of the most important primary preventative measures to reduce the disease burden associated with influenza A (H1N1) infection [3] . Several high risk groups have been identified as "the priority group" to receive the influenza A (H1N1) vaccination and among these, healthcare workers have been identified "as a first priority" to be vaccinated against influenza A (H1N1) by the World Health Organization [4, 5] . Although it is considered essential for all healthcare workers to be immunized against influenza A (H1N1) to prevent the spread of influenza A (H1N1) to patients as the pandemic evolves, previous studies that have examined the acceptability of seasonal influenza vaccination among healthcare workers have generally demonstrated a low acceptance rate of vaccination in this group [6, 7] . Among all healthcare workers, nurses constitute the largest group with the highest frequency of contacts with patients and staff [8] . Previous findings of the acceptability of seasonal influenza vaccination in nurses showed that their acceptance of vaccination was lowest among all healthcare workers [6, 7, 9, 10] . Acceptability of influenza A (H1N1) vaccination in healthcare workers has been shown to be low [11] [12] [13] . A survey conducted in Greece found that only 17% of hospital healthcare were willing to receive influenza A (H1N1) vaccination [11] . Of all healthcare workers, nurses were found to have the lowest rate of acceptability of influenza A (H1N1) vaccination [12, 13] . A study of Italian healthcare workers showed 31% of nurses willing to accept vaccination compared to 67% of physicians [12] . In a study conducted of Hong Kong healthcare workers in hospitals, it was found that only 25% of nurses were willing to accept influenza A (H1N1) vaccination, compared with 47% of doctors and 29% of allied professionals [13] . General practitioners working in the community in France also report a high rate of acceptability of influenza A (H1N1) vaccination at 62% [14] . It is therefore not surprising that a recent online poll conducted in the UK suggested that nurses may be unwilling to receive pandemic influenza vaccination [15] . In a cross-sectional survey that was conducted on experienced nurses who were members of the nursing professional organizations in Hong Kong, the vaccination rate for seasonal influenza vaccination was about 50% [16] . In a more recent survey that explored influenza A (H1N1) acceptance rate in the same group of nurses [17] , it was found that only 13% were willing to accept vaccination for influenza A (H1N1) compared to 38% who plan to receive the seasonal influenza vaccination. However, in the study, there was a low response rate of 28% of nurses with different clinical settings. There is a lack of studies in Hong Kong looking at influenza A (H1N1) vaccination acceptability particularly in the community setting. Nurses who work in the community may be the first group to be in contact with patients who are affected with the influenza A(H1N1) infection. A recent study [18] showed differences in the concerns in using new vaccines during a pandemic than using established vaccine in a non-crisis situation. Therefore, we undertook the current study to examine the willingness of frontline registered nurses who work in the community in Hong Kong to receive vaccination against influenza A (H1N1) at the time of a pandemic. All participants in this study were specially trained nurses, who provided nursing care and treatment for patients in their own homes (also known as Community Nursing Service) in Hong Kong. The responsibility of these community nurses is to provide nursing care and health education to patients through home visits. CNS nurses are employed by Hospital Authority in Hong Kong and provide continuity of care for patients who have been discharged from hospitals such that patients can recover in their own homes. Community nurses were chosen because of their frequent contacts with patients in their homes which is likely to increase their risk for exposure to influenza. We have only included CNS nurses who provide medical services in the study. The rest of the CNS nurses (around 100 nurses) provide psychiatric services in the community. Currently, there are a total of 401 nurses who provide medical related services for the Community Nursing Service (CNS) centres that are distributed among the 7 geographical clusters in Hong Kong (in Hong Kong, public hospital and primary care services are organized in 7 clusters that covers all of Hong Kong). In this study, twelve major CNS centres were contacted first and all CNS nurses were invited to participate in the current study through these 12 major centres. All 12 centres responded to this study and 270 questionnaires were returned with 267 completed questionnaires [19] . The response rate for this study was 67% and all questionnaires were received within a 2week period at a time when there was widespread H1N1 in the community. The survey was sent out from June 24 th to June 30 th , 2009 when the WHO influenza pandemic alert level assigned to H1N1 was phase 6. Phase 6 signifies a widespread human infection, indicating that the virus has caused sustained community level outbreaks in at least one other country in another WHO region (WHO pandemic phase description). The pandemic in Hong Kong started on 1 st May, when a Mexican traveller was confirmed with influenza A (H1N1). Till the end of our data collection, there were 1389 confirmed cases and no death were reported. All general managers of the involved community nursing centres were contacted through telephone to obtain approval to send questionnaires to their nursing staff. In total, 401 self administered, anonymous questionnaires were posted to general managers of centres who then passed these questionnaires to the community nurses in their centres. The general managers of centres were then reminded via telephone during the period from 2 nd July and 8 th July one week after the questionnaires were sent out and advised to return the completed questionnaires within the week. Once completed, questionnaires were collected and returned by their supervisors, except for one of the (Sau Mau Ping) sub-offices, where nurses mailed back their questionnaires individually. All centres sent their questionnaires back after one telephone reminder. The last pile of completed questionnaires was received on 14 th July, 2009. The questionnaire consisted of six parts with 44 questions and the full questionnaire can be accessed by contacting the authors. The first four parts were based on a conceptual framework developed by Patel et al [20] to guide systematic planning for community primary care service response to pandemic influenza with modifications to make it more relevant for nurses. We added a fifth part on psychological responses to pandemic influenza and a sixth part on demographics of respondents which were based on two studies previously published (one on general practitioners' response to SARS and one on general public response to swine flu) [21, 22] . In summary, these sections were 1) clinical services change as a response to pandemic influenza; 2) internal environment changes as a response to pandemic influenza e.g. wearing of mask; 3) macro-environmental changes as a response to pandemic influenza e.g. use of guideline etc; 4) professional and public health responsibilities with respect to pandemic influenza; 5) attitude and psychological responses to pandemic influenza; and 6) demographics and year of education of respondents. The willingness to accept influenza A(H1N1) vaccination was asked in the professional and public health responsibility sections and the question "will you receive the new influenza A (H1N1) vaccine when it is available" was asked with a dichotomous "yes" or "no" response. For those who answered no, they were further asked to give their reasons for refusing to receive the vaccine. Only results on willingness of accept influenza A (H1N1) vaccination and information related to the analysis on willingness to accept vaccine are reported in this paper. Other results from this survey will be presented in a separate report. Descriptive results were cross-tabulated. χ 2 test was used to examine characteristics between nurses who were willing to accept influenza A (H1N1) vaccination against those who were not willing to accept vaccine. Univariate analysis was performed with demographic information (age, post year education and working district), personal protective behaviour (hand washing practice), experience of taking care of SARS patients, and influenza vaccination in the previous 12 months as independent variables. Dependent variables were the willingness to receive pandemic influenza vaccination. Multiple logistic regression analysis was conducted to examine the relationship between pre-defined factors that we think might be associated with the acceptance of the influenza A (H1N1) vaccine when constructing the questionnaire and the dependent variable. The level of statistical significance was set at a p-value of ≤ 0.05. Among the respondents ( Table 1) , most of them were females who had worked an average of 8.8 years as a community nurse (ranging from 2 months to 32 years) and having been a registered nurse for 16.5 years (ranging from 1 year to 36 years). The mean age of respondents was 39.1 years and about a third (30%) had had the experience of dealing with SARS. One third of them had received vaccination for seasonal influenza in the past 12 months. Nurses from each geographical cluster in Hong Kong participated, with 11% of respondents working in Hong Kong Island, 47% working in Kowloon and 42% working in the New Territories (Hong Kong is geographically divided into Hong Kong Island, Kowloon peninsula and the New Territories). Overall, 194 (73%) participants do not want to receive new influenza A (H1N1) vaccine when it is available. The reasons for their not intending to receive vaccination when it is available are summarised in Table 2 . The characteristics of respondents who were willing to accept influenza A (H1N1) vaccination and with those who were not willing to accept influenza A (H1N1) influenza vaccination were compared by χ2 test and were presented Table 1 . Nurses who were willing to receive influenza A (H1N1) vaccine were different from nurses who were not willing to receive influenza A (H1N1) vaccine with respect to "being vaccinated against seasonal influenza vaccination in the previous 12 months". There were no statistical significant differences in other characteristics as analyzed by chi-square test. The relationship between demographic and other characteristics of the nursing respondents and their willingness to accept vaccination were analyzed further using forced entry logistic regressions (Table 3) . Having seasonal vaccination in the past 12 months was significantly independently associated with the willingness to accept influenza A (H1N1) vaccination (OR = 4.03; 95% IC: 2.03-7.98). Washing hands before and between patient contact, however, was negatively independently associated with willingness to accept influenza A (H1N1) vaccination (OR = 0.49; 95% IC: 0.23-1.06). To confirm the results, we have also conducted backward logistic regression and the results also indicated that having seasonal vaccination in the past 12 months was significantly associated with the willingness to accept influenza A (H1N1) vaccination (OR = 3.56, 95% CI: 1.87-6.80, p < 0.001). Consistent with findings from previous surveys conducted in hospital healthcare workers and nurses [13, 17] , we have shown that the majority of nurses from community nursing services in Hong Kong were not willing to be vaccinated against H1N1 influenza when the vaccine becomes available. Similar to findings from previous studies in healthcare workers [13, 17, 23, 24] , we showed that the major concerns for vaccination against pandemic influenza was fear of side effects and concern of efficacy of the new vaccine (Table 2) . Moreover, influenza vaccination in the previous 12 months was significantly associated with their willingness to accept the pandemic influenza vaccination. We also showed that in addition to previous vaccination with seasonal influenza, preventive behaviours such as frequent hand washing practice were independently associated with nurses' willingness to accept influenza A (H1N1) vaccination. We showed that "have been washing hands between and before patient contact" was negatively associated with willingness to accept vaccination independently although the reason for this is unclear and be a result of our relatively small sample. We can only postulate that the barrier to pandemic influenza vaccination is probably not related to the willingness of nurses to protect themselves against infections or their personal hygiene in general. Researchers [17] have suggested one of the barriers to pandemic influenza vaccination in nurses was misconceptions about the purpose of vaccinations in which nurse might think that the aim of vaccination was for self protection rather than to protect at risk populations in contact with them [17, 23, 25] . Specific vaccination policy for health care workers may improve vaccination in this group as nurses have different concerns and priorities when compared to the general public's concerns [17, 25] . Although some may suggest that more educational programs for healthcare workers may be a solution to the low vaccination uptake [13] , studies have reported low influenza vaccination rates among healthcare workers even when educational programs were implemented [10] . Other studies including randomized controlled trials also failed to show that better knowledge or educational programmes (27) Other concerns (i.e. pregnancy, poor health status, and the severity of the epidemic of H1N1) 10 (<0.1) Note: The total percentage exceeds 100% because multiple responses were allowed. were effective in increasing acceptability of vaccination in healthcare workers [26] . Indeed, some suggested that educational campaigns based on the Health Belief Model were unlikely to be enough to change healthcare workers' acceptability of vaccination as evidence showed that perceived seriousness of infection, acknowledgement of increased risk of infection and knowledge of vaccine being safe were unrelated to vaccine uptake in healthcare workers [26] . Others suggested that educational programmes may be counter-productive as many of these healthcare workers do not perceive themselves to be at risk for contracting the infection. Recently, Ofstead et al [25] suggested that an ecological model, which included engaging organizations, communities and policy makers to create environments that were more conducive to risk reduction, might be more effective in increasing vaccination rates in healthcare workers. To our knowledge, this is the first study to explore the willingness of nurses who work in the community to be vaccinated for pandemic influenza and our results confirmed that their acceptability of influenza A (H1N1) vaccination is low. A strength of our study is our response rate of 67% which is higher than similar report conducted in Hong Kong with a response rate of 28% [18] . A limitation of our study is that we have only documented nurses' intentions of when a vaccine is available and not the actual uptake of vaccination. Furthermore, all data from this study were from self-reports and recall bias, such recalling influenza vaccination in the previous year, might have occurred. A possible contributory factor e.g. recent episode of influenza-like illness which may influence the willingness of vaccination was not enquired. Our analysis of results was limited by the relatively small sample size in nurses who are part of the Community Nursing Service in Hong Kong with no information available on non respondents. However, our results are similar to recent studies conducted in hospital healthcare workers [13] and members of professional nursing organizations [17] in Hong Kong. Consistent with previous findings which were conducted in healthcare workers and nurses [13, 17] , we confirm that the acceptance rate of pandemic influenza vaccination is low amongst community nurses. Since community nurses are at high risk of contracting influenza infection, and play a significant role in caring for community cases, special attention should be paid to this group as successful vaccination strategy has been shown to be beneficial in disease transmission [27] . Future work, including interventional studies evaluating potential interventions based on the ecological model or interventions that aim to increase awareness among nurses on the importance of vaccination in healthcare workers to protect vulnerable populations [16] is needed. The need to address low influenza vaccination rates in this high-risk group is urgent in the context of pandemic response.

On epidemic modeling in real time: An application to the 2009 Novel A (H1N1) influenza outbreak in Canada

BACKGROUND: Management of emerging infectious diseases such as the 2009 influenza pandemic A (H1N1) poses great challenges for real-time mathematical modeling of disease transmission due to limited information on disease natural history and epidemiology, stochastic variation in the course of epidemics, and changing case definitions and surveillance practices. FINDINGS: The Richards model and its variants are used to fit the cumulative epidemic curve for laboratory-confirmed pandemic H1N1 (pH1N1) infections in Canada, made available by the Public Health Agency of Canada (PHAC). The model is used to obtain estimates for turning points in the initial outbreak, the basic reproductive number (R(0)), and for expected final outbreak size in the absence of interventions. Confirmed case data were used to construct a best-fit 2-phase model with three turning points. R(0 )was estimated to be 1.30 (95% CI 1.12-1.47) for the first phase (April 1 to May 4) and 1.35 (95% CI 1.16-1.54) for the second phase (May 4 to June 19). Hospitalization data were also used to fit a 1-phase model with R(0 )= 1.35 (1.20-1.49) and a single turning point of June 11. CONCLUSIONS: Application of the Richards model to Canadian pH1N1 data shows that detection of turning points is affected by the quality of data available at the time of data usage. Using a Richards model, robust estimates of R(0 )were obtained approximately one month after the initial outbreak in the case of 2009 A (H1N1) in Canada.

Epidemics and outbreaks caused by emerging infectious diseases continue to challenge medical and public health authorities. Outbreak and epidemic control requires swift action, but real-time identification and characterization of epidemics remains difficult [1] . Methods are needed to inform real-time decision making through rapid characterization of disease epidemiology, prediction of shortterm disease trends, and evaluation of the projected impacts of different intervention measures. Real-time mathematical modeling and epidemiological analysis are important tools for such endeavors, but the limited public availability of information on outbreak epidemiology (particularly when the outbreak creates a crisis environment), and on the characteristics of any novel pathogen, present obstacles to the creation of reliable and credible models during a public health emergency. One needs to look no further than the 2003 SARS outbreak, or ongoing concerns related to highly pathogenic avian influenza (H5N1) or bioterrorism to be reminded of the need for and difficulty of real-time modeling. The emergence of a novel pandemic strain of influenza A (H1N1) (pH1N1) in spring 2009 highlighted these difficulties. Early models of 2009 pH1N1 transmission were subject to substantial uncertainties regarding all aspects of this outbreak, resulting in uncertainty in judging the pandemic potential of the virus and the implementation of reactive public health responses in individual countries (Fraser et al. [2] ). Multiple introductions of a novel virus into the community early in the outbreak could further distort disease epidemiology by creating fluctuations in incidence that are misattributed to the behavior of a single chain of transmission. We sought to address three critical issues in real time disease modeling for newly emerged 2009 pH1N1: (i) to estimate the basic reproduction number; (ii) to identify the main turning points in the epidemic curve that distinguish different phases or waves of disease; and (iii) to predict the future course of events, including the final size of the outbreak in the absence of intervention. We make use of a simple mathematical model, namely the Richards model, to illustrate the usefulness of near realtime modeling in extracting valuable information regarding the outbreak directly from publicly available epidemic curves. We also provide caveats regarding inherent limitations to modeling with incomplete epidemiological data. The accuracy of any modeling is highly dependent on the epidemiological characteristics of the outbreak considered, and most epidemic curves exhibit multiple turning points (peaks and valleys) during the early stage of an outbreak. While these may be due to stochastic ("random") variations in disease spread, and changes in either surveillance methods or case definitions, turning points may also represent time points where epidemics transition from exponential growth processes to processes that have declining rates of growth, and thus may identify effects of disease control programs, peaks of seasonal waves of infection, or natural slowing of growth due to infection of a critical fraction of susceptible individuals. For every epidemic, there is a suitable time point after which a given phase of an outbreak can be suitably modeled, and beyond which subsequent phases may be anticipated. Detection of such "turning points" and identification of different phases or waves of an outbreak is of critical importance in designing and evaluating different intervention strategies. Richards [3] proposed the following model to study the growth of biological populations, where C(t) is the cumulative number of cases reported at time t (in weeks): Here the prime "′" denotes the rate of change with respect to time. The model parameter K is the maximum case number (or final outbreak size) over a single phase of outbreak, r is the per capita growth rate of the infected population, and a is the exponent of deviation. The solution of the Richards model can be explicitly given in terms of model parameters as Using the Richard model, we are able to directly fit empirical data from a cumulative epidemic curve to obtain estimates of epidemiological meaningful parameters, including the growth rate r. In such a model formulation, the basic reproduction number R 0 is given by the formula R 0 = exp(rT) where T is the disease generation time defined as the average time interval from infection of an individual to infection of his or her contacts. It has been shown mathematically [4] that, given the growth rate r, the equation R 0 = exp(rT) provides the upper bound of the basic reproduction number regardless of the distribution of the generation interval used, assuming there is little pre-existing immunity to the pathogen under consideration. Additional technical details regarding the Richards model can be found in [5] [6] [7] . Unlike the better-known deterministic compartmental models used to describe disease transmission dynamics, the Richards model considers only the cumulative infected population size. This population size is assumed to have saturation in growth as the outbreak progresses, and this saturation can be caused by immunity, by implementation of control measures or other factors such as environmental or social changes (e.g., children departing from schools for summer holiday). The basic premise of the Richards model is that the incidence curve of a single phase of a given epidemic consists of a single peak of high incidence, resulting in an S-shaped cumulative epidemic curve with a single turning point for the outbreak. The turning point or inflection point, defined as the time when the rate of case accumulation changes from increasing to decreasing (or vice versa) can be easily pinpointed as the point where the rate of change transitions from positive to negative; i.e., the moment at which the trajectory begins to decline. This time point has obvious epidemiologic importance, indicating either the beginning of a new epidemic phase or the peak of the current epidemic phase. For epidemics with two or more phases, a variation of the S-shaped Richards model has been proposed [6] . This multi-staged Richards model distinguishes between two types of turning points: the initial S curve which signifies the first turning point that ends initial exponential growth; and a second type of turning point in the epidemic curve where the growth rate of the number of cumulative cases begins to increase again, signifying the beginning of the next epidemic phase. This variant of Richards model provides a systematic method of determining whether an outbreak is single-or multi-phase in nature, and can be used to distinguish true turning points from peaks and valleys resulting from random variability in case counts. More details on application of the multi-staged Richards model to SARS can be found in [6, 7] . Readers are also referred to [8, 9] for its applications to dengue. We fit both the single-and multi-phase Richards models to Canadian cumulative 2009 pH1N1 cumulative case data, using publicly available disease onset dates obtained from the Public Health Agency of Canada (PHAC) website [10, 11] . PHAC data represent a central repository for influenza case reports provided by each of Canada's provinces and territories. Onset dates represent best local estimates, and may be obtained differently in different jurisdictions. For example, the province of Ontario, which comprises approximately 1/3 of the population of Canada, and where most spring influenza activity was concentrated, replaces onset dates using a hierarchical schema, whereby missing onset dates may be replaced with dates of specimen collection (if known) or date of specimen receipt by the provincial laboratory system, if both dates of onset and specimen collection are missing. Data were accessed at different time points during the course of the "spring wave (or herald wave)" of the epidemic in May-July of 2009, whenever a new dataset is made available online by the PHAC. By sequentially considering successive S-shaped segments of the epidemic curve, we estimate the maximum case number (K) and locate turning points, thus generating estimates for cumulative case numbers during each phase of the outbreak. The PHAC cumulative case data is then fitted to the cumulative case function C(t) in the Richards model with the initial time t 0 = 0 being the date when the first laboratory confirmed case was reported and the initial case number C 0 = C(0) = 1, (the case number with onset of symptoms on that day). There were some differences between sequential epidemic curves in assigned case dates. For example, data posted by PHAC on May 20 indicated an initial case date of April 13, but in the June 3 data this had been changed to April 12, perhaps due to revision of the case date as a result of additional information. Model parameter estimates based on the explicit solution given earlier can be obtained easily and efficiently using any standard software with a least-squares approximation tool, such as SAS or Matlab. Daily incidence data by onset date were posted by PHAC until June 26, after which date only the daily number of laboratory-confirmed hospitalized cases in Canada was posted. For the purpose of comparison, we also fit the hospitalization data to the Richards model in order to evaluate temporal changes in the number of severe (hospitalized) cases, which are assumed to be approximately proportional to the total cases number. The case and hospitalization data used in this work are provided online as Additional file 1. We fit the model to the daily datasets, acquired in real time, throughout the period under study. The leastsquared approximation of the model parameter estimation could converge for either the single-phased or the 2-phase Richards models. For the sake of brevity, only four of these model fits are presented in Table 1 to demonstrate the difference in modeling results over time. The resulting parameter estimates with 95% confidence intervals (CI) (for turning point (t i ), growth rate (r), and maximum case number (K)), time period included in the model, and time period when the data set in question were accessed, is presented in Table 1 . Note that all dates in the tables are given by month/day. We also note that the CI's for R 0 reflect the uncertainty in T as well as in the estimates for r, and does not reflect the error due to the model itself, which is always difficult to measure. In order to compare the 1-phase and 2-phase models, we also calculate the Akaike information criterion (AIC) [12] for the first, third, and fourth sets of data in Table 1 , where there is a model fit for the 2-phase model. The results, given in Table 2 , indicates that whenever there is a model fit for the 2-phase model, its AIC value is always lower than that of the 1-phase model and hence compares favorably to the 1-phase model. Parameter estimates fluctuate in early datasets, and the least-squared parameter estimations diverge within and between 1-phase and 2-phase models in a manner that seems likely to reflect artifact. In particular, for the earliest model fits, using data from April 13 to May 15, the estimated reproductive number for the second phase is far larger than that obtained in the first phase, and that obtained using a single-phase model, and illustrating the pitfalls of model estimation using the limited data available early in an epidemic. Estimates stabilize as the outbreak progresses, as can be seen with the final data sets (April 11 to June 5 and April 12 to June 19). For comparison, we plot the respective theoretical epidemic curves based on the Richards model with the estimated parameters described in the table above in Figure 1 . As noted above, model can be used to estimate turning points (t i ) and basic reproductive numbers (R 0 .), if the generation time T is know. We used T = 1.91 days (95% CI: 1.30-2.71), as obtained in [2] by fitting an age stratified mathematical model to the first recognized 2009 influenza A (H1N1) outbreak in La Gloria, Mexico. Estimates are presented in Table 1 . We also conducted sensitivity analyses with R 0 # calculated based on longer generation times (T = 3.6 (2.9, 4.3)) for seasonal influenza in [13] (see last column in Table 1 ). Excluding implausibly high estimates of R 0 generated using initial outbreak data (April 13 to May 15), we obtain the estimates of R 0 for the 2-phase model that range between 1.31 and 1.96. Inasmuch as Richards model analyzes the general trends of an epidemic (e.g., turning point, reproductive number, etc.), it can be used to fit any epidemiological time series for a given disease process, as long as the rate of change in the recorded outcome is proportional to changes in the true number of cases. As such, for comparison, we fit our model using the time series for 2009 pH1N1 hospitalizations in Canada posted by PHAC on July 15 [11] (that last date these data were made available) ( Table 3) . This time series was easily fit to a one-phase model ( Figure 2) . Further examples of using hospitalization or mortality data to fit the Richards model can be found in [14] . We used the Richards model, which permits estimation of key epidemiological parameters based on cumulative case counts, to study the initial wave of 2009 influenza A (H1N1) cases in Canada. In most model fits, April 28-29 and May 4-7 were identified as early turning points for the outbreak, with a third and final turning point around June 3-5 in models based on longer time series. Although this modeling approach was not able to detect turning points using some earlier data sets (e.g., those limited to the period from April 12 to May 27), in general the turning points identified were consistent across multiple models and time series. Perhaps the most important divergence between models occurred with the detection of an April 29 turning point in the case report time series, but not in the time series based on hospitalized cases. We believe this may be attributable to the small number of hospitalizations, relative to cases, that had occurred by that date, as well as the fact that hospitalization data only became available on April 18. The turning point can correspond to the point at which disease control activities take effect (such that the rate of change in epidemic growth begins to decline) or can represent the point at which an epidemic begins to wane naturally (for example, due to seasonal shifts or due to the epidemic having "exhausted" the supply of susceptibles such that the reproductive number of the epidemic declines below 1). This quantity has direct policy relevance; for example, in the autumn 2009 pH1N1 wave in Canada, vaccination for pH1N1 was initiated at or after the turning point of the autumn wave due to the time taken to produce vaccine; as the epidemic was in natural decline at that point, the impact of vaccination has subsequently been called into question. Although the Richards model is able to capture the temporal changes in epidemic dynamics over the course of an outbreak, it does not define their biological or epidemiological basis. As such, determining the nature of these turning points requires knowledge of "events on the ground" for correlation. We suspect that the last Note that all dates in the tables are given by month/day. Dates of posting are listed in parentheses. Model duration indicates whether they fit a 1-phase or 2phase model. Note that the maximum case number is rounded off to the nearest integer. R 0 # is obtained using the generation interval of T = 3.6 (2.9, 4.3) for seasonal influenza [13] . Table 2 Comparison of Akaike information criterion (AIC) values between 1-phase and 2-phase models for time periods with 2-phase model fit in Table 1 Time Table 1 , last line) and a 1-phase model using hospitalization data (June 11), this lag in turning points would actually be expected, due to the time from initial onset of symptoms until hospitalization, which was reported to have an interquartile range of 2-7 days in a recent study from Canada [15] . Timelines for the 2-phase model for case data of 4/12-6/19 and the 1-phase model for hospitalization data are presented graphically in Figure 3 . In addition to identifying turning points, the Richards model is useful for estimation of the basic reproductive number (R 0 ) for an epidemic process, and our estimates derived using a Richards model were consistent with estimates derived using other methods. For example, our R 0 agrees almost perfectly with that of Tuite et al., derived using a Markov chain Monte Carlo simulation parameterized with individual-level data from Ontario's public health surveillance system [16] . Our estimates of R 0 is smaller than that derived by Fraser et al. [2] using Mexican data, but such differences could relate in part to the different age distributions of these two countries [17] , and may also reflect the fact that our estimate is obtained Canadian data at a national level, while empirical Mexican estimates were based on data from the town of La Gloria with only 1575 residents. Most epidemic curves in the early stage of a novel disease outbreak have multiple phases or waves due to simple stochastic ("random") variation, mechanisms of disease importing, initial transmission networks and individual/community behavior changes, improvements in the performance of surveillance systems, or changes in case definitions as the outbreak response evolves. However, changes in phase (signified by the presence of turning points identified using the Richards model) may also pinpoint the timing of important changes in disease dynamics, such as effective control of the epidemic via vaccination or other control measures, depletion of disease-susceptible individuals (such that the effective reproductive number for the disease decreases to < 1), or the peak of a "seasonal" wave of infection, as occurs with [4, 18, 19] , some competing methods require more extensive and detailed data than are required to build a Richards model, which requires only cumulative case data from an epidemic curve. As we also demonstrate here, the Richards model produces fairly stable and credible estimates of reproductive numbers early in the outbreak, allowing these estimates to inform evolving disease Table 1 , derived using early case data accessed on May 20, closely approximate our final estimates (Table 1, last row) . Thus, while early estimation with the Richards model failed to correctly detect turning points or accurately estimate the final outbreak size, it was nonetheless useful for rapid estimation of R 0 within a month of first case occurrence in Canada. As with any mathematical modeling technique, the approach presented here is subject to limitations, which include data quality associated with real-time modeling (as data are often subject to ongoing cleaning, correction, and reclassification of onset dates as further data become available), reporting delays, and problems related to missing data (which may be non-random). In our current study, the hierarchical approach used by Canada's most populous province (Ontario) for replacement of missing data could have had distorting effects on measured disease epidemiology: the replacement of missing onset dates with dates of specimen collection could have resulted in the artifactual appearance of early turning points identified by our model, due to limitations in weekend staffing early in the outbreak. If, as we believe to be the case, public health laboratories did not have sufficient emergency staffing to keep up with testing on weekends such that weekend specimen log-ins declined sharply, this would have created the appearance of epidemic "fade out" on weekends. Other factors that might distort the apparent epidemiology of disease include changes in guidelines for laboratory testing of suspected cases, improved surveillance and public health alerts at later stages of the outbreak leading to increased case ascertainment or over-reporting of cases [20] . However, the quality of the time series will tend to improve with the duration of the epidemic, both because stochastic variation is "smoothed out", and also because small variations become less important as the cumulative series becomes longer. We note that a further application of the Richards model in the context of influenza would relate to comparison of the epidemiology of the 2009 influenza A H1N1 epidemic to past Canadian epidemics, though such an endeavor is beyond the scope of the present study. In summary, we believe that the Richards model provides an important tool for rapid epidemic modeling in the face of a public health crisis. However, predictions based on the Richards model (and all other mathematical models) should be interpreted with caution early in an epidemic, when one need to balance urgency with sound modeling. At their worst, hasty predictions are not only unhelpful, but can mislead public health officials, adversely influence public sentiments and responses, undermine the perceived credibility of future (more accurate) models, and become a hindrance to intervention and control efforts in general. Additional file 1: Electronic Supplementary Material. 2009 Canada novel Influenza A(H1N1) daily laboratory-confirmed pandemic H1N1 case and hospitalization data.

Antiviral and Neuroprotective Role of Octaguanidinium Dendrimer-Conjugated Morpholino Oligomers in Japanese Encephalitis

BACKGROUND: Japanese encephalitis (JE), caused by a mosquito-borne flavivirus, is endemic to the entire south-east Asian and adjoining regions. Currently no therapeutic interventions are available for JE, thereby making it one of the most dreaded encephalitides in the world. An effective way to counter the virus would be to inhibit viral replication by using anti-sense molecules directed against the viral genome. Octaguanidinium dendrimer-conjugated Morpholino (or Vivo-Morpholino) are uncharged anti-sense oligomers that can enter cells of living organisms by endocytosis and subsequently escape from endosomes into the cytosol/nuclear compartment of cells. We hypothesize that Vivo-Morpholinos generated against specific regions of 3′ or 5′ untranslated regions of JEV genome, when administered in an experimental model of JE, will have significant antiviral and neuroprotective effect. METHODOLOGY/PRINCIPAL FINDINGS: Mice were infected with JEV (GP78 strain) followed by intraperitoneal administration of Morpholinos (5 mg/kg body weight) daily for up to five treatments. Survivability of the animals was monitored for 15 days (or until death) following which they were sacrificed and their brains were processed either for immunohistochemical staining or protein extraction. Plaque assay and immunoblot analysis performed from brain homogenates showed reduced viral load and viral protein expression, resulting in greater survival of infected animals. Neuroprotective effect was observed by thionin staining of brain sections. Cytokine bead array showed reduction in the levels of proinflammatory cytokines in brain following Morpholino treatment, which were elevated after infection. This corresponded to reduced microglial activation in brain. Oxidative stress was reduced and certain stress-related signaling molecules were found to be positively modulated following Morpholino treatment. In vitro studies also showed that there was decrease in infective viral particle production following Morpholino treatment. CONCLUSIONS/SIGNIFICANCE: Administration of Vivo-Morpholino effectively resulted in increased survival of animals and neuroprotection in a murine model of JE. Hence, these oligomers represent a potential antiviral agent that merits further evaluation.

The genus Flavivirus is composed of more than 70 different closely related species [1] . Many flaviviruses are arthropod-borne and causes significant human diseases. Among these, the four serotypes of dengue virus (DENV), yellow fever virus (YFV), West Nile virus (WNV) and Japanese encephalitis virus (JEV) are categorized as emerging global pathogens [2] . JEV is a mosquitoborne, positive sense, single stranded RNA virus, responsible for frequent epidemics of encephalitis, predominantly in children, in most parts of Southeast Asia and adjoining regions. It is the causal factor for 30,000-50,000 cases of encephalitis occurring every year and accounts for about 10,000 deaths annually with serious neurological squeal in the survivors [3] . JEV has been expanding its 'geographical footprint' into previously non-endemic regions and with several billion people at risk, Japanese encephalitis (JE) represents an internationally emerging concern in tropical and sub-tropical countries. Currently three types of JE vaccine are in use-the inactivated mouse-brain derived, the inactivated cellculture derived and the live attenuated cell-culture derived. However, there are limitations for their usage in terms of availability, cost and safety [4] . At present, chemotherapy against JEV is largely supportive and not targeted towards the virus. A lot of avenues has been explored in the past and are also being currently tried, so as to develop a safe and effective molecule that would be able to prevent the virus from replicating within the host. The JEV genome is approximately 11 kb in length that carries a single long open reading frame (ORF) flanked by a 95-neucleotide 59 untranslated region (59 UTR) and a 585-neucleotide 39 UTR. The ORF encodes a polyprotein which is processed by viral and cellular proteases into three structural and seven non structural proteins [5, 6] . The 59 and 39 UTRs of the JEV genome contain conserved sequence elements and can form conserved stem loop structure. 59 UTR contain secondary structures which are required for the formation of translation pre-initiation complex [7] . JEV requires long range RNA-RNA interaction between 59 and 39 regions of its genome for efficient replication; one such interaction occurs between a pair of 10 complementary nucleotides, located in coding sequence for the capsid protein at 136-146 nucleotides from 59 end of the genome, and 39 cyclization sequence, commonly denoted as 39CSI (39 conserved sequence I) located at 104-114 nucleotides from 39 end of the genome [8, 9] . The 39CSI is highly conserved across members of JEV serocomplex, indicating the possibility that RNA elements within the 59 and 39 UTRs in JEV genome are essential for its replication. Anti-sense oligonucleotides have been shown to be effectively used as therapeutic agents against viral infection. In one such study siRNA generated against the cd loop-coding sequence in domain II of the viral Envelope protein (which is highly conserved among all flaviviruses because of its essential role in membrane fusion) has been found to protect against lethal encephalitis [10] . Similarly siRNAs has also been generated against various nonstructural proteins of JEV and were found to be effective in inhibiting viral replication [11, 12] . Anti-sense approach has also been employed to inhibit flaviviral replication by generating anti-sense molecules against RNA elements within the 59 and 39 UTRs in flaviviral genome. In one such approach, DNAzyme against 39 UTR of JEV genome has be found to be effective in controlling virus infection in a murine model [13] . Under the same approach but with different kind of anti-sense oligonuleuotide called Morpholino, flaviviral replication has been inhibited in cultured cells as well as in animal models [14, 15] . Morpholino oligomers are single stranded DNA analogues containing same nitrogenous bases as DNA but joined by backbone consisting of morpholine rings and phosphorodiamidate linkages [16] . For efficient delivery into cells these Morpholino are often conjugated with arginine rich peptide [17] . However, in the current study we have used a different type of Morpholino oligomer called Vivo-Morpholino against 39CSI and one of the secondary structures present in 59 UTR of the JEV genome. Vivo-Morpholino are specialized type of non-peptide Morpholino oligomers, conjugated with a new transport structure that provides effective delivery into a wide variety of tissues in living animals, thereby raising the possibilities of their use as therapeutic agents. The transporter comprises of a dendritic structure assembled around a triazine core which serves to position eight guanidinium head groups in a conformation effective to penetrate cell membranes. Vivo-Morpholinos have also been shown to effectively enter and function within cultured cells [18] . Vivo-Morpholinos are also cost effective, non immunogenic, and stable under physiological conditions as compared to other types of Morpholinos. This study was designed to evaluate whether the use of Vivo-Morpholinos as therapeutic agents, is possible in an experimental model of JE. We intend to show that these specifically designed Vivo-Morpholinos are effective in countering the viral load in the body, thereby imparting significant protection to the animals that were infected with a lethal dose of JEV. All animal experiments were approved by the institutional animal ethical review board named ''Institutional Animal and Ethics Committee of National Brain Research Centre''. The animal experiment protocol approval no. is NBRC/IAEC/2007/ 36. Animals were handled in strict accordance with good animal practice as defined by Committee for the Purpose of Control and Supervision of Experiments on Animals (CPCSEA), Ministry of Environment and Forestry, Government of India. Vero cells (a kind gift from Dr. Guruprasad Medigeshi, Translational Health Science and Technology Institute, Gurgaon, India) and Neuro2A (obtained from National Centre for Cell Science, Pune, India) cells were grown in DMEM (Dulbecco's modified Eagles medium, supplemented with 10% fetal bovine serum (FBS) and antibiotics. The GP78 strain of JEV was propagated in suckling BALB/c mice and their brains were harvested when symptoms of sickness were observed. A 10% tissue suspension was made in MEM (minimum essential medium), followed by centrifugation at 10,0006g and finally filtered through a 0.22 m sterile filter [19] . JEV was titrated by plaque formation on Vero cell monolayer. Vero cells were seeded in six-well plates to form semi-confluent monolayer in about 18 h. Cell monolayer were inoculated with 10-fold serial dilutions of virus samples made in MEM containing 1% FBS and incubated for 1 h at 37uC with occasional shaking. The inoculum was removed by aspiration and the monolayers were overlaid with MEM containing 4% FBS, 1% low-melting-point agarose and a cocktail of antibiotic-antimycotic solution (Gibco, Invitrogen Corporation, Grassland, NY, USA) containing penicillin, streptomycin, and amphotericin B. Plates were incubated at 37uC for 72-96 h until plaques became visible. To allow counting of the plaques, the cell monolayer was stained with crystal violet after fixing the cells with 10% formaldehyde. All Vivo-Morpholino (MO) oligos were commercially procured from Gene Tools LLC, (Philomath, OR, USA). MOs were Japanese encephalitis (JE) is caused by a flavivirus that is transmitted to humans by mosquitoes belonging to the Culex sp. The threat of JE looms over a vast geographical realm, encompassing approximately 10 billion people. The disease is feared because currently there are no specific antiviral drugs available. There have been reports where other investigators have shown that agents that block viral replication can be used as effective therapeutic countermeasures. Vivo-Morpholinos (MOs) are synthetically produced analogs of DNA or RNA that can be modified to bind with specific targeted regions in a genome. In this study the authors propose that in an animal model of JE, MOs specifically designed to bind with specific region of JE virus (JEV) genome, blocks virus production in cells of living organisms. This results in reduced mortality of infected animals. As the major target of JEV is the nerve cells, analysis of brain of experimental animals, post treatment with MOs, showed neuroprotection. Studies in cultured cells were also supportive of the antiviral role of the MOs. The potent anti-sense effect in animals and lack of obvious toxicity at the effective dosage make these MOs good research reagents with future therapeutic applications in JE. Vivo-Morpholino in Japanese Encephalitis www.plosntds.org designed to be complementary to sequences in the JEV (GP78 strain) genome, as shown in Table 1 . These oligonucleotides targeted specific regions in the 39 and 59 UTRs of the JEV genomic RNA (Figure 1) . A 21 base scrambled MO of random sequence (SC-MO) was used as a negative control in all the experiments. All MO sequences were screened with BLAST (http://www.ncbi.nlm.nih.gov/BLAST) against primate and murine mRNA sequences and the SC-MO was additionally screened against all flaviviral sequences. All MOs were procured in 300 nanomole quantities as a liquid of 0.5 mM stock (approximately 4 mg/mL) in buffered saline. They were diluted with sterile 16 PBS to achieve desired concentrations, and stored at 4uC as aliquots. Five to six weeks old BALB/c mice of either sex were randomly distributed into 5 groups-Sham, JEV-infected, JEV-infected and treated with scrambled Morpholino (JEV+SC-MO), JEV-infected and treated with Morpholino against viral 39 conserved region (JEV+39 MO) and JEV-infected and treated with Morpholino against secondary structure in the 59UTR of viral RNA (JEV+59 MO). Initially each group contained 8 animals. Animals belonging to all groups except Sham were infected with 3610 5 plaque forming units (PFU) of JEV (GP78 strain) and that day was considered as day zero [20] . Animals of Sham group received equal volume of filtered MEM. Starting from 3 h post infection on day zero, 100 mg of SC-MO, 39 MO and 59 MO, diluted in 0.1 mL of sterile 16PBS (corresponding to 5 mg/kg body weight), were administered to animals belonging to JEV+SC-MO, JEV+39 MO and JEV+59 MO groups respectively, once per day, for 5 consecutive days. Animals belonging to the Sham-treatment group received equal volumes of sterile 16 PBS only. Survivality of animals in each group following JEV infection and Morpholino treatment were monitored daily upto 15 days post JEV infection (or till their death, whichever was earlier). Toxicity of the Morpholinos in mice was evaluated by weight loss and abnormal behavioral & clinical observations (including tremors, ruffled fur, hunching, ataxia, gait abnormalities), in a masked manner to minimize bias [14, 20] . Mouse cytokine bead array (CBA) kits were used to quantitatively measure cytokine levels in mouse whole-brain lysates. 50 mL of bead mix containing a population of beads with distinct fluorescence intensities that have been coated with capture antibodies for different cytokines, and 50 mL of whole-brain lysates were incubated together, along with equal volume of phycoerythrin (PE)-conjugated detection antibodies, for 2 h at room temperature, in dark. The beads were then washed and resuspended in 300 mL of supplied 16 wash buffer. The beads were acquired using Cell Quest Pro Software in FACS Calibur and analyzed using BD CBA software (Becton Dickinson, San Diego, CA). Standard curve was prepared by incubating 50 mL of supplied mouse inflammation standards with 50 mL of bead mix and PE-conjugated detection antibodies [21] . Protein concentrations of whole brain lysates were estimated by Bradford method. Sample volumes containing 20 mg of protein were electrophoresed on polyacrylamide gel and transferred onto nitrocellulose membrane. After blocking with 7% skimmed milk, the blots were incubated overnight at 4uC with primary antibodies against JEV E-glycoprotein (Abcam, USA), and JEV NS5 (a kind gift from Dr. Chun-Jung Chen, Taichung Veterans General Hospital, Taichung, Taiwan), iNOS (Upstate-Chemicon, USA), HSP-70, SOD-1 (Santa Cruz Biotechnology, CA, USA), TRX (AB Frontiers, Korea; a kind gift from Dr. Ellora Sen, NBRC), phospho NFkB, phospho ERK1/2, total ERK1/2 and phos-phoP38 MAP kinase (Cell Signaling, USA) at 1:1000 dilutions. After extensive washes with PBS-Tween, blots were incubated with appropriate secondary antibodies conjugated with peroxidase (Vector Laboratories, CA, USA). The blots were again washed with PBS-Tween and processed for development using chemiluminescence reagent (Millipore, USA). The images were captured and analyzed using Chemigenius, Bioimaging System (Syngene, Cambridge, UK). The blots were stripped and reprobed with antib-tubulin (Santa Cruz Biotechnology, USA) to determine equivalent loading of samples [22] . For immunohistochemical staining, brains from scarified animals were excised following repeated transcardial perfusion with ice-cold saline and fixed with 4% paraformaldehyde. Twenty micron thick cryosections were made with the help of Leica CM3050S cryostat and processed for immunohistochemical staining to detect presence of JEV antigen in the brain and to label activated microglia. Sections were incubated overnight at 4uC with mouse anti-JEV antigen (Nakayama, 1:250) (Chemicon, CA, USA) and rabbit anti-Iba-1 (1: 500; Wako, Osaka, Japan), respectively. After washes, slides were incubated with FITCconjugated anti-mouse or anti-rabbit secondary antibodies (Vector laboratories Inc. Burlingame, USA) and following final washes, sections were sections were cover slipped after mounting with 49-6diamidino-2-phenylindole (DAPI, Vector laboratories Inc.). The slides were observed under Zeiss Axioplan 2 fluorescence microscope and Zeiss Apotome microscope (Zeiss, Gottingen, Germany), respectively [21] . Cryosections of brain from Sham-treated, JEV-infected and JEV-infected and MO treated animals were rinsed in de-ionized water followed by incubation with the thionin dye. The excess dye The level of ROS produced within brain tissue of each treatment groups were measured by the cell permeable, nonpolar, H 2 O 2 -sensitive probe 5(and 6)-chlromethyl-20,70-dichlorodihydrofluoresceindiacetate (CM-H2DCFDA; Sigma, USA). CM-H2DCFDA diffuses into cells, where its acetate groups are cleaved by intracellular esterases, releasing the corresponding dichlorodihydrofluorescein derivative. Subsequent oxidations of CM-H2DCFDA yields a fluorescent adduct dichlorofluorescein that is trapped inside the cell. Brain homogenates were treated with 5 mM solution of CM-H2DCFDA followed by incubation in dark at room temperature for 45 min and then the relative fluorescence intensity were measured with the help of Varioskan Flash multimode reader (Thermo Electron, Finland) at excitation 500 nm and emission 530 nm. The fluorescence intensity of intracellular CM-H2DCFDA is a linear indicator of the amount of H 2 O 2 in the cells. The measured mean fluorescence intensity was then normalized to equal concentrations of protein in each sample [23] . Nitric oxide released from brain homogenates following MO treatment was assessed using Griess reagent as described previously. Briefly, 100 mL of Griess reagent (Sigma, St. Louis, USA) was added to 100 mL of brain homogenate and incubated in dark for 15 min. The intensity of the color developed was estimated at 540 nm with the help of a Benchmark plus 96-well ELISA plate reader (Biorad, CA, USA). The amount of nitrite accumulated was calculated (in mM) from a standard curve constructed with different concentrations of sodium nitrite [21] . Mouse neuroblastoma cells (N2a) were plated in five 60 mm plates at a density of 5610 5 cells/plate, and were cultured for 18 h. After 6 h in serum free DMEM, cells were either mock- Vivo-Morpholino in Japanese Encephalitis www.plosntds.org infected with sterile 16PBS or infected with JEV at multiplicity of infection (MOI) of 5. After 1K h, cells were washed twice with sterile 16PBS to remove non-internalized virus. Three of the four plates that were infected with JEV, were treated with SC-MO, 39 MO and 59 MO at 10 mM concentrations and all plates were incubated for 24 h in serum free media. After two washes with 16 PBS, cells were first fixed with BD cytofix solution (BD Biosciences) for 15 min and permeabilized by resuspending in permeabilization buffer (BD Cytoperm plus; BD Biosciences) and incubated at 25uC for at least 10 min. Cells were then washed twice in wash buffer (PBS containing 1% bovine serum albumin) then resuspended in wash buffer at 1610 6 cells per 100 mL. Primary antibody (JEV Nakayama strain; Chemicon, USA) were added in 1:100 dilutions and incubated for 30 min at 25uC. The cells were washed with wash buffer and pelleted by centrifugation followed by incubation with FITC conjugated secondary antibody for 30 min. After final wash with wash buffer, cells were resuspended in 400 mL FACS buffer and analyzed on a FACS Calibur. The percentage of population of JEV-positive cells was calculated after gating the populations on a Dot plot using Cell Quest Pro Software (BD Biosciences). Statistical analysis was performed using SIGMASTAT software (SPSS Inc., Chicago, IL, USA). Data were compared between groups using one-way analysis of variance followed by post hoc test. Differences upto p,0.05 were considered significant. MOs confer protection to animal from Japanese encephalitis MO treatment conferred significant protection to mice following JEV infection. The survival of mice following JEV infection was dramatically increased with treatments of both 39 and 59 MO. Approximately 90% of all the animals that were treated with 39 MO survived as compared to 75% survival of those animals that were treated with 59MO, post infection with JEV ( Figure 2A ). Infection with JEV was accompanied with distinct symptoms and weight loss whereas treatments with both 39 and 59 MO post JEV infection, prevented animals from suffering. Not much considerable changes in the average body weights of JEVinfected animals treated with both 39 and 59 MO were observed when compared to animals belonging to JEV and JEV+ SC-MO groups showing significant reductions in their body weights 6 days post infection ( Figure 2B ). The symptoms associated with JE in murine model were observed on daily basis and scores were attributed accordingly. The animals that had most symptoms received the highest scores. It was observed that 39 and 59 MO treated animals scored lesser than those belonging to the JEVinfected or JEV+SC-MO groups ( Figure 2C ). To assess whether the MOs has any effect on reduction of viral load in brain, homogenized brain samples from all the treatment groups were subjected to plaque assay as described in materials and methods section. Number of PFU/mL of the brain homogenate was found to be significantly higher in both JEV and JEV+SC-MO groups when compared to Sham (p,0.001). Viral PFUs were found to be significantly reduced following 39 MO and 59 MO treatment when compared to only JEV-infected or JEV+SC-MO group (p,0.001) ( Figure 3A) . To further validate the results obtained from the plaque assay, immunoblot for some of the JEV-specific proteins were performed. The expression of NS5, a non structural protein of JEV, was significantly increased in JEV and JEV+SC-MO groups when compared to Sham (p,0.01), but its level were found to be significantly reduced after both 39 and 59 MO treatments when compared to JEV-infected group (p,0.01). Similarly, E glycoprotein level showed significant increase in JEV-infected and JEV+SC-MO groups when compared to Sham (p,0.01) which were then drastically reduced following 39 and 59 MO treatments (p,0.01) (Figure 3B-D) . Immunostaining of brain sections showed greater presence of JEV antigen in JEV-infected and JEV+SC-MO groups, whereas 39 and 59 MO treatments resulted in lesser presence ( Figure 3E ). To further characterize the inhibitory effects of MO on JEVinduced neuronal death, brain sections from all the treatment groups were subjected to thionin staining. Numerous healthy cells were seen in sections obtained from Sham, JEV+39 MO and JEV+59 MO groups when compared to sections belonging to only JEV-infected or JEV+SC-MO groups which contained numerous unhealthy/dying neurons with altered morphology ( Figure 4A ). Microglial activation and increased proinflammatory cytokine production are the hallmarks of JEV infection [24] . To see whether MO treatment helps in vitiation of these effects, immunostaining for microglial specific marker Iba-1 was performed in brain sections of all treatment groups. In brain sections of JEV and JEV+SC-MO groups the number of activated microglia with characteristic morphology, appeared to be more frequent when compared to sections belonging to Sham, JEV+39 MO and JEV+59 MO groups ( Figure 4B ). CBA performed to check the proinflammatory cytokines levels in the brain homogenates obtained from different treatments showed that levels of MCP-1, IFN-c, TNF-a, and IL-6 were found to be significantly increased in both JEV and JEV+SC-MO groups when compared to Sham infected groups (p,0.01). The elevated levels of these proinflammatory cytokines were drastically reduced with 39 and 59 MO treatments (p,0.01) ( Figure 4C -F). Increased oxidative stress in CNS is a major outcome of JEV infection [20] . To evaluate whether MO treatment of mice resulted in abrogation of oxidative stress following JEV infection, we measured ROS and NO levels in brain homogenate obtained from all treatment groups. Two fold increases were observed in the ROS levels in the brain samples of JEV and JEV+SC-MO groups when compared to Sham (p,0.01), significant reduction in the ROS levels were observed in JEV+39 MO and JEV+59 MO groups when compared to only JEV-infected groups (p,0.01). Although ROS levels has decreased in JEV+39 MO group when compared to JEV group, it remained significantly higher than that of Sham (p,0.01) ( Figure 5A ). Superoxide dismutase 1 (SOD-1) and Thioredoxin (TRX-1) are the proteins associated with oxidative stress. SOD-1 levels were found to be elevated approximately 2-and 3-fold in JEV-infected and JEV+SC-MO groups respectively when compared to Sham (p,0.01). Its levels in JEV+39 MO and JEV+59 MO groups were reduced significantly when compared to JEV-infected group (p,0.01). TRX-1 levels were also found to be increased significantly in JEV-infected and JEV+SC-MO groups when compared to Sham (p,0.01) but it were significantly reduced in brain samples obtained from JEV+39 MO and JEV+59 MO groups when compared to only JEVinfected groups (p,0.01) (Figure 5B, D&E) . HSP-70 is a heat shock Vivo-Morpholino in Japanese Encephalitis www.plosntds.org protein that has been associated with intracellular stress. Significant twelve fold increases in the levels of HSP-70 were observed in JEV and JEV+SC-MO groups when compared to Sham (p,0.01), this drastic increases in the levels of HSP-70 in JEV and JEV+SC-MO groups were reduced in JEV+39 MO and JEV+59 MO groups (p,0.01) ( Figure 5B&C ). JEV infection leads to increased nitric oxide (NO) production in CNS [25] . Significant two fold increases were seen in the NO levels in brain samples obtained from JEV-infected and JEV+SC-MO groups when compared to those obtained from Sham (p,0.01). NO levels subsequently got down to significantly lower levels following 39 and 59MO treatments (p,0.01) ( Figure 5F ). Immunoblot analysis showed nearly 8-fold increases in levels of iNOS in JEV-infected and JEV+SC-MO groups when compared to Sham (p,0.01). iNOS levels showed significant decreases in JEV+39 MO and JEV+59 MO groups when compared to only JEV-infected groups (p,0.01) ( Figure 5G&H ). Western blot analysis demonstrated a significant inhibition in the expression of different stress related proteins whose levels were elevated following JEV infection. Upon MO treatments there were approximately 4-fold increases in the levels of pNFkB in JEV and Figure 2 . Mice are protected from JEV following MO treatment. The survival of mice following JEV infection was dramatically increased with treatments of both 39 and 59 MO, though the survival in 39 MO treated mice was greater (,90%) than those treated with 59MO (75%) (A). Considerable changes in the average body weights of JEV-infected animals treated with both 39 and 59 MO were not observed when compared to animals belonging to JEV and JEV+ SC-MO groups that showed significant reductions in their body weights from 6 th day post infection till their death. Black arrows points to the days by which all the animals died. (B). Infection with JEV was accompanied with distinct symptoms that were alleviated following treatments with both 39 and 59 MO. Animals were assigned scores according to the symptoms, in a blinded manner. The graph was plotted by taking the scores of one animal that was considered as the representative of that group (C). n = 8 for all experiments; data shown are representative of duplicate sets of experiments. doi:10.1371/journal.pntd.0000892.g002 Vivo-Morpholino in Japanese Encephalitis www.plosntds.org JEV+SC-MO groups when compared to Sham (p,0.01). The levels of pNFkB were found to be significantly reduced in JEV+39 MO and JEV+59 MO groups when compared to only JEV-infected groups (p,0.01) ( Figure 6A&B ). Phospho p38 MAPK levels also showed significant 3-fold increases in JEV and JEV+SC-MO groups when compared to Sham (p,0.01), its levels were also found to be reduced significantly following treatment with 39 and 59 MO when compared to only JEV-infected groups (p,0.01) ( Figure 6A&C ). Both phospho ERK1 and ERK2 levels were found to be significantly increased in JEV and JEV+SC-MO groups when compared to Sham (p,0.01). The levels of phospho ERK1 and ERK2 showed considerable decreases in JEV+39 MO and JEV+59 MO groups when compared to only JEV-infected groups (p,0.01) ( Figure 6A&D ). To assess whether MO has any effect on viral load in vitro N2a cell lysates from all the treatment groups were subjected to plaque assay. PFU/mL of the cell lysates was found to be significantly higher in both JEV and JEV+SC-MO groups when compared to mock-infected cells (p,0.01). Viral loads were found to significantly reduced in both JEV+39 MO and JEV+59 MO groups when compared to only JEV-infected group (p,0.01) ( Figure S1A ). To further ascertain the results obtained from plaque assay, intracellular staining of JEV antigen in N2a was performed and number of JEV-positive N2a cells was then counted by flow cytometry. Only 16% and 9% of the total gated cells were found to be positive for JEV antigen in JEV+39 MO and JEV+59 MO groups respectively as compared to 30% in JEV-infected group, and 34% in JEV+SC-MO group ( Figure S1B ). Use of anti-sense molecules for targeted inhibition of viral replication has been under investigation for quite sometime. Though the application of these molecules has raised the possibilities of their future use as novel therapeutic agents, there Vivo-Morpholino in Japanese Encephalitis www.plosntds.org are many issues regarding their effectiveness in terms of their stability and delivery to targeted cells. Recent studies are involved in developing techniques to minimize or eliminate these issues so that anti-sense therapy can be employed to a wide variety of intractable diseases such as splice-modifying genetic defects and viral diseases. The role of various anti-sense molecules in the inhibition of replication of JEV has been reported with positive outcomes [10, 11, 12, 13, 26] . Morpholino oligomers are single stranded anti-sense molecules that exert their action by steric blocking of complementary RNA. Unlike other types of anti-sense oligonucleotides, Morpholinos provide all the desired properties of stability, nuclease resistance, high efficacy, long-term activity, water solubility, low toxicity, and exquisite specificity. Morpholino oligomers has been used previously for the inhibition of flaviviral replication [14, 27] including JEV [15] though all of them has utilized peptide based Morpholinos. The peptide based Morpholinos contain delivery moiety evolved from natural peptides whose active components are 6-9 arginine residues in a bio-available 6-aminohexanoicspaced structure [28] . However, these arginine-based peptides are not commercially available for research purposes and their greatest efficacies have been in delivering Morpholinos to the cytosol of tissues like liver [29] or leaky muscle [30] , which would be considered as easily deliverable. As a result, the reach of peptide based Morpholinos into a wide spectrum of tissues remains questionable [18] . Also, owing to the peptidic nature, degradation of the peptide portion of the conjugates was found to be time and tissue dependent [31] . Furthermore, the applications of the arginine-rich peptide transporters are limited due to their high cost, scalability and stability. Added to that are the risks of immune responses against the peptides which limits repeated administrations for diseases requiring long-term treatment [32] . Increases were observed in the ROS levels in the brain samples of JEVinfected and JEV+SC-MO groups in comparison to Sham, that were reduced following 39 and 59 MO treatments. Although ROS levels were decreased in JEV+39 MO group when compared to JEV-infected group, it remained significantly higher than that in Sham (A). Approximately 13-fold increases were observed in the levels of HSP-70 in JEV-infected and JEV+SC-MO groups as compared to Sham. These drastic increases were significantly reduced in JEV+39 MO and JEV+59 MO groups, the levels remained significantly higher than Sham (B&C). SOD-1 levels were found to be elevated 2and nearly 3-fold in JEV-infected and JEV+SC-MO groups respectively compared to Sham. 39 and 59 MO treatment caused significant reduction of SOD-1 levels compared to JEV-infected group (B&D). Alterations in TRX-1 levels were similar to that observed in SOD-1 except that its level in JEV+ SC-MO was not significantly different than only JEV-infected group (B&E). Nearly 2-fold increases were observed in NO levels of JEV-infected and JEV+SC-MO groups when compared to those obtained from Sham. NO levels were subsequently reduced to significantly lower levels following 39 and 59MO treatments (F). iNOS expression was found to increase 8-fold in JEV-infected and JEV+SC-MO groups when compared to Sham. Following 39 and 59MO treatments iNOS levels decreased significantly as compared to JEV-infected group (G&H). ( * p,0.01 for JEV when compared to Sham; ** p,0.01 for JEV+SC-MO when compared to JEV-infected only; # p,0.01 for JEV+ 39MO To minimize the problems encountered by the peptideconjugated Morpholinos, octaguanidinium dendrimer-conjugated Morpholino oligomers have been developed that are commonly referred to as Vivo-Morpholino (MO). These custom-sequence anti-sense molecules have been reported to enable Morpholino applications in adult animals. MO was our choice of anti-sense molecule as this enabled us to test the specifically designed oligonucleotides in both animal as well as cell culture models. Though 'outstanding' results have been reported to be achieved by intravenous (i.v.) administration of the MO, we preferred the intraperitoneal route via which modest systemic delivery can be achieved. This was so done because brain has been reported to be an ineffective tissue when MOs are administered i.v. [33] , though there is no direct evidence showing that MOs can cross blood brain barrier, when administered via other routes. According to the manufacturer's (Gene Tools LLC) instructions the maximum suggested dosage in mammals is 12.5 mg/kg in a 24 hour period. Our aim was to determine the minimum dose at which our desired effects could be achieved. Initially we had chosen two doses of 5 mg and 10 mg per kg body weight (b.w.) of the animals. We found that there was no significant difference between the survival rate of JEV-infected and MO treated mice in either dose (data not shown for 10 mg/kg b.w.). The survival rate was approximately 90% in those JEV-infected animals that were treated with 39 MO and 75% in animals treated with 59 MO. Thus we decided to proceed with the 5 mg/kg b.w. dose for all subsequent experiments. Plaque assay from the brain homogenates of animals of all groups revealed that the number of infective viral particle production was dramatically reduced following 39 and 59 MO treatment. The 39 MO was generated against the 39 CSI region of the JEV genome that interacts with 59 CS region located in coding sequence for capsid protein at 136-146 nucleotides from 59 terminal of the genome. This interaction results in cyclization of JEV genome that is necessary for its efficient replication. The 59 MO was targeted towards one of the secondary structures of the 59 UTR that are required for the formation of translation preinitiation complex. Blocking of these two sites in the JEV genome leads to the most likely effect, i.e. inhibition of replication and translation of viral genome. This was further corroborated by the decrease in the expressions of viral proteins (NS5, E glycoprotein Vivo-Morpholino in Japanese Encephalitis www.plosntds.org and general flaviviral envelop protein) in the brain. Flaviviral NS5 is known to possess guanylyltransferase activity that helps in the synthesis of methylated cap structure at the 59 end of the viral genome that plays a crucial role in the translation and stability of mRNAs [34] . The JEV E glycoprotein is believed to be involved in viral adhesion and entry into host cells, hemagglutination, cellular tropism, viral virulence, and the induction of protective immune responses [35] . Decreased expression of these proteins indicates that viral replication and production of new infective viral particles are inhibited due to the MOs. Immunohistochemical staining for viral antigen also provided visual confirmation of the fact that JEV antigen was detected at much lower amounts in the brain following MO treatment. However, these data does not prove that MOs directly inhibit infective viral particle production in the brain itself, as it cannot be conclusively stated whether the MOs can reach brain. These data merely suggests that the number of replication-competent infective JEV in the brain was significantly reduced, which subsequently leads to neuroprotection. It is well known that JEV infection causes microglial activation. Activated microglia releases an array of chemical mediators that are detrimental for the neurons in brain [24] . Since there was reduction in the production of infective viral particles following 39 and 59 MO treatments, we studied the effect on microglial pathophysiology in mouse brain. Our results show that there was significantly reduced number of activated microglia in the brain sections of both 39 and 59 MO-treated animals as compared to only JEV-infected or JEV-infected and SC-MO treated animals. Since there were little or no activation of microglia, proinflammatory cytokine levels in the brain were found to be significantly downregulated. Histochemical staining also revealed that neuronal population and morphology remained largely unaffected in 39 and 59 MO-treated animals' brains as compared to only JEV-infected or JEV-infected and SC-MO treated animals. Generation of ROS with the generation of oxidative damage has been implicated in neurodegenerative diseases and in the degradation of nervous system functions and are also reported to increase following JEV infection [22] . Increase in ROS levels initiates various responses within the cell, including damage to proteins, DNA and lipid [36] . In this study, ROS levels were found to be many-fold increased in JEV-infected or JEV-infected and SC-MO treated animals that were then found to be counteracted by the treatment of 39 and 59 MO. The levels of stress related proteins such as SOD-1, HSP-70 and TRX where also found to be positively modulated following 39 and 59 MO treatment. NO is a known antagonist of JEV. It has been shown that NO inhibits JEV infection by preventing viral replication [37] . In our study NO levels were increased in the brain in response to JEV infection possibly due to the upregulation of inducible nitric oxide synthase (iNOS). Treatments with 39 and 59 MO caused a decrease of NO to basal levels as observed in Sham-treated animals. Activation of pNFkB regulates apoptotic genes, especially the TRAF1 and TRAF2, and thereby checks the activities of the caspases, which are central to most apoptotic processes. JEV is known to activate pNFkb via a PI3K-dependent pathway in the brain of infected animals, which is associated with apoptosis [38] . JEV infection has also been shown to activate stress kinases, which in turn results in activation of ERK1/2, and p38 MAPK pathway leading to apoptotic death of neurons [39] . In accordance with the established results, here also we found that there was similar activation pattern of these molecules in JEV-infected and JEVinfected and SC-MO treated animal brain samples. Treatment with the MOs resulted in abrogation of those changes that led to greater survivality of brain neurons as observed by histochemical staining. Activation of p38MAPK is also related to the transcriptional activation of proinflammatory genes in the brain [40] . Thus the decrease in phophoP38 MAPK levels correlates with the decreased levels of proinflammatory cytokine levels in obtained from the brain. To confirm the anti-viral and neuroprotective property of the MOs observed in in vivo models, cultured neuroblastoma cells were infected with JEV, followed by MO treatment. Though the MOs are specifically developed for in vivo studies, they are also known to be taken up by cells in culture conditions [18] . There was a significant decrease in viral titer in samples obtained from the cells that were treated with 39 and 59 MOs as compared to either JEVinfected or JEV-infected and SC-MO treated cells, as revealed by plaque assay. This data was supported by FACS analysis following intracellular staining for JEV antigen. This study was undertaken to determine the antiviral and neuroprotective efficacy of Vivo-Morpholinos in an experimental model of JE so that it can be considered as a therapeutic agent in the near future. There have been studies regarding the anti-JEV effects of other types of Morpholino oligomers though none of them are yet to be considered for therapeutic purposes. This is the first study that investigates the role of Morpholino oligomers specially designed for effective delivery into live animal models. Generally, the i.p route of administration of any drug is preferred in animal studies over any other routes. However, the efficacy of these antisense molecules needs to be checked by administering through other applicable routes, as i.p. administration in humans is uncommon, though not unheard of. The amounts of oligomers required and the route of administration in this study marks these molecules as practicable therapeutic agents in JE, though further studies are required before these can be recommended for clinical trials. Figure S1 MO treatments decrease viral load in vitro. Mouse neuroblastoma cell (N2a) lysates from all the treatment groups were subjected to plaque assay in order to determine viral loads. PFU/mL was found to be significantly higher in both JEV and JEV+SC-MO groups when compared to Sham. Viral loads were found to be significantly reduced in both JEV+39 MO and JEV+59 MO groups when compared to only JEV-infected group (* p , 0.01 for JEV and JEV+SC-MO when compared to Sham; # p , 0.01 for JEV+39 MO and JEV+59MO when compared to only JEV-infected group) (A). Intracellular staining for JEV antigen in N2a was performed and number of JEV-positive N2a cells was then sorted by flow cytometry. 30% of the total gated cells were found to be positive for JEV antigen in JEV-infected group as compared to 34% in JEV+SC-MO group. Only 16% and 9% of the total gated cells were found to be positive for JEV antigen in JEV+39 MO and JEV+59 MO groups respectively (B). Found at: doi:10.1371/journal.pntd.0000892.s001 (0.25 MB TIF)

Severe novel influenza A (H1N1) infection in cancer patients

Background: The natural history and consequences of severe H1N1 influenza infection among cancer patients are not yet fully characterized. We describe eight cases of H1N1 infection in cancer patients admitted to the intensive care unit of a referral cancer center. Patients and methods: Clinical data from all patients admitted with acute respiratory failure due to novel viral H1N1 infection were reviewed. Lung tissue was submitted for viral and bacteriological analyses by real-time RT-PCR, and autopsy was conducted on all patients who died. Results: Eight patients were admitted, with ages ranging from 55 to 65 years old. There were five patients with solid organ tumors (62.5%) and three with hematological malignancies (37.5%). Five patients required mechanical ventilation and all died. Four patients had bacterial bronchopneumonia. All deaths occurred due to multiple organ failure. A milder form of lung disease was present in the three cases who survived. Lung tissue analysis was performed in all patients and showed diffuse alveolar damage in most patients. Other lung findings were necrotizing bronchiolitis or extensive hemorrhage. Conclusions: H1N1 viral infection in patients with cancer can cause severe illness, resulting in acute respiratory distress syndrome and death. More data are needed to identify predictors of unfavorable evolution in these patients.

Patients with malignancies are more susceptible for acquisition of infections than the general population and are thought to potentially develop more complications [1] . Due to many disruptions in both innate and acquired immunity, even organisms with low virulence potential are able to cause significant morbidity and mortality in cancer patients [2] . During April 2009, a novel swine-origin influenza A (H1N1) virus (S-OIV) was identified in California and in Mexico as the cause of human respiratory disease, originating a pandemic [3, 4] . As of February 2010, 700 000 laboratory-confirmed cases of novel H1N1 influenza virus and 14 000 deaths have been reported globally. Most affected patients present with influenza-like symptoms and have a benign course [5] . However, patients with comorbidities as cancer and chronic diseases may show a serious clinical presentation, characterized by respiratory failure with variable severity [6, 7] . Although the presence of a malignancy is considered to have a negative impact on the H1N1 disease severity, there are few reports on clinical outcomes in cancer patients affected by the disease. Redelman-Sidi et al. [8] recently described a cohort of 45 patients with solid and hematological malignancies with H1N1 infection. From this population, one single patient required intensive care admission and there were no viral infection-related deaths. On the other hand, fatalities have been described in oncologic patients, particularly in the ones with hematological malignancies [9] . So far, there are no reports describing clinical characteristics of cancer patients with a severe form of the disease. These patients represent a vulnerable population, which require rapid diagnostic work-up and intensive management in specialized units. Therefore, we believe that it would be important to report on the clinical characteristics of cancer patients with a severe form of the H1N1 infection. During the 2009 Southern Hemisphere H1N1 pandemics, Sao Paulo was among the cities with the highest incidence of disease in Brazil. In this study, we describe the clinical and pathological findings in critically ill patients with cancer and respiratory failure related to novel H1N1 infection admitted to original article an oncologic intensive care unit (ICU) in a reference cancer center in Sao Paulo, Brazil. This analysis and report was approved by the Institutional Medical Ethical Committee. Clinical and laboratorial data at admission and during ICU stay are described, including the management of respiratory support and nonventilatory strategies. Data for the Simplified Acute Physiology Score II (SAPS II) and for the Acute Physiology And Chronic Health Evaluation (APACHE II) were reported as the worst value within 24 h after ICU admission. A daily evaluation of organ function according to the Sequential Organ Failure Assessment (SOFA) score was performed Radiological findings were evaluated through X-ray in all patients and computerized tomography when appropriate. The in vivo diagnosis of H1N1 infection was confirmed by real-time RT-PCR (rRT-PCR) test using nasopharyngeal swab specimens, in accordance with guidelines from the Centers for Disease Control and Prevention (CDC) [10] . autopsies Five patients who died had their autopsies performed in the Department of Pathology of the Universidade de Sao Paulo. The pathological findings on part of this population have been previously described by our group [11] . Tissue fragments were formalin fixed, paraffin embedded and hematoxylin-eosin stained. For lung sections, Grocott, Brown-Hopps and Ziehl-Neelsen stainings were performed for the identification of fungi, bacteria and acid-fast bacilli, respectively. Lung fragments were sent for microbiological investigation using rRT-PCR to the Instituto Adolfo Lutz in Sao Paulo. Seasonal influenza A and swine influenza A detection was performed using the CDC protocol [10] . The RT-PCR test used for bacteria identified DNA from Haemophilus influenzae and Streptococcus pneumoniae. results baseline characteristics Table 1 shows the baseline characteristics of the eight patients. Patient ages ranged from 55 to 65 years (median, 58 years). Four patients (50%) were male. All patients (100%) had preexisting medical conditions other than the neoplasm. Most of the patients presented no functional impairment before the infection-seven patients (87.5%) had Karnofsky scale >70. Five patients (62.5%) had solid neoplasms and three patients (37.5%) had hematological malignancies. Two patients had been submitted to chemotherapy in the last 4 weeks (Table 1) . However, most patients were considered as having active disease regarding cancer status (seven patients, 87.5%). Four patients (50%) had metastatic disease. The patient with myelofibrosis had been submitted to stem-cell transplantation 1 year ago. All patients presented cough and fever, most patients had myalgia and dyspnea (87.5%). Hemoptysis, rhinorrhea and wheezing were present in 25% and diarrhea was related by 12.5% of patients. At admission, all patients presented with signs of systemic inflammatory response syndrome, defined as two or more of the signs and symptoms described in Table 2 . All patients had fever or hypothermia and tachypnea. Leukocytosis was present in 75% of cases. At admission, only two patients were hypotensive (25%). However, hypoxemia was present in 100% of cases, and four patients (50%) had oxygen saturation <90% at admission (Table 2) . Initially, lung disease was in most cases localized in one or two quadrants of lung (75%). However, most patients developed a more extensive and progressive In all the eight patients, the diagnosis of acute respiratory distress syndrome (ARDS) could be established based on the presence of bilateral pulmonary infiltrates, a ratio of the partial pressure of arterial oxygen to the fraction of inspired oxygen (PO 2 /FiO 2 ) £200 and no clinical evidence for an elevated left atrial pressure. Due to rapidly progressive hypoxemia (PO 2 /FiO 2 117) and the rapid worsening of lung infiltrates (Figure 1 ), five patients (62.5%) needed invasive mechanical ventilation. Four patients were intubated in the first 24 h of ICU and one patient 48 h after unsuccessful noninvasive ventilation (NIV). NIV was successfully used in three patients (37.5%), who had milder forms of disease as showed by a computerized tomography that revealed sparse bilateral infiltrate ( Figure 2 ). Patients who needed mechanical ventilation were managed with pressure-cycled ventilation, with a low tidal volume (target 6 ml/kg) open-lung strategy of ventilation, and a positive endexpiratory pressure (PEEP) titrated based on FiO 2 for goal plateau pressure (Pplat) < 30 cm H 2 O and SpO 2 88%-90% according to ARDS Network protocol. In some cases, due to refractory hypoxemia, PEEP levels of 14 to 16 cm H 2 O were applied. In all patients, recruitment maneuvers were usedcontinuous positive airway pressure 35-40 cm H 2 O for 30 swith short-term improvements in oxygenation in three of five patients (initial mean oxygen saturation rate increased from 88% to 94%). Neuromuscular blockade was used in two patients ( Table 2) . Despite the aforementioned ventilatory strategy, this particular group of patients developed refractory persistent hypoxemia. Lung mechanics of patients showed very low static and dynamic compliance and high airway resistance. Most patients presented risk scores that predicted high-risk mortality (APACHE II 24 and SAPS II 52). Respiratory failure was the most commonly encountered dysfunction, followed by cardiovascular, renal and hematological failures. Two patients needed dialysis. Most patients needed vasopressor and inotropic drugs. Patients presented with a serious, rapidly Oseltamivir was initiated early in all patients. Corticosteroids were administered in all patients admitted in the ICU, at a dose of methylprednisolone 2 mg/kg/day. Antibiotics were empirically administered in all patients due to clinical suspicion of bacterial coinfection. Despite early treatment, these patients had a high mortality rate, with five deaths (62.5%). The median period from admission to death was 3 days (1-8). Interestingly, three patients who died showed persistent positive rRT-PCR in nasopharyngeal swabs after 5 days of oseltamivir. Hematological evaluation showed leukocytosis (median 16 100/mm 3 ) and anemia (median hemoglobin 9.3 g/dl) in majority of cases. Also, signs of tissue hypoxia as acidosis (median pH 7.2), high levels of lactate (median 4.2 mmol/l) and low levels of base excess (median 4.5 mEq/l) were noted in patients since admission until death. Reactive protein-C ranged from 99 to 447, revealing inflammatory response associated or not with coinfection (Table 3) . During evolution, bacterial coinfection was diagnosed in seven patients: Staphylococcus aureus in blood stream in one patient, urinary tract infection due to Enterococcus faecalis in one patient and pneumonia due to Pseudomonas aeruginosa in one case and due to S. pneumoniae in four other cases. Also, one patient presented urinary tract infection due to Candida. An autopsy was performed in the five patients who died, and a summary of the findings is presented in Table 4 . All patients presented extensive pathological alterations in the lungs; lungs were heavy, diffusely edematous and with variable degrees of hemorrhage. Diffuse alveolar damage (DAD) was present in most of the patients, but as previously described, there were three distinct patterns of pulmonary pathological changes [11, 12] : (i) four patients had classic exudative DAD, with alveolar and interstitial edema, hyaline membranes and reactive pneumocytes; (ii) one patient (esophagus neoplasm) had DAD and severe necrotizing bronchiolitis (NB) characterized by extensive necrosis of the bronchiolar wall and dense neutrophilic infiltrate within the bronchiolar lumen and (iii) one patient (myelofibrosis) presented with exudative DAD with an intense hemorrhagic component (Figure 3 ). Only one of the five patients did not present acute interstitial changes. In this patient with esophageal cancer, death was secondary to pulmonary thromboembolism and bacterial pneumonia. Two patients had pulmonary thromboembolism (two patients with esophagus neoplasm). In four of five patients, bronchopneumonia coinfection due to S. pneumoniae was confirmed at autopsy ( Figure 3) . Metastatic disease was present in all patients with solid neoplasms, and two of those had lung involvement by tumor. All patients had atrophic or nonreactive white pulp in the spleen (Figure 3 ). In the lymph nodes, nonreactive follicles and sinusoidal erythrophagocytosis were found. The liver showed erythrophagocytosis and a few mononuclear inflammatory cells in the sinusoids in all patients and variable degrees of shockrelated centrilobular necrosis. The bone marrow was hypocellular in four of five patients. All patients had mild/ moderate kidney acute tubular necrosis. No patient had histological signs of encephalitis, myocarditis or myositis. We report the clinical and pathological findings from eight patients with cancer and severe H1N1 infection who were admitted to an oncologic ICU during the winter period of the 2009 pandemic in Sao Paulo, Brazil. These patients were characterized by having active malignant disease, comorbidities and difficult to manage rapidly progressive acute respiratory failure. Viral infections may represent up to 26% of the infections identified in cancer patients with pulmonary infiltrates. In a large study examining viral infections in hematological cancer patients, pneumonia was observed in 31% of the cases, and the overall mortality was 15% [12] . In this study, influenza A was the most common isolated agent, and the only independent predictor of fatal outcome was an absolute lymphocyte count £200 cells/ml. Clinical outcomes in cancer patients with H1N1 infection have not been fully characterized yet. Due to potential severity of H1N1 infection in this group of patients, Crawford et al. [13] recommended during the 2009 pandemic that patients with cancer who are receiving chemotherapy and develop fever should be admitted to a hospital and receive oseltamivir after swab collection. In addition, patients with febrile neutropenia should be treated according to the usual protocol with the addition of oseltamivir. Kharfan-Dabaja et al. [14] first reported H1N1 infection in two allogeneic hematopoietic cell transplantation recipients. In one case, the patient presented with fever, myalgia, sore throat and diarrhea without evidence of hypoxemia or lung progressive infiltrates. After treatment with oseltamivir, ciprofloxacin and doxycicline, the symptoms resolved without sequelae. In the second case, pulmonary symptomatology continued to deteriorate despite aggressive polymicrobial treatment, requiring mechanical ventilation and ultimately the patient died from respiratory failure [14] . Redelman-Sidi et al. [8] recently described 45 patients with cancer and/or hematological conditions and H1N1 infection at the Memorial Sloan-Kettering Cancer Center, with no reports of mortality or serious morbidities. In this report, only one patient was admitted in the ICU and did not need mechanical ventilation. In these patients with cancer, mortality was very high. The differences in outcomes may be related to specific characteristics of our population, including high prevalence of metastatic disease, comorbidities and high incidence of bacterial coinfection. Differently from the MSKCC experience [8] , most patients in our series initially presented with signs and symptoms of lung disease-hypoxemia, dyspnea and tachypnea-instead of only influenza symptoms. The serious clinical presentation of the novel Influenza A (H1N1) infection in some cancer patients should be expected. Patients with cancer now live longer and immunosupression from malignant disease or its treatment renders many susceptible to infections [15] . Indeed, most of the patients in this series developed bacterial coinfections. The altered immunological response of these patients may contribute to the original article Annals of Oncology development of more severe forms of disease. In our study, three cases still excreted virus after 5 days in the ICU, as already reported [16] . Prolonged periods of viral shedding could be associated to disease severity [17] and perhaps with an oseltamivir-resistant strain of the virus [18] . Respiratory failure occurred in all patients who required ICU care and was characterized by rapidly progressive bilateral lung infiltrates with refractory hypoxemia and low left atrial pressure-ARDS. Most patients needed mechanical ventilation (five of eight), and they were all treated with protective strategies according to the ARDS Network protocol [19] . In all these patients, recruitment maneuvers were applied, with just transient improvement in oxygenation (initial median oxygen saturation rate increased from 88% to 94%), reflecting the extensive lung involvement of disease. In some cases of severe H1N1 infection, extracorporeal membrane oxygenation [20] has been proposed as an alternative support therapy with promising results [21, 22] . Based on previous cases of H1N1 infection in noncancer patients, Ramsey et al. [6] described a significant proportion of patients with hypoxemic respiratory failure, specially when associated with comorbidities. Early intubation and admission in the ICU is prudent, given the rapid progression of hypoxemia. Based on current evidence, authors recommend patients should be managed with a low tidal volume open-lung strategy of ventilation, with PEEP titrated based on FiO 2 to achieve adequate SpO 2 and low Pplat [21, 22] . This strategy was performed in all studied patients of our series; however, despite that there was no adequate response and all patients who required mechanically ventilated died with refractory hypoxemia and multiple organ failure. Similar to our findings, in a Mexican series of S-OIV cases, of 12 patients who needed mechanical ventilation, 7 died [23] . Three patients presented with a milder form of lung disease, responding well to NIV, and with progressive improvement of hypoxemia. There is some controversy around the clinical Annals of Oncology original article efficacy of NIV in novel influenza A (H1N1) infection. Whereas NIV may be considered as a mode of ventilation for hypoxemic respiratory failure, there are concerns about its usefulness in an infectious epidemic [24] . In the Canadian experience, 30% of patients were noninvasively ventilated on admission, but 85% of these patients required subsequent intubation and invasive ventilation [6] . In addition, NIV machines have no bacterial or viral filters and do generate aerosols. There are often leaks from the mask, and it has to be removed sometimes for nursing care [24] . During the pandemic, NIV was associated with transmission of disease to health care workers [25] . In our study, we detected bacterial coinfection in seven of the eight patients. The CDC reported that 29% of fatal cases in the United States presented at least one bacterial coinfection [26] . Mauad et al. [11] found evidence of bacterial coinfection in 38% of fatal cases in Sao Paulo. The presented incidence of 87% of bacterial coinfection might explain in part the high mortality rate of this group, despite early antimicrobial therapy. Interestingly, biochemical and hematological data could not discriminate H1N1 infection from sepsis of bacterial origin in this group of patients. Corticosteroids were administered to all patients as Meduri et al. [27] recommend in ARDS cases with low PO 2 /FiO 2 . Steroid use has been reported in some cases of H1N1associated ARDS without any adverse outcome [28] . However, some authors discuss the potential adverse effects of steroids in H1N1 infection, including higher mortality possibly related to virus spreading [29] . Although there is no consensus about its efficacy in this disorder, and despite the belief that corticosteroid could reduce pulmonary inflammation and fibrosis in severe cases, our poor results suggest that this strategy will need to be carefully reevaluated in the future. By performing autopsies in the fatal cases in this population we could determine that the cause of death in all patients was extensive involvement of the lungs and alterations secondary to multiple organ failure in major organs such as kidney and liver [11] . Patients had severe DAD associated with severe NB and alveolar hemorrhage. Further, autopsy results showed that patients had metastatic disease and signs of cellular immunosuppression as depletion of white pump view on spleen analysis. Certainly, autopsies contributed to a better characterization of these patients. This report has some limitations. The unicentric characteristics of the study and the small sample size do not allow for definite conclusions about severe H1N1 presentation in oncologic patients. We characterized H1N1 infection in a selected population of patients with neoplasm, with a high incidence of metastatic disease and who needed ICU care. Our findings certainly describe the most serious presentation of disease. The importance of describing this serious form of disease in patients with cancer is to reinforce prevention strategies in this group, as to recommend vaccine, hygienic measures, prophylactic antiviral treatment in cases of contact and adequate isolation in cases of hospital admission due to H1N1 infection [30, 31] . Also, a better understanding of clinical and pathological findings in the group of cancer patients could guide ventilator management and nonventilatory strategies to obtain lower rates of mortality. In summary, our report of cancer patients highlights the severity of the Influenza A (H1N1) pandemic in this vulnerable population and the urgent need to establish specific protocols of care and management strategies designed to face this health care challenge. funding Conselho Nacional de Desenvolvimento Científico e Tecnoló gico (CNPq).

The Tennessee Children's Respiratory Initiative: Objectives, design and recruitment results of a prospective cohort study investigating infant viral respiratory illness and the development of asthma and allergic diseases

Background and objective: The ‘attack rate’ of asthma following viral lower respiratory tract infections (LRTI) is about 3–4 fold higher than that of the general population; however, the majority of children who develop viral LRTI during infancy do not develop asthma, and asthma incidence has been observed to continuously decrease with age. Thus, we do not understand how viral LRTI either predispose or serve as a marker of children to develop asthma. The Tennessee Children's Respiratory Initiative has been established as a longitudinal prospective investigation of infants and their biological mothers. The primary goals are to investigate both the acute and the long‐term health consequences of varying severity and aetiology of clinically significant viral respiratory tract infections on early childhood outcomes. Methods: Over four respiratory viral seasons, 2004–2008, term, non‐low birth weight previously healthy infants and their biological mothers were enrolled during an infant's acute viral respiratory illness. Longitudinal follow up to age 6 years is ongoing. Results: This report describes the study objectives, design and recruitment results of the over 650 families enrolled in this longitudinal investigation. The Tennessee Children's Respiratory Initiative is additionally unique because it is designed in parallel with a large retrospective birth cohort of over 95 000 mother–infant dyads with similar objectives to investigate the role of respiratory viral infection severity and aetiology in the development of asthma. Conclusions: Future reports from this cohort will help to clarify the complex relationship between infant respiratory viral infection severity, aetiology, atopic predisposition and the subsequent development of early childhood asthma and atopic diseases.

The Tennessee Children's Respiratory Initiative (TCRI) has been established as a longitudinal prospective investigation of term, non-low birth weight otherwise healthy infants and their biological mothers. The primary goals of the study are: (i) to investigate both the acute and the long-term health consequences of varying severity and aetiology of clinically significant viral respiratory tract infections on the outcomes of allergic rhinitis (AR) and early childhood asthma; and (ii) to identify the potentially modifiable factors that define children who are at greatest risk of developing asthma following infant respiratory viral infection. This study is unique, in that it was designed in parallel with our Tennessee Asthma Bronchiolitis Study (TABS), which is a retrospective birth cohort study of over 95 000 infants and their biological mothers similarly designed to elucidate the factors predisposing to childhood asthma and allergic This report describes the study objectives, design and recruitment results of the Tennessee Children's Respiratory Initiative, designed to investigate the role of respiratory viral infection severity and etiology in asthma development. Future reports will address the complex relationship between infant respiratory viral infection and asthma and atopic diseases. diseases, but lacking biospecimens. Thus, we designed the prospective TCRI to establish a base for the evaluation of both the risks and benefits of documented significant infant viral respiratory infection of varying severity and aetiology and other environmental exposures on childhood atopy outcomes and to establish a biospecimen repository for analyses including biomarker testing and genotyping. The prospective cohort has the longitudinal design properties that may overcome potential limitations intrinsic to retrospective studies, such as our TABS cohort. [1] [2] [3] [4] [5] It is our eventual goal that the findings from these investigations, in conjunction with other investigations that have helped to elucidate genetic and environmental factors associated with asthma development, will help in the identification of primary and secondary prevention strategies for asthma. This report describes the study objectives, design and recruitment results of this study cohort. The TCRI is a prospective cohort of mother-infant dyads enrolled in a longitudinal investigation of the relationship of infant viral respiratory infection severity and aetiology and the interaction of other risk factors on the development of childhood asthma and allergic diseases. The study was approved by the Vanderbilt Institutional Review Board, and parents provided written informed consent for both their and their child's study participation. Term, non-low birth weight, otherwise healthy infants were enrolled along with their biological mothers, at a single academic institution, Vanderbilt Children's Hospital, at the time of an acute visit (hospitalization, emergency department or unscheduled outpatient visit) for presumed viral bronchiolitis or upper respiratory tract infection (URI) during respiratory viral seasons September through May 2004-2008. Inclusion and exclusion criteria are outlined in Table 1 . Because of the grant funding start date, the first study season did not begin until November 2004. Recruitment was solely hospital-and clinic-based, and was performed 7 days/week during the first 2 years of cohort accrual, and 5 days per week for the two subsequent years. Screening and recruitment were done by experienced Congenital or acquired chronic heart or lung disease, prior requirement for mechanical ventilation for cardiac or pulmonary disease, immunodeficiency, neurologic disease with possible aspiration, significant gastroesophageal reflux disease felt to contribute to pulmonary disease, tracheomalacia Ever received one or more doses of RSV-IVIG or palivizumab Prior study inclusion Fever and neutropenia Children whose parents or guardians were not able to understand the consent process, or a language barrier † research nurses using computerized medical charts to screen infants with presumed respiratory viral illness. The components and time line of the subject visits are outlined in Table 2 Mothers underwent prick skin testing to eight common aeroallergens and a blood specimen was obtained by venepuncture for serum immunoglobulin E (IgE) and DNA. A structured abstraction form was used to obtain information from the medical record regarding the index enrolment visit: current infant weight, confirmation of birth weight, room air pulse oximetry, requirement for supplemental oxygen, medication administration, prior wheezing episodes and detailed information on the current illness and hospital course. Following discharge from the hospital or outpatient setting, the final discharge diagnosis and results of culture data were obtained through chart review. There are three phases of cohort follow up. First, mothers and children undergo an in-person wellchild follow-up visit during the child's second year of life conducted in the Vanderbilt Clinical Research Center, or during a home visit. Second, families are re-contacted every 12-18 months by phone and/or mailings for purposes of cohort retention, and to provide reminders about the remaining study components. Third, mothers, or the current guardians, undergo a phone interview during the fourth and sixth years of life to identify children with asthma, transient wheezing, AR and eczema. The 2-year in-person well-child visit is conducted in the Vanderbilt Clinical Research Center, or during a home visit offered to those unable to return to the study institution. During the visit, the ISAAC questionnaire is administered, blood samples are obtained from children and mothers (if not previously collected) and a buccal swab is collected for DNA if blood cannot be obtained. A structured telephone questionnaire is administered to the mother/parent when their child is 4 and again at 6 years. Trained interviewers employed in the Vanderbilt Survey Research Shared Resource use a web-based computer system to conduct structured telephone interviews, which capture detailed information on asthma and atopy diagnoses and symptoms, extensive environmental exposure history, physical activity, and comorbidities. Asthma, AR and atopic dermatitis outcomes are determined using the ISAAC questionnaire. For children with report of asthma and/or asthma symptoms in the previous 12 months, the Asthma Therapy Assessment Questionnaire is administered, and information on asthma medications, and asthma-related health-care visits are sought. 12, 13 Biospecimen collection and laboratory analyses Table 3 outlines the details of cohort biospecimen collection, repository and testing, which includes infant nasopharyngeal, urine, blood and nasal epithelial cell sample collection, and maternal prick skin testing and blood samples. The discharge diagnosis and supporting clinical parameters of the infant acute respiratory illness visit were reviewed to confirm whether each child had lower respiratory tract infections (LRTI) or URI (n = 628) or another diagnosis (n = 46). LRTI and URI were defined using both the physician discharge diagnosis, as well as post-discharge chart review, and those cases that were not clearly identified as either LRTI or URI were reviewed by a panel of paediatricians who determined whether the illness represented an LRTI, URI, croup or other, which included those that could not be categorized with the available clinical information. Acute respiratory illness severity was determined using the ordinal bronchiolitis score that incorporates admission information on respiratory rate, flaring or retractions, room air oxygen saturation and wheezing, into a score ranging from 0 to 12 (12 being most severe). 27, 28 highly sensitive and specific qRT-PCR assays for many common respiratory viruses, including hMPV, HCoV, RSV, influenza A and B, parainfluenzaviruses 1-3 and rhinovirus. Real-time RT-PCR is performed using the Cepheid Smart Cycler II. All specimens are first tested for GAPDH to confirm integrity of RNA and monitor for potential PCR inhibitors. Negative and positive controls are included with each run, including RNA runoff transcripts to generate a quantitation curve. [14] [15] [16] [17] [18] [19] [20] [21] If rapid antigen and/or culture for RSV or Influenza were performed at the discretion of the admitting physicians during the index visit, these results are also captured and entered in the database. The cells are collected using Mid Turbinate Peds Nylon Flocked Swabs (MicroRheologics, Brescia, Italy), and placed into collection and digestion media, followed by processing and plating onto collagen IV coated tissue culture and grown at 37°C in a 5% CO2 incubator. To develop techniques for isolation, short-term culture, and in vivo modelling of epithelial and stromal cells, a xenograph model has been developed. Following short-term growth in culture, cells are transferred to denuded rat tracheas and implanted subcutaneously in nude mice. Follow-up well-child visit Mother and child DNA is collected from both the mother and child during the blood collection, or using a buccal swab if blood can not be obtained. DNA is extracted by the Vanderbilt Center for Human Genetics Research Core Laboratory and stored following extraction for future studies. Enrolment Infant Urine is collected from hospitalized infants during the acute infant illness at study enrolment, and from a convenience sample of outpatient subjects. Urinary measurements including, leukotrienes C4/D4/E4 (LTC4/D4/E4), and urinary metabolite of the isoprostane, 15-F2t-IsoP (8-iso-PGF2a) will be measured by a gas chromatographic, and other biomarkers and the remainder of the urinary biospecimens will be maintained in the repository. 25, 26 IgE, immunoglobulin E; LRTI, lower respiratory tract infections; URI, upper respiratory tract infection. The family history of atopy was obtained using a family tree. Maternal atopy will be categorized as evidence of atopy by skin testing or specific IgE, and/or clinical symptoms of an atopic disease as assessed by the ISAAC questionnaire. Atopic status of the child will be determined by laboratory evidence of specific IgE during the second year of life, and by clinical evidence based on the above definitions. The diagnosis of asthma will be determined at age 6 years based on responses to the ISAAC questionnaire. [6] [7] [8] The following criteria will define asthma during the sixth year of life: (i) 12-month prevalence of symptoms of asthma (current wheeze) or the presence of exercise-induced wheeze or dry cough at night not due to a cold or chest infection; and (ii) physician diagnosis as determined by the ISAAC questionnaire using either parental reported physician diagnosis of asthma or chronic use/ prescription of asthma-specific medications. Probable asthma will be defined as physician diagnosis only and analysed separately. Transient early wheezing will be defined as wheezing episodes present in the first 4 years of life, but not meeting the definition for childhood asthma at age 4 and 6 years. 29 Allergic rhinitis will be determined through the ISAAC core questions on AR. 7, 8 Children will be considered to have definite AR if each of three conditions is present between age 5 and 6 years: (i) a history of nasal congestion, runny nose, itchy watery eyes, sinus pain or pressure or headaches, sneezing, blocked nose, loss of sense of smell; (ii) substantial variability in symptoms over time or seasonality; and (iii) diagnosed as having allergic rhinitis by a physician or on medications for AR. Probable allergic rhinitis will be defined as meeting two of the three criteria, or only criteria 3. Atopic dermatitis will be determined through the ISAAC core questions on atopic dermatitis, which are based on a list of major and minor criteria widely applied in clinical studies. 8, 30, 31 As eczema is probably more readily confirmed by objective tests than either asthma or rhinitis, patients will be considered to have definite atopic dermatitis if between age 5 and 6 years they report ever having an itchy rash that comes and goes for at least 6 months, and being diagnosed with eczema by a physician. 30, 31 Probable atopic dermatitis will be defined as one of the two above criteria. In order to standardize and monitor the quality of data collection and processing, all study personnel received training and were certified for all the study procedures. Information is recorded on paper case report forms, data are entered and then checked by a second reviewer. Logical data checks are programmed and additionally performed by our systems analyst, investigators and again by our biostatisticians. For laboratory analyses, blind qualitycontrol samples are included in each biospecimen run. Telephone interviewers complete classroom training, orientation to the study population, computer modules, role play interviewing and training on study-specific protocols, and are formally evaluated at the end of training. A verbatim-recording of the interviewer and participant responses, and 10% participant re-contact allows quality-control staff to verify responses. The outcome variables of interest are the incidence of asthma and allergic rhinitis. The primary exposure variables of interest are the severity and aetiology of the infant viral illness and maternal and familial atopic status and other environmental exposures. Cumulative asthma incidence over time, taking into account loss to follow up, will be used for illustrating incidence data. Incidence of asthma and allergic diseases among the enrolled infant population with viral respiratory illness will be calculated by dividing the number of incident asthma cases by the person time of follow up. A Kaplan-Meier plot of cumulative incidence over time, taking into account loss to follow up, will be used for illustrating incidence data. Incidence rates will be calculated by dividing the patient population into quartiles/quintiles of bronchiolitis severity scores. The adjusted risk of asthma and allergic rhinitis with bronchiolitis severity will be evaluated using the Cox proportional hazard regression model. To assess the relationship between biomarker concentrations and increased risk of infant bronchiolitis, and early childhood asthma outcomes, we will conduct a nested case-control study. Geometric means of urinary biomarker concentrations for those who develop and do not develop asthma will be calculated separately and compared using the paired t-test. The association between these biomarkers and the risk of asthma and allergic rhinitis will be assessed using odds ratios and their corresponding 95% confidence intervals from adjusted logistic regression models. The potential confounders that will be considered in multivariable analyses will include demographic and exposure characteristics. Subject recruitment for the TCRI Study occurred over 4 years, and was completed in May 2008. Overall, 9329 visits were screened, representing 7632 unique infants, and 2986 of these infants met study eligibility requirements. From the 2986 eligible infants, 674 infants and their biological mothers were enrolled (Fig. 1) . Among the 2312 subjects who were available during the recruitment periods, the major reasons for non-response were refusal (22%), insufficient time/ unwilling to stay for the visit (outpatients) (39%), conflict with or already enrolled in another study (20%), and other (18%) (includes language barrier, mother/ guardian not present, previously enrolled and other miscellaneous). In 99.9% of the cohort, the nasal/ throat swab was obtained, and in 79% of the hospitalized infants one spot urine sample was obtained at hospital admission. Weekly enrolment into the cohort is depicted in Figure 2 . The TCRI is a large and comprehensive prospective epidemiologic study of mothers and their biologic children enrolled during infancy with a clinically significant LRTI or URI who are being followed through early childhood. This study will provide important information about the role of infant respiratory viral infection severity, aetiology, biomarkers and predictors important in the development of early childhood asthma and allergic rhinitis. The TCRI is additionally unique because it is designed in parallel with a large retrospective birth cohort of over 95 000 mother-infant dyads with similar objectives to investigate the role of respiratory viral infection severity and aetiology in the development of asthma. As evidence suggests that the development of asthma may result in part from respiratory viral infection during infancy, which has a predilection for infecting, destroying and/or in some way biologically altering lower airway epithelium, this study will help to delineate whether the severity of that infection and other early-life events impact the risk of asthma and allergic disease development in later childhood. 3 Despite a high attack rate of developing asthma following viral bronchiolitis, the majority of children who have infant bronchiolitis do not develop childhood asthma. Thus while viral respiratory infections may alter lung physiology and target the inflammatory response to the lower airway, this may only occur during a vulnerable time period during development of the immune system or lung, or in the presence of other risk factors. This developmental component may further reflect important gene-environment interactions that regulate both short-and long-term airway physiological alterations that manifest themselves clinically as childhood asthma. Efforts to determine and define the role of these factors, including disease severity, maternal atopy and other environmental exposures, such as second-hand smoke, to asthma pathogenesis are the focus and goal of the TCRI. Several limitations of this study should be noted. First, the study sample was not randomly selected from the general population, but instead was recruited from a single hospital and clinic-based setting, the Vanderbilt Children's Hospital. While Vanderbilt hospitalizations represent greater than 90% of Davidson County/Nashville infant hospitalizations, it represents a smaller proportion of emergency department visits (51%), and likely even fewer paediatric acute care visits. 2 The relatively low participation rate among eligible subjects is multifactorial and the result of the long-term nature of the study, the lack of study personnel to enrol all eligible subjects, as well as lower willingness among outpatients to extend their visit in order to participate in the study. While this impacts the demographics and exposures of the study population, and thus generalizibility of our study results, it should not impact the findings of the role of infant viral infection on the outcomes of interest. Next, while airway hyperreactivity is not assessed in making the diagnosis of childhood asthma, the identification of incident asthma cases will take into consideration the positive response to a validated written questionnaire that has been compared with bronchial provocation testing. 6, 8 While such an ascertainment strategy might result in the underdiagnosis of asthma, it is unlikely to result in false positive diagnoses during the sixth year of life. Finally, as with many studies, where all eligible participants were approached for participation, difficulty was encountered in follow up of those currently age-eligible for follow up. Strategies to address this include study personnel doing a significant number of follow-up visits at the subject's home, and shipping follow-up materials and requests to paediatricians of subjects who have moved from the region. Future reports from this cohort will help to clarify the complex relationship between infant respiratory viral infection severity, aetiology, atopic predisposition, and the development of early childhood asthma and other atopic diseases. Ultimately, this study, along with the companion TABS cohort, has the potential to provide new approaches to identify infants at high risk of developing early childhood asthma and allergic diseases, as well as provide important information that may contribute to the development of prevention strategies.

Eosinophilic infiltrate in a patient with severe Legionella pneumonia as a levofloxacin-related complication: a case report

INTRODUCTION: Legionella pneumonia can appear with different levels of severity and it can often present with complications such as acute respiratory distress syndrome. CASE PRESENTATION: We report the case of a 44-year-old Caucasian man with Legionella pneumonia with successive development of severe acute respiratory distress syndrome. During his stay in intensive care the clinical and radiological situation of the previously observed acute respiratory distress syndrome unexpectedly worsened due to acute pulmonary eosinophilic infiltrate of iatrogenic origin. CONCLUSION: Levofloxacin treatment caused the occurrence of acute eosinophilic infiltrate. Diagnosis was possible following bronchoscopic examination using bronchoaspirate and transbronchial biopsy.

Since the pneumonia epidemic that struck the delegates of the American Legion Convention in Philadelphia in 1976, Legionella spp. has become a relatively frequent cause of community acquired pneumonia [1] . Legionella may appear in different forms, from subclinical presentations to Legionnaires' disease, which has a mortality rate as high as 30 to 50% in cases of hospital infections and in cases of complications such as acute respiratory distress syndrome (ARDS). The fatality rate is 5 to 25% even in patients who are immunocompetent [2] . Other complications are rare, although a significant number of drugs used in the treatment of Legionella pneumonia can be associated with the appearance of pulmonary eosinophilic infiltrates, especially non-steroidal anti-inflammatory drugs (NSAIDs) and antibiotics [3] . The diagnosis is mainly based on the temporal correlation between the administration of drugs and the appearance of the clinical condition, but it is often not easy to determine the etiologic agent with certainty. This report concerns the case of a man with Legionella pneumonia that evolved into ARDS and then became complicated with eosinophilic infiltration as an effect of treatment with levofloxacin. Usually this drug is safe, though in some cases can cause eosinophilic pneumonia [4] . A 44-year-old Caucasian man presented to our hospital for hyperpyrexia (over 39°C) for about a week, with general weakness and strong headaches; he had been treated by his general practitioner with amoxicillin and clavulanate administrated orally with no improvement. His case history revealed that he was a smoker (20 packs/year). No other pathologies or trips abroad had been registered in the last 6 months. On admission, he had hyperpyrexia (38.9°C), headache, dry cough, diarrhea, general weakness and sinus tachycardia (100 beats/minute); his oxygen saturation was 95% (no oxygen supplement). The results of a physical examination of his chest were reduced vesicular respiration and crackling in the median axillary line to the left and in front; a chest X-ray showed extensive inconsistent parenchymal consolidation at the fissure of the left upper lobe ( Figure 1A ). The results of initial laboratory examinations revealed his white blood cell count was 2020 cells/mm 3 , total bilirubin level was (1.6 mg/dL), he had reduced albuminemia (2.7 g/dL), increased alkaline phosphatase (382 U/L), γ-glutamyl transferase (69 U/L) and creatine phosphokinase (422 U/L). His serology test results were negative for Hepatitis B virus, Hepatitis C virus and HIV. His initial blood culture test results were negative for aerobic and anaerobic germs and mycetes. Our patient began treatment with intravenous piperacillin and tazobactam (13.5 g/day) and clarithromycin orally (1 g/day). On the third day the results of his urinary antigen test were found to be positive for Legionella serogroup 1, so clarithromycin was suspended and substituted with intravenous levofloxacin (750 mg/day). We maintained the piperacillin and tazobactam treatment to help prevent secondary infection from other Gram-positive and Gram-negative bacteria. On the sixth day, his clinical condition worsened. After consultation with an infectious disease specialist, we added rifampicin (900 mg/day) to support the levofloxacin action against Legionella pneumonia. On the ninth day he showed respiratory distress (40 breaths/minute). An Arterial Blood Gas analysis in room air gave the following results: partial O 2 pressure (pO 2 ) of 50 mmHg, partial CO 2 pressure (pCO 2 ) of 30 mmHg, pH 7.50 and oxygen saturation (SaO 2 ) of 86%. A computed tomography (CT) scan of his chest revealed multiple areas of parenchymal consolidation in the entire upper left pulmonary lobe, mixed with ground-glass areas and abundant pleural effusion. In the right lung, in the dorsal and basal regions, there were ground-glass areas mixed with consolidation areas (Figure 1B) . On the 10th day PaO 2 /fraction of inspired O 2 (FiO 2 ) ratio was 101 and he was moved to our intensive care There was an absence of pleural effusion and no cardiomegaly. B) A chest CT scan taken on the ninth day: consolidation areas can be seen on the whole superior left lobe, mixed with ground-glass areas and air bronchogram. There was an absence of pleural effusion. C, D) A CT scan taken on the 21st day: on the left there is parenchymal consolidation with air bronchogram and pneumothorax, and several areas of parenchymal consolidation on the right superior lobe. There was an absence of pleural effusion. unit. Here he was placed on a ventilator on continuous positive airway pressure modality, with noticeable improvement of the respiratory parameters (PaO 2 /FiO 2 ratio of 254). On the 17th day, levofloxacin was suspended in order to allow wash-out and taking of further blood cultures. On the 19th day levofloxacin was resumed; after advice from an infectious diseases specialist intravenous levofloxacin 1500 mg per day together with intravenous fluconazole 800 mg per day were given On the 21st day, after an initial improvement, he showed respiratory distress. A CT scan showed increased parenchymal consolidation with left pneumothorax ( Figure 1C, D) . On the 22nd day, because of the unexpected occurrence of muscular exhaustion, orotracheal intubation was performed and he was placed on a mechanical ventilator in synchronized intermittent mandatory ventilation mode associated with appropriate kinetic therapy on a reclining bed. A fibrobronchoscopy study, carried out with bronchoalveolar lavage (BAL) for bacteriological reasons and in order to define the cytological profile, revealed the presence of numerous macrophages (32%), lymphocytes (26%; CD4/CD8 ratio 0.8), neutrophilic granulocytes (40%) and some eosinophilic granulocytes (2%). Protozoa, fungus and neoplastic cells were absent. On the 23rd day, methylprednisolone (120 mg/day intravenously) was added to the therapy. On the 26th day, he underwent another bronchoscopy, with BAL and transbronchial biopsy in the basal segments of the lower right lobe, which revealed a histological condition compatible with acute eosinophilic pneumonia (Figures 2 and 3) . The BAL confirmed the presence of eosinophils 28%, macrophages 57%, lymphocytes 15%, neutrophilic granulocytes 2% and a CD4/CD8 ratio of 1. Incidental findings showed masses of finely pigmented macrophages (due to our patient's smoking habit). Serum levels of total IgE were within normal limits, and the specific IgE antibody results for allergens (food, pollen, fungal) were also negative. Fecal and serological test results were negative for parasites. On the 27th day, his steroid therapy was increased (methylprednisolone 1 g/day) while levofloxacin was suspended. His response to steroid therapy was rapid, with a general improvement starting from the fifth day of treatment (the 32nd day overall), associated with accompanying improvement of respiratory exchange and subsequent return to spontaneous breathing on the 41st day (PaO 2 /FiO 2 ratio of 357). On the 51st day, a chest X-ray showed that the pneumonia bilateral consolidation had completely resolved ( Figure 4 ). ARDS is a common medical emergency and is usually a complication of a previous illness, which is the etiological cause [5] . In our patient, the unusual fact was the overlapping of acute eosinophilic infiltrate in legionellosis. Eosinophilic pneumonias include a wide range of pulmonary pathologies, characterized by alveolar and peripheral blood eosinophilia. Peripheral eosinophilia may be absent, in particular in the early stages of acute idiopathic eosinophilia pneumonia or in patients taking systemic corticosteroids. It may occur with extremely variable forms of seriousness, from asymptomatic pulmonary infiltrates to acute respiratory distress syndrome associated with respiratory insufficiency. The possible causes, such as drugs or parasitic infections, have been widely studied, but are, in most cases, idiopathic [6] . In our opinion, in accordance with the findings of other [7] , early low-dose steroid therapy leads to a better outcome of pneumonia with severe respiratory distress; however it could determine a delayed onset of eosinophilic pneumonia. In our patient, we are inclined to consider it as having an iatrogenic etiopathogenesis. Other causes were excluded by laboratory tests for differential diagnosis options (serum total and specific IgE, fecal and serologic examinations for parasite infections). Eosinophilic pneumonia has been linked to more than 80 drugs, although only 20 of these (for the most part NSAIDs and antibiotics) can be considered as common causes of this pathology [6] . All the drugs administered in the weeks prior to the appearance of eosinophilic infiltrate should be suspected as a possible cause of the pathology. Iatrogenic eosinophilic infiltrates usually develop progressively, with dyspnea, cough and fever in subjects who have taken certain drugs for weeks or months. The diagnosis of drug-induced eosinophilic pneumonia is mainly based on a detailed history of drug exposure, evidence of eosinophil accumulation in the lung and exclusion of other causes. Numerous methods have been studied in order to demonstrate sensitivity to one or more drugs. One of the most commonly applied methods is the lymphocyte stimulation test (LST), which measures the proliferation of T lymphocytes in response to a drug in vitro, in order to diagnose a previous reaction in vivo. This concept was confirmed by the finding of drug-specific T lymphocyte clones that can interact with cellular receptors without being metabolized and without bonding to protein carriers [8] . We did not consider it necessary to carry out the LST with our patient because this method is not specific and sensitive, and it has the major drawback of being difficult to interpret [8] . With regard to the challenge test in vivo, this was not performed because of the serious clinical condition of our patient, who in any case did not give his consent. However, voluntary challenge may cause life-threatening adverse reactions and it should be limited to rare situations [9] . Among the possible causes we considered, the first was levofloxacin. There are some reports in the literature regarding the possibility of development of eosinophilic pneumonia during the course of levofloxacin therapy [4] ; moreover, it was the drug administered to our patient for the greatest number of days (21 in total). Other points can be taken into account: (1) the drug was suspended for four days in order to allow for washout and subsequent blood culture; afterwards, the same drug was resumed. At the same time, the clinical radiological findings became worse, with an unintentional challenge effect. (2) The BAL on the 22nd day, as some other authors have reported, still showed compatibility with ARDS Legionella, [10] while the following BAL showed eosinophilia (28%) compatible with an acute eosinophilic pneumonia [6] , which histological exams confirmed ( Figure 3) . With regard to the other drugs administered, there are reports of isolated cases of eosinophilia associated with parenchymal infiltrates as a consequence of rifampicin therapy [11] . There is only one reported case where clarithromycin may have led to eosinophilic pneumonia [12] , but our patient was only treated with this drug for two days. Moreover, it is possible that eosinophilic pneumonia could be an adverse reaction to smoking in predisposed subjects: this sometimes happens to patients who have recently started smoking or who have modified their 'way' of smoking (for example, increasing or changing type of smoking). Our patient, however, did not report any changes, either in quantity or in quality, in his smoking habits, so this would seem to exclude any relation to smoking [13] . However, it is plausible that smoking could have acted as a cofactor (together with the drugs) in triggering the clinical condition, because it is a known fact that acute eosinophilic infiltrates are often frequent in smokers [14] . In conclusion, levofloxacin may be the most probable cause of the occurrence of acute eosinophilic infiltrate in this patient. It is important to emphasize that we decided to change the diagnostic and therapeutic approach only when the presence of eosinophilic infiltrate was proven by transbronchial biopsy. Published studies dealing with risks of invasive endoscopic procedures in a patient who was critically ill on mechanical ventilation showed a higher incidence of complications such as hemorrhage and pneumothorax. Correlating the endoscopic risk to the percentage of correctly carried out diagnoses, which varies from 33% to 76%, with consequent change in therapeutic strategy, it may be stated that the risk/benefit ratio of the endoscopic procedure in terms of therapeutic response is surely in its favor and it is, therefore, recommended [15] . Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

Functional analysis of the SRV-1 RNA frameshifting pseudoknot

Simian retrovirus type-1 uses programmed ribosomal frameshifting to control expression of the Gag-Pol polyprotein from overlapping gag and pol open-reading frames. The frameshifting signal consists of a heptanucleotide slippery sequence and a downstream-located 12-base pair pseudoknot. The solution structure of this pseudoknot, previously solved by NMR [Michiels,P.J., Versleijen,A.A., Verlaan,P.W., Pleij,C.W., Hilbers,C.W. and Heus,H.A. (2001) Solution structure of the pseudoknot of SRV-1 RNA, involved in ribosomal frameshifting. J. Mol. Biol., 310, 1109–1123] has a classical H-type fold and forms an extended triple helix by interactions between loop 2 and the minor groove of stem 1 involving base–base and base–sugar contacts. A mutational analysis was performed to test the functional importance of the triple helix for −1 frameshifting in vitro. Changing bases in L2 or base pairs in S1 involved in a base triple resulted in a 2- to 5-fold decrease in frameshifting efficiency. Alterations in the length of L2 had adverse effects on frameshifting. The in vitro effects were well reproduced in vivo, although the effect of enlarging L2 was more dramatic in vivo. The putative role of refolding kinetics of frameshifter pseudoknots is discussed. Overall, the data emphasize the role of the triple helix in −1 frameshifting.

Ribosomal frameshifting is a translational recoding mechanism that allows the synthesis of multiple proteins from a single mRNA. During this process a certain proportion of the ribosomes is forced to move one or two nucleotides backwards (-1 or À2 frameshift) or forwards (+1 or +2 frameshift) whereafter they continue translation in the new reading frame. As a result the stop codon of the first open-reading frame is bypassed and a fusion protein is synthesized [reviewed in (1, 2) ] Frameshifting is frequently used by RNA viruses, in particular by those with a single genome, and is thought to lead to precise ratios of viral proteins, which is crucial for successful infection (3, 4) . Frameshifting is occasionally used by the eukaryotic cell, e.g., to regulate expression of antizyme (5) or by the prokaryotic cell to regulate production of release factor RF2 (6) and synthesis of the gamma subunit of DNA polymerase (7) . The signal that makes a ribosome shift comprises two elements: a slippery sequence, where the ribosome switches the reading frame, and an adjacent stimulatory signal, usually a specific RNA structure. For À1 frameshifting, the slippery sequence usually consists of a heptanucleotide motif X XXY YYZ, where X can be any three identical nucleotides, Y can be three A's or U's, and Z is not G (8) . The slippery sequence has been shown to be shifty on its own in vitro, up to 2%, but is strongly stimulated, up to 40-fold, by the presence of a hairpin, a pseudoknot, a three-way junction (9) or an antisense oligonucleotide (10, 11) , located 5 to 8 nt downstream of the slippery sequence. Despite recent progress (12) (13) (14) , the exact mechanism by which a downstream RNA structure stimulates À1 frameshifting remains unclear. The current view is that the downstream RNA element forms a physical barrier that causes a fraction of ribosomes to stall in their translocation step and puts tension on the mRNA-tRNA interaction (15) . This tension is relieved by realignment of A-site and P-site tRNAs in the 5 0 -direction, whereafter the ribosome resumes translation in the À1 reading frame. Several data also point to a role for the E-site tRNA in stimulating frameshifting (16, 17) . In general, a pseudoknot is more efficient in stimulating frameshifting than a hairpin of the same sequence (18, R.C.L. Olsthoorn unpublished data). This difference is likely related to a higher thermodynamic stability of the pseudoknot. Indeed, from thermodynamic analyses it appears that pseudoknots are more stable than their hairpin counterparts (19) (20) (21) . Recent studies employing mechanical 'pulling' of frameshifter pseudoknots have shown a correlation between the mechanical strength of a pseudoknot and its frameshifting capacity (13, 14) and influence of major groove and minor groove triplex structures (22) . The higher strength of a pseudoknot can be primarily attributed to the formation of base triples between the lower stem S1 and loop 2 (Figure 1 ), making it more resistant against unwinding by an elongating ribosome (15, 23) . Base triples in the pseudoknots of Beet western yellows virus (BWYV) (24) , Pea enation mosaic virus (PEMV-1) (21) and Sugarcane yellow leaf virus (ScYLV) (25) have been shown to play an essential role in frameshifting. We previously solved the solution structure of the pseudoknot present in the overlapping region of gag and pol genes of Simian retrovirus type-1 (SRV-1) by NMR (26) . The structure has a classical H-type fold and is further characterized by interactions between the minor groove of stem 1 and loop 2, which forms a triple helix by various tertiary interactions and extensive stacking ( Figure 1A) . Here we present a detailed mutational analysis of the SRV-1 pseudoknot addressing the role of the triple helix for À1 frameshifting in vitro and in vivo. The data not only emphasize the functional role of the triple helix in À1 frameshifting but also suggest a role for refolding kinetics of frameshifter pseudoknots. Mutations in the SRV-1 frameshifting signal were made in plasmid SF2 that is derivative of pSFCASS5 (8), a frameshift reporter construct used in earlier phenotypic studies. SF105-113 [except SF110 (A28G) see SF210] were obtained by a two-step PCR mutagenesis procedure using degenerated primers on plasmid pSF103 that contains the sequence of the 'NMR pseudoknot' (26) . The appropriate mutants were selected by dideoxy sequencing. SF202-206 and SF209-211 were constructed as follows. First a BglII-NcoI fragment from the SF2 vector was replaced by a short DNA fragment obtained by hybridization of oligonucleotides (5 0 -GATCTTAATACGACT CACTATAGGGCTCAGGGAAACTGATCA-CGTGG C-3 0 ) and 5 0 -CATGGCCACGTGATCAGTTTCCCTGA GCCCTATAGTGAGTC-GTATTAA-3 0 ) creating a unique BclI restriction site just downstream of the slippery sequence. The resulting plasmid pSF201 was opened at the BclI and NcoI sites and sets of complementary oligonucleotides corresponding to mutants SF202-206 were inserted. In mutant SF207, which codes only for the À1 reading frame product, the BglII-NcoI fragment was replaced by two short complementary oligonucleotides: SF218/SF219: 5 0 GATCTGGCCACTAGTAC/5 0 CATGG TACTAGTGGCCA). Note that this in-frame product is 21 aa shorter than the frameshifted À1 product. A C B Figure 1 . Structure of the 'NMR' SRV-1 frameshift pseudoknot and the effect of substitutions in L2 on its frameshift-inducing capacity. This pseudoknot differs from the viral pseudoknot by having G2C18-to-CG, C10G32-to-UA, G20-to-C and deletion of GCU between C24 and A25 substitutions (26) . (A) The 3D model of the NMR pseudoknot (PDB 1E95); L2 bases, magenta; S1 and S2 base pairs in cyan, L1 base in red, 3 0 single-stranded sequence AC in yellow. (B) Substitutions in the NMR pseudoknot sequence. Dashed lines illustrate base triples. (C) Rabbit reticulocyte lysate translation products of mRNAs derived from BamHI-digested templates were separated on a 17.5% SDS polyacrylamide gel and detected by fluorography. The migration of the 19 kDa 0-frame product (NFS) and the 22 kDa 'frameshift' product (FS) are indicated. SF103, wild-type; see Table 1 for the explanation of other constructs. SF213-224 were constructed as follows. First the entire BglII-NcoI fragment of pSF2 was replaced by a synthetic dsDNA fragment (5 0 -GATCTTAATACGACTCAC TATA-GGGCTCATTTAAACTAGTTGAGGGGCCA TATTTCGC-3 0 and 5 0 -CATGGCGAAATATGGCCC CTCAACTAGTTTAAATGAGCCCTATAGTGAGTC GTATTAA-3 0 , sequences forming a SpeI restriction site are underlined). The resulting plasmid pSF208 was opened at the BglII and SpeI sites and oligonucleotides SF226 (5 0 -GATCTTAATACGACTCACT-ATAGGGCTCAGG GAAA-3 0 ) and SF227 (5 0 -CTAGTTTCCCTGAGCCCT ATA-GTGAGTCGTATTAA-3 0 ) were inserted. This yielded pSF212, which was subsequently digested with SpeI and NcoI to insert sets of complementary oligonucleotides corresponding to mutants SF213-224. The sequences of these oligonucleotides are available on request. All mutants were checked by dideoxy sequencing and are listed in Table 1 . In vitro transcription DNA templates were linearized by BamHI digestion and purified by successive phenol/chloroform extraction and column filtration (Qiagen). SP6 polymerase directed transcription was carried out in a 50-ml reaction containing 1-3 mg linearized DNA, 1 mM NTPs, 40 mM Tris-HCl (pH 7.9), 10 mM NaCl, 10 mM DTT, 6 mM MgCl 2 , 2 mM spermidine, 6 units of Rnase inhibitor (RNAguard, Pharmacia) and 15 units of SP6 polymerase (Promega). After an incubation period of 2 h at 37 C, samples were taken and run on agarose gels to determine the quality and quantity of the transcripts. Appropriate dilutions of the reaction mix in desalted and sterilized water were directly used for in vitro translations. Alternatively, transcripts were purified by phenol/ chloroform extraction and isopropanol precipitation as described earlier (27) . Reactions contained 4 ml of RNA solution, 4.5 ml of reticulocyte lysate (Promega), 1 ml of 35 S methionine (ICN, in vitro translation grade), 0.5 ml of 1 mM amino acids (ÀMet) and were incubated for 60 min at 28 C. Samples were boiled for 3 min in the Laemmli buffer and loaded onto 12% SDS polyacrylamide gels. Gels were dried and exposed to phosphoimager screens. Band intensity of 0 frame and À1 frame products was measured using Molecular Imager FX (Biorad) and Quantity One software. Frameshift percentages were calculated as the amount of À1 frame product divided by the sum of 0 and À1 frame products, corrected for the number of methionines, multiplied by 100. Pseudoknot mutants were cloned in KpnI/BamHI digested pDUAL-HIV(0) (4). The GGGAAAC slippery sequence was changed to the more efficient UUUAAAC (28) to obtain a better read-out. In these constructs the stopcodon of the first open-reading frame, Renilla luciferase, is located downstream of the pseudoknot. A non-frameshifting control was constructed by changing the slippery sequence to UUUAAGC, while in the in-frame control a C-residue was added directly downstream of this heptamer. In both constructs the pseudoknot was the NMR-pseudoknot. A list of oligonucleotide sequences is available on request. All constructs were confirmed by DNA sequencing. HeLa cells (24-well plate) were transfected with 250 ng of plasmid using lipofectamine-2000 (Invitrogen). Cells were lysed 20 h after transfection and luciferase activities were measured in a Glomax-multidetector (Promega) according to manufacturer's protocol. The NMR structure of the SRV-1 pseudoknot (26) revealed a number of interesting features that may be relevant to the process of frameshifting. These features include stacking of adenosines A21-A26 in loop L2 and flipping out of cytosine 24, base-base and/or base-sugar interactions of A21, A26, U27 and A28 from L2 with the minor groove of stem S1 ( Figure 1A ). To test the importance of these interactions in frameshifting, mutations were introduced in these regions of the pseudoknot and their frameshifting capacity was measured in rabbit reticulocyte lysates and compared to the NMR wild-type pseudoknot ( Table 1) . C20. For NMR purposes the wild-type G-residue at this position was replaced by a C (26) . Back mutation to a G (SF105) resulted in a 1.2-fold increase in frameshifting ( Figure 1C ). An adenosine at this position (SF106) was even more beneficial for frameshifting showing a 1.5-fold increase relative to the control NMR pseudoknot ( Figure 1C , compare lanes 'SF103 0 and '106 0 ). A uracil at this position was not tested since this would lead to a premature stopcodon in the À1 frame. Insertion of UUU (SF114) between C19 and C20 led to a slight decrease in frameshifting: 88% relative to the NMR wild-type (SF103). Even though in the latter mutant two additional base pairs might be formed with the 5 0 single-stranded sequence (AU and GU) these are probably melted when the ribosome is stalled over the slippery sequence. A21. This nucleotide forms a base triple with the C2-G18 base pair. Mutation of A21 to G did not affect frameshifting, but replacement with C or U decreased the relative frameshifting activity to 60 and 63%, respectively (Table 1) . Although a guanine cannot substitute for the triple interaction with the C2-G18 base pair, its stacking properties may compensate for this. The two pyrimidines may be too small to bridge the distance here to form an interaction with C2-G18. C24. This nucleotide is extruded from loop 2 by the stacking interaction of the continuous run of adenines and is not involved in any interaction. Thus, from a structural point of view, the presence of C24 in loop L2 seems superfluous or even a nuisance by hindering the stacking of the A21-A26 adenosines. Changing C24 to U (SF111) did not affect frameshifting ( Figure 1C) , suggesting that also a uridine at this position is flipped out. Replacing C24 by G (SF112) or deleting it (SF202) had no significant effect on frameshifting while an adenosine at this position slightly enhanced frameshifting activity (SF113), possibly as a result of extended adenosine stacking. These results show, as expected, that C24 is irrelevant and dispensable for frameshifting. A26. A26 forms three hydrogen bonds with stem 2. Two hydrogen bonds from the N3 and hydroxyl group of A23 to the C16 2 0 OH together form a single ribose-zipper motif. The third hydrogen bond is between the N1 and amino group of G4. A pyrimidine seems too small to fulfil these interactions and indeed replacing A26 by C or U showed an approximately 3-fold lower frameshifting activity (Table 1, SF217 and SF216 ). G would be capable of forming the ribose zipper motif though a considerable clash between its N1 and the G4 amino group can be expected. The relatively high frameshifting activity, 70% of the A26G pseudoknot, suggests that, at this position in L2, stacking properties assisted by the ribose interaction may be more important than formation of the zipper motif itself. U27. U27 forms only a single hydrogen bond via O2 to the amino group of G15, putting less structural restraints on frameshifting. Still mutation of this residue reduced frameshifting 1.4-to 2.5-fold ( Table 1 ). The 2-fold reduction seen by substituting C for U27 is not easily explained as this pyrimidine has an identically positioned O 2 to hydrogen bond to the amino group of G15. A28. A28 is a key nucleotide in stabilization of the pseudoknot fold by anchoring L2 to the S1/S2 junction through formation of two hydrogen bonds with the 2 0 -hydroxyl group of G14. Replacement by U (SF108) or G (SF210) resulted in almost 2-fold decrease whereas a C (SF109), though having a similarly positioned N1 and amino groups as an adenosine, resulted in $4-fold drop in frameshifting ( Figure 1C) . Apparently, the relatively small pyrimidine ring is too far away to form interactions with the 2 0 OH of G14. In addition, the lower frameshift activity of the C-mutant could also be caused by formation of a base pair with G14, which could change the overall structure around the S1/S2 junction. In summary, the effect of single-nucleotide mutations in L2 show the functional, critical importance of the A26:G4-C16 and A28:C5-G15 triples that were previously identified by NMR spectroscopy (26) . Base triples at these positions, stabilizing the S1/S2 junction have been recognized to be essential for frameshifting in other pseudoknots as well, be it with different hydrogen bond patterns (21, 24, 25, 29) . To provide additional evidence for the importance of the specific interactions between L2 and S1 in frameshifting, we accumulated a number of the adverse mutations into one construct. Therefore, a construct was designed with a pyrimidine-rich L2 sequence CUCUCGUG (SF229, 'L2-mut') that included the A21U, C24U, A26G and A28G mutations. SF229 showed an 8-fold lower activity in frameshifting (Table 1) , which is approximately the sum of the single-nucleotide changes. G1-C19. A previous sequence comparison showed that most frameshifter pseudoknots start with a 5 0 G-C3 0 base pair (30). Changing this base pair, which is not involved in any loop interaction, to C-G (SF209) affected frameshifting efficiency somewhat (73%). As stacking energies of this base pair are also lower (À2.4 versus À3.3 kcal/mol) this confers to the general notion that helix stability is an important determinant in frameshifting. Changing G1-C19 into an A-U base pair (SF211) or a G-G mismatch (SF107) reduced the efficiency even further to about 40%, again emphasizing the role of S1 stability. G3-C17. This pair is also not involved in any triple interaction, but replacing it with A-U (SF342) resulted in a 60% drop in frameshifting. Again this indicates that the stability of S1 plays an important role in the frameshifting capacity of the pseudoknot. Previously, changing G3-C17 in the wild-type pseudoknot to A-U resulted in an almost 5-fold drop in frameshifting (31) . Since we do not have high-resolution structural information of the wild-type pseudoknot (L2: GAAACAAGCUUA), we cannot exclude that the G3-C17 base pair forms a base triple in this context, which could further attribute to the enhanced effect of mutating this base pair. G4-C16. This base pair is involved in a base triple with A26, mutation of which had quite some impact on frameshifting (see above). Changing G4-C16 into A-U (SF204) is predicted to result in the loss of one hydrogen bond (G24N2H-A26N1), similar to changing A26 to G (SF218), which was still 71% active in frameshifting. The observed dramatic low activity of the A-U base pair mutant (13%) cannot be solely attributed to the loss of one hydrogen bond and must therefore also be due to compromising the stability of the S1 stem. C5-G15. This pair forms a triple interaction with U27, mutation of which had at most a one-third lower frameshifting activity. Changing the base pair to U-A (SF203) caused an almost 4-fold drop in frameshifting activity, again showing the importance of stem stability. C6-G14. This pair forms a base triple with A28 via the hydroxyl group of G14. This interaction is not expected to change by mutating it to a U-A bp (SF205). The 2.5-fold decrease must therefore be a consequence of reduced stem stability and/or stacking properties with the S2 U-A pair at the junction. This large decrease again illustrates the critical importance of the A28:C6-G15 triple interaction. As with the single-nucleotide substitutions, the effect of the base pair mutations also correlates with the NMR structure, thereby illustrating the functional importance of the S1-L2 interactions. Also the stability of S1 as a primary determinant of frameshifting is clearly confirmed. Using the above knowledge we attempted to create a supershifting pseudoknot (Figure 2 ). To this end we substituted G for C20, removed C24 and restored the G2-C18 and C10-G32 base pairs, as present in the SRV-1 pseudoknot. To our surprise this mutant was 3-fold less active in frameshifting than the parent SF206 pseudoknot. Close inspection of the sequence indicated a possible alternative structure containing two hairpins competing with pseudoknot formation (Figure 2 ). To prevent this alternative structure, we mutated C10-G32 to U-A (SF223) or G-C (SF224). In these constructs the U-A pair is predicted to destabilize the 4-and 7-bp hairpins and the G-C base pair is predicted to completely disrupt the 4-bp hairpin and destabilize the 7-bp hairpin. Frameshift levels of SF224 increased approximately 2-fold compared to SF222, while that of SF223 increased more than 3-fold suggesting that the role of alternative structure formation should not be underestimated. Although SF224 is a slightly better shifter than SF206 (1.11-fold) with an absolute frameshift frequency of 24% it still cannot be considered a supershifter. Previous experiments with SRV-1 pseudoknot mutants yielded better shifters (31) . One of these, SF67, which differed from . Figure 2 . Putative alternative structure of the 'supershifter' pseudoknots. The boxed G-C base pairs and the circled G are also present in the wild-type SRV-1 pseudoknot. The position of the C24 deletion is indicated by the dot and arrowhead. SF222 by still having C24, was capable of shifting 30% of the translating ribosomes. Thus, it appears that depending on the context deleting C24 is not always beneficial for frameshifting. At present it is difficult to interpret these results. One explanation could be that C24 or other L2 nucleotides also play a role in refolding of the pseudoknot after ribosome passage. It can be envisaged that a faster refolding pseudoknot might be a more efficient frameshifter, because in order to cause a frameshift, a stable pseudoknot needs to be regenerated before it encounters the next ribosome. If refolding kinetics of the pseudoknot is a determinant in frameshift efficiency the length of L2 is also expected to have an effect on frameshifting. To test this possibility we changed the length of L2 to 3 and 27 nt in mutants SF348 and SF350, respectively. The absolute frameshifting activity of SF348 was only 1.3%, i.e. a 16-fold decrease compared to SF206. SF350 showed 8.4% frameshifting, a 2.5-fold drop. Of course, when introducing such dramatic changes, one should also consider their effects on the pseudoknot structure and stability. In SF348 most of the base triples are presumably lost and a 3-nt loop crossing the minor groove of stem S1 may not be sterically and thermodynamically favorable. In fact, the very large-16-fold-reduction in frameshifting suggests that the pseudoknot is not formed at all (the presence of a pseudoknot structure was not checked by e.g. structure probing). With SF350, containing the 27-nt L2, the situation is different. In this mutant some of the triples may still form but the additional unstructured 18 nt are expected to introduce a substantial entropic penalty into the system, which increases DG and therefore reduces frameshifting. In addition, a large loop may lead to a longer refolding time of the pseudoknot once a ribosome has passed through this structure. To investigate whether the mutations that affected frameshifting efficiency in vitro would also exert their effect in vivo, we introduced some of the above mutations into a dual luciferase reporter plasmid and assayed their activity in vivo ('Materials and Methods' section). The frameshifting efficiency of the wild-type SRV-1 pseudoknot and GGGAAAC slippery sequence (SF400) was only $8%, against a background of 2% of a non-frameshifting control (data not shown). This is a factor of 3 lower than that reported in vitro (28) . The NMR pseudoknot showed $11% of frameshifting (SF402, data not shown) that increased to $19% in combination with the UUUAAAC slippery sequence (SF404). This 1.7-fold increase is very similar to the previously reported increase from 23 to 40% in vitro for changing the slippery sequence (28) . Mutations were made in the pseudoknot of plasmid SF404. As can be seen in Figure 3 , the substitutions had almost the same effect on frameshifting in vivo as in vitro. The only exception seems to be the effect of enlarging L2, which is consistently more detrimental for frameshifting in vivo than in vitro. This could be due to refolding kinetics of the pseudoknot, what in combination with heavier ribosomal traffic in vivo would result in a lower fraction of properly folded pseudoknots. However, other possibilities, like enhanced RNase susceptibility or alternative folds involving L2, may also account for this effect. The interactions between L2 and S1 in the SRV-1 pseudoknot as previously determined by NMR (26) turn out to be important for frameshifting. Although single alterations of the base triples found in the NMR study had at most a 3-fold deleterious effect on frameshifting, combining several of these changes into one mutant, L2-mut, led to an 8-fold decrease in vitro and a 6-fold decrease in vivo. This is quite remarkable as the mutated sequence should still be able to fold as a pseudoknot, but apparently critical stem-loop interactions are lacking. As found for other frameshifter pseudoknots, the composition and length of L2 can greatly influence frameshifting. Most frameshifter pseudoknots have adenine-rich L2 loops, presumably because of the need for stable triple helix formation, which is best performed by adenosine's N1 and C6-amino groups (2) . Deletion of the bulging C24 from the loop L2 did not affect frameshifting significantly in vitro and in vivo. This adds to the notion that this nucleotide can be safely squeezed out to allow stacking of adenosines in L2. It is noteworthy that other frameshifter pseudoknots also have a cytosine flanked by adenines in L2: GAACAAA in Figure 3 . Comparison of in vivo and in vitro frameshifting activities relative to that of the NMR pseudoknot. White bars, in vivo; gray bars, in vitro. The efficiency of the NMR pseudoknot ('wild-type') is set at 100%. Experiments were done at least three times in triplicate. BWYV (29) , UAAAAACAC in ScYLV (25) , ACUCAAA A in MMTV (32) and AAACAA in HIV-1 type O (33) . Interestingly, L2 of the human telomerase pseudoknot, not involved in frameshifting, consists of CAAACAAA. This pseudoknot has recently been shown to function in frameshifting in vitro (22) . Also here the adenosines formed triple interactions with base pairs in stem S1 and were essential for frameshifting (22, 34) . The effect of substituting the second C in L2 has not been investigated; its presence may be required to prevent slippage of the RNA polymerase. The main role of the triple interactions seems to be in stabilizing the pseudoknot such that high frequencies of frameshifting are possible. Mutations that remove triple interactions in general lead to lower frameshifting. Conversely, it seems that frameshifting pseudoknots with weaker stems need to be stabilized by such triples. For instance in the BWYV and PEMV-1 pseudoknots, S2 is only three base pairs but this is compensated by an additional triple between L1 and S2. The high number of triples in luteovirus pseudoknots may explain why for instance the ScLYV pseudoknot with a relatively high number of A-U base pairs in the bottom of S1 is still capable of inducing $15% of frameshifting (25) . On the other hand, when S1 is increased to 11 base pairs, triple interactions do not seem to contribute anymore as frameshifting becomes independent of the sequence of L2 (C-H. Yu and R.C.L Olsthoorn, unpublished data). If we assume that ribosomes first sense the bottom of stem S1 we would expect base triples to be present at the bottom of the stem or in the 5 0 end of L2. This would be in line with the observation that the first three base pairs of a frameshift pseudoknot are usually G-C (30). However, most triples are found near the 3 0 end of L2 close to the junction with S2. Possibly, triples at the junction render the pseudoknot more resistant to forced unwinding by the ribosomal helicase, which would otherwise serve as a pivot. Because the mere presence of a pseudoknot structure with a certain minimal thermodynamic stability might not be the sole primary determinant of frameshifting a number of alternative models have been proposed in which mechanical resistance of the pseudoknot to forced unwinding is the key (12, 15, 23) . In these models a pseudoknot trapped in the mRNA entry tunnel resists mechanical unwinding, which causes the ribosome to pause. Indeed, recent optical tweezers experiments, in which rupture forces of RNAs are measured by pulling 5 0 and 3 0 ends apart, showed a correlation between frameshifting and mechanical strength of the pseudoknot and confirmed a positive effect of base triple interactions on frame shift efficiency (13, 14, 22) . Frameshifting by the trapped pseudoknot is thought to be either passive, by stalling the ribosome for a sufficient amount of time, or active, by building up tension at the mRNA-tRNA interface by a counteracting ribosome, which can be relieved by a À1 frameshift (15) . Observations that ribosomes can stall for seconds to minutes without frameshifting to occur (35) and pausing appears to be necessary, but not sufficient for frameshifting (36, 37) , rather point to an active role. Although it now seems mechanical resistance to RNA unfolding is the key to ribosomal frameshifting, how the ribosome exactly 'chokes' on such a structure is still far from being understood. Remaining questions are for instance if torsional restraint of the pseudoknot resisting ribosomal unwinding is essential (23) , whether the pseudoknot truly plays an active role in for instance overbending the tRNA in the P-site (12) and at which stage of translational elongation frameshifting actually occurs. One should also keep in mind the possible formation of alternative structures that can affect the frameshifting efficiency by lowering the fraction of properly folded pseudoknots. This possibility has not been investigated in detail but our own data ( Figure 2 ) suggest that this may be a relevant issue. Especially during heavy ribosomal traffic, as may be the case in vivo, fast and correct refolding of pseudoknots could be an important parameter for frameshifting. Under such circumstances it is conceivable that a hairpin is more preferred as it probably refolds faster than a pseudoknot. A recent study on the HIV-1 frameshift signal has shown that a decrease in the translation initiation frequency can lead to an increase in the frequency of frameshifting (38) . Also a previous study in yeast has shown a correlation between frameshifting and translation initiation frequency (36) . Our data with the enlargement of L2 also hint at a possible role for folding kinetics. Although the time scale of folding (milliseconds) may at first seem irrelevant compared to the time scale of translation elongation (seconds) it has recently been reported that the conversion rate between two comparatively stable hairpins ranges from a few seconds to several minutes (39) . Also, slow-folding pseudoknots, 5-350 s, limit the activity of the Hepatitis Delta virus ribozyme (40) . Thus the efficiency of a frameshifter pseudoknot may not be solely dictated by its stability but also by its kinetic properties.

A molecular clamp ensures allosteric coordination of peptidyltransfer and ligand binding to the ribosomal A-site

Although the ribosome is mainly comprised of rRNA and many of its critical functions occur through RNA–RNA interactions, distinct domains of ribosomal proteins also participate in switching the ribosome between different conformational/functional states. Prior studies demonstrated that two extended domains of ribosomal protein L3 form an allosteric switch between the pre- and post-translocational states. Missing was an explanation for how the movements of these domains are communicated among the ribosome's functional centers. Here, a third domain of L3 called the basic thumb, that protrudes roughly perpendicular from the W-finger and is nestled in the center of a cagelike structure formed by elements from three separate domains of the large subunit rRNA is investigated. Mutagenesis of basically charged amino acids of the basic thumb to alanines followed by detailed analyses suggests that it acts as a molecular clamp, playing a role in allosterically communicating the ribosome's tRNA occupancy status to the elongation factor binding region and the peptidyltransferase center, facilitating coordination of their functions through the elongation cycle. The observation that these mutations affected translational fidelity, virus propagation and cell growth demonstrates how small structural changes at the atomic scale can propagate outward to broadly impact the biology of cell.

Analyses of a growing number of high resolution ribosome structures coupled with kinetic studies are revealing that the ribosome is highly dynamic, capable of assuming >40 different conformational states through the translation elongation cycle [reviewed in (1) ]. Thus, a critical question is how the ribosome coordinates all of these states to ensure the directionality and fidelity of protein synthesis. The general answer lies in allostery: the formation and breaking of specific intermolecular contacts in response to different ligand binding states serves as a series of switches to ensure that the ribosome is optimally configured to proceed to the next functional state. The current challenge is to identify and functionally map the specific allosteric switching components. Numerous researchers have employed elegant biochemical, biophysical, structural, genetic, and computational approaches to this problem, a few of which are highlighted here. For example, biochemical approaches have been employed to define the kinetic parameters governing every step of the elongation cycle, revealing that selection of the appropriate aminoacyl-tRNA (aa-tRNA) depends on a two step kinetic process [reviewed in (2) ]. Biophysical methods have revealed that conformational switching by rRNA and ribosomal protein L1 ensures exit of deacylated tRNA from the ribosome (1,3), while real time single molecule fluorescence and force measurements are revealing dynamic motions of the ribosome and tRNAs, and directly probing the forces stabilizing ribosomal complexes [reviewed in (4) ]. Biochemical, computational and high resolution structural methods have been employed to map changes in rRNA structures during the transit of tRNAs through the ribosome and during the peptidyltransfer reaction (5) (6) (7) (8) (9) (10) . Pertinent to this study, we have previously used a combination of molecular genetics, biochemical and biophysical approaches to identify the contributions of specific ribosomal proteins and rRNA bases involved in coordinating the stepwise process of accommodation of aa-tRNA into the ribosomal A-site, activation of the peptidyltransferase center, and recruitment of the trans-acting translocase (EF-G in bacteria, eEF2 in eukaryotes) using yeast ribosomes as a model (11) (12) (13) (14) (15) (16) . Ribosomal protein L3 in particular appears to have an important function in this process. Figure 1A shows L3 within the context of the large ribosomal subunit (LSU), including the following structural and functional elements; the peptidyltransferase center (PTC), the aa-tRNA accommodation corridor which lies in between Helix 89 and the complex Helix 90-92 structure, Helix 95 (the Sarcin/Ricin Loop or SRL), and the GTPase-associated center (GAC). The latter two structures interact with the eEF1A-aa-tRNA-GTP ternary complex and eEF2 (the eukaryotic translation elongation factor homologs of bacterial EF-Tu-aa-tRNA-GTP ternary complex and EF-G, respectively). The 3D rendering in Figure 1B shows that L3 contains a globular domain that interfaces with the solvent side of the LSU, and two structures, the N-terminal extension and a central loop that extend deep into the central core of the LSU. The central loop can be subdivided into two domains, the 'tryptophan finger' (W-finger) positioned at the tip of the central extension, and a cluster of basic amino acids that protrudes from the center of the internal loop like a thumb roughly perpendicular to the W-finger that we call the L3 'basic thumb'. Figure 1C shows that this basic thumb is nestled in the core of a cagelike structure formed by elements of three different 25S rRNA helices: H61, H73 and H90. We previously proposed a 'rocker switch' model describing how structural rearrangements of the N-terminal extension and the W-finger of L3 function to coordinate the stepwise processes of translation elongation (17) . Missing from the prior analyses was an explanation of how the movements of these extensions of L3 are communicated to functional centers of the ribosome. In this current study, analysis of the basic thumb of L3 illuminates this question. Mutagenesis of the indicated amino acids to alanines followed by genetic, biochemical and structural analyses suggests that the basic thumb acts as a molecular clamp to play a role in allosterically communicating the ribosome's tRNA occupancy status to the elongation factor binding region and the peptidyltransferase center, thus facilitating coordination of their functions through the elongation cycle. Escherichia coli DH5a was used to amplify plasmid DNA. Transformation of E. coli and yeast, and preparation of yeast growth media (YPAD, synthetic drop out medium, and 4.7 MB plates for testing the Killer phenotype) were as reported earlier (18) . Restriction enzymes were obtained from MBI Fermentas (Vilnius, Lithuania). The QuikChange XL II site-directed specific mutagenesis kit was obtained from Stratagene (La Jolla, CA, USA). Macrogen Inc. (Piscataway, NJ, USA) performed DNA sequence analysis. Oligonucleotide primers were purchased from IDT (Coralville, IA, USA). The yeast strains used in this study were all derived from the rpl3-gene disruption (rpl3D) strain JD1090 (MAT ura3-52 lys2-801 trp1 leu2 À his3 RPL3::HIS3 pRPL3-URA3-CEN6 [L-A HN M 1 ]) (19) . Mutants of rpl3 were generated using the wild-type RPL3 gene in pJD225 (19) , synthetic oligonucleotides, and the QuikChange XL II kit. The pYDL dual luciferase reporter series of plasmids for monitoring programmed ribosomal frameshifting (PRF) were used as described earlier (20) . (36) . L3 is indicated in green. The aa-tRNA accommodation corridor is framed by Helix 89, and the complex structure formed by Helices 90-92. Elongation factors bind to the GTPase Associated Center (GAC) and the Sarcin/Ricin Loop (SRL) at the tip of Helix 95. The peptidyltransferase center (PTC) is in the center of the large subunit. (B) The 3D view of isolated L3, heat map colored from the N-terminus (blue) to the C-terminus (red). The N-terminal extension and the basic thumb and tryptophan (W) finger of the central extension are indicated. (C) The 3D view of interactions between amino acid residues of the L3 basic thumb investigated in this study. It is surrounded by a cagelike structure formed by 25S rRNA Helices 61-64, H73 and H90. Amino acids mutated in this study are labeled. The killer virus assay was carried out as described earlier (18) . Briefly, yeast colonies were replica plated to 4.7MB plates newly seeded at an optical density at 595 nm (OD 595 ) of 0.5 of the 5 Â 47 killer indicator strain per plate. After 2-3 days at 20 C, killer activity was scored as a zone of growth inhibition around the Killer + colonies. To monitor programmed À1 frameshifting using the dual luciferase reporter plasmids, glass beads were used to prepare lysates from cells expressing the 0-frame, À1 (L-A derived), or +1 (Ty1 derived) dual luciferase plasmids (20) . After clarification of the lysates by centrifugation, typically 5 ml was used in a total volume of 100 ml of dual luciferase assay reagents (Promega, Madison WI, USA), and Renilla and firefly luciferase activities were quantitated using a TD20/20 luminometer (Turner Designs, Sunnyvale, CA, USA). Frameshifting efficiencies were calculated by dividing the firefly/Renilla luminescence ratios from lysates of cells expressing the PRF test reporters by the same ratio obtained from lysates of cells expressing the zero-frame control reporter. All assays were replicated enough times to achieve >95% confidence levels, and statistical analyses were performed as described earlier (21) . Aminoacyl-tRNA synthetases were purified as described earlier (22, 23) .Yeast tRNA Phe was aminoacylated with unlabeled phenylalanine or with [ 14 C]Phe to make Phe-tRNA Phe and [ 14 C]Phe-tRNA Phe , respectively. [ 14 C]Phe-tRNA Phe was used to monitor enzymatic binding to the A site of poly(U) primed ribosomes, and acetylated-[ 14 C]Phe-tRNA Phe (Ac-[ 14 C]Phe-tRNA Phe ) was generated to monitor nonenzymatic P-site binding using poly(U) primed salt washed ribosomes. Phe-tRNA Phe and Ac-Phe-tRNA Phe were used in SHAPE structure probing experiments (see below). Yeast tRNA Phe was aminoacylated as described earlier (24) Reaction mixtures were incubated for 30 min at 30 C, and proteins were removed by extraction with acid-phenol-chloroform. Charged tRNAs were ethanol precipitated and purified using G25 spin columns. [ 14 C]Phe-tRNA Phe was separated from uncharged tRNA by high-performance liquid chromatography (HPLC) as described earlier (25) with the following modifications. Samples were loaded onto a 4.6-by 250-mm JT Baker wide-pore butyl column equilibrated with buffer A (20 mM NH 4 Cl, 10 mM MgCl 2 , 400 mM NaCl; pH 5.0) at 1 ml/min. The column was washed with 10 ml of buffer A, conditions under which free phenylalanine and aminoacyladenylate are eluted from the column. Uncharged tRNAs and residual free [ 14 C]Phe and nucleotides were eluted by isocratic elution of 19 ml at 15% of buffer B (20 mM NH 4 Cl, 10 mM MgCl 2 , 400 mM NaCl, 60% methanol; pH 5.0). [ 14 C]Phe-tRNA Phe was eluted using a programmed binary gradient of buffers A and B. Elution of aa-tRNA was monitored by OD 260 readings, and [ 14 C]Phe-tRNA Phe concentrations and specific activities were determined. The presence of aa-tRNA in the eluted material was confirmed by TLC (24) . Ac-[ 14 C]Phe-tRNA Phe was obtained in a similar manner. Yeast tRNA Phe was charged with [ 14 C]Phe as above, extracted with phenol and purified using G25 columns. Reaction mix (4 ml) contained 200 mM NaOAc, pH 5.2 and 7 nmol of [ 14 C]Phe-tRNA Phe . Acetylation was carried out by addition of 64 ml of acetic anhydride at 1 h intervals for 2 h on ice. After incubation, NaOAc concentration was raised to 300 mM and Ac-[ 14 C]Phe-tRNA Phe was ethanol precipitated. Ac-[ 14 C]Phe-tRNA Phe was further purified by HPLC as described earlier. Sulfolink resin was charged with cysteine as described earlier (26) . Yeast cells were grown in YPAD media to mid log phase, collected and washed with binding buffer (10 mM Tris-HCl, pH 7.5; 5 mM MgCl 2; 60 mM NH 4 Cl; 2 mM DTT). Cells were suspended in binding buffer and disrupted using a Mini Bead Beater. Lysates were centrifuged at 30 000g for 30 min in Beckman MLS 50 rotor. Supernatant (2 ml) was removed and added to 2 ml of cystein-charged Sulfolink slurry (50% resin equilibrated with binding buffer) and incubated on ice for 15 min with mixing as resin sediments. After incubation resin was spun down at 1500g for 0.5 min. Supernatants were removed and resin washed five times with 5 ml of binding buffer. After washing, resin was suspended in 1 ml of elution buffer (10 mM Tris-HCl, pH 7.5; 10 mM MgCl 2; 0.5 M KCl; 1 mg/ml heparin; 2 mM DTT) and incubated for 5 min on ice with occasional mixing. The suspensions were centrifuged, supernatant collected, and elution was repeated two more times. Supernatants were combined (3 ml total volume) and GTP and pH neutralized puromycin were added to 1 mM each. After incubation at 30 C for 30 min, reaction mixtures were loaded on top of a 1 ml glycerol cushion (10 mM Tris-HCl, pH 7.5; 10 mM MgCl 2; 0.5 M KCl; 2 mM DTT; 25% glycerol) and centrifuged at 100 000g for 16 h. Ribosome pellets were suspended in 2 ml of elution buffer without heparin, loaded on top of 2 ml glycerol cushions and centrifuged at 100 000g for 16 h. Ribosome pellets were resuspended in storage buffer [50 mM HEPES-KOH pH 7.6; 5 mM Mg(CH 3 COO) 2; 50 mM NH 4 Cl; 1 mM DTT; 25% glycerol] at 5-10 pmol/ ml (1 OD 260 = 2 pmol) and stored at À80 C. Complex C [ribosome-poly(U)-AcPhe-tRNA] was formed in 400 ml of binding buffer (80 mM Tris-HCl, pH 7.4, 160 mM ammonium chloride, 11 mM magnesium acetate, 2 mM spermidine and 6 mM b-mercaptoethanol) containing 0.4 mM GTP, 500 pmol ribosomes, 0.4 mg/ml poly(U) and 700 pmol Ac-[ 14 C]Phe-tRNA. Mixtures were incubated for 20 min at 30 C and then placed on ice. Complexes were purified from free Ac-[ 14 C]Phe-tRNA Phe by centrifugation through a glycerol cushion (0.5 ml; 20% glycerol in binding buffer by centrifugation at 50 000 rpm for 2 h in MLS 50 rotor). Ribosome pellets were rinsed twice with 1 ml of binding buffer and suspended in 1.15 ml of binding buffer. For puromycin reactions, 1.15 ml of complex C extract was pre-incubated at 30 C for 5 min, and reactions were initiated by adding pH neutralized puromycin (100 mM stock) to final concentrations of 10 mM. Aliquots of 100 ml were removed, and reactions were terminated at the indicated time intervals by addition of 100 ml of 1.0 N NaOH. Reaction products were extracted with 0.4 ml of ethyl acetate, 0.2 ml of organic phase was transferred to scintillation vials, and radioactivity was determined by scintillation counting. A 50-ml aliquot of initial reaction mixture was also transferred to scintillation vials, and total radioactivity (N o ) was determined. Controls without puromycin were included in each experiment, and the values obtained were subtracted as background. The percent of the bound Ac-[ 14 C]Phe-tRNA Phe converted to Ac-[ 14 C]Phe-puromycin was corrected with the extent factor a (determined if complex C were allowed to react for 1 h; C o = aN o ), as described earlier (27, 28) . The reaction plots were fit to a first-order exponential equation, and values of K obs (the apparent rate constant of entire course of reaction at a given concentration of puromycin) were calculated by using Graphpad Prism software. Soluble protein factors were prepared as described earlier (28, 29) . aa-tRNA binding to the A-site of the ribosome was carried out as described earlier (13) . Ribosome mixtures (50 ml) contained 80 mM Tris-HCl, pH 7.4, 160 mM NH 4 Cl, 11 mM Mg(CH 3 COO) 2 ]Phe-tRNA Phe were added to ribosome mixtures and incubated for 20 min at 30 C. Aliquots were then applied onto nitrocellulose membranes, filters were washed with 6 ml of binding buffer, and radioactivity was measured by scintillation counting. Background levels of radioactivity were determined using a blank sample (without ribosomes) and subtracted from test samples. K d values were determined assuming single binding sites using Graphpad Prism software. eEF2 binding 6xHis-tagged eEF2 was purified from TKY675 yeast cells (kindly provided by Dr T. Kinzy) as described earlier (30) with the following modifications. EDTA was added to 5 mM to eluted eEF2 just before dialysis to bind leached Ni 2+ ions and prevent precipitate formation during dialysis due to aggregation of His-tagged protein. eEF2 concentration was determined by [ 14 C]ADP-ribosylation with diphtheria toxin (see below). Each preparation of eEF2 showed linear concentration response curves in the range of eEF2 amounts used in binding experiments. For eEF2-binding experiments, reaction mixes (25 ml) containing 12.5 pmol of salt washed 80S ribosomes and various concentrations of 6xHis-tagged eEF2 in binding buffer (50 mM Tris-HCl, pH 7.5, 50 mM ammonium acetate, 10 mM magnesium acetate, 2 mM DTT, 100 mM GDPNP) were incubated for 20 min at room temperature. Estimation of bound eEF2 was carried out as follows by assuming that ribosome bound eEF2 is not susceptible to ADP-ribosylation by diphtheria toxin (31) (32) (33) . Free (unbound) eEF2 was estimated by ADP-ribosylation of eEF2: 100 pmol [ 14 C] NAD + and 0.2 mg of diphtheria toxin were added to each reaction mix and incubated for 30 min at 30 C. Total eEF2 in each reaction mix was determined by ADP-ribosylation reaction after bound eEF2 was released by adding EDTA to 10 mM. Reaction mixes were precipitated with TCA, and amounts of [ 14 C]ADP-ribosylated eEF2 were determined by liquid scintillation counting. Control values (lacking diphtheria toxin) were subtracted. Ribosome bound eEF2 was calculated by subtracting free values from total amount. K d values were determined assuming single binding sites using Graphpad Prism software. To prime ribosomes with poly(U), reaction mix (100 ml) in binding buffer (80 mM Tris-HCl, pH 7.4, 100 mM NaCl, 15 mM Mg(CH 3 COOH) 2 , 6mM b-mercaptoethanol) containing 55 pmol ribosomes and 50 mg poly(U), was incubated for 20 min at 30 C. Next, for P-site complex, 200 pmol of Ac-Phe-tRNA Phe was added. For structure probing of ribosomes with occupied A sites, ribosomal P sites were blocked with 4Â excess of deacylated tRNA Phe and Phe-tRNA Phe (200 pmol), GTP (0.5 mM) and 5 ml of crude elongation factor mix were added. Reaction mixes were incubated for 20 min at 30 C. Reactions were divided into two parts of 50 ml each (control and modification tubes) and 75 ml of binding buffer was added to each tube. 25 ml of 1M7 (130 mM in DMSO) was added to modification tubes. Control samples contained 25 ml of DMSO. After incubating for 85 min at 30 C ribosomes were precipitated with 450 ml ethanol. Ribosomal RNA was extracted using RNAqueous kit (Ambion). Pellets were dissolved in 100 ml of RNAqueous Lysis Buffer and processed according manufactures instructions. RNA was eluted in 50 ml volume and concentration adjusted to 1 mg/ml with elution buffer. Reverse transcriptase (RT) primer extension analyses of modified RNAs were performed as described (34) . The site-specificity of charged tRNA binding was confirmed using the puromycin reaction (35) . The cryo-electron microscopy (cryo-EM) reconstruction of Thermomyces lanuginosus modeled with Saccharomyces cerevisiae rRNA and ribosomal proteins (36) was visualized using PyMOL (DeLano Scientific LLC). The L3 basic thumb mutants affect cell growth, programmed À1 ribosomal frameshifting and yeast Killer virus maintenance rpd3Dcells harboring wild-type pRPL3-Ura were transformed with mutant rpl3 alleles expressed from a low copy TRP1 vector, and the viability of mutants was assessed by their ability to grow in the presence of 5-FOA. All of the L3 basic thumb single mutants (R232A, K236A, K237A, K241A, R244A, R247A and R248A) were viable as the sole forms of L3 with the exception of R240, which was lethal (Figure 2A , summarized in Table 1 ). Among the viable single mutants, only R232A and R247A conferred noticeable growth defects. A series of double mutants were constructed based on the viable mutants and their physical locations relative to one another. This analysis revealed that the K236A/R247A and K241A/R244A mutants significantly affected cell growth, while the R247A/ R248A mutant was inviable. The K236A/K237A double mutant did not grossly affect cell growth. Most strains of S. cerevisiae harbor a symbiotic virus called 'Killer', a bipartite dsRNA viral system composed of the L-A helper virus and the M 1 satellite [reviewed in (37) ]. The 4.6-kb dsRNA L-A viral genome encodes the viral coat protein (Gag), and the Gag-pol replicase that is encoded by a programmed À1 ribosomal frameshift (À1 PRF) (38) . The M 1 satellite dsRNA is encapsidated and replicated inside of L-A encoded viral particles. The M 1 encoded preprotoxin is processed by the Kex1p and Kex2p cellular proteases into the mature secreted toxin. Cells infected by L-A and M 1 can kill uninfected cells, but are themselves immune, hence the name 'Killer'. The presence of Killer can be easily assayed by replica plating test cells onto a lawn of diploid uninfected (indicator) cells: Killer + cells will kill the nearby indicator cells, resulting in a ring of growth inhibition around the test cells. One of the first yeast mutants cloned, mak8-1, was first identified by its inability to maintain the Killer phenotype (Mak À phenotype) and encodes an allele of rpl3 (39) . Analyses of the L3 basic thumb mutants revealed that three of the single mutants (K236A, K237A and R247A) were completely unable to maintain the Killer virus, while R232A and R244A had weak Killer phenotypes. None of the double mutants were able to maintain the virus ( Figure 2B , summarized in Table 1 ). The efficiency of À1 PRF determines the relative ratio of structural Gag to enzymatic Gag-pol available for viral particle self-assembly, and changing À1 PRF strongly inhibits virus maintenance (18, 40) . Previous studies have identified numerous L3 mutants that promoted altered rates of L-A promoted programmed À1 ribosomal frameshifting (À1 PRF), but did not affect Ty1 mediated programmed +1 ribosomal frameshifting (+1 PRF) (13, 17, 19, 23) . However, while all previous L3 mutants analyzed to date enhanced À1 PRF efficiency, all of the viable L3 basic thumb mutants promoted decreased À1 PRF, ranging between $55% and $75% of wild-type rates ( Figure 2C , summarized in Table 1 ). The significance of these changes in À1 PRF is confirmed by loss of the Killer virus, maintenance of which is known to be sensitive to even small decreases in À1 PRF rates (18, 41) . Consistent with prior studies none of the basic thumb mutants affected +1 PRF (summarized in Table 1 ). The simultaneous slippage model of À1 PRF requires that both the aa-tRNA in the ribosomal A-site, and the peptidyl-tRNA in the P-site must shift on the mRNA (42), while in Ty1 mediated +1 PRF, only the peptidyl-tRNA slips (43) . Consistent with the frameshifting data, all of the L3 basic thumb mutants promoted decreased affinity for aa-tRNA to the A-site ( Figure 3A and B, and summarized in Table 1 ), but did not affect binding of Ac-aa-tRNA to the P-site ( Figure 3C and D, summarized in Table 1 ). Specifically, R232A, which does not directly contact any rRNA bases, had the smallest effect on aa-tRNA binding (K d values $126 nM compared to $94 nM for wild-type, i.e. 1.4-fold increase), while the K236A and K237A mutants, which participate in only a few rRNA contacts ( Figure 4B ), had moderate effects ($165 nM each, 1.8-fold wild-type). In contrast, R247A, which contacts both H61 and H90 had a very strong effect on aa-tRNA binding ($300 nM, 3.3-fold wild-type). The double mutants, which also affected multiple rRNA contacts had comparable effects on aa-tRNA binding (from $245 nM to $340 nM). Eukaryotic elongation factor 2 (eEF2) catalyzes translocation and binds to the same site as the aa-tRNA-eEF1A-GTP ternary complex. Given the effects of the mutants on aa-tRNA binding, the effects of seven mutants on eEF2 binding were assayed ( Figure 3E and F, summarized in Table 1 ). For wild-type ribosomes, the dissociation constant for eEF2 was $383 ± 93 nM. Similar values were observed for R232A, K236A and K237A mutant ribosomes, but the R247A mutant promoted an $8-fold increase in affinity for eEF2 (K d $ 47 nM). Among the double mutants assayed, the K237A/K236A mutant promoted the largest increase in affinity for eEF2 ($84 nM, $4.5-fold increase), followed by K236A/R247A ($110 nM, $3.5-fold increase), and K241A/R244A ($229 nM, $1.7-fold increase). Single round assays of peptidyltransferase activity were performed on puromycin treated salt washed ribosomes pre-loaded with Ac-[ 14 C]Phe-tRNA Phe and purified through glycerol cushions (Complex C) as described in the 'Materials and Methods' section. The observed K obs = 0.34 min À1 in wild-type ribosomes ( Figure 3G , Table 1 ) is comparable to similar reactions using E. coli ribosomes and the Ac-Phe-tRNA substrate (44) , confirming that these relatively low rates are determined by Ac-Phe-tRNA as a poor substrate for the peptidyltransferase reaction, and thus represent true measurements of peptidyltransferase activity, as opposed to peptidyl-tRNA turnover or other artifacts. The R232A, K236A and R247A mutants all promoted decreased rates of peptidyltransfer to approximately two-thirds of wild-type levels, while the double mutants (K236A/K237A, K236A/R1247A, and R247A/K248A) had stronger effects, decreasing rates to $50% of wild-type ( Figure 3G and H, Table 1 ). Unexpectedly, the K237A mutant enhanced the rate of peptidyltransfer by almost 2-fold above wild-type levels. The L3 basic thumb mutants promote changes in 25S rRNA structure both locally and in elements associated with aa-tRNA and eEF2 related functions Inspection of atomic resolution ribosome structures reveals that the L3 basic thumb participates in a highly conserved set of interactions with the PTC proximal bases of Helix 73, and Helix 90, with bases on both sides of Helix 61, and with bases in a complex loop structure connecting Helices 61-64 of the LSU rRNA ( Figures 1C, 4B and D) (6, 36, (45) (46) (47) . Strikingly, while 2D maps of the LSU suggest that these structural elements are in physically separate domains from one another in (Figure 4B A-minor motif in the A-site of the peptidyltransferase center (48) . For ease of comparison, yeast 25S rRNA bases are listed with their E. coli homologs in Table 2 . To assess the effects of the viable mutants on rRNA structure, SHAPE [Selective 2 0 -Hydroxyl Acylation and Primer Extension, (34, 49) ] using 1M7 [1-methyl-7nitroisatoic anhydride, (50) ] was used to probe wild-type and selected mutant ribosomes containing either Ac-aa-tRNA Phe at the P-site alone, or both tRNA Phe and aa-tRNA Phe at the P-and A-sites respectively (Figure 4) . Inspection of the results revealed two general trends. First, that the mutants promoted significant changes in rRNA structure in the peptidyltransferase center (U2953, U2955, G2977), along the path taken by aa-tRNA as it accommodates into the LSU (accommodation corridor, G2912, U2924, C2929, A2934), in Helix 94 where it interacts with the globular domain of L3 (G3003, A3006, G3009), and in Helix 95 (U3019, U3023, A3033). Secondly, while the mutants only altered the Helix 73-Helix 95 region when only the P-site was occupied by Ac-aa-tRNA, the majority of changes in the Helix 91-93 region were observed when both A-and P-sites were occupied. Detailed analyses reveal that the R247A mutant conferred the largest number of changes in rRNA structure, promoting deprotection of C2985 in Helix 91, G3003, A3006, G3009 in Helix 94, and U3019, U3023 and A3033 in Helix 95 when only the P-site was occupied ( Figure 4A and B). In contrast, this mutant promoted increased deprotection of G2912 in Helix 92, U2924 in the A-loop, A2934 in the bulge between Helix 92 and Helix 90, and U2955 in the peptidyltransferase center when the A-site was occupied by aa-tRNA. This mutant also caused hyperprotection of A2987 in Helix 73 when the A-site was occupied. Some of the other mutants had similar effects on some but not all of the same bases, e.g. K237A caused deprotection of G2912 and U2955, and enhanced protection of A2987 when the A-site was occupied. Other mutant specific effects were observed. For example, when only the P-site was occupied K247A promoted enhanced protection of G2977, while both K247A and K236A promoted deprotection of A2995 and U2953 under these conditions ( Figure 4A and B). K237A also promoted increased protection from 1M7 at G2977 when both the A-and P-sites were occupied. R232A had significant effects in the 3 0 loop between H90 and H92, and in the peptidyltransferase center (U2955). The observation that the bulged A2971 in Helix 93 was generally deprotected when only the P-site was occupied by Ac-aa-tRNA Phe , but became protected from chemical attack upon loading of aa-tRNA Phe into the A-site (Figure 4 ) serves as an important control, as this site occupancy-specific conformational change has also been observed for bacterial ribosomes (51) . Interestingly, this pattern was also observed for A2926 in the A-loop, the possible significance of which is discussed below. How is information flow coordinated through the ribosome to ensure the directionality of protein synthesis? Although the bulk of the ribosome is comprised of rRNA, and indeed, many of its critical functions are mediated through RNA-RNA interactions, it is clear that protrusions of ribosomal proteins, which can be thought of as loops, hooks and fingers, function to help 'switch' the ribosome between different conformational/functional states. For example, the C-terminal extension of E. coli S13 is thought to help coordinate movement of the peptidyl-tRNA with structural rearrangements at the ribosome interface that are critical for translocation by sampling the tRNA occupancy status at the decoding center (52, 53) . The N-terminal 'hook' of yeast L10 (E. coli L16) is believed to play an active role to coordinate switching of the ribosome between the pre-and post-translocational states (16) . Ribosomal protein L2, which is intimately intertwined with multiple domains of the LSU, is thought to coordinate long range interactions between tRNAs and the ribosome (14) . Ribosomal protein L3 is of particular interest because of its function as a 'gatekeeper' to the ribosomal A-site (13) . A follow-up study suggested that two critical structures of L3, the W-finger and the N-terminal extension, function together as a 'rocker switch' to coordinate LSU associated functions (17) . The L3 basic thumb is of interest because it appears to provide the structural link in this rocker switch mechanism. Protruding roughly perpendicular from the L3 W-finger toward the intersubunit face of the LSU, it is surrounded by a cagelike structure formed by a large bulge framed between Helices H61-64, Helix 73, Helix 90 and Helix 94. It is in the center of a nexus connecting the W-finger with the L3 globular domain, the peptidyltransferase center, the aa-tRNA accommodation corridor, and the SRL. Furthermore, it is proximal to the B5 intersubunit bridge, which involves multiple contacts involving bases in Helices 62 and 64 (54) . Unlike the small subunit (SSU), where the four rRNA domains are largely physically distinct, the six rRNA domains of the LSU are highly intertwined (54) . With regard to the current study, the loop bounded by Helices 61-64 lie in domain IV, the PTC and Helices 90-93 and Helix 73 are in domain V, and Helices 94 and 95 are in domain IV. Previously, we demonstrated that the conformationally dynamic nature of the W-finger enables the central extension of L3 to function like a lever, and as such contribute to allosteric repositioning of rRNA structural elements (13, 17) . However, it was not clear how a small radial movement of thin, essentially planer element, could have such large and long ranging effects on rRNA structure and ribosome function. The perpendicular orientation of the basic thumb may answer this: we propose that it amplifies the action of this lever by adding three-dimensionality to the central extension in the form of a platform upon which structural elements from three different domains of 25S rRNA are anchored. In support of this, comparison of the L3 structures between EF-Tu and EF-G bound Thermus thermophilus ribosomes (55) reveals displacement of the a-carbon backbone of the W-finger and basic thumb structures by $2-3 Å , and of some individual sidechains by as much as 5 Å (Supplementary Figure S1) . This model explains how The large number of basically charged amino acids in the basic thumb enables it to participate in numerous hydrogen bonding/electrostatic interactions with rRNA bases, phosphates and riboses, functioning as a 'molecular clamp' to bridge these three domains. The current study focusing on eight amino acids in the L3 basic thumb neutralized their positive charges with alanine substitutions both singly and in selected pairs. The lethality of the R240A mutant as the sole form of L3 indicates that it may function to 'glue' the PTC proximal stem of H90 with the H61-H64 loop, and may well aid in coordinating formation of the C2876-G2922-G2951 Aminor motif that is critical for the 'induced fit' function of the peptidyltransferase center (5, 6) . Similarly, R247, which bridges Helix 61 with Helix 90 had profound effects on rRNA structure, ribosome biochemistry, translational fidelity, and cell growth, suggesting that it too is a critical bridging component between these two domains. Interestingly, the R247A/R248A double mutant was lethal, while R248A had a wild-type phenotype. This confirms the role of R247 and also indicates that the molecular defect conferred by the R247A single mutant might be partially complemented by the positively charged R248 adjacent to it. R240A and R247A appear to be the exception rather than the rule however, as the other single mutants had significantly lesser effects on these parameters. This is consistent with mutagenesis studies on other residues of L3, and with other LSU proteins and rRNA bases (11) (12) (13) (14) (15) 17, 23, (56) (57) , suggesting that the ribosome is an elegantly evolved molecular machine containing multiple levels of functional redundancy. Furthermore, all of the double mutants tested had strong effects on binding of aa-tRNA and eEF2, and on peptidyltransfer (Figure 3 ), suggesting that it multiple defects are generally required to disrupt the functionally redundant interactions between the basic thumb and the LSU rRNA. As an aside, the observation that A2926 (base paired to U2920) was protected from 1M7 modification in wild-type ribosomes when both Aand P-sites were occupied by tRNAs but deprotected when only the P-site was occupied ( Figure 4A ) is unique to yeast. To our knowledge, this has not been observed in E. coli ribosomes (51), which contains a G-C base pair (C2551-G2557) at this position, a conformational difference that suggests a potentially novel antibiotic target. Interestingly, Haloarcula marismortui appears to split the difference with a G-U base pair (U2576-G2582). Some specific changes in rRNA structure are particularly telling. The base stack of aa-tRNA C74 with H. marismortui U2590 (E. coli U2555, yeast U2924) is thought to promote the induced fit of peptidyltransfer (6) . The enhanced deprotection of this base in R247A mutant ribosomes when both A-and P-sites were occupied by tRNAs is consistent with its strong effect on peptidyltransferase activity by this mutation (Figure 3 H) . Interestingly, this base was also strongly deprotected under the same conditions in K237A mutant ribosomes, but in this case peptidyltransferase activity as monitored using the puromycin reaction was actually enhanced ( Figure 3H ). The rates of peptidyltransfer Figure 4 . Continued arrowheads, and those deprotected relative to wild-type are indicated by black arrowheads. Bases marked in gray (A2926 and A2971) were deprotected when the A-site is unoccupied relative to when it contains aa-tRNA. (B) rRNA protection patterns of the L3 basic thumb mutants mapped onto the 2D diagram of 25S rRNA. Arrowheads indicate relatively protected and deprotected bases as above. Colored boxes indicate bases that interact with specified L3 basic thumb amino acid side chains. A2926 and A2971 are circled in gray, and C2925, which is the first gate in the aa-tRNA accommodation corridor, is circled in purple. The three bases participating the Type II A-minor motif that stabilizes the PTC are boxed and indicated. (C) Data from panels A and B mapped onto the 3D structure of the yeast ribosome. Indicated bases colored black correspond to bases deprotected in the mutants, while those colored gray are hyperprotected. Bases participating in the Type II A-minor motif (A m ) are colored purple. Helical structures and the PTC are color coded as indicated. Note that the loop formed between H61-H64 was removed from this figure because it obscures the L3 basic thumb. Table 2 . Homologous yeast 25S rRNA and E. coli 23S rRNA bases pertinent to this study observed in the current study ($0.3 min À1 ) and in similar analyses using E. coli ribosomes are significantly lower than naturally occurring rates [estimated to be >300 s À1 , (58) ] because Ac-Phe-tRNA is a poor substrate for this reaction. However, this property actually enables us to tease out the effects of local structural changes on PTC activity. Recent molecular dynamics simulations of portions of the ribosome reveal that individual bases can undergo a conservable degree of structural mobility due to local Brownian movements [reviewed in (59) ], suggesting that specific bases in the PTC are relatively free to assume either the induced or uninduced conformations in the absence of tRNAs, and that the equilibrium between these two states is influenced by the presence or absence of aa-tRNA in the A-site. Thus, we suggest that R247A mutant drives this equilibrium toward the uninduced conformation, while the K237A mutant favors the induced arrangement. The observation of distinctly different patterns of rRNA protection/ deprotection (e.g. compare G2912, C2929, U2953, G2977, C2985, A2995, G3003, A3006, G3009, U3019, U3023 and A3033 between the K237A and R247A mutants in Figure 4A ) is consistent with the idea that they also have opposing effects on PTC conformation and functionality. In addition, early studies demonstrated that while empty ribosomes are heterogeneous in their affinity for eEF2, consisting of two sub-populations having K d 's for eEF2 ranging from subnanomolar to hundreds of nanomoles (60), the affinity for eEF2 strongly depends on the functional status of the ribosome as determined by the occupancy status of the A-and P-sites (61, 62) . This suggests that the R247A and the double mutants that increased eEF2 affinity shift this equilibrium as well, possibly stabilizing ribosomes in the pre-translocation state, and that the interactions between the 25S rRNA bases and L3 amino acid residues investigated here are involved in transitions between the pre-and post-translocational states. As discussed above, the effects of the K237A mutant are locally confined to the PTC, while those conferred by R247A are more global. The latter is reflected in their different effects on ligand binding to the A-site: R247A promoted very strong effects on both aa-tRNA and eEF2 binding as compared to the much weaker effects conferred by K237A. Examination of the ligand binding data reveals a reciprocal relationship between affinities for aa-tRNA and eEF2, i.e. increased aa-tRNA K d correlates with decreased eEF2 K d (Figure 3 ). This is consistent with the model that L3 plays a central role as an allosteric switch to coordinate binding of elongation factors, opening and closing of the accommodation corridor, and PTC activity to ensure the unidirectionality of protein synthesis (13, 17) . Interestingly, all of the mutants affected peptidyltransfer, consistent with observations that this process is highly sensitive to even minor structural changes in the ribosome [reviewed in (63) ]. In the end, the most important parameter is life; i.e. how does the L3 basic thumb contribute to the fitness of the organism? While only some of the mutants had gross effects on cell growth, they all affected translational fidelity as monitored by decreased rates of À1 PRF (Figure 2 ). In fact, the effects on À1 PRF correlated well with the A-site aa-tRNA-binding data. For example, R247A and K241A/R244A, which had the most pronounced effects on À1 PRF, also promoted >3-fold decreases in affinity for aa-tRNA. In contrast, R232A and K236A, which had the smallest effects on aa-tRNA binding, also promoted the smallest decreases in À1 PRF. As noted above, the mutants investigated in this report are unique in that they are the first examples that promoted decreased À1 PRF. This trend had only previously been observed with anisomycin, a competitive inhibitor for aa-tRNA 3 0 binding to the PTC (41) . We have suggested that the majority of À1 PRF occurs after aa-tRNA accommodation into the A-site, and prior to peptidyltransfer, while a smaller fraction can occur during translocation [reviewed in (64) ], a view that is supported by a recent study coupling kinetic modeling of À1 PRF within the translation elongation cycle with mass spectroscopic analyses of frameshifted peptide products (P.-Y. Liao et al., submitted for publication). By this model, decreased rates of aa-tRNA accommodation into the A-site should decrease the steady state abundance of substrate for À1 PRF, thus inhibiting this reaction. Changes in À1 PRF in turn alter the relative amounts of viral protein products available for viral particle self assembly, a ratio that is critical for virus propagation (18, 40) . This illustrates how minute changes in ribosome structure at the atomic scale can propagate outward, affecting ribosome biochemistry, translational fidelity, and the ability of cells to replicate viruses.

Hybridization properties of long nucleic acid probes for detection of variable target sequences, and development of a hybridization prediction algorithm

One of the main problems in nucleic acid-based techniques for detection of infectious agents, such as influenza viruses, is that of nucleic acid sequence variation. DNA probes, 70-nt long, some including the nucleotide analog deoxyribose-Inosine (dInosine), were analyzed for hybridization tolerance to different amounts and distributions of mismatching bases, e.g. synonymous mutations, in target DNA. Microsphere-linked 70-mer probes were hybridized in 3M TMAC buffer to biotinylated single-stranded (ss) DNA for subsequent analysis in a Luminex® system. When mismatches interrupted contiguous matching stretches of 6 nt or longer, it had a strong impact on hybridization. Contiguous matching stretches are more important than the same number of matching nucleotides separated by mismatches into several regions. dInosine, but not 5-nitroindole, substitutions at mismatching positions stabilized hybridization remarkably well, comparable to N (4-fold) wobbles in the same positions. In contrast to shorter probes, 70-nt probes with judiciously placed dInosine substitutions and/or wobble positions were remarkably mismatch tolerant, with preserved specificity. An algorithm, NucZip, was constructed to model the nucleation and zipping phases of hybridization, integrating both local and distant binding contributions. It predicted hybridization more exactly than previous algorithms, and has the potential to guide the design of variation-tolerant yet specific probes.

Microbial genomes can be highly variable because of high mutation rates. Because of this extreme variability, it is often difficult to identify regions within a specific virus genome that are sufficiently evolutionarily conserved to serve as targets for specific detection primers and probes. RNA viruses are especially variable. The influenza virus, a negative sense, single-stranded RNA (ssRNA) virus with a highly variable RNA genome, for example, has been known to cause the diagnostic problem that is at the basis of this article, because of a high rate of mutation and genetic drift. In such situations, optimal detection primers and probes would be broadly targeted yet specific, and remain functional even if the genome sequence changed because of genetic drift. Diagnostic nucleic acid hybridization probes are constructed from the most conserved portions of genes from viruses commonly causing infection. Long probes have a large inherent tolerance to microbial variation. The introduction into the introduction into the probe design of a base that can hybridize with all four normal bases (a universal base), or of multiple nucleotides (degenerations; wobbles) in a single position, can induce tolerance to natural viral variation (mismatch). The naturally occurring (1-4) nucleotide (nt) deoxyribose-Inosine (dInosine) is one of many more or less generally hybridizing nt (5,6) known as universal bases. All four normally occurring DNA bases can hybridize to dInosine. The general trend in decreasing hybridization stability is I:C > I:A > I:T & I:G > I:I when using 1 M NaCl, 10 mM sodium cacodylate and 0.5 mM EDTA pH 7 (7, 8) . However, dInosine is readily available and can be recognized as a G by polymerases (5, 9, 10) . Alternative universal bases, e.g. 5-nitroindole, also exist. 3 M tetramethylammonium chloride buffer (TMAC) is a hybridization buffer that selectively raises the stability of A:T base pairs to approximately that of G:C base pairs (11) (12) (13) (14) . It was used in these studies to reduce the effect of sequence composition when comparing different probes of the same length. The term nucleation site is used in this article to indicate a stretch of contiguous perfectly matching nt, capable of initiating hybridization (15) . The aim of these studies was to improve understanding of the design of probes to be used in a TMAC buffer system, by investigating variability and the inclusion of dInosines, other universal bases and wobbles. Specifically, we examined (i) variation (i.e. mismatch) tolerance, (ii) sensitivity to different mismatch distributions, (iii) utilization of dInosine as an nt analog, and (iv) specificity; we also present a new algorithm for prediction of hybridization results. In additional experiments, the question of the use of degeneracy versus a universal base was addressed. Furthermore, we investigated the use of the derived design criteria for detection of rotavirus RNA in clinical samples. A 70-mer nucleic acid hybridization probe, named the InflA probe, was constructed from the most conserved portion of the matrix gene in segment 7 of the Influenza A H3N2 virus. Properties important for the design of variation (mismatch)-tolerant yet specific probes were investigated by studying the interaction between a set of virus-derived probes and complementary targets with different degrees and distributions of mismatch. The 70-mer DNA probes were coupled to color-coded microspheres, hybridized with biotinylated target nucleic acids, incubated with streptavidin-phycoerythrin and analyzed in a Luminex Õ 200 TM system. The hybridization reaction was performed at a standard non-saturating concentration (0.2 nM target) in 3 M TMAC buffer. In further experiments, in an attempt to make a probe with an extended mismatch tolerance, a series of dInosinecontaining probes were synthesized and analyzed for hybridization in the 3 M TMAC buffer system. A limited number of experiments comparing dInosine with an alternative universal base (5-nitroindole) or with wobbles (degenerations) were also conducted. Many viruses harbor synonymous mutations (sm), which means that the third base in a codon can wobble without changing the amino acid in the protein. Targets with regions where every third base was mutated, to resemble the common phenomenon of sm (i.e. the nt sequence is varied without affecting the coding for a specific amino-acid code), were of special interest, since this is often the cause of the variation in coding viral sequences. Rotavirus is a dsRNA virus that causes gastroenteritis. Clinical fecal samples, previously confirmed to contain rotavirus, were used to test the design strategies of this report. An asymmetric PCR, with biotinylated rev-primer in excess, was set up for the VP6 segment. The primers were designed using knowledge gained from this report, i.e. both degeneration and dInosine were used to create moderately degenerated probes with uninterrupted matching stretches as long as possible, in the most conserved regions. The biotinylated reverse (rev-) primer was made 77-nt long to be as tolerant to variation as possible, while the shorter forward (fw-) primer (23 nt) contained four locked nucleic acids (LNA) to increase the hybridization strength (represented by melting temperature, T m ) to match that of the rev-primer. Single stranded 70-mer oligonucleotides with a C12 aminolink at the 5 0 end, with or without dInosines/ 5 0 -nitroindole/N wobbles, were obtained from Biomers.net (Ulm, Germany). The design of the probes was based on the programs BLASTn (16) , ClustalX (17) and ConSort ß (J.Blomberg et al., unpublished results). Briefly, matching viral sequences were retrieved from the GenBank database at NCBI, NIH, using the viral sequence of interest as a query, and alignments were performed by a BLASTn search. BLASTn and ClustalX alignments were analyzed in ConSort ß to define the most suitable probe sequence. ConSort ß provides the frequency of variation and the variation of nt composition in each base position, the number of aligned sequences, and a majority consensus sequence. The proposed probe sequence was further analyzed for its predicted Tm and probable homodimer and hairpin interactions using Mfold at the IDT OligoAnalyser site (http://eu.idtdna .com/analyzer/Applications/OligoAnalyzer/) [http:// mfold.bioinfo.rpi.edu/ (18) ] and Visual OMP TM 7.0 (DNA Software). Visual OMP TM 7.0, which uses the Nearest Neighbor (NN) algorithm [DNA Software (19) ], was used to estimate the change in Gibb's free energy associated with hybridization of the two strands (ÁG) for the interaction between probes and targets at 45 and 55 C in 3M TMAC. IDT OligoAnalyser was used to estimate ÁG for the interaction between probes and targets using 50-mM Na + and 2-mM Mg 2+ . Each combination was tested with the heterodimer-formation function (http://eu.idtdna.com/analyzer/Applications/ OligoAnalyzer/). To create the 70-nt InflA probes, nt 725-794 of the matrix protein 2 gene segment 7 of Influenza A H3N2, accession nr CY023083, was used as query in BLASTn (GenBank database at NCBI, NIH). The Norovirus probe sequence comprises nt 646-715 of the capsid gene of the Norwalk-like virus, accession nr AY274264. The probes for detection of rotavirus were made using sequence EU372725 of the Human rotavirus A strain CMH171/01 inner capsid protein (VP6) gene as query. The probe region chosen, nt 112-178, was analyzed using the haplotype function of ConSort ß . Two differently degenerated probes containing dInosines were designed from each of the two major haplotypes, which potentially covered all rotavirus group A variations recorded in GenBank. Synthetic targets, complementary to the consensus sequence of the InflA, Norovirus and Rotavirus probes, with various numbers of mismatches, were purchased as 70-mer oligonucleotides with biotin attached to a 2-aminoethoxy-ethoxyethanol linker at the 5 0 end (Biomers.net, Ulm, Germany). Specific synthetic 5 0 amine-C12 modified 70-mer probes for influenza A were designed and solid-phase coupled to xMAP carboxylated color-coded microspheres (Luminex Corp., Austin TX, USA), according to the protocol of the Luminex corporation (Austin TX, USA). Briefly, 2.5 Â 10 6 stock microspheres were collected by centrifugation and resuspended in 25 ml of 0.1 M MES, pH 4.5 (2N-morpholino-ethanesulfonic acid, Sigma). Subsequently, 0.2 nmol of the probe and freshly made 2 ml 10 mg/ml EDC [1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (water-soluble carbodiimide; Pierce; sold by Nordic Biolabs AB, Sweden)] in H 2 O were added to the microspheres and the suspension was incubated in the dark for 30 min at room temperature. Care was taken to store the EDC in a dry condition, in aliquots. After addition of another 2 ml (10 mg/ml) EDC in H 2 O and repeated incubation, the microspheres were washed with 0.5 ml of 0.02% Tween-20. The coupled microspheres were pelleted by centrifugation at 8000 Â g for 2 min and resuspended in 0.5 ml of 0.1% SDS. After a second spin at 8000 Â g for 2 min, the final pellet was resuspended in 50 ml of TE, pH 8.0. An amount of 5 ml of 2.0 nM synthetic biotin-labeled target was mixed with hybridization buffer consisting of 33 ml 3 M TMAC buffer (3 M tetramethylammonium chloride, 0.1% Sarkosyl, 50 mM Tris-HCl, pH 8.0, 4 mM EDTA, pH 8.0; Sigma) and 12 ml 1Â TE-buffer pH 8.0 and 0.05 ml ($2500 microspheres) for each probe-coupled Luminex microsphere. The mixture was heated at 95 C for 2 min to denature the DNA targets and probes, followed by hybridization at 45 or 55 C for 30 min while shaking on the Thermostar (BMG LabTech; Offenburg, Germany) microplate incubator. An amount of 2 ml (0.05 mg/ml) of streptavidin-R-phycoerythrin (QIAGEN, Hilden, Germany) was added to the mixture, which was further incubated at 45 or 55 C for 15 min before analysis for internal microsphere and R-phycoerythrin reporter fluorescence on the Luminex Õ 200 TM system (Luminex corporation, Austin, TX). The amount of biotinylated target that hybridizes to the microsphere-bound probes is directly proportional to the Median Fluorescence Intensity (MFI) reported by the instrument (in all experiments, fluorescence was measured from a minimum number of each type of microsphere: a set of 100 beads). The term total MFI describes the hybridization signal from a perfectly matching probetarget duplex, while the percentage (%) of the total MFI describes the ratio between the hybridization signal of a mismatching probe-target duplex and the total MFI of that particular probe. Titration experiments established that $60% of the maximum hybridization capacity of the microsphere-bound probes was reached using the conditions under which total MFI was measured (i.e. microsphere-bound probes were not saturated). An MFI of 100 was used as the lower limit of detection (LLOD). Fecal samples were obtained from children with gastroenteritis, from the Children's Hospital ward at Uppsala Academic Hospital. Samples were handled anonymously according to the rules of the ethical committee at the Academic Hospital. The study used samples that were positive in a rotavirus antigen detection test; 100 ml of the sample was diluted in 900 ml 1Â TE buffer. After centrifugation, 400 ml was added to a lysis buffer and total nucleic acid was extracted as described by the manufacturer (easyMag Õ , bioMe´rieux). The samples were eluted in 110 ml buffer and stored at À70 C. A reverse transcriptase (RT)-PCR was set up to amplify and biotin-label the nucleic acid of human rotavirus A from these clinical samples. The 545 sequences obtained from the rotavirus query in BLASTn were analyzed in the ConSort ß program to construct fw-and rev-primers. The fw-primer (nt 1-23), 5 0 -G GCTTTW+AAA+CGAA+GTC+TTCR-3 0 (+A, +C, +G, +T are LNA residues) and the biotinylated rev-primer (nt 502-426), 5 0 -TATGGAAATATATTAGG TTTATGAAAAACAAATCCIGTACGTTGTCTTCT ITTITGIARRTTCCAITTITCIATRTA-3 0 , resulted in a PCR product of 502 nt. After nucleic acid extraction in the easyMag Õ , the extracts were heated at 97 C for 5 min, followed by snap cooling on ice for 2 min to obtain ssRNA from the rotaviral dsRNA. The PCR reaction contained 5 ml of nucleic acid extract, 25 ml 2Â RT-PCR iScript buffer, 200-nM Fw-primer, 600-nM biotinylated rev-primer, nuclease-free water and 1 ml iScript RT enzyme (total volume 50 ml). The samples were run at 50 C for 30 min, 94 C for 10 min, 50 cycles at 94 C for 30 s, 55 C for 30 s and 72 C for 30 s, ending with 72 C for 7 min. An amount of 5 ml of the PCR product was used in the hybridization experiment with the microsphere-coupled Rotavirus probes (sequences shown in Supplementary Table S4A and B. The PCR products were sequenced with a 3130 Genetic Analyzer (Applied Biosystem) by utilizing the same fw-and rev-primers described above. The discovery of the importance of long uninterrupted perfect matches and the long-range effects of mismatch (see 'Results' section), which are not embodied in current NN hybridization theories (19) , led us to formulate a simple new descriptive theory. The new algorithm includes aspects of NN theory (8, (20) (21) (22) but extends this to longer hybridizing segments and includes the effects of dInosine in the long oligonucleotides. Visual Omp TM 7.0 predicted that some 70-mer combinations would not hybridize, yet they did hybridize. In order to better predict hybridization, we investigated predictive strategies that took into account the matching nt, its neighbors, the length of the matching region, and the cooperativity of neighboring matching regions. We finally settled for an algorithm which attempts to model the nucleation and zipping stages during the hybridization process. The new model was termed NucZip ( Figure 5B ). The results obtained using this model was then correlated with those obtained using Visual OMP TM 7.0 and both were compared with experimental data. NucZip simulates the hybridization process, starting with potential nucleation sites and then proceeding to 'zip' in both directions. The NucZip algorithm (written by JB in Visual FoxPro) was implemented as a module (procedure) in the ConSort ß sequence analysis program. The procedure is relatively simplistic, and does not contain the thermodynamic and secondary structure analyses provided by more sophisticated programs such as Visual OMP TM 7.0. The algorithm starts by searching for perfectly matching hexa-, hepta-, octa-and nonamers, as potential nucleation sites (the 'Nuc' part). Every matching oligomer is given the number of matching nt as its NucScore. If the oligomer contains dInosine, the number of dInosines is subtracted from the score but the segment is still counted as uninterrupted, regardless of the dInosines. The NucScore is then used to select the two highest scoring potential nucleation sites which will undergoing zipping, up and downstream. The score for the 'Zip' portion of the model is obtained from the number of consecutive matching trimers, tetramers etc, up to pentadecamers, each counted with equal weight, within a contiguous matching segment. Thus, ZipScore modex = AE k = 1 k = kmax (AE n = 3 n = 15 S n ), where kmax is the number of uninterrupted matching segments, including the chosen nucleation site, and S n is the number of successive segments of length n (varying from 3-15), i.e. the number of full length trimers, full length tetramers etc. up to full length pentadecamers, which fit into the matching segment. The same Zip scoring system was performed in two modes, counting dInosines either as matching (mode 1) or as mismatching (mode 2). In the second mode, dInosines will shorten the length of matching segments, decreasing the score. The final ZipScore was calculated as a weighted mean of ZipScore mode1 and ZipScore mode2 , where the weighting factor was based on the empirical data presented in the current report (Figures 2-4) . Since dInosine hybridizes more strongly to C than to the other nts, the algorithm adds a contribution based on the number of dI:C pairs weighted by an InoCfactor. The upstream and downstream ZipScores were obtained as: ZipScore downstream =ZipScore mode1 -dInosinefactor (ZipScore mode1 -ZipScore mode2 )+(dInoCnr * dInoCfactor) and ZipScore upstream = ZipScore mode1 -dInosinefactor (ZipScore mode1 -ZipScore mode2 )+(dInoCnr * dInoCfactor). The ZipScores from up-and downstream zipping were then added; the final NucZipScore = ZipScore downstream+ZipScore upstream. In this way, the contribution from longer matching segments was factored in with the contribution from the nearest neighbors (approximated by the trimer part of the algorithm). Figure 8 summarizes the principle of this computational work. The probability of a match in a degenerated nt position is approximately predicted from the ConSortß analyses of target sequences. The behavior of the relatively few oligonucleotides with degeneration that were tested is approximately in line with NucZip reasoning, which places a premium on long uninterrupted matching stretches. However, a larger number of degenerated probes need to be analyzed before the contribution of probabilities of contiguous stretches extending through degenerated positions can be estimated and included into NucZip. The programming code is included in Supplementary Data. ConSortß was used to demonstrate the variation in the nt sequence of segment 7 from 7333 genomes of Influenza A (GeneBank database at NCBI, NIH). All H and N influenza A types, HxNy, are represented in the alignment ( Figure 1 ). In comparison with the InflA probe designed to match a 70-nt region of the Influenza A H3N2 virus, the H5N1 virus differed in 5-nt positions, and the H1N1 virus in three other nt positions (Table 1) . Thus, if a detection probe could tolerate mismatches in nine positions, including the variant nt positions of the H1N1 and the H5N1 viruses, it would fully cover 67 different H and N combinations of Influenza A, i.e. nearly all recorded variants in the chosen region, as demonstrated in the BLASTn search. The InflA probe was tested against 70 nt target molecules with 3, 5, 7, 9, 11, 12, 13, 14, 15, 16 and 21 point mutations (pm) and 21 grouped mutations (gm) (Figure 2A -D; nt sequences of probes and targets can be found in Supplementary Tables S1A and B). The positions of the pm were based on the variations found by comparing H3N2, H5N1 and H1N1 viruses. Two targets had the same number of mutations, but different distributions: the 21 pm target had 21 evenly distributed pm, and the 21-gm target had seven groups of three mutations interspersed by 5 to 7 conserved nt. The InflA probe, coupled to Luminex microspheres, was allowed to hybridize with one of the biotinylated ssDNA targets in 3 M TMAC at two different temperatures: 45 C ( Figure 2A ) and 55 C ( Figure 2B ). The hybridization, given as MFI, was analyzed in the Luminex flow meter. Introducing an increasing number of evenly distributed pm in the target had a negative effect on hybridization, as reflected in decreasing MFI; see Figure 2A 12-nt, one 7-nt and four 5-nt matching regions) still hybridized at 68% (MFI 4496, 45 C, Figure 2A ) and 71% (MFI 4001, 55 C, Figure 2B ) of the total MFI (i.e. of the MFI of the perfectly matching InflA target/InflA probe: MFI 6604, 45 C, Figure 2A ; and MFI 5657, 55 C, Figure 2B ) while the 15 pm target (containing one 6 nt, four 5 nt, and two 4 nt matching regions) hybridized at 13% (MFI 854, 45 C) and 1% (MFI 54, 55 C) of the total MFI. The InflA probe failed to hybridize with the 16-pm target (containing one 6-nt, three 5-nt, and two 4-nt matching regions) at either temperature, providing MFI values that were 6% (MFI 403, 45 C) and 0.5% (MFI 29, 55 C) of the total MFI. The longest perfectly matching sequence between the mismatches in the InflA probe/21-pm target combination was 3 nt in length. No hybridization was detected at either temperature. In contrast, for the InflA probe/21-gm target combination, where the distribution of the 21 mutations created seven stretches of 5-7 perfectly matching nt between the mismatches, hybridization with the InflA probe was restored (Figure 2A-D) . The MFI signal increased to 20 and 30% (Figure 2A Table 1 sequence of H3N2). The number of dInosines included in the probes is indicated in the names Ino3-Ino21. Similarly, the wobbN_21 probe contains 21 N wobbles in the same position as the dInosines in the Ino21 probe. The complementary target of the InflA probe is named the InflA target. These targets have 0-21 pm, as indicated by the names; the 21-gm target has seven groups of three mismatches ( Figure 2D ). All target molecules were biotinylated at their 5 0 end. Each sample contained 0.2 nM of one of the synthetic biotinylated Continued and 7 and 8% ( Figure 2B and C, 55 C) of the total MFI, when the InflA probe hybridized with the 21-gm target instead of the 21-pm target. In conclusion, a target with one to nine evenly distributed mismatches, preserving multiple contiguous matching stretches of at least 5 nt, has little reducing effect on hybridization with a 70-mer probe, while a target with >14 evenly distributed pm hybridizes inefficiently or not at all. Hybridization can be improved by utilizing a less stringent (lower) hybridization temperature. Furthermore, grouped mismatches tend to strengthen the hybridization compared to evenly distributed mismatches. Consequently, for hybridization between 70-mer strands with 10-20 mismatches, the distribution of the mismatching nt affects the hybridization more than the number of mismatches, indicating that the length and number of perfectly matching stretches are of greatest importance. A panel of five 70-mer probes, Ino3-21, containing 3, 5, 7, 9 or 21 dInosines, was designed based on the InflA probe sequence. The dInosines were placed to match the positions of the pm in the above pm targets ( Figure 2D ). Introduction of 3-9 dInosines in the probe resulted in only a small reduction in MFI when binding to the InflA target; i.e. the MFI signal decreased in the order InflA probe & Ino3 > Ino5 > Ino7 > Ino9. The Ino9 probe hybridized with the InflA target by as much as 76% (MFI 4970, 45 C) and 59% (MFI 3340, 55 C) of the total MFI, comparable with the hybridization of the InflA probe to the 9-pm target. In fact, all the Ino3-9 probes hybridized as efficiently with targets that had up to the same number of pm as the number of dInosines, including two mismatches not covered by dInosines, as they did with the InflA target. When the Ino probes hybridized with targets containing more than 12 pm, the probes with many dInosines worked better than the InflA probe; e.g. the Ino7 and Ino9 probes resulted in a signal 1.2-1.9 times higher than that for the InflA probe for targets with 13-16 pm at 45 C. Interestingly, even the Ino21 probe hybridized quite strongly with the InflA target, at 26 and 39% (Figure 2A and C, 45 C) and 11 and 13% ( Figure 2B and C, 55 C) of the total MFI. Importantly, the Ino21 probe was able to restore hybridization with the 21 pm target (11 and 37% of total MFI) to which the InflA probe had totally failed to bind (Figure 2A and C, 45 C). The Ino21 probe hybridized with all the matching 3-21 pm targets with almost the same efficiency (mean 29.5% of the total MFI, 1946 MFI, at 45 C, Figure 2A; and 8% of the total MFI, 435 MFI, at 55 C, Figure 2B ), which is in the same range as the hybridization of the InflA probe with the 21-gm target (30.4% of the total MFI, 2005 MFI, 45 C, Figure 2A ; and 7.8% of the total MFI, 444 MFI, 55 C, Figure 2B ). The Ino21 probe and 21 gm target combination had 13 mismatches outside the dInosine positions and failed to hybridize (Figure 2A , B and C). In conclusion, the presence of dInosine in the probe decreased hybridization with a perfectly matching target, but a dInosine-containing probe bound more strongly than a dInosine-free probe when the target had many mismatches juxtaposed to the dInosine residues. A minimum length of perfectly matching nt sequences is required for hybridization The importance of uninterrupted matching regions was analyzed by making targets with different lengths of perfectly matching sequences at the 5 0 end or at both the 5 0 and 3 0 ends, in combination with a long region of 26, 33 or 74% randomly distributed mutations ( Figure 3A , B and D, nt sequences in Supplementary Table 2A and B) . As expected, hybridization of the InflA probe to targets with 26% random mutations in a region between two flanking regions of 5 (26%5F), 7 (26%7F), 9 (26%9F) or 15 (26%15F) perfectly matching nt showed that shorter perfectly matching flanking regions reduced the MFI. All the 26% targets hybridized with the InflA probe at 45 C but, at 55 C, the 26%5F target failed to hybridize (5.6% of the total MFI, 271 MFI). Like the 33%12F and the 16-pm target, the 26%5F target contained 16 mutations (Figures 2A, B, 3A and B). The 33%12F target (45% of the total MFI, Figure 3A ), with its two regions of 12 uninterrupted nt, hybridized much more strongly than the 26%5F (27% of the total MFI) or 16-pm (6.1% of the total MFI) targets, which both contained several shorter matching regions, of 6, 5, 4 and 3 nt ( Table 2) . The same effect, i.e. that a few long perfectly matching regions result in better hybridization than several shorter regions, was also seen with the three targets that had 14 mismatches: the 26%9F, the 33%15F and the 14-pm targets at both temperatures ( Table 2 ). The 74%9F target, with only 12 matching nt dispersed in the central region, did not hybridize to the InflA probe (1.4% of the total MFI, 86 MFI, 45 C; and 0.6% of the total MFI, 30 MFI, 55 C), while the 26%9F target, with 38 matching nt (three regions of 5 nt and one region of 6 nt), did hybridize (44% of the total MFI, 2583 MFI, 45 C; and 24% of the total MFI, 1125 MFI, 55 C). This shows that the two flanking regions of nine matching nt did not cause hybridization alone. Furthermore, the 74%12F, with 11 matching nt dispersed in the central These experiments, taken in conjunction with the InflA probe hybridizing to the 21-gm but not to the 21-pm target, confirm that both the number of mismatches and their distribution are important. It is reasonable to assume that perfectly matching sequences of a minimum length function as nucleation sites (15) which initiate hybridization between the probe and the target. Hybridization of the InflA probe with targets containing 74% mismatching 70-mers with one perfectly matching end of various lengths (74%xnt) was compared with hybridization of the InflA probe with short, perfectly matching targets (12-22 nt_free; Figure 3A , B and D; nt sequences in Supplementary Table S2B ) to analyze the effect of long mismatching ends. Utilizing the InflA probe / InflA target as reference, the 12nt_free target did not hybridize at 45 C (1.8% of the total MFI, 106 MFI) but the 15-, 18-and 22-nt_free targets gave successively stronger MFI signals (62, 98 and 123% of the total MFI; 3669, 5770 and 7269 MFI, respectively) than the 74%15nt, 74%18nt and 74%22nt targets (17.4, 55 and 62% of the total MFI; 1029, 3222 and 3637 MFI, respectively). Thus, the MFI is higher when the short, perfectly matching targets (xnt_free) hybridize with the InflA probe with only one long end protruding from the hybridized portion of the probe. Previous reports have demonstrated that 1-5 nt single dangling ends tend to stabilize duplex formation (23) (24) (25) (26) . This study shows that two long mismatching ends destabilize the hybridization of the matching part of the duplex. The long free mismatching ends could form intramolecular secondary structures that could have an effect on the hybridizing duplex. % MFI (InflA probe against target/InflA probe against InflA target). Alternatively, the Brownian movements of the two long non-hybridized sections could mechanically stress the remaining base pairs. The tolerance of the probe against sm was tested using targets (70 nt) with every third nt harboring an sm (referring to the reading frame of the matrix 2 protein of the H3N2 Influenza A) in regions of different length. The 33%9F, 33%12F and 33%15F targets have a region containing 18, 16, or 14 sm between two flanking regions of 9, 12, or 15 perfectly matching nt, respectively. The 33%9nt, 33%12nt and 33%15nt targets have a region of 20, 19 or 18 sm in combination with one region of 9, 12 and 15 perfectly matching nt at the 5 0 end ( Figure 3D , nt sequences in Supplementary Table 2A and B) . As demonstrated in Figure 3A and B (and confirmed in Figure 3C ), the targets 33%9nt and 33%9F failed to hybridize with the InflA probe, while the 33%15F target hybridized at both hybridization temperatures (60% of the total MFI, 3517 MFI, at 45 C; 39% of the total MFI, 1934 MFI, at 55 C). The 33%12F target with its two 12 nt flanking regions only hybridized at 45 C (45% of the total MFI, 2635 MFI). The 33%15nt, with one contiguous region of 15 perfectly matching nt also hybridized only at 45 C (23% of the total MFI, 1352 MFI; MFI result taken from Figure 3A and B). Thus, for a 70-mer probe to hybridize with a target containing a relatively long stretch of sm, it must have (i) one uninterrupted perfectly matching region of at least 15 nt at 45 C, and longer than 15 nt at 55 C, or (ii) two uninterrupted matching regions of at least 12 nt at 45 C or 15 nt at 55 C. dInosine-containing probes restore hybridization to targets containing sm The targets containing sm were further tested against a set of dInosine-containing probes: one probe (the Ino18 probe) contained 18 dInosines matching the sm in the 33%9F target, two probes (21Ino_9nt5 0 and 21Ino_9nt3 0 ) contained 21 dInosines positioned as sm leaving a region of matching 9 nt at either the 5 0 or 3 0 end; and one probe (Ino24) contained 24 dInosines in every third base throughout the whole 70-mer probe. Importantly, when dInosines matching the positions of the sm in the targets were included in the probes, the Ino18 probe hybridized with all the sm-containing targets, although with slightly varying MFI ( Figure 3A , B and confirmed in C). The 33%9F target, which did not hybridize with the InflA probe, was able to hybridize with the Ino18 probe (66% of the total MFI, 3894 MFI at 45 C, Figure 3A ; 55% of the total MFI, 2808 MFI at 55 C, Figure 3B ). Even the 33%9-15nt targets, with two mismatching nt at the 3 0 end not covered by dInosines, hybridized well with the Ino18 probe (48-51% of the total MFI, 2818-2994 MFI at 45 C, Figure 3A ; and 16-32% of the total MFI, 778-1463 MFI at 55 C, Figure 3B) . Thus, as shown in the previous series of pm matched to dInosine, when the sequence is interrupted too frequently by mismatches, leaving no suitable regions of perfect match, a probe with dInosines at the positions of variation will restore the hybridization by effectively creating the required longer matching region. Introducing a dInosine in every third position throughout the Ino24 probe decreased the hybridization dramatically for all the 33% targets (6-16% of the total MFI, 361-920 MFI at 45 C; 1% of the total MFI, 53-79 MFI at 55 C). When utilizing the 21Ino_9nt5 0 probe with the 33%xnt targets, the MFI decreased because of the mismatches created by the two sm at the 3 0 end of the target (4-10% of the total MFI, 232-415 MFI at 45 C). In comparison, the 21Ino_9nt5 0 probe and the 33%xF targets, with no mismatch, hybridized at 26-43% of the total MFI (1692-2898 MFI at 45 C). Furthermore, the 21Ino_9nt3 0 , which did cover the two sm at the 3 0 end of 33%xnt, hybridized more strongly (10-21% of total the MFI, 450-695 MFI at 45 C) than the 21Ino_9nt5 0 probe (4-10% of the total MFI, 232-415 MFI at 45 C). Both these results demonstrate that using the less stringent temperature (45 C) permits hybridization even when a large number of Inosines is present (21 dInosines in a 70-mer probe), as long as a matching region of at least 9 nt is formed in the hybrid. The hybridization of a long probe containing dInosines is comparable with that of a long degenerated probe with the same number of N wobbles, under lower stringency conditions The effects of probes with dInosine or wobbles in the same positions were also investigated in 3M TMAC. The presence of a dInosine in a specific position instead of a wobble would theoretically decrease the degeneration of the probe and subsequently increase the concentration of the particular probe variant. A probe with 21 N wobbles, wobbN_21, at the same positions as the dInosines in the Ino21 probe, was tested. The surprising result was that the probe containing N wobbles hybridized very well with the InflA target (29% of the total MFI, 2220 MFI, Figure 2C ) and the 21-pm target (29% of the total MFI, 2190 MFI, Figure 2C ). This is in the same range as hybridization of the Ino21 probe with the InflA (26% of the total MFI, 1976 MFI, Figure 2C ) and 21-pm (11% of the total MFI, 794 MFI, Figure 2C ) targets (39 and 37% of the total MFI versus the InflA and the 21-pm targets, see Figure 2A ). These results also demonstrate that the wobb_N21 probe is not affected to the same extent as the Ino21 probe by increasing the hybridization temperature from 45 to 55 C ( Figure 2C ). The test was repeated by comparing an Ino18 with a wobbN_18 probe ( Figure 3C and D) . At 45 C, the Ino18 probe hybridized at least as well as the wobbN_18 probe while, at 55 C, the wobbN_18 probe hybridized better than the Ino18 probe. Interestingly, a probe containing 24 wobbles still hybridized better with all 33%xnt and 33%xF targets (30-45% of the total MFI at 45 C; 13-22% of total MFI at 55 C; Figure 3C ) compared with Ino24 (6-16% of the total MFI at 45 C; 1% of the total MFI at 55 C; Figure 3A and B). Obviously, the 70-mer probes can accommodate multiple degenerate positions and still hybridize because the majority of probe molecules will contain several long perfectly matching stretches created by chance. This is further deliberated under Discussion section. The hybridization of a long probe containing dInosines is stronger than that of a long probe containing the same amount of 5-Nitroindole, at either high or low temperatures 5-Nitroindole is a second-generation universal base nt analog that was chosen for comparison with the first-generation dInosine with respect to hybridization properties in 3M TMAC. According to Loakes and Brown (1994) , 5-nitroindole is less destabilizing than its 4-and 6-isomers (27) and than 3-nitropyrrole (9). A probe with 18 5-nitroindole residues (5-NitroInd_18) was designed; the nt analogs were distributed to match the pattern of the dInosines in the Ino18 probe ( Figure 3D ). The probes were allowed to hybridize with the InflA target and the set of targets with sm, 33%_xF and 33%_xnt ( Figure 3C ). At 45 C, hybridization of 5-NitroInd_18 with the InflA (44% of the total MFI), 33%_xF (23-37% of the total MFI), and 33%_xnt (4-7% of the total MFI) targets resulted in hybridization signals that were much lower than those seen with the Ino18/InflA (73% of the total MFI), Ino18/33%_xF (54-78% of the total MFI), and Ino18/33%_xnt (47-59% of the total MFI) probes ( Figure 3C, 45 C) . Increasing the temperature to 55 C destabilized the 5-NitroInd_18 probe even more, resulting in hybridization of only 1-8% of the total MFI. In conclusion, dInosine functions much better than 5-nitroindole as a universal nt analog, under 3 M TMAC buffer conditions. Hybridization of a probe containing dInosine is sensitive to mismatches neighboring the dInosine position, aiding specificity It has been shown above that when the dInosine and the mismatch have the same distribution pattern, i.e. the dInosine masks the mismatch, hybridization can be restored (Figures 2A, B, 3A and B). We analyzed how many mismatches outside the rescuing position of dInosine (mismatch outside dInosine; mmoi) a probe can tolerate. Norovirus, Ino18 and InflA probes were tested against a set of targets whose sequences were designed to range successively from a Norovirus sequence to the InflA sequence, allowing different amounts of mismatch and mmoi to be analyzed. Norovirus is a highly variable, positive-sense RNA virus belonging to the Caliciviridae, which causes 'winter vomiting disease'. The Norovirus sequence chosen (the capsid gene of the Norwalk-like virus, accession nr AY274264), after Blastn with the InflA probe sequence, has a short region of 8 nt that perfectly matches the end region of the InflA probe and has 10 dispersed matching nt ( Figure 4A ). The Norovirus target (70) 0_36_52 (0.8) (this code is explained below, and in the legend to Figure 4 ) was gradually changed to resemble the InflA target (0.8) 51_0_0 (70) by altering the central nt sequence. A set of targets was also created where the nine nt at the 5 0 end were changed into an InflA sequence and the central region was gradually changed from the Norovirus to the Influenza sequence, starting with (0.61) 9_27_42 (9.8). The targets were named according to the number of matching and mismatching nt in comparison with the three probes: (nt matching those of the Norovirus probe at 5 0 and 3 0 ) mismatching nt versus the Norovirus probe_Ino18 probe_InflA probe (nt matching those of the InflA probe at 5 0 and 3 0 ) (see Figure 4A ; nt sequences in Supplementary Table S3A and B) . Two targets had 26 mismatches, with different dispersion patterns, after hybridization with the Norovirus or InflA probes; the distributions are shown in the (10.10) 26_10_26 (0.8) and (0.11) 26_10_26 (9.8) targets in the upper and lower panels of Figure 4A . The InflA probe did not hybridize with either of them, while the Noro probe hybridized weakly with both: 13% of the total MFI for the (0.11) 26_10_26 (9.8) target and 20% of the total MFI for the (10.10) 26_10_26 (0.8) target. The% MFI for the Noro probe was calculated by comparing the MFI with that of the Noro probe/Noro target hybridization. The stronger hybridization to the Noro probe than to the InflA probe can be explained by the distribution and length of the matching sequences between the 26 mismatches; the perfectly matching regions of 8 and 9 nt were not long enough to induce hybridization to the InflA probe, while one region of 11 nt (together with a 5-nt and a 6-nt region) or two longer flanking regions of 10 nt in the two targets was enough to induce hybridization with the Noro probe. When the Ino18 probe (middle panel of Figure 4A ) was used with a target with 1 mmoi [i.e. (0.10) 35_1_17 (9.8)] or even 10 mmoi [(10.10) 26_10_26 (0.8)] located at the 5 0 and 3 0 ends of the target, outside the central region containing dInosines, there was no or little inhibition of hybridization at 45 C (71% of the total MFI, 4741 MFI, and 41% of the total MFI, 2773 MFI, respectively). Interestingly, when the 10 mmoi were evenly distributed within the region of 18 dInosines, as in the (0.11) 26_10_26 (9.8) target, hybridization was lost (0.9% of the total MFI, 59 MFI). The sensitivity to mmoi adjacent to dInosine was therefore investigated further. Targets with increasing numbers (2, 4 or 5) of mmoi neighboring the positions of dInosines successively reduced the hybridization signals: 2319 MFI (35% of the total) for the (10.10)_24_12_28_(0.8) target, 927 MFI (13%) for the (10.10)_22_14_30_(0.8) target, and 375 MFI (5.4%) for the (10.10)_21_15_31_(0.8) target, all at 45 C. Thus, dInosine is sensitive to neighboring mismatches. A comparison of the sensitivity of a dInosine-free probe (InflA) and a probe containing dInosine (Ino18) to neighboring mismatches showed that the dInosine-free InflA probe can hybridize to a target with 17 evenly distributed mismatches between two perfectly matching flanking sections of 9 and 8 nt, respectively [the (0.10) 35_1_17 (9.8) target, 24% of the total MFI, 1614 MFI at 45 C]. However, 7 mmoi adjacent to the dInosines completely destroyed hybridization between the Ino18 probe and the (10.10) 19_17_33 (0.8) target: 0.5% of the total MFI, 34 MFI, at 45 C. Results for 55 C are shown in Supplementary Figure S4B and Table 3 . The Ino18 probe failed to hybridize when 2 mmoi were placed next to the dInosines PAGE 11 OF . Thus, the hybridization capacity of a dInosine-containing sequence is severely reduced when the mismatch is adjacent to the dInosine. Figure 4A also shows how introduction of many dInosines affects the specificity. There was no hybridization between the influenza probe Ino18 with its 18 dInosines and the Norovirus target (70) 0_36_52 (0.8). Furthermore, although the dInosines mask 16 mismatches in the (0.11) 26_10_26 (9.8) target, the 10 mmoi that are in close proximity to the dInosines abolish hybridization (0.9% of total MFI, 59 MFI). In contrast, the other target containing 26 mismatching nt and 10 distant mmoi, (10.10) 26_10_26 (0.8), did hybridize to the Ino18 probe (41% of total MFI, 2773 MFI). Thus, the cross hybridization of a foreign (unrelated) dInosine-containing probe is dependent to a certain extent on the amount of mismatch, but is even more dependent on the distribution of mmoi. Figure 4B and Supplementary Table S6 show the origin of the nt that are not covered by the dInosines when using the Ino18 probe (MFI values from Figure 4A ). They demonstrate the number of InflA-matching nt outside the dInosine position (moi) needed for hybridization and the number of Norovirus moi causing cross hybridization. At least 37-38 InflA moi were needed to induce hybridization with the Ino18 probe but, as mentioned above, the distribution is at least as important as the actual number of matching and mismatching nt. Of the fewer than 30-31 nt that were of Norovirus origin in a target that hybridized with the Ino18 probe, 16 nt were common to both Norovirus and Influenza virus. If more than 31 nt were of Norovirus origin, hybridization to the Influenza Ino18 probe failed. The region chosen for the 77-nt rotavirus probe, positions 112-178 in the alignment, was analyzed using the haplotype function of ConSortß, which decomposes highly variable stretches into a small number of less variable stretches (haplotypes). This resulted in two major haplotypes and probes which potentially covered all rotavirus group A variations recorded in GenBank. The two haplotype probes contained 14 and 8 dInosines, respectively, in combination with four degenerations. They were called Ino14_w4 and Ino8_w4 ( Figure 5C , Inosine as yellow and wobbles as light grey boxes). Two additional probes with fewer dInosines and more degenerations were also created: Ino11_w7 and Ino5_w7 ( Figure 5C ). The consensus sequence and the pattern of variation of the region chosen for the probe are shown in Figure 5A . The sequences of the four probes are shown in Supplementary Table S4A and B and (schematically) in Figure 5C . The degenerated LNA-containing fw-primer and the long degenerated dInosine-containing biotinylated rev-primer generated a single band of the correct size, 502 nt, when analyzed by electrophoresis using EtBr-stained agarose gel in all five clinical samples (data not shown). All four microsphere-bound probes detected the consensus synthetic rotavirus target as well as the amplified rotavirus nucleic acid from all five clinical samples. Interestingly, the four probes hybridized almost equally well within each sample. It was found that an asymmetric PCR of the clinical samples was necessary in order to obtain an MFI of reasonable strength from the probes (data not shown). This was probably because the complementary strand of the PCR product outcompeted the probe due to an affinity between the two strands that was higher than that between a dInosine-containing degenerated probe and the target strand. The data in Figure 5B are from one of the experiments using samples run in duplicate. The PCR products were sequenced (Supplementary Table S4 ), revealing that the Ino14_w4 and Ino11_w7 probes, which belonged to the same haplotype, covered the variations in all positions in all samples. However, the other pair of probes (Ino8_w4 and Ino5_w7) had one mismatch against clinical samples 1 and 2 and two mismatches against clinical samples 3, 4 and 5, as well as the consensus synthetic rotavirus target (magenta colored boxes in targets in Figure 5C , Supplementary Table S4 ). In conclusion, the long dInosine-containing degenerated probes worked well as variation-tolerant probes, covering variations, accepting a few mismatches, and still remaining specific (neither of the Rotavirus probes hybridized with the InflA target). Once we had these experimental data, we tried to develop a unifying view of them. The ÁG predicted by the Visual Omp TM 7.0 software was compared with the percentage of the total MFI for each probe and target combination, including dInosine-containing probes (Figure 2A , 3A Figure 4B shows the total number of nt in each target that originated from either Noro (white triangles) or InflA (grey squares) and that are outside the position of the 18 dInosines when hybridized with the Ino18 probe, compared with the MFI (from Supplementary Figure S1A) for each combination. and 4A). For the sake of simplicity, data from 5-nitroindole and N-wobble-containing probes were omitted. The results demonstrated that some probe/ target combinations that hybridized well in practice, had very low predicted ÁG values ( Figure 6B ), e.g. the InflA probe hybridizing to the 74%12F (43% of the total MFI, ÁG = -20.62), 74%15F (77% of the total MFI, ÁG = À26.7), 74%18 nt (55% of the total MFI, ÁG = -29.24), and 74%22 nt (61% of the total MFI, ÁG = -34.55) targets. When the results of the new NucZip scoring system were scored against the % MFI in Figure 6A , which shows all the target and probe combinations plotted in Figure 6B , it was found that they were more highly correlated with the experimental data than the predicted ÁG. To investigate these differences, each outlier in Figure 6B was connected to its plot position in Figure 6A ; see Figure 7A and B. Figure 7 shows that probe-target combinations containing many mismatches and dInosines were the main causes of the lower correlation between predicted and observed hybridization in Figure 6B . However, hybridizations between a long probe and a short target were not included in Figures 6 and 7 . Nor were data from probes containing 5-nitroindole or N-wobbles, because a full investigation such as this would require many more observations and would be out of the scope of this article. The NucZip results are further discussed in the Discussion section. Thus, when the actual degree of hybridization for the entire data set (265 probe-target combinations) was matched with the predicted ÁG in Visual Omp TM 7.0, an only moderately precise correlation was obtained. The hybridization of combinations involving many mismatches and many dInosines was poorly predicted. However, when the NucZip algorithm was used, a higher degree of correlation was observed. The adjusted determination coefficient (R a 2 ) was 0.8636, indicating that 87% of the variation was explained by the NucZip algorithm, while the best fit of MFI% to the Visual Omp TM 7.0 predictions gave a determination coefficient of 0.7505, indicating that 75% of the variation was explained by NN theory (as embodied in Visual Omp TM 7.0). NN theory was thus insufficient for predicting hybridization under the hybridization conditions of our study. A high number of mismatches and dInosines gave hybridization predictions in Visual Omp TM 7.0 that were too low ( Figure 6B ). The NucZip algorithm, which takes into account the length of matching segments and cooperativity effects within and between matching oligonucleotide segments, increased the accuracy of hybridization prediction. dInosines were scored intermediate between matches and mismatches. Other hybridization prediction algorithms are available on the Internet. However, when we compared the delta G predictions obtained from IDT Oligo Analyzer (http:// eu.idtdna.com/analyzer/Applications/OligoAnalyzer/), which uses a proprietary algorithm, with our experimental data, the correlation was poor (Supplementary Figure 2) . Although the exact experimental conditions (3M TMAC and 45 C) were not represented, this is not likely to have caused the low correlation. Nucleic acid hybridization is fundamental to many molecular biology applications, and is expected to grow in significance as nanomedicine joins molecular medicine at the cutting edge of research (28) . In particular, biomedical applications of hybridization such as detection of variable viral target sequences are highly dependent on a precise understanding of the process involved. A probe that has a broad detection spectrum should be as specific as current narrower probes while retaining the ability to cover the biologically or clinically relevant sequence variants of specific microbes. The design of long mismatch-tolerant probes demands knowledge about hybridization in the presence of mismatches, degeneracy and nt analogs. In pursuit of this level of understanding, and in order to obtain reliable hybridization data, we chose to use the Luminex suspension array system in our studies. The inherent ability of the system to report the median of a high number of measurements (i.e. measuring hybridizations signals from 100 different beads) provides highly reliable data. Moreover, hybridization equilibrium is reached more rapidly using the suspension array system (taking around 15 min) than by solid phase hybridizations such as micro arrays (often overnight). A probe length of 70 nt was selected for our studies because of the elevated mismatch tolerance of this length compared with shorter probes (29) . However, the advantage of the extended length of the probe could possibly be countered by a loss of specificity. Hybridization studies using long (50 or more nt) probes in a 3 M TMAC buffer system have not been reported previously; in reports using other hybridization systems, however, it has been suggested that 50-mer or 70-mer probes should contain no more than 15-20 contiguous nt complementary to non-targets (29, 30) . Previously, the hybridization properties of long probes have been analyzed using microarrays, with overnight (16) and ConSortß, to find suitable regions for primers to be used in reverse-transcription PCR and a conserved region for a detection probe. The length of the black bars represents the frequency of variation as an average percentage conservation at each nt position (y-axis). The figure shows the alignment and variation of 214 Rotavirus A sequences in the nt position of the probes displayed in Figure 5C . ConSortß was used to group the variations of the probe region into two haplotypes, which were then used to construct the probes shown in Figure 5C and Supplementary Table S4A Letowski et al. found that, under microarray conditions, mismatches grouped at the 5 0 or 3 0 end of a 50-mer probe affected the binding to a target less than if the mismatches were distributed throughout or centered in the probe. Furthermore, the 50-mer probes with mismatches distributed along the whole probe were more destabilized than the probes that had mismatches centered in the duplex (34) . When Deng et al. studied mismatches in 50-mer microarray probes with 1-7 pm in different distribution patterns, they concluded that the signal intensity was decreased more by evenly distributed than by randomly distributed pm (32) . When 60-mer Figure 2A , 3A and 4A were plotted against the predicted ÁG calculated in Visual OMP TM 7.0 (DNA Software). The predicted ÁG was obtained for the interaction between probes and targets in 3M TMAC buffer at a hybridization temperature of 45 C. The percentage MFI is the MFI signal of a probe hybridized with a study target divided by the MFI signal of the same probe hybridized with its perfectly matching target (e.g. InflA probe against InflA target) at the same temperature. Regression lines were calculated using the SigmaPlot dynamic curve fitting system. A five parameter sigmoidal function gave the highest correlation [f = y 0 +a/ (1+exp (-(xx 0 ) oligonucleotides were hybridized in a microarray, mismatches located near the middle of the probe resulted in a greater reduction of signal intensity than those located at the ends (33) . Additionally, microarray experiments with short oligonucleotides (16-40 nt) and one mismatch or a nt insertion in all positions show that the hybridization signal decreases when the mismatching portion is centered (35, 36) . The matching segments are shorter when mismatches are centered than when they are located peripherally. Our results, using microsphere-bound 70-mer probes and Luminex technology in a buffer containing 3 M TMAC, confirm that the distribution of the mismatches is of great importance and that hybridization is stronger when there are a few longer uninterrupted sequences than when there are many short sequences. These effects are formulated in the NucZip algorithm. Furthermore, it has previously been shown that two different oligonucleotides of 18 nt, complementary to the inner and outer portions of a 25 nt probe, could hybridize in solution with equal efficiency but, when the probe was coupled to a solid phase via a C6 linker, the 18-nt target complementary to the outer part of the 25-nt probe bound more efficiently than the 18-nt target complementary to the inner part, close to the solid phase (37-39), cf (33) . On our probes, the 5 0 end was coupled to the microspheres via an amino C12 linker. We did not observe any significant differences between matching stretches close to the bead surface and far from it and conjecture that perhaps the long linker allowed for greater accessibility. It is possible to create a probe against a target with high nt variation, such as an RNA virus, by using degenerated bases at variable positions but the degeneracy of the probe is dramatically increased by each wobbling base, thus decreasing the effective probe concentration. For instance, a sequence with two wobbling bases present in nine positions of variation would give a degeneracy of 512 unique sequence combinations, while a target with 14 variations including A, T, C or G would demand a set of 268 Â 10 6 unique probes (degeneracy 268 Â 10 6 ). Honoreé t al. successfully used 18-23 nt probes with a degeneracy of up to 512 in 3 M TMAC buffer (40) . In the more stringent PCR buffers, the usage of probes with a degeneracy greater than 10 is not often reported (41) . Degenerated primers have the property of being 'forgiving' (41) (42) (43) . This is because the amplimer from a previously successful primer is a target for the same pool of primers in the next round of amplification, leading to an accumulation of amplifiable targets. However, the situation for a probe is different. A degenerate probe will always face the same target variation. Therefore, universal bases like dInosine may be more useful than degenerated sites for probes, as long as the hybridization strength (represented by T m or -ÁG) is good enough. The introduction into a probe of a universal base like dInosine instead of a wobbling base reduces the complexity of the oligonucleotide mixture and increases the actual number of hybridizing oligonucleotides. Previously, Honore´et al. introduced up to three dInosines in radioactively labelled short oligonucleotide probes, 18-23 nt, in dot-blot hybridization, using a buffer containing 3 M TMAC (40) . The aim was to reduce the degeneracy in probes used for screening cDNA libraries. They found that dInosines had a slightly destabilizing effect on hybridization, especially when hybridizing against A, G and T, but that this could be minimized by reducing the hybridization temperature. However, the behavior of dInosine in long probes in 3 M TMAC has, to our knowledge, not previously been systematically explored. The 3 M TMAC is known to increase the binding contribution of the A:T base pairs, resulting in a similar contribution to T m to that from the G:C base pairs. The high ionic strength makes this an environment of relatively low hybridization stringency. The general trend shown in our study (e.g. Figure 2A and B), that dInosine in the probe decreases hybridization in 3 M TMAC, indicates that TMAC did not enhance the binding strength of dInosine base pairs as much as that of A:T base pairs; cf (8) . In a segment with dInosines at every third base, such as in the Ino18 probe, every matching nt neighbors a dInosine, i.e. it will not bind as strongly as a probe containing neighboring matches. Clearly, dInosine matches cause less destabilization than mismatches, and allow hybridization of probe/ target combinations with many short matching segments, like the InflA probe / 21-pm target combination. It is reasonable to assume that when a probe fails to hybridize due to a high number of mismatches in the target, dInosines at these positions will restore hybridization, since dInosine appears to bridge adjacent matching stretches, increasing their ability to nucleate. Our results confirm the findings of Honore´et al., despite differences in (i) methods of detection, (ii) hybridization time, and (iii) length of probes. The experience gathered in this work indicates that dInosine base pairing can be considered intermediate between a match and a mismatch, when carried out in 3 M TMAC. Furthermore, our results indicate that dInosine causes less destabilization when hybridized with a C and an A, than when hybridized with a G and a T (7, 8, 40) , in 3 M TMAC; e.g. the Ino18 probe hybridized more strongly with the 33%9nt and 33%15nt targets than with the 33%12nt target ( Figure 3C and D and Supplementary Table S5 ). To lessen this effect and to be able to use the same hybridization temperature for a panel of probes containing no or different amounts of dInosine, it is preferable for the probes to be long, like the 70-mer probes investigated here. The effects of the universal base dInosine were also compared with those of N wobbles. Thus, at the lower temperature, a dInosine-containing probe hybridized more strongly and, at the higher temperature, the N wobble probe hybridized more strongly. To understand how the highly degenerated probes, wobbN_21, wobbN_18 and wobb_N24, hybridized so well, we calculated the probability of randomly achieving an extension of the matching regions at the 5 0 and 3 0 ends of the wobbN_18 probe (Table 4 ). The probability that the closest N wobble to either the 5 0 or 3 0 end would be a perfect match is 0.5. Thus, 50% of the pool of degenerated probes have a 3 nt longer perfect match (12+9 or 9+12 matching nt at the 5 0 and 3 0 flanking regions) which, according to our results with non-degenerated probe/target combinations, should lead to rather good hybridization of the wobbN_18. In fact, wobbN_18 hybridization was similar in strength to that of the InflA probe to the 33%12F or 33%15nt targets. Furthermore, the probability of having several additional 5-nt matching regions in the central region is also high, probably giving rise to many more combinations in the same pool that matched There is a high probability of several additional matching regions of 5 nt in the central region, which will contribute to hybridization. x, matching nt; N, wobble of A, C, G, or T. even better. By restricting the wobbles to 3 (e.g. a D or a B) or 2 (e.g. a Y or a T) nt, the probability of a match becomes even greater. Thus, in a highly degenerated probe with at least one continuous region of perfectly matching nt, a large part of the pool will extend this region and contribute to nucleation, zipping and hybridization. The behavior of the highly degenerated probes is encouraging and in accordance with the NucZip model, which predicts that the high likelihood of several matching stretches of 5 nt or longer will result in significant hybridization. One of the aims of this study was to investigate the binding capacity of dInosine in 3M TMAC. 5-NitroIndole was chosen as a comparative universal base. The results shown in Figure 3C demonstrate that 5-NitroIndole in the 5-NitroInd_18 probe had a much greater destabilizing effect than dInosine in the Ino_18 probe, without the same capacity to rescue hybridization with a target containing many sm. Furthermore, the 5-NitroInd_18 probe was more affected than the Ino_18 probe when the hybridization temperature was raised from 45 to 55 C. It is concluded that, under 3 M TMAC buffer conditions, dInosine is a better choice than 5-NitroIndole when designing a variation-tolerant probe with as little degeneration as possible. Recently, Majlessi et al. (15) studied the nucleation process during double helix formation of short probes, 18-28 nt, with RNA or DNA targets of varying lengths and number of mismatches, in a buffer containing lithium succinate and lithium lauryl sulfate at pH 5.1. Hybridization is initiated by random collisions, but occasionally the complex is stable enough to nucleate the hybridization process. After investigating their model, they suggested that one nucleation region of 9 nt is not enough for further zipping and formation of a double helix. Instead, the first nucleation site needs a second nucleation site so that they can then cooperatively induce the zipping mechanism. They reported that inactivation of one of the 9-nt sites reduced hybridization >2-fold. Interestingly, one complete turn of a dsDNA molecule consists of 10.4 nt (44) . It is conceivable that the first (often temporary) contact between two single nucleic acid strands (nucleation) should not exceed one turn of the dsDNA helix in length, to avoid torsional disturbance and faulty interlocking of the strands. From this point of view, the nucleation site should be long enough to minimize false contacts, and short enough to have minimal steric effects on the strands. Nucleation sites of 6-9 nt fulfil these criteria. The chance of two random single strands matching at a hexanucleotide is 1/16 394, and at a nonanucleotide is 1/ 1 048 576. A matching nonanucleotide will thus reduce the ratio of random successful to unsuccessful nucleations, i.e. those which do not lead to further hybridization in the subsequent zipping phase, a million-fold. The subject is far beyond the scope of this article; however, our data, using 70-mer probes in 3 M TMAC buffer, reveal that a target with two separate regions of 9 nt was enough for efficient hybridization when several shorter regions of 5-6 nt were available between mismatches during the hybridization process (InflA probe/26%9F, 2583 MFI). Increasing the number of mismatches, i.e. shortening the matching regions between the 9 nt flanking regions, caused failure of hybridization (InflA probe/33%9F, 133 MFI and InflA probe/74%9F target, 86 MFI). In contrast, the Ino18 probe, with dInosines covering the mismatches and nine matching nt at the 5 0 and 3 0 ends, hybridized strongly (Ino18/33%9F, 3894 MFI). Furthermore, the InflA probe/74%15nt or InflA probe/74%18nt combinations showed that one region of 15-18 nt was enough to induce and sustain hybridization reproducibly, even if the rest of the 70-mer probe contained 74% mismatches. Having two perfectly matching regions of 15 nt (15F) compared to one region of 15 nt (15 nt) gave 2.6-fold higher MFI for 33%15F compared to 33%15 nt and 4.5-fold higher MFI for 74%15F compared to 74%15 nt, at 45 C. The Ino24 probe, with no dInosine-free matching trimers, hybridized inefficiently with all targets (361-920 MFI at 45 C), showing that dInosine is relatively inefficient in creating nucleating regions. Thus, two nucleation sites of 9 nt are enough to cause hybridization in 3 M TMAC when they are placed next to each other to form a longer region or when there are enough shorter matching regions of 5-6 nt between them. Alternatively, dInosines could bridge the mismatches between the 9-nt regions. Earlier work has indicated that a region of 15 nt in a 50-mer probe or 20 nt in a 70-mer probe could cause significant cross hybridization in microarray hybridization experiments (30, 31) and our data agree with this. A 70-mer probe is able to hybridize with a region of 15 nt in an otherwise highly mismatching target in 3 M TMAC (74%15 nt, Figure 3A ). To summarize, the current study shows that a dInosine-free probe of 70 nt needs (i) at least three regions of at least six perfectly matching nt, (ii) two stretches of 12-15 perfectly matching nt, or (iii) one stretch of 15-18 perfectly matching nt to result in measurable hybridization. Probes with a high number of dInosines positioned at sites of variation need shorter matching regions than dInosine-free probes. It is suggested that this is probably because dInosine participates in nucleation and zipping during the hybridization process. Thus, a probe with 18 dInosines which match mismatches in the target needs either (i) two regions of 9 nt if the hybridization temperature is 55 C, or (ii) one region of nine perfectly matching nt at 45 C. As also shown in the study, the risk of cross hybridization when using an nt analog like dInosine is minimal, since dInosine is sensitive to a mismatch in the position next to it and >5 mmoi will reduce hybridization. On the other hand, one should be aware that if an unintended target has many mismatches covered by dInosine and only a limited number of mmoi (<5 mmoi), this could lead to cross hybridization and false positivity. Furthermore, the assumption that sm all differ at the third codon base is an oversimplification. Some synonymous codons also differ at the first and second bases. Thus, even if the Ino18 probe could hybridize when 11 of 17 trimers were intact, with perfect matches at bases 1 and 2, the tolerance to mmoi of a highly dInosine-substituted probe like Ino18 is limited. Around half of its six surplus trimers must be reserved for sm occurring at codon positions 1 and 2. This leaves three trimers available for non-synonymous mismatches. However, a long probe is more likely than a short probe to have matches not neighboring mismatches. The NN theory was developed for hybridization of short oligonucleotides in solution (45) . Its application to surface-bound oligonucleotides has not been precisely studied. Hooybergs et al. studied hybridization of 30-mer surface-bound oligonucleotides with a 20-mer linker to 30-mer targets in solution, with no, one or two evenly spaced mismatches (46) . Although NN theory was approximately corroborated, NN factors had to be recalculated to give an approximate fit to experimental data. Moreover, the adsorptive (Langmuir) behavior deviated from expectation at high target concentrations. Thus, many unresolved questions regarding hybridization behavior remain. The concept of NucZip, with both local and distal cooperativity contributions, is an attempt to predict the hybridization behavior of most 70-mer probe-target combinations under the given conditions ( Figures 6A and 8) . The NucZip algorithm is now under revision to include probes containing 1, 2, 3 or 4 nt wobbles, as well as taking into account the results with the universal base 5-nitroindole. The algorithms, schematically described in Figure 8 , (i) were based on our experimental data (described above) and (ii) included highly matching nucleation sites extending beyond the neighboring nt. Thus, unlike the NN theory, NucZip takes the effects of matches at longer distances into account. The unique property of 3 M TMAC to provide a roughly equal contribution to hybridization by A:T and G:C pairs justifies simple computational approaches. The well known additivity of binding contributions per nt inherent in ÁG calculations according to the NN theory (22, 45) indicates that hybridization can be treated in a relatively simplistic way. Our concepts were based on the finding that the longer the sequence of uninterrupted matching nt, the more stable is the hybridization. Our calculations were thus focused more on binding than on destabilization, i.e. the use of positive rather than negative contributions. One of the weaknesses of the NN theory is that it adds all contributions, positive or negative, to a grand sum. The negative contributions are subtracted for the whole molecule, instead of in the local context where they belong. In our approach, the binding is first assessed locally, mimicking nucleation, and then extended cooperatively to the whole molecule, mimicking the zipping process. It is binding that keeps the hybrid together, and it is thus logical to focus on binding. NN theory does, to some degree, predict that the distribution of the mismatches is an important factor insofar as it affects the nearest neighbors. However, the concentration of this theory on the nearest neighbor disregards the importance of longer matching stretches. The importance of mismatch distribution and uninterrupted matches is exemplified by the stronger signal for hybridization of the InflA probe with the 21-gm target (30% of the total MFI, 2005 MFI at 45 C; 8% of the total MFI, 443 MFI at 55 C) compared to the abolished signal when using a 21-pm target. This is demonstrated in Table 2 by the InflA probe hybridizations with targets containing 14 or 16 mismatches in different distribution patterns, where long uninterrupted perfectly matching sequences favoured hybridization. The cooperativity in hybridization beyond the nearest neighbor that was noticed when two or more matching trimers neighbored each other strengthened hybridization more than when the same numbers of matching trimers were separated by mismatches. In Figure 3A (45 C), the InflA probe hybridized better with 74%12nt (10% MFI), 74%15nt (26% MFI), or 75%18nt (57% MFI) targets than with the 74%9F target, with its 9 nt+9 nt of perfect match (3% MFI). Furthermore, at the higher temperature ( Figure 3B , 55 C), the 74%18nt (37% MFI) and 74%22nt (66% MFI) targets resulted in good hybridization while the 74%12F target, with its 12 nt +12 nt perfectly matching regions, failed (3% MFI). Three matching trimers account for approximately one turn [10.4 nt (44) ] of the helix of a dsDNA molecule. It is thus likely that two long DNA strands become entangled or conformationally committed when at least one turn of the helix has been completed. Nucleation has to proceed rapidly, without too much torsion and entanglement, to allow many encounters in a short time. We envision that contact between two strands extending to 9 nt allows rapid comparison between strands with only local torsional disturbance. If the brief contact does not achieve binding strength over a certain threshold (the nucleation threshold), the strands separate and new comparisons are made. If the nucleation threshold is exceeded at initial contact, Total ZipScore mode1 : 25+12+12=49 Total ZipScore mode2 : 25+12+5+1=43 ZipScore downstream : 49-dInosinefactor*(49-43)+(dInoCnr*dInoCfactor) contiguous tri-to pentadecamers Figure 8 . The zipping component of NucZip. When a 6-9-bp sequence fulfilling the nucleation criteria has been detected, hybridization up and downstream of the nucleating site is attempted (zipping). The figure shows the downstream zipping process, with successive accumulation of score within a matching segment arising from the trimers, tetramers, etc. up to pentadecamers (each scoring equally) which fit into it. In this way, a longer matching segment gets more than a linear increase in score relative to a shorter one. Zipping extends from the potential nucleation site, terminating with a mismatch or the end of one of the strands. dInosines are counted as intermediate between a match and a mismatch, as described in the 'Materials and Methods' section. If several consecutive matching segments are encountered, their scores are added. binding extends further up and downstream, i.e. zipping occurs. This proceeds as long as the binding strength remains sufficient. The strands are held together chemically by base-base interactions and topologically by the multiple turns of intertwinement. It was a challenge to model this process. In the NucZip model, the program first tests for possible nucleation sites and selects the two highest scoring sites for further evaluation. Zipping is then performed up and downstream from each suggested nucleation site. The zipping algorithm symbolizes the successive cooperativity of binding by adding the number of successive tri-to pentadecamers for each matching segment, each of which ends either in a mismatch or at the end of one or two of the oligonucleotides. The contribution of added dInosine molecules is counted less than those of proper matches. The highest scoring nucleation point is chosen as the result. The results using degenerated probes were in line with this NucZip theory. A long matching segment created by a wobble position increased hybridization strength beyond the contribution of the additional single match. At present, the NucZip model is intended for equally long nt segments. Exceptions to the model found in the experimental section of this work can be illustrated by the remarkable difference in hybridization between the short perfectly matching targets of 15-22 nt (15 nt_free, 18 nt_free and 22 nt_free) and the longer 70-mer targets (74%_15 nt, 74%_18 nt and 74%_22 nt) ( Figure 3A , B and D). Several groups have analyzed the effect of short so-called dangling ends of 1-5 nt (23) (24) (25) (26) , and report that they appear to stabilize hybridization of the duplex. Doctycz et al. observed that the dangling nt closest to the duplex contributed most to the stabilizing effect (47) . The lengths of our long mismatching ends ranged from 55 to 48 nt with a 74%-nt mismatch and no matching trimers. We speculate that the destabilization seen in a duplex with two long mismatching ends, one from the probe and one from the target, compared with a duplex between a long and a short oligonucleotide, with only one long protruding oligonucleotide, is due to shearing stress on the matching nt, or to competition from intra-strand secondary structures at the separate ends. Although it can predict many aspects of long oligonucleotide hybridization in 3 M TMAC at 45 C, the NucZip concept has to be amended to include other temperatures, oligonucleotide concentrations and buffers in order to be generally useful. The NN model, which was developed with great precision by SantaLucia et al. (8, (20) (21) (22) and is used in the Visual OMP TM 7.0 computer program, is much more sophisticated than our procedure. It was considered out of the scope of this article to evaluate the Visual OMP TM 7.0 program in detail; however, importantly, its hybridization conditions can be adjusted. When Visual OMP TM 7.0 was used to calculate ÁG (Figure 6 ), the buffer and temperature conditions were set at 3 M TMAC and 45 C. Comparison of the NucZip and Visual OMP TM 7.0 models showed that the distribution of the targets with the long dangling ends (74%_Xnt) was corrected to a certain extent by NucZip; however, neither NucZip nor Visual OMP 7.0 accurately predicted the hybridization behavior of the short perfectly matching probe-target combinations of the InflA probe with the 15nt_free, 18nt_free and 22nt_free targets. In conclusion, we have demonstrated that the distribution of mismatches greatly affects probe hybridization. A minimum number of continuous, perfectly matching stretches (nucleation sites) is needed to initiate hybridization. Thus, if the target contains many variations and no long uninterrupted matching segments, use of a dInosine-containing probe, which partially overcomes the obstacles caused by mismatches, will be beneficial. With respect to hybridization prediction algorithms, a simple statement of percentage mismatch, as used in many such algorithms, does not adequately reflect the hybridization properties of a long duplex. The insertion of dInosines in the positions of variation could detect target nucleic acid that a consensus probe would fail to catch. Hence, while high dInosine content in a probe decreases the probe's binding capacity, it also covers mismatches, as required when creating mismatch-tolerant, broaddetection probes against highly variable target nucleic acid sequences such as those seen in RNA viruses. The dInosine probes can be made even more forgiving by using a lower hybridization temperature. Furthermore, the probability of cross hybridization is low because the mmoi neighboring the dInosines have a destabilizing effect on hybridization. While the aim of this study was to improve our understanding of variability and the effects of dInosines, other universal bases and wobbles in the design of probes to be used in a 3 M TMAC buffer system, with more exploratory work the results may also be relevant to other hybridization systems and could aid the development of hybridization-based diagnostic tools, including nanotechnological applications such as the volume-amplified magnetic nanobead detection assay (48, 49) . Supplementary Data are available at NAR Online.

Asymmetry in the Presence of Migration Stabilizes Multistrain Disease Outbreaks

We study the effect of migration between coupled populations, or patches, on the stability properties of multistrain disease dynamics. The epidemic model used in this work displays a Hopf bifurcation to oscillations in a single, well-mixed population. It is shown numerically that migration between two non-identical patches stabilizes the endemic steady state, delaying the onset of large amplitude outbreaks and reducing the total number of infections. This result is motivated by analyzing generic Hopf bifurcations with different frequencies and with diffusive coupling between them. Stabilization of the steady state is again seen, indicating that our observation in the full multistrain model is based on qualitative characteristics of the dynamics rather than on details of the disease model.

In this work we study the stability of a multistrain disease model in two coupled populations. Multistrain diseases are diseases with multiple coexisting strains, such as influenza (Andreasen et al. 1997) , HIV (Hu et al. 1996) , and dengue (Ferguson et al. 1999b) . We consider two kinds of strain interactions: cross immunity and antibody-dependent enhancement. When the disease infects an individual, his or her immune system creates serotype-specific antibodies, which will protect the individual against that serotype. 1 However, there is evidence that antibodies also give some cross-protection to the other serotypes (Halstead 2007) . This reduced susceptibility to the other serotypes is temporary. When the temporary cross immunity wanes, heterologous secondary infections are possible. Low level antibodies developed from primary infections are believed to form complexes with the virus so that more cells are infected, and viral load is increased (Vaughn et al. 2000) . This effect is called antibody-dependent enhancement (ADE), and it has been observed in vitro in diseases such as Ebola (Takada et al. 2003 ) and dengue (Halstead and O'Rourke 1977) . Throughout this work we make the hypothesis that ADE increases the infectiousness of secondary infective cases due to the higher viral load Schwartz et al. 2005) . Alternative views of ADE as an increase in mortality associated with secondary infectives can be considered (Kawaguchi et al. 2003) . We focus in this paper on the multistrain disease dengue, which is believed to exhibit both temporary cross immunity and antibody-dependent enhancement (Halstead 2007) . Dengue is a subtropical mosquito-borne disease that exhibits up to four serotypes. It is widespread in tropical regions of southeast Asia, Africa, and the Americas, infecting an estimated 50 to 100 million people every year (World Health Organization Website 2006) . Primary infections are sometimes asymptomatic, while secondary infections are more severe, with about 5% of secondary infections leading to dengue hemorrhagic fever (DHF) or dengue shock syndrome (DSS), the potentially fatal forms of the disease (World Health Organization Website 2006) . An effective vaccine against dengue is very difficult to achieve. Because of ADE, infection with an unvaccinated strain following a single-strain vaccination could lead to the more severe symptoms associated with secondary infections (Halstead and Deen 2002) . Therefore, an effective vaccine must protect against all four serotypes simultaneously. A recent theoretical study on a 2 strains system (Billings et al. 2008) has shown that eradication of dengue using only single-strain vaccines is unlikely. Because a tetravalent vaccine is not immediately forthcoming, deepening our understanding of the dynamics of dengue in more realistic models is of great importance. The dynamics of dengue in a single, well-mixed population has been studied in several recent publications, including but not limited to Ferguson et al. (1999a) , Schwartz et al. (2005 , Wearing and Rohani (2006) , Bianco et al. (2009) . ADE and cross immunity have been shown to play a fundamental role in mathematical models for the spreading of dengue, causing instability, desynchronization of serotypes, and chaotic outbreaks Cummings et al. 2005; Bianco et al. 2009 ). However, real populations may be spatially heterogeneous. To gain further insight into the wide spectrum of possible dengue dynamics, we relax the assumption of a well-mixed population. The division of a population into spatially distinct patches simulates the potentially heterogeneous environment in which the disease spreads. Spatial heterogeneity has been invoked in the past to account for the persistent character of some infectious diseases (Lloyd and May 1996; Hagenaars et al. 2004) and to explain in-phase and out-of-phase dynamics of diseases (He and Stone 2003) . A multipatch model with age structure has been used to explain bi-and tri-annual oscillations in the spread of measles (Bolker and Grenfell 1995) . Human mobility patterns have a significant influence on the spreading of infectious disease. We will assume here that the coupling between patches is via migration, movement of individuals from one patch into another (as in Liebovitch and Schwartz 2004; Ruan et al. 2006; Sattenspiel and Dietz 1995; Grais et al. 2003 ). An alternative coupling strategy is mass action coupling, in which susceptibles in one patch are assumed to interact with infectives in another patch, introducing nonlinear coupling terms (Bolker and Grenfell 1995; Lloyd and May 1996; Rohani et al. 1999) . Stochastic coupling can be modeled (see Keeling and Rohani 2002 and references therein, and Colizza and Vespignani 2008) . The purpose of this work is to analyze the stability of a model for multistrain diseases with interacting strains, using dengue as an example, in a system of two coupled patches. Since chaotic outbreaks are likely to produce a higher number of infected individuals, understanding the stability properties may play an important role in public health. The case of non-identical parameters in the two patches will be of particular interest, as this serves as a model for spatial heterogeneity. The paper is divided as follows. In Sect. 2 we introduce the epidemic model. Section 3 summarizes the bifurcation structure for a single patch. In Sect. 4 we present numerical results for bifurcations in two coupled patches. Section 5 motivates these results via analysis of a simple, lower dimensional model, and Sect. 6 concludes. We use a compartmental model for multistrain disease spread with cross immunity and antibody-dependent enhancement, previously studied in a single, well-mixed population (Bianco et al. 2009 ). In this model, individuals can develop a primary infection with any of the serotypes. Immediately after recovering, the individual experiences a period of temporary partial cross immunity to all other serotypes. When the cross immunity wears off, immunity to the primary infecting strain is retained, but the individual can develop a secondary infection with a different serotype. Infectiousness of the secondary infectives is increased due to antibody-dependent enhancement. After the secondary infection, complete immunity to all serotypes is assumed. A flow diagram for the single patch model with two serotypes is shown in Fig. 1 for simplicity, but we present results here for all four serotypes. We extend the model to two spatially distinct patches, which are coupled by linear migration terms. For two patches (indexed by q) and n strains (n arbitrary), the model is as follows: where the variables are s q , the fraction of susceptibles in patch q; x q,i , the fraction of primary infectives with strain i in patch q; c q,i , the fraction of individuals in patch q with partial cross immunity to strain i; r q,i , the fraction of individuals in patch q that are recovered from a primary infection with strain i and no longer have cross immunity to the other strains; and x q,ij , the fraction of individuals in patch q recovered from strain i and currently infected with strain j . The parameters are the number of strains n, the contact rate in patch q β q , the recovery rate σ , the ADE factor φ, the strength of cross immunity , the rate θ for cross immunity to wear off, the birth and mortality rate in patch q μ q , and the migration rate between patches ν. A list of parameters appears in Table 1 . For clarity, we describe in detail the terms appearing in (5) describing the time evolution of the secondary infectives x q,ij . These are individuals who were previously infected with strain i and are currently infected with strain j . Individuals recovered from an infection with i and no longer retaining cross immunity (r q,i ) can become infected by any infective with strain j . We use mass action contact between infectives and non-invectives, contributing a term β q r q,i (x q,j + φ k =j x q,kj ) (where the secondary infectives have enhanced contagion given by the ADE factor φ). Cross immune individuals c q,i become infected with a reduced probability (1 − ). Linear terms −σ x q,ij and −μ q x q,ij appear as secondary infectives recover with rate σ or are lost to natural death with rate μ q . Finally, individuals are exchanged between patches q and q with rate ν, yielding migration terms −νx q,ij + νx q ,ij . For simplicity, the birth and death rates in a patch are set equal to each other so that the total population of each patch is constant. The model of (1)-(5) allows for one reinfection. Tertiary infections are not considered (Nisalak et al. 2003) . The parameter determines how susceptible the cross immune compartments c i are to (Bianco et al. 2009 ). Note the reduction of susceptibility to a secondary infection through the cross immunity factor (1 − ) and the enhancement of secondary infectiousness due to the ADE factor φ. Mortality terms for each compartment are not included in the diagram for ease of reading Ferguson et al. (1999b) β, transmission coefficient, years −1 ∼200 Ferguson et al. (1999b) σ , recovery rate, years −1 50 Rigau-Perez et al. (1998) θ, rate to leave the cross 2 Wearing and Rohani (2006) immune compartment, years −1 φ, ADE factor ≥1 S c h w a r t z e t a l . ( 2005) , strength of cross immunity 0-1 Bianco et al. (2009) ν, migration rate, years −1 0-0.05 n, number of strains 4 a secondary infection, where = 0 means no cross immunity (the infection rate is identical for compartments c i and r i ) and = 1 confers complete cross immunity (cross immunes are immune to a secondary infection for an average time θ −1 ). We allow to take any value between 0 and 1. The ADE factor φ is the enhancement in the infectiousness of secondary infectives. φ = 1 means that secondary infectives are as infectious as primary infectives, while φ = 2 means that secondary infectives are twice as infectious as primary, and so forth. In contrast to Aguiar and Stollenwerk (2007), we will not consider values of φ smaller than 1. The migration rate from patch q to q , and from patch q to q, is ν. For simplicity, we assume that all individuals migrate with equal probability, independent of their infection status. This assumption may be relaxed in a future study. The migration rate is assumed to be slow compared to the infection spread. For convenience, we put it on the same order as the birth/death rate. We assume that the social parameters, which are the contact rate β q and birth/death rate μ q , may vary between patches, while the epidemic parameters are the same in all regions. Because the social parameters depend on human factors (and in the case of the contact rate also include mosquito levels, which are weather-dependent), these parameters are the most likely to be different in adjacent regions. We use parameter values compatible with dengue fever, which are summarized in Table 1 . Our contact rate β corresponds to a reproductive rate of infection R 0 of 3.2 − 4.8, which is consistent with previous estimates (Ferguson et al. 1999b; Nagao and Koelle 2008) . A similar model to the one of (1)-(5) has recently been used to analyze the dynamics of dengue fever in a single well-mixed population (Bianco et al. 2009 ). The ADE φ and cross immunity strength were varied as bifurcation parameters. In the absence of cross immunity ( = 0), ADE alone generates instability, desynchronization, and ultimately chaotic outbreaks Schwartz et al. 2005) . A Hopf bifurcation is observed for a critical value of the ADE factor φ, above which oscillatory solutions are obtained. Weak cross immunity stabilizes the system, while strong cross immunity triggers instability and chaos even in Table 1 the absence of ADE (Bianco et al. 2009 ). In the latter case, destabilization occurs via a Hopf bifurcation for a critical value of the cross immunity strength . At the bifurcation, three identical complex pairs of eigenvalues of the Jacobian simultaneously become unstable. Although Bianco et al. (2009) discusses the full two parameter bifurcation structure (in and φ), we will consider the cross immunity and ADE effects separately in the present work. Figure 2a shows the bifurcation behavior of the single patch model with no cross immunity as the contact rate β is varied. The bifurcation structure was computed using a continuation routine (Doedel et al. 1997 ). The ADE value at which the bifurcation occurs increases slightly as β increases, and the dependence is approximately linear. The period of the periodic orbit, shown in Fig. 2b for a fixed ADE value, also varies with β. (Note that multistrain models with ADE can display subcritical Hopf bifurcations, Billings et al. 2007 , so the periodic orbit shown in Fig. 2b exists throughout the range of β values shown.) Similarly, varying the contact rate β in the absence of ADE affects the location of the Hopf bifurcation in cross immunity and its characteristic frequency of oscillation. Likewise, varying the other social parameter, the birth rate μ, affects the location and frequency of the Hopf bifurcations in φ and (data not shown). We next examine the effect of coupling between distinct regions on the dynamics previously observed in the single patch model. In particular, we consider how migration between non-identical patches affects the stability of the steady state. We investigate the dynamics of the coupled systems by numerically integrating (1)-(5) and by tracking the bifurcations using a continuation routine (Doedel et al. 1997) . As mentioned in the previous section, the model has a Hopf bifurcation at critical values of the bifurcation parameters φ, the ADE factor, and , the cross immunity strength. We Fig. 3 Critical values of the parameters (a) and (b) φ at which the Hopf bifurcation occurs for coupled patches, as a function of the contact rate in patch 1, for two values of the migration rate, ν = 0.02 (solid black) and ν = 0.05 (dashed gray). The contact rate in patch 2 is fixed at β 2 = 200. In (a), φ = 1 (no ADE). In (b), = 0 (no cross immunity). μ 1 = μ 2 = 0.02 and other parameters are as in Table 1 study the effect of coupling and asymmetry on the stability by observing their effect on the location of the Hopf bifurcations. We consider cross immunity and ADE separately; that is, we analyze the system either with ADE and no cross immunity or with cross immunity and no ADE. Including both cross immunity and ADE together leads to qualitatively similar behavior to that reported here, at least locally when the asymmetry is not too large. We proceed by fixing the patch-specific parameters in patch 2 and varying a social parameter (β 1 or μ 1 ) in patch 1 to observe the effect of increasing asymmetry on the dynamics. As previously mentioned, changing the social parameters modifies the natural frequency of the system. For each value of asymmetry, we look for the critical points at which the coupled system loses stability. We also consider several migration rates ν. The effect of asymmetry in the contact rate is shown in Figs. 3(a) and 3(b) for two migration rates, namely ν = 0.02 (black) and ν = 0.05 (gray). The value of the critical parameter ( Fig. 3(a) ) or φ (Fig. 3(b) ) at which the Hopf bifurcation occurs is plotted against the varying contact rate. The symmetric case is at the bottom of the curve. We see that two identical patches have the same bifurcation point as a single, well-mixed population even when migration is present (cf. Fig. 2a) . However, a striking difference in the dynamics appears when we make the two patches weakly asymmetric. The values of φ and at which the Hopf bifurcation occurs are dramatically different from the symmetric case. The steady state stability persists well into the parameter regime where a single patch would display chaotic oscillations (Billings et al. 2007; Bianco et al. 2009 ). The location of the single patch Hopf point does not depend strongly on β, and decreasing the contact rate is actually destabilizing (Fig. 2a) , so the stabilization observed here is clearly a result of the coupling between asymmetric systems. Slightly increasing the migration rate (from ν = 0.02 to ν = 0.05 in Fig. 3 ) further increases the stability of the system. Fig. 4 Critical values of the parameters (a) and (b) φ at which the Hopf bifurcation occurs for coupled patches, as a function of the birth/death rate in patch 1, for ν = 0.02. The birth/death rate in patch 2 is fixed at μ 2 = 0.02. In (a), φ = 1 (no ADE). In (b), = 0 (no cross immunity). β 1 = β 2 = 200 and other parameters are as in Table 1 Similar results are obtained if the asymmetry occurs in the other social parameter, the birth rate, as depicted in Fig. 4 . Again, for asymmetric patches either stronger cross immunity or stronger ADE is needed to destabilize the steady state through a Hopf bifurcation. We now motivate the results of the preceding section using a simple lower dimensional model and show that the qualitative features depend only on the bifurcation structure and characteristic frequencies rather than on details of the epidemic model. In this section we study the general behavior of two coupled systems, each displaying a Hopf bifurcation, but with different characteristic frequencies. The generic form for a Hopf bifurcation is the following, in polar coordinates (Strogatz 2001 A Hopf bifurcation occurs at α = 0, where periodic oscillations with frequency ω are observed. We now couple two systems of the sort in (6), each with a potentially different frequency ω q , via linear migration terms with migration rate ν. To lowest order (not displaying cubic terms), the coupled system in cartesian coordinates iṡ x 1 = αx 1 − ω 1 y 1 − νx 1 + νx 2 y 1 = αy 1 + ω 1 x 1 − νy 1 + νy 2 x 2 = αx 2 − ω 2 y 2 − νx 2 + νx 1 y 2 = αy 2 + ω 2 x 2 − νy 2 + νy 1 Stability is determined by evaluating the Jacobian of (7) at the steady state (x 1 , y 1 , x 2 , y 2 ) = 0. At the Hopf bifurcation, the real part of the largest eigenvalue crosses zero, with nonzero imaginary part. The roots of the characteristic polynomial f (λ) of the Jacobian are the eigenvalues {λ}. The four eigenvalues are but it is not easy to see from these expressions where the Hopf bifurcation occurs. Instead, we solve directly for the critical value of α at the Hopf point. At a Hopf bifurcation, a pair of eigenvalues has zero real part. Thus to obtain the Hopf point, we can set λ = ib, where b is real. If ib is an eigenvalue, we require and If α = ν, (11) has nonzero roots b = ± 1 2 4α 2 − 8αν + 2ω 2 1 + 2ω 2 2 . When this b is substituted into (10), we obtain four potential roots for α at the Hopf point. Two are complex, which is unphysical and can be ignored. The one corresponding to the Hopf bifurcation is α c = ν − 1 2 4ν 2 − (ω 1 − ω 2 ) 2 . (The fourth root is larger and corresponds to loss of stability of the second pair of eigenvalues.) This α c is real and thus gives the Hopf point location when |ω 1 − ω 2 | ≤ 2ν. When ω 1 − ω 2 = 0 and ν > 0, α c is positive, indicating stabilization due to the asymmetry and coupling. When α = ν, (11) is always satisfied. In that case, we must turn to (10) to determine b. Equation (10) has four roots, b = ± 1 2 (ω 1 + ω 2 ) ± 1 2 (ω 1 − ω 2 ) 2 − 4ν 2 . By our assumption, b must be real, and this occurs when |ω 1 − ω 2 | ≥ 2ν, which is precisely the case not covered by the above result. Thus when |ω 1 − ω 2 | ≥ 2ν, the Hopf point is at α c = ν. Summarizing, the Hopf bifurcation occurs at In Fig. 5 we show the location of the Hopf bifurcation given by (12) as a function of the asymmetry between the two systems for ν = 0.05. Comparison with Figs. 3 and 4 shows the qualitative agreement of the theoretical results for the lower dimensional system with the full multistrain system. Increasing the migration rate ν increases the value of the bifurcation parameter to which the Hopf point saturates, as in Fig. 3 . It is also worth noticing that, in the case of identical frequencies ω 1 = ω 2 , the Hopf bifurcation occurs at α c = 0, the location in the absence of coupling. This is consistent with the numerical results for the multistrain system in the case of symmetric patches. We have studied the endemic steady state stability properties for a multistrain epidemic model on two migration-coupled patches. Interactions between strains in the model were governed by temporary partial cross immunity and antibody-dependent enhancement. In the absence of coupling, the system displayed Hopf bifurcations in two epidemic parameters. Coupling between patches with non-identical parameters, which gave them non-identical characteristic frequencies of oscillation, was shown to shift the Hopf bifurcations, stabilizing the steady state. This behavior was observed for the Hopf bifurcation obtained by sweeping the cross immunity in the absence of ADE and the bifurcation obtained by sweeping the ADE in the absence of cross immunity. It occurred for asymmetry in either of our two social parameters, the birth rate and the contact rate. To motivate this result, we diffusively coupled two low dimensional Hopf bifurcations with different characteristic frequencies and analyzed the stability of the steady state. We again saw that coupling between asymmetric systems led to stabilization. This indicates that the stabilization in the epidemic model is a result of the underlying dynamics, rather than the details of the model. We suggest that the stabilization may occur as a result of the two different coupled frequencies generating oscillations that tend to cancel each other because of phase differences. This topic will be studied in more detail in a future work. Bifurcations from steady state to oscillatory behavior can be associated with an increased number of infection cases, particularly if chaotic oscillations occur, as in previous dengue models (Billings et al. 2007; Bianco et al. 2009 ). Our results suggest that if control strategies in one region are able to generate enough asymmetry, this could lead to a stabilization of the outbreaks, which would have a positive effect on adjacent regions. Asymmetry could be generated in the effective contact rate by mosquito control, which could include reducing mosquito breeding sites (Slosek 1986) , or through new genetic controls which are under development (Barbazan et al. 2008) . Asymmetry in the birth rate could be generated by lowering the effective birth rate through vaccination of new susceptibles once a vaccine is available. However, because the bifurcation point saturates rather than increasing indefinitely as asymmetry increases, such a strategy would be successful only if the epidemic parameters in the real system are moderately close to the bifurcation. (Extreme asymmetry may even be destabilizing, so this strategy works best locally when the asymmetry is not too strong.) Real systems exhibiting oscillatory behavior may already be well into an unstable region of parameters. For such systems, our suggested control strategy may not apply since asymmetry between adjacent regions is already likely, thus reducing that impact that an increase in the asymmetry could have. On the other hand, in the case of dengue fever a recent study (Cummings et al. 2009 ) has highlighted a decreasing trend in both the force of infection and birth and death rates in the population of Thailand during a time span of about 30 years. This phenomenon could potentially drive the system closer to the bifurcation point and make it more amenable to control. In addition, the role of seasonality in exciting oscillations should not be ignored. Seasonal variations in the contact rate have been included in previous dengue models . The interplay of seasonality and coupling is a topic for future study. The results of the present work may also be useful in estimation of the parameter values for ADE and cross immunity. ADE especially is difficult to measure in vivo and must be estimated by other means. Since recent publications have suggested that epidemiological data from Thailand show chaotic outbreaks , and given the certainty of human migration and asymmetry between adjacent regions of the country, it is possible that the actual parameter values for ADE and cross immunity are higher than the ones estimated by studying a single, well-mixed population. Finally, the work discussed here shows a potential effect of human movement between heterogeneous regions. As spatial effects are further studied in epidemic models, it remains to be seen how this phenomenon will extend to more complicated spatial geometries, including more patches and perhaps non-symmetric coupling terms. This work represents a first step towards understanding the role of migration and spatial heterogeneity in dynamical properties of dengue observed in epidemiological data, such as traveling waves of infection in Thailand (Cummings et al. 2004 ). Furthermore, because the migration-induced stabilization depends only on the existence of a Hopf bifurcation in the model, it is expected that the stabilization will be observed in other population models that also contain Hopf bifurcations (e.g., Fussmann et al. 2000; Greenhalgh et al. 2004) .

Clinical factors associated with severity in hospitalized children infected with avian influenza (H5N1)

OBJECTIVE: The World Health organization received reports of 478 laboratory-confirmed cases of influenza A (H5N1) from 15 countries between November 2003 and February 2010. More than 50% of these cases involved patients <20 years of age. Determining an association between the clinical factors at the time of hospital admission and prognosis may be useful for timely and adequate consultation and treatment. It has been difficult to obtain these clinical factors adjusted with other confounding factors, such as age and sex, as published studies of H5N1 virus infection usually reported only a few cases. So, we performed a pooled analysis of the reported cases. METHODS: Five case reports (36 patients <18 years of age) of H5N1 infection from four countries published between 2004 and 2009 were assessed based on available individual clinical data. Using the pooled data for all patients, we investigated the associations between patients’ prognosis and available laboratory findings, such as white blood cell (WBC) counts, platelet (PLT) counts, and serum levels of aspartate aminotransferase (AST) and alanine aminotransferase (ALT) by adjusting for age and/or sex. RESULTS: The linear regression analysis revealed that mortality was negatively associated with WBC and PLT counts adjusted with age and sex. Increased log AST tended to be associated with a poor prognosis (p = 0.054), but there was no significant association between survival and log ALT level. CONCLUSIONS: Both decreased WBC and PLT counts can be considered to be common predictors of poor prognosis in H5N1 influenza patients <18 years of age. Further studies are needed for clarification.

The 2009 H1N1 influenza pandemic spread throughout the world, but since the beginning of 2010, the number of cases of this influenza has been declining. Reports of avian influenza (H5N1) infection in humans have also been reported, raising the threat of a new pandemic through a novel combination of the H5N1 virus with the 2009 H1N1 strain. Cases of the highly pathogenic avian influenza A virus (H5N1) causing influenza symptoms in humans were observed in the Hong Kong Special Administrative Region, People's Republic of China in 1997 [1] . A total of 478 laboratory-confirmed cases of influenza A (H5N1) infection were subsequently reported to the World Health Organization (WHO) from 15 countries between November 2003 and February 2010 [2] . More than 90% of these cases were reported in five countries, namely, Indonesia, Vietnam, Egypt, China, and Thailand. Patients \20 years of age accounted for 63% of the cluster cases and 49% of the sporadic cases [3] . The case-fatality rate (CFR) in all cases was 62.1. Of the 383 cases reported between November 2003 and May 2008 from these 15 countries, 10 .4% involved patients aged\9 years, 24% involved patients between ages 10 and 19 years, and 25% involved patients aged 20-29 years [4] . The overall CFR of H5N1 infection was 65%, and the CFR for patients aged 10-19 years was 78%, which was highest among all age groups [4] . Patients for whom the disease was ultimately fatal showed respiratory failure and progressed to pediatric acute respiratory distress syndrome (ARDS), which in patients infected with H5N1 influenza virus consisted of severe viral pneumonia accompanied by diffuse alveolar damage that was induced by intense cytokine reactions and inflammation [5] . Wiwanitkit investigated hematologic findings among reported H5N1infected patients and observed lymphopenia and anemia, but he did mentioned that the effects of the infection on platelet count were controversial [6] . A knowledge of the clinical factors present at admission to the hospital that are associated with prognosis may be useful when the aim is timely and adequate consultation and treatment. As many of the studies on the H5N1 virus have each only reported a few cases of H5N1 influenza, it has been difficult to get identify these clinical factors adjusted with other confounding factors, such as age and sex. Here, we report our investigation of the early indicators of prognosis in patients with H5N1 influenza aged \18 years which we identified from a pooled analysis of case reports. For the case-investigation of H5N1 human infection reports containing individual clinical data, we selected five case reports from four countries published between 2004 and 2009 for investigation. From each report, we selected patients aged \18 years; ultimately, our study population consisted of 36 patients who were included in the pooled analysis. Four patients were from Tran et al. [7] (Vietnam), 12 were from Kawachi et al. [5] (Vietnam), 7 were from Chotpitayasunondh et al. [8] (Thailand), 8 were from Oner et al. [9] (Turkey), and 5 were from Kandun et al. [10] (Indonesia). H5N1 case-patients were identified by reverse transcriptase (RT)-PCR. The patients reported in Kawachi et al. [5] who showed severe ARDS were identified in the National Hospital of Pediatrics (NHP), and those in other reports were from passive surveillance. We investigated factors related with patients' prognosis from available laboratory findings, such as white blood cell (WBC) counts, platelet (PLT) counts, and serum levels of aspartate aminotransferase (AST) and alanine aminotransferase (ALT). The AST and ALT levels were not available for the cases of Kandun et al. [10] and five cases from Kawachi et al. [5] , and the leukocyte and platelet counts were not available for one case from Kawachi et al. [5] . Data were analyzed using Wilcoxon and Kruskal-Wallis tests for continuous variables and Fisher's exact test for categorical variables. Linear regression analysis was used for investigating associations between laboratory findings and patients' prognosis. All statistical analyses were done by SPSS ver. 14.0J (SPSS, Chicago, IL). We summarized the characteristics of patients from each report in Table 1 . The mean age of the pooled patients from all studies was 8.0 ± 4.4 years, and mean ages of the patients were significantly different among the different reports (p = 0.032). For the pooled data, the proportion of patients who showed ARDS or respiratory failure during hospitalization was 75%; this proportion was also significantly different among the different reports (p = 0.014). Overall, the CFR was 69.4%, and no significant difference between reports was observed. For the pooled data, median number of days from onset to admission was 6.0 and the median number of days from onset to outcome was 14.0; there no significant differences among the reports for both data sets. In terms of the laboratory data at hospital admission, for the pooled data the median WBC count was 2900 (/mm 3 ), and the median PLT count was 125 (910 3 /mm 3 ); a significant difference in PLT count was observed among reports ( Table 2) . Based on available data on the liver function tests (three reports), the median AST and ALT levels were 280 and 52 IU/L, respectively; the levels of both liver enzymes were significantly different among reports (Table 2) . Differences between laboratory data at hospital admission between fatal and nonfatal cases are shown in Table 3 . Leukopenia and thrombocytopenia at hospital admission were present in the fatal cases, but there were no differences in the serum levels of AST and ALT between fatal and nonfatal cases. In the linear regression analysis performed to investigate the association between blood cell counts and prognosis, WBC and PLT were dependent variables, and age, sex, and prognosis were independent variables. In the linear regression analysis performed to investigate the association between log-transformed liver function data and prognosis, log AST and log ALT levels were dependent variables, and age and prognosis were independent variables as the sample size was small enough to enable one covariate to be added. Standardized regression coefficients for the results of these linear regression analyses are shown in Table 4 . Death (the value of prognosis = 1) was negatively associated with WBC and PLT counts adjusted with age and sex. In contrast, increased log AST tended to be associated with poor prognosis, but there was no significant association between survival and log [1] with the results from the pooled data of our five studies reveals that the CFR for patients\19 years of age was 56.9% in the WHO report and 69.4% in our pooled data. The median time from onset of symptoms to admission was 4 days (mean 4.6 days) among all of the patients reported in our five studies and 5 days in patients reported to have died in the WHO report. The median time from onset to admission was 6 days in the pooled data. Cases were mainly identified through passive surveillance in the WHO report, whereas we also included cases collected through hospital medical records in our study. Thus, the CFR in the pooled data was higher than that of the WHO report, and the median time from onset to admission in the pooled data was longer than that in the WHO report. Our pooled data was biased to include more severe cases. Significant variations in patterns that may reflect differences in exposure related to social behavior, cultural and religious practices, access to care, and treatment or virulence of the viruses among countries have been pointed out in previous reports [4] . With respect to the virulence of viruses, viruses capable of infecting humans have been divided into groups from clades and subclades 0, 1, 2.1, 2.2, 2.3, and 7 from the phylogenetic tree for the hemagglutinin gene of highly pathogenic avian influenza A (H5N1) viruses. Clade 1 viruses were found to predominate in Vietnam, Thailand, and Cambodia in the early phase of the outbreak (2004) (2005) , and clade 2.1 viruses are endemic in Indonesia (2005) (2006) . Clade 2.2 viruses were associated with the outbreak of avian infection in Central and South Asia, the Middle East, Europe, and Africa and with human infection in western Asia, the Middle East, and Africa (including the Turkish outbreak from 2005 to 2006). The median time from onset to hospitalization were found to be unchanged among clades [11] . Differences in the proportion of fatalities were observed, but it is difficult to determine whether this difference was due to the difference in viral virulence, as other variations existed among countries. Our pooled data also included these significant variations. Although our study is characterized by a number of the limitations shown above, our results may help generate hypothetical candidates for early indicators of prognosis in patients \18 years of age with H5N1 influenza. In most of the earlier reports on H5N1 cases, the number of cases was not large enough to adjust for potential confounders. However, several published studies did investigate factors that predicted poor outcome, adjusting the confounders for the whole-age spectrum of the subjects. Kandun et al. [12] analyzed 127 confirmed infections from June 2005 through February 2008 in Indonesia and found that an individual case was significantly associated with the mortality. Liem et al. [13] reviewed medical records and analyzed 67 of 93 cases of H5N1 infection in Vietnam from January 2004 through December 2006. Applying a backward step-wise variable selection strategy for logistic regression analysis, these researchers were able to show that the presence of both neutropenia and raised ALT level predicted a poor outcome. Yu et al. [14] collected 26 H5N1 cases from hospital medical records of H5N1 cases in China from October 2005 through April 2008 and found that decreased PLT count, elevated lactic dehydrogenase level, ARDS, and cardiac failure were associated with mortality. In our study, decreased WBC and PLT counts were significantly associated with the prognosis adjusted with sex and age, but these results coincided with the results of individual previous reports. Both decreased WBC and PLT counts are considered to be common predictors of poor prognosis in H5N1 influenza patients \18 years of age. Further studies that include larger numbers of patients are needed to clarify these results.

Human monoclonal IgG selection of Plasmodium falciparum for the expression of placental malaria-specific variant surface antigens

Pregnancy-associatedPlasmodium falciparum malaria (PAM) is a major cause of morbidity and mortality in African women and their offspring. PAM is characterized by accumulation of infected erythrocytes (IEs) that adhere to chondroitin sulphate A (CSA) in the placental intervillous space. We show here that human monoclonal IgG antibodies with specificity for variant surface antigens (VSA) specifically expressed by CSA-adhering IEs (VSA(PAM)) can be used in vitro to select parasites from nonpregnant donors to express VSA(PAM) and that this selection for VSA(PAM) expression results in preferential transcription of var2csa. The results corroborate current efforts to develop PAM-specific vaccines based on VAR2CSA.

Children living in areas of stable Plasmodium falciparum transmission acquire substantial protective immunity against malaria during the first decade of life. Protection is mediated to a large extent by variant surface antigen (VSA)specific IgG. Nevertheless, women in such areas remain highly susceptible to P. falciparum infection if they become pregnant, as the parasites can switch to expression of particular VSA (called VSA PAM ), which allow infected erythrocyte (IE) sequestration in the placenta, but which are not compatible with parasite survival in a nonpregnant host. Acquired immunity to PAM is mediated by VSA PAMspecific IgG that either opsonizes IEs for phagocytosis or interferes with chondroitin sulphate A (CSA)-specific adhesion of IEs (1) . VAR2CSA, which is a member of the P. falciparum erythrocyte membrane protein 1 (PfEMP1) family of VSA, appears to be the dominant or only pregnancy-associated Plasmodium falciparum malaria (PAM) type VSA in the P. falciparum genome, and is therefore the main target of current efforts to develop a vaccine against PAM (2) . To assist this work, we have developed a panel of VSA PAM -specific human IgG1 monoclonal antibodies (3) , which can opsonize VSA PAMexpressing IEs for phagocytosis and interfere with IE adhesion to CSA (Barfod et al . unpublished data) . These antibodies can be used to enrich for VSA PAM expression in parasites not expressing VSA PAM but obtained from women with PAM (3) . In the present study we used two of the above-mentioned monoclonal antibodies to select for VSA PAM expression in parasites from nonpregnant donors. One antibody (PAM1·4) was chosen because it appears to react with most or all parasites expressing VSA PAM . The other antibody (PAM8·1) was chosen because it reacts with a well-defined, but inter-clonally variant, epitope in the DBL3X domain of VAR2CSA. We used eight VSA PAM -specific monoclonal IgG1 antibodies generated as described elsewhere (3; 4) . VSA PAM is defined here as IE surface-expressed VSA, which are significantly better recognized by plasma IgG from P. falciparumexposed multigravidae than from sympatric men, and where the recognition by plasma IgG from these men is not significantly different from recognition by plasma IgG from nonexposed controls. In contrast, typical non-PAM VSAs are better recognized by plasma IgG from P. falciparumexposed adults than from nonexposed donors, without marked sex-dependent differences. Seven of the monoclonal antibodies used here are specific for inter-clonally variant epitopes in either the DBL3-X or the DBL5-ε domain of VAR2CSA (3) . The exact specificity of the last antibody (PAM1·4) remains undefined, but it appears to recognize a conformational, and possibly discontinuous, epitope in VAR2CSA that is difficult to reproduce in recombinant constructs (3) . PAM1·4 recognizes VSA PAM expressed by most or all P. falciparum genotypes, whereas the VAR2CSA DBL3-X epitope recognized by PAM8·1 is present in some, but not all P. falciparum clones (3). Both PAM1·4 and PAM8·1 can opsonize VSA PAM -expressing IEs for phagocytosis and interfere with their adhesion to CSA (Barfod et al. in preparation) . We used the two long-term in vitro -adapted parasites 3D7 and HB3. The 3D7 clone was originally derived from NF54 parasites isolated from a Dutch girl near Amsterdam airport (5) . It was chosen here because it can be selected for expression of VSA PAM that react with PAM1·4 but not PAM8·1 (3). HB3 was cloned from the Honduras I/CDC parasite strain (6) , and was chosen because it can be selected for expression of VSA PAM reactive with PAM1·4 and PAM8·1. All parasites were grown in 0 + erythrocytes as described (3) . Selection for VSA PAM expression was done essentially as described elsewhere (7) . In brief, monoclonal PAM1·4 or PAM8·1 antibodies were immobilized on Protein A-coated magnetic beads (Dynal) and mixed with 3D7-or HB3-IEs. IEs adhering to the antibody-coated beads were isolated in a strong magnetic field and subsequently returned to in vitro culture. Selection protocol was repeated when multiplication of the antibody-selected parasites allowed it. VSA expression was assessed by flow cytometry analysis as described elsewhere (8) . We used plasma from 10 P. falciparum -exposed multigravidae, 10 sympatric men, and 10 nonexposed controls to assess the sex-specificity of VSA recognition. According to the original criteria (9), sex-specific recognition requires that (i) levels of IE-surface reactive IgG are significantly higher in P. falciparum -exposed multigravidae than in sympatric men, and that (ii) the difference in IgG levels in P. falciparumexposed men and nonexposed controls is not statistically significant. We used term plasma from 30 P. falciparum -exposed women (10 pregnant for the first time, 10 for the third time, and 10 for the fifth time) to assess parity-dependency, which requires a statistically significant relationship between IgG levels and number of pregnancies in P. falciparumexposed women (9) . Parasite isolates were only considered to express VSA PAM if plasma IgG recognition of IEs was both sex-specific and parity-dependent. Late-stage IEs were isolated by magnetic separation as described (8) and returned to culture overnight to obtain ring-stage IEs. Genomic DNA was isolated with a QIAamp blood kit (Qiagen), and total RNA extracted (TRIzol, Invitrogen) and treated with DNase I (Invitrogen) for 30 min (10) . The absence of DNA in RNA samples was confirmed as described (11) . Reverse transcription was performed using Superscript II (Invitrogen) and random hexamer primers, followed by real-time PCR to quantify var transcript abundances as described (11) . We used primer pairs specific for the 59 var genes in the 3D7 genome (12) and for 45 var genes in the HB3 genome (Table S1 in Supporting Information). The latter were tested on genomic DNA dilutions to ascertain appropriate fragment size, melting temperature, and amplification efficiency compared to the internal control seryl-tRNA synthetase . All primer pairs varied less than two Ct values from that of the internal control and amplified single fragments of the expected sizes. Differences in antibody levels among plasma from P. falciparumexposed multigravidae, P. falciparum -exposed men, and nonexposed control donors were analysed by one-way anova , or by one-way analysis of ranks for non-normal distributed data. If statistically significant ( P < 0·05) overall differences were detected, significant pair-wise The VSA expressed on the surface of unselected 3D7-and HB3-IEs were significantly better recognized by plasma IgG from P. falciparum -exposed donors of both sexes than by IgG from nonexposed donors, whereas levels in exposed men and multigravidae were not significantly different from each other. Furthermore, IE surface-reactive IgG levels did not depend on parity (Table 1 and Figure 1a ). Thus, unselected 3D7 and HB3 both showed typical non-PAM-type plasma antibody recognition patterns of IE surface-expressed VSA. This was confirmed by the nonreactivity of 3D7-and HB3-IEs with all eight VSA PAM -specific monoclonal antibodies (Figure 1d ). When VSA expression was re-assessed after three rounds of selection (about 6 weeks after the first round of selection), the recognition of 3D7-and HB3-IEs selected on either PAM1·4 or PAM8·1 all showed an indeterminate VSA phenotype, where one but not both criteria for sex-specific antibody recognition were met (Table 1 and Figure 1b ). PAM8·1-selected 3D7 remained nonreactive with the four monoclonal antibodies used for testing at this time (PAM1·4, PAM3·10, PAM4·7, and PAM8·1), whereas the other three parasite lines showed reactivity with at least one of them (Figure 1e ). These results suggested that further rounds of selection might lead to definite VSA PAM expression, at least for PAM1·4-selected 3D7, PAM1·4-and PAM8·1-selected HB3. Indeed, PAM1·4-selected 3D7, as well as PAM1·4-and PAM8·1-selected HB3 had all acquired a typical sex-specific and parity dependent VSA PAM expression pattern after four additional rounds of selection (Table 1 and Figure 1c ), and reacted with all the monoclonal antibodies except the VAR2CSA DBL5-ε -specific PAM4·7 (Figure 1f ). In contrast, 3D7 selected seven times for reactivity with PAM8·1 retained an indeterminate sex-specificity pattern also seen after three rounds of selection, did not acquire the parity-dependent pattern typical of VSA PAMexpressing lines (Table 1) , and did not react with any of the eight VSA PAM -specific monoclonal antibodies (Figure 1f ). Thus, PAM8·1 could not be used to select 3D7 for VSA PAM reactivity, consistent with the absence of the predicted epitope for this antibody in the VAR2CSA DBL3-X domain of this parasite (3). Transcripts of var2csa (PFL0030c) constituted 7% of total measured var gene transcripts in unselected 3D7 parasites, increasing to 80% after seven rounds of PAM1·4 selection. In contrast, PAM8·1 selection did not affect the proportion of var2csa transcripts in 3D7 (Figure 1g ). The HB3 genome contains two var2csa paralogs, var2csa-A and var2csa-B (13) . In unselected HB3 parasites, var2csa-A transcripts constituted 1% of the measured var transcripts, increasing to 28% and 4% after seven rounds of selection by PAM1·4 and PAM8·1, respectively (Figure 1h ). Transcript levels of the var2csa-B gene increased from 23% to 57% and 92% after seven rounds of selection by PAM1·4 and PAM8·1, respectively (Figure 1h ). No var transcript other than a Human monoclonal VSA PAM -specific antibody used for selection. b Determination of sex-specificity involves a two-step procedure: first establishment that IE surface-reactive IgG levels in at least one of the three donor categories (exposed multigravidae, exposed men, and unexposed controls) differs from at least one other category. Only if this is the case (as it was here: P < 0·001 in all cases), can sex-specificity be confidently assessed by post hoc testing for significant (P < 0·05) pair-wise differences. The result of this post hoc testing is indicated here as No: none of the two post hoc criteria for sex-specificity were met, Indeterminate: only one of the criteria was met. Yes: both criteria were met. See Materials and Methods for further details. var2csa showed marked changes, and var2csa was therefore the dominant transcript in all the parasites following antibody selection. The dominance of var2csa-B transcripts relative to var2csa-A in HB3 after PAM8·1 selection, suggested that the DBL3-X domain in the protein encoded by var2csa-B might be of the FCR3-type recognized by PAM8·1, whereas the corresponding domain encoded by var2csa-A might be of the 3D7-type (PFL0030 c) not recognized by PAM8·1 (2). Recognition of IEs by plasma IgG from 10 P. falciparum-exposed multigravidae (Exp. MG), 10 sympatric men (Ex. men) and 10 nonexposed controls (Unexp. ctrls) before selection (a), and after three (b) or seven (c) rounds of selection by human monoclonal IgG antibody PAM1·4. Results in panels A-C are presented as medians (horizontal line), central 50% of data points (boxes), central 80% of data points (whiskers) and outliers (•). In addition, statistically significant (P < 0·05) pair-wise differences are indicated by heavy horizontal bars along the top of the panels. Recognition of IEs by human monoclonal VSA PAM -specific IgG antibodies before selection (d), and after three (e) or seven (f) rounds of selection. Results in panels D-F are presented as individual data points. Negative cut-off, defined as the upper level of recognition of unselected parasites, is indicated as a dashed horizontal line. The proportion of var2csa transcripts among all var transcripts in 3D7 (g) and HB3 (h) before selection and after seven rounds of selection. Panel (i) shows the amino acid sequence of the PAM8·1-specific region of VAR2CSA DBL3-X in the two VAR2CSA paralogs in HB3, FCR3, and 3D7 parasites. Amino acid differences between the two HB3 sequences and the PAM8·1-reactive FCR3 sequence are underlined. However, var2csa-A and var2csa-B encode an identical amino acid sequence in the region spanning the PAM8·1 epitope, and this sequence was of the FCR3-type (Figure 1i ). The var2csa-A : var2csa-B transcripts may therefore reflect a founder effect. The VSA subset VSA PAM is expressed by P. falciparum involved in pregnancy-associated malaria (PAM). VSA PAMspecific IgG mediates acquired immunological protection against PAM, and the PfEMP1 variant VAR2CSA appears to be the main or only target of these antibodies. VAR2CSA is therefore the leading candidate for development of vaccines against PAM. We used monoclonal human IgG antibodies to select erythrocytes infected by two genotypically distinct laboratory P. falciparum clones derived from nonpregnant donors for expression of VSA PAM . Parasites acquiring expression of VSA PAM following selection showed increased levels of transcripts encoding the PfEMP1 variant VAR2CSA, which appears to be the only PAM-type VSA in the P. falciparum genome. The results obtained in the study are important for several reasons. First they support the hypothesis that all P. falciparum parasites have the capacity to express VSA PAM . This hypothesis is supported by previously published data that all P. falciparum genomes appear to contain at least one paralog of the gene encoding the only known VSA PAM -type antigen, VAR2CSA (11, 14) . Furthermore, this gene is selectively transcribed by placental parasites (and following selection for adhesion to CSA in vitro) and VAR2CSA is expressed on the IE surface (15) (16) (17) . Second, they indicate that P. falciparum parasites regularly and spontaneously switch to expression of VSA PAM in the absence of an external signal, for example pregnancyassociated hormonal changes. It has been speculated that switching to VSA PAM expression requires signals from the pregnant host, for example hormones, and that selection therefore might not be possible unless the parasite is derived from such a host. It has also been argued that selection of parasites by panning on CSA in vitro might result in expression of antigens of dubious relevance to the antigens expressed as a result of in vivo selection occurring in the pregnant woman. Our data show that switching to expression of genuine VSA PAM can occur in vitro in the absence of external signals, in line with other recent evidence (18) . By extension, these findings suggest that switching to VSA PAM in vivo also occurs spontaneously regardless of the pregnancy status of the host. In a pregnant host, such parasites will often be at a selective advantage (because of the frequent absence of VSA PAM -specific immunity in women of low parity) (19) , whereas they appear to be unable to survive in a nonpregnant host (20) . Third, our results support the hypothesis that although interclonal variation in VAR2CSA is the result of antibodydriven positive selection, the diversity of functionally important antibody epitopes in the molecule is constrained (21) . Thus, some parasites can express VAR2CSA without being vulnerable to recognition by certain VAR2CSAreactive antibodies, because of variation in defined parts of VAR2CSA. The persistent nonrecognition by PAM8·1 of VAR2CSA-expressing 3D7 is a case in point. At the same time, the PAM8·1 antibody, originally identified by its reactivity with VSA PAM -expressing FCR3-IEs (3), was effectively selected for VSA PAM expression in the genetically unrelated HB3 clone. The significance of these findings is underscored by the fact that PAM8·1 efficiently opsonizes IEs for phagocytosis as well as interferes with IE adhesion to CSA (Barfod et al. in preparation) . Our study also corroborates existing evidence that PAM1·4 recognizes most if not all VSA PAM -expressing parasites by recognizing a functionally highly constrained and conformation-dependent discontinuous epitope in VAR2CSA (3). Like PAM8·1, PAM1·4 is an opsonin and capable of interfering with IE adhesion to CSA (Barfod et al. in preparation) . Therefore, the PAM1·4 epitope is a highly attractive candidate for development of a PAM-specific vaccine, and its characterization is a matter of the highest priority in our current research. Although the theoretical possibility of another, non-PfEMP1 target of PAM1·4 remains, available evidence point to VAR2CSA. In conclusion, our evidence support current efforts to develop PAM-specific vaccines based on VAR2CSA and highlight the versatility of human monoclonal antibodies generated from clinically immune donors in these investigations.

Stimulation of ribosomal frameshifting by antisense LNA

Programmed ribosomal frameshifting is a translational recoding mechanism commonly used by RNA viruses to express two or more proteins from a single mRNA at a fixed ratio. An essential element in this process is the presence of an RNA secondary structure, such as a pseudoknot or a hairpin, located downstream of the slippery sequence. Here, we have tested the efficiency of RNA oligonucleotides annealing downstream of the slippery sequence to induce frameshifting in vitro. Maximal frameshifting was observed with oligonucleotides of 12–18 nt. Antisense oligonucleotides bearing locked nucleid acid (LNA) modifications also proved to be efficient frameshift-stimulators in contrast to DNA oligonucleotides. The number, sequence and location of LNA bases in an otherwise DNA oligonucleotide have to be carefully manipulated to obtain optimal levels of frameshifting. Our data favor a model in which RNA stability at the entrance of the ribosomal tunnel is the major determinant of stimulating slippage rather than a specific three-dimensional structure of the stimulating RNA element.

Programmed ribosomal frameshifting is a translational recoding event that increases the versatility of gene expression. It is mainly utilized by eukaryotic RNA viruses (1) (2) (3) , though some prokaryotic (4) and mammalian genes (5) (6) (7) are also controlled by ribosomal frameshifting. The requirements for À1 ribosomal frameshifting are the presence of a slippery heptanucleotide sequence X XXY YYZ (where X can be A, U, G or C; Y can be A or U; and Z does not equal Y; the spaces indicate the original reading frame) (8) followed by a downstream structural element, such as a pseudoknot, a hairpin or an antisense oligonucleotide duplex [for reviews, see (9) ]. Although the mechanism of frameshifting is still elusive, a promising model has been proposed by Brierley and co-workers using cryo-electron microscopy to image mammalian 80S ribosomes (10) . In their model, the ribosome is paused by its inability to unwind a pseudoknot structure resulting in a blockage of the A-site by eEF-2. During translocation, the P-site tRNA is bent in the 3 0 -direction by opposing forces. To release the tension, the P-site tRNA may un-pair and subsequently re-pair in the À1 frame with a certain frequency, followed by A-site tRNA delivery into the new À1 reading frame. These and other recent data obtained by mechanical unfolding of frameshifter pseudoknots suggest that mRNA secondary structures with certain conformational features that resist ribosomal helicase-mediated unwinding and eEF-2 catalyzed translocation are key players in ribosomal frameshifting. Small oligonucleotides have been used for several years to regulate gene expression by RNaseH-dependent RNA degradation (11) , blocking translation (12) , or re-directing splicing (13) . More recently, microRNAs (miRNAs) (14) and small interfering RNAs (siRNAs) have appeared on the scene of post-transcriptional gene regulation (15) . siRNAs may be effective in treatment of chronic hepatitis-B virus infection (16) , HIV infection (17) , cancer (18) and age-related macular degeneration (19) . Very few antisense oligonucleotides, for example against the bcl-2 oncogene have reached the stage of clinical trials (20) or have actually been approved by the FDA, for instance for the treatment of human cytomegalovirus retinitis (21) . Enhancing the stability of small oligonucleotides to prolong circulation and meanwhile increasing target specificity are major concerns for therapeutic applications. Various kinds of modifications in backbones, sugars or even analogs have already been studied extensively [for reviews, see (22, 23) ] to meet these requirements. Locked nucleic acid (LNA) is a rather novel nucleic acid analog comprising a class of bicyclic high-affinity RNA analogs in which the furanose ring of LNA monomers is conformationally locked in an RNA-mimicking C3 0 -endo/ N-type conformation (24) . The LNA modification also resists degradation by cellular nucleases. Furthermore, introducing LNA into DNA or RNA oligonucleotides improves the affinity for complementary sequences and increases the melting temperature by several degrees (25) . A recent study showed that LNA/DNA mix-mers against miRNA-122 can be acutely administered at high dosage with long lasting effects without any evidence of LNA-associated toxicities or histopathological changes in the studied animals (26) . These data suggests that LNA is a promising candidate for small oligonucleotide applications. We and others have demonstrated that small RNA oligonucleotides are able to mimic the function of frameshifter pseudoknots or hairpins by redirecting ribosomes into new reading frames (27, 28) . In this article, we have investigated the length and concentration of RNA oligonucleotides for optimal frameshifting, as well as the effects of introducing LNA-type sugars in DNA oligonucleotides. The À1 ribosomal frameshifting events were monitored by the SF reporter construct described earlier (27) . Complementary oligonucleotides (Eurogentec, Liege, Belgium) SF462 (CTAGTTGACCTCAACCCTTGG AA) and SF463 (CATGTTCCAAGGGTTGAGGT CAA) and SF468 (CTAGTTGAGCGCGCTGGAGGC CATGG) and SF469 (CATGCCATGGCCTCCAGCGC GCTCA) were annealed and ligated into SpeI/NcoI digested SF reporter to construct the SF462 and SF468 templates, respectively. All constructs were verified by DNA sequencing on an ABI PRISM Õ 3730xl analyzer (LGTC, Leiden, The Netherlands). RNA oligonucleotides (except for RNA13 which was obtained from Invitrogen) were purchased from Dharmacon (Lafayette, USA). The RNAs from Dharmacon carried a 2 0 -O-ACE protection group, which was removed by incubation with 100 mM acetic acid pH 3.8 and TEMED at 60 C for 30 min. The sequences of RNA oligos were as follows: RNA6: GCGCGC, RNA9: CCAGCGCGC, RNA12: C CUCCAGCGCGC, RNA15: UGGCCUCCAGCGCGC, RNA18: CCAUGGCCUCCAGCGCGC, 18RNA: GCG CGCUGGAGGCCAUGG, and RNA13: CCAAGGGG UUGAGG. DNA and LNA/DNA mix-mers were synthesized by Eurogentec. Custom oligonucleotides were extracted by phenol/chloroform followed by ethanol precipitated before use. The sequences of DNA and LNA/DNA mix-mers were as follows (lower case represents the LNA modification and capital represents DNA): Plasmids were linearized by BamHI and purified by phenol/ chloroform extraction followed by ethanol precipitation. In vitro transcription was conducted by SP6 RNA polymerase and carried out in the 30 ml reaction mixture of: 1 mg of linearized template, 5 mM of rNTPs, 20 units of RNase inhibitor and 15 units of SP6 RNA polymerase with buffer (all from Promega, Benelux). After 2 h incubation at 37 C, the integrity and quantity of transcripts were checked by agarose gel and appropriate amount of the RNA were diluted in nuclease free water for in vitro translation. In vitro translations were carried out in nuclease treated rabbit reticulocyte lysate (RRL) (Promega). The amount of mRNA was 0.025 pmol and different amounts of oligonucleotides (0.025-15.625 pmol) were mixed with template for 20 min at room temperature. After incubation, 4 ml of RRL, 0.01 mM amino acids mixture except methionine, 2 mCi of 35 S methionine (10 mCi/ml, MP Biomedicals, in vitro translational grade) were added in total volume of 10 ml and incubated at 28 C for 1 h. After translation, samples were mixed with 2Â Laemmli buffer, boiled at 90 C for 5 min and resolved by 13% SDS polyacrylamide gels. Gels were fixed in 10% acetic acid and 30% methanol for 20 min, dried under vacuum, and exposed to phosphoimager screens (Biorad). The screen was scanned and the 0 frame and À1 frameshift protein products were quantified by Quantity One software (Biorad). Frameshift percentages were calculated by dividing the amount of À1 frameshift product by the amount of 0-frame and À1 frameshift products after correction for the number of methionines in the protein sequence, multiplied by 100. Determination of the melting temperature of oligonucleotide duplexes RNA oligonucleotide 18RNA (5 0 GCGCGCUGGAGGC CAUGG3 0 , Dharmacon, USA) was mixed in a 1:1 molar ratio with RNA18, DNA18 or one of the various DNA/ LNA mix-mers, in UV-melting buffer (100 mM NaCl, 10 mM Cacodylate acid, pH 6.8). The analysis was performed on a Varian Cary 300 spectrophotometer using temperature ramps of 0.25 C /min during heating and cooling. The absorbance at 260 nm was recorded and normalized to the blank control. Although antisense oligonucleotides were found to induce ribosomal frameshifting (27, 28) , the optimal number of base pairs has not been addressed yet. To investigate this we designed antisense RNA oligonucleotides that are 6, 9, 12, 15 and 18 bases complementary to the region downstream of an UUUAAAC slippery sequence in our reporter plasmid SF468 ( Figure 1 ). First, titration with RNA6 and RNA9 oligonucleotides revealed that a 625-fold molar excess of oligonucleotides over mRNA resulted in the highest level of frameshifting ( Figure 2a) ; this ratio was used in the following experiments. The shortest oligonucleotide, RNA6, was not capable of inducing significant levels of frameshifting (Figure 2b) , whereas RNA9 induced $3.5% of frameshifting. Maximum levels were obtained with RNA12, RNA15 and RNA18; all three induced $12% of frameshifting. In the following experiments oligonucleotides between 12 and 18 nt in length were used. Since we have absent knowledge about the efficacy of LNA-induced ribosomal frameshifting, LNA/DNA mix-mers of 18 nt in length were designed to investigate this ( Figure 3) . A DNA oligonucleotide, as expected, was less capable (3.5%) of inducing frameshift due to the lower thermodynamic stability of RNA-DNA duplexes, see also below. Surprisingly, substituting the 3 0 -cytosine and guanosine in this DNA oligonucleotide by their LNA analogs enhanced its frameshift inducing capacity to 8.7%, i.e. as high as an RNA oligonucleotide (8.8%). Increasing the LNA content of this oligonucleotide further did not lead to higher frameshifting. On the contrary, the efficiency of LNA4 was with 7.7% lower than that of LNA2 and that of LNA6 was a mere 1.1%. Since the overall translation efficiency seemed not affected by LNA6 we suspected an effect of the oligonucleotide itself (see below). To demonstrate that the enhanced effect of LNA oligonucleotides is a general feature we designed another construct (SF462) in which the target sequence was replaced by an unrelated sequence ( Figure 4 ). LNA/DNA mix-mers were designed in which nucleotides starting from the 3 0 -end were gradually replaced by LNA ( Figure 4 ). Increasing the number of LNAs from one to two and four in these DNA oligonucleotides improved their frameshift inducing ability, reaching an apparent optimum of 7.0% with four LNA substitutions. Further increase of the LNA content to 6 nt (LD6) did not improve frameshift efficiency, but, on the other hand, LD6 also did not lead to the dramatic decrease as observed above for the LNA6 oligonucleotide applied in the SF468 construct. We suspected that (partial) self-complementarity may be limiting the effective concentration of free LNA/DNA oligonucleotides. To check this possibility, we ran all the oligonucleotides on a non-denaturing polyacrylamide gel. Figure 5 showes that the LNA6 oligonucleotide indeed migrated more slowly indicative of partial dimer formation, presumably by intermolecular base pairing of the palindromic GCGCGC sequences in each oligonucleotide (compare the migration to that of the full dimer formed by annealing of oligonucleotides DNA18 and 18DNA). The LD series, as predicted, migrated as monomers. We noted that LD2, though loaded in equal amount, based on its UV absorbance, showed a higher affinity to ethidium bromide than its counterparts. At present we have no explanation for this unexpected behavior of LD2, since its migration and therefore its conformation was identical to the other LNA/DNA mix-mers. These results demonstrate that LNA modifications indeed enhance the antisense-induced frameshifting efficiency probably due to higher thermodynamic stability and RNA-like structural properties. This phenomenon appears to be general, at least in our experiments. To investigate which positions in a DNA oligonucleotide would exert the largest effect when substituted by an LNA analog, we designed LNA/DNA mix-mer mutants based on LNA2, which is the most efficient LNA/DNA mix-mer in our experiments and would give a good read-out. When the two LNA substitutions were moved two positions more inward (L2-1), compared to LNA2, frameshift efficiency decreased to 6.7% ( Figure 6 ). However, when the LNA modifications were moved another two positions more inward (L2-2), activity dropped to 2.7% ( Figure 6 ) which is comparable to an unmodified DNA oligonucleotide. Similarly, when the LNA groups were introduced at the other end of the oligonucleotide, activity was as low as DNA18 ( Figure 6 ). Finally, L2-4, in which the first and fourth position were LNA, was only half as efficient as LNA2. These results indicate that the choice of the location of the LNA modifications is crucial for the frameshift-inducing efficiency of an oligonucleotide. Theoretically the position effect of the LNA substitutions could simply be explained by differences in thermodynamic stability of the resulting mRNA/oligonucleotide duplexes. To investigate this possibility we carried out Table 1 . The T m of the 18RNA/RNA18 duplex was the highest with 82 C in agreement with its high frameshifting efficiency. The 18RNA/DNA18 duplex had a much lower T m of 72 C, which is expected for an RNA/DNA hybrid, and also agreed with the lower frameshifting efficiency. The LNA substituted oligonucleotides, all had higher T m s (+4 to +9 C) than DNA18. The T m of L2-2 was with 81 C almost as high as that of RNA18. Remarkably there was no correlation between the T m of the LNA oligonucleotides and their frameshifting inducing capacity. For example, the T m of LNA2 was rather low with 76 C but it had the highest frameshifting activity, and L2-2, which had the highest T m , actually had the lowest frameshifting activity. T m s of L2-3 and L2-4 were identical but their frameshifting activities were 2.3 and 4.6%, respectively. We also noted that both L2-2 and L2-3 were comparable to DNA18 in frameshifting activity but formed far more stable duplexes. These data suggest that the position effect of the LNA substitutions is related to the mechanism of frameshifting and not per se to their thermodynamic stability. Previously, we have demonstrated that antisense oligonucleotides can induce high levels of À1 frameshifting (27) . The optimal length of small antisense oligonucleotides, however, was not investigated. Understanding the optimal length of trans-acting oligonucleotides that can induce the most efficient frameshifting and, at the same time, escape RNAi interference will be an important issue for future in vivo applications. Here we found that maximum levels of frameshifting were obtained with oligonucleotides of 12 nt and more. This is comparable to the stem lengths (S1+S2) of known examples of highly frameshift inducing H-type pseudoknots, such as the 6+6 bp of the Simian retrovirus typeÀ1 pseudoknot (29) , the 11+6 bp of the minimal Infectious Bronchitis virus (IBV) pseudoknot (30) , and the 6+6 bp chimeric Mouse Mammary Tumor virus (MMTV)-IBV pseudoknot (31) . In addition, in known examples of hairpin-induced frameshifts, the stem length of hairpins is around 12 bp (32) (33) (34) . This may imply that a full helical turn of an RNA helix either in one single stem or in two stacking stems of a pseudoknot (S1+S2) is selected by viruses to induce efficient ribosomal frameshifting. In addition to RNA oligonucleotides, we demonstrated that LNA/DNA mix-mers are also capable of stimulating efficient À1 ribosomal frameshifting in contrast to DNA oligonucleotides. Replacing 2 nt in a DNA oligonucleotide by LNA was already sufficient to reach the same level of frameshifting as with a comparable RNA oligonucleotide. However, the excellent affinity of LNA oligonucleotides could be a double-edged sword in certain cases. In our experimental system, the oligonucleotides are partly self-complementary and this resulted in the formation of dimers ( Figure 5 ), which were apparently unable to induce frameshifting (LNA6, Figure 3 ). Hence, LNA substitutions should be optimized in a sequence that is prone to form dimers. In our SF462 construct (Figure 4) , the optimal number of LNA substitutions to induce the most significant amount of frameshifting is four. LD6 with two additional LNA substitutions did not improve the efficiency. Thus, our results suggest that the first 4 bp are critical for antisense-induced frameshifting. A likely explanation is that when a ribosome that is translating the slippery sequence, the helicase active site is around position +11, with respect to the first nucleotide of the P-site, which is close to the first base pair of the mRNA/ oligonucleotide duplex (35) . Increasing the local thermodynamic stability in this region may prevent ribosomes to unwind RNA structures, causing ribosomal pausing at the slippery sequence, and finally results in a higher frequency of ribosomal frameshifting. Our data also showed that a single LNA modification is not sufficient to turn a DNA oligonucleotide into an efficient frameshift inducer but that a second LNA is needed. The best position for the second modification appeared to be also close to the 3 0 -end of the oligonucleotide. Although one could expect that spacing of two LNA groups by two non-modified sugars as applied in probes for miRNAs, results in the optimal induction of the 3 0 -endo conformation in the neighboring sugars (36), this was not the case in our frameshift assays. Here such a spacing was less efficient (see data for L2-4, Figure 6 ). However, we have not investigated if possible differences of self-dimerization behavior of these oligonucleotides accounts for the different stimulating activities, since such effects were only observed when six LNA modifications were introduced in an oligonucleotide of this sequence. The observation that different positions of LNA substitutions induced different levels of ribosomal frameshifting is interesting. Even though the overall thermodynamic stability of these oligonucleotides is roughly the same, they still create different degrees of barriers for ribosomes to unwind and these differences could be the reason for different level of induced frameshifting. The finding that local stability at the 3 0 -end of the LNA/ DNA mix-mers is important for frameshifting is in agreement with the observation that in natural examples of frameshift stimulators, most of them have high GC content in the first few nucleotides (1) . Hence, our data support the notion that the stability of the 3 0 -end of the oligonucleotide, which may reside in the active site of the ribosomal helicase, is critical for frameshift-inducing structural elements. In pseudoknots this stability is probably attained by triple interactions, since nature has no other way to increase the stability of a GC-rich A-type helix. Triplex structures have been documented for a number of frameshifter pseudoknots, e.g. BWYV (37), SRVÀ1 (38) and in a telomerase pseudoknot (39) . Several models of ribosomal frameshifting have been proposed (1, 40, 41) . The consistency from these studies is that ribosomal pausing at shifty sites by downstream structural elements is important but that pausing caused by RNA secondary structure, does not always result in frameshifting. In addition, a lack of correlation between the extent of pausing and the efficiency of frameshifting by IBV pseudoknots has been observed (42) . A recent study also showed that pseudoknots with a similar global structure can still induce very different levels of frameshifting although their thermodynamic stabilities were different (43) . These data complicate the view on the role of the downstream structure. Experiments involving simple oligonucleotides such as shown here may be better alternatives to elucidate the role of the downstream element. Several groups have correlated the mechanical force of unfolding of a pseudoknot with its frameshifting efficiency by using optical tweezers (39, (44) (45) (46) and suggest that frameshift efficiency is dependent on the unfolding force rather than on differences of thermodynamic stability between folded and unfolded states. Since we showed here that antisense oligonucleotides can induce frameshifting presumably by serving as a physical barrier for the elongating ribosome, it will be interesting to measure the strength of these linear oligonucleotides in complex with (a piece of) mRNA by optical tweezers and see if there is a correlation with their frameshifting efficiency. Finally, several properties of LNA, including its good aqueous solubility, low toxicity, highly efficient binding to complementary nucleic acids, high biostability, and, improved mismatch discrimination relative to natural nucleic acid (47) make LNA a promising candidate for in vivo applications of antisense-induced frameshifting. Funding for open access charge: Leiden Institute of Chemistry, Leiden university.

A flexible loop in yeast ribosomal protein L11 coordinates P-site tRNA binding

High-resolution structures reveal that yeast ribosomal protein L11 and its bacterial/archael homologs called L5 contain a highly conserved, basically charged internal loop that interacts with the peptidyl-transfer RNA (tRNA) T-loop. We call this the L11 ‘P-site loop’. Chemical protection of wild-type ribosome shows that that the P-site loop is inherently flexible, i.e. it is extended into the ribosomal P-site when this is unoccupied by tRNA, while it is retracted into the terminal loop of 25S rRNA Helix 84 when the P-site is occupied. To further analyze the function of this structure, a series of mutants within the P-site loop were created and analyzed. A mutant that favors interaction of the P-site loop with the terminal loop of Helix 84 promoted increased affinity for peptidyl-tRNA, while another that favors its extension into the ribosomal P-site had the opposite effect. The two mutants also had opposing effects on binding of aa-tRNA to the ribosomal A-site, and downstream functional effects were observed on translational fidelity, drug resistance/hypersensitivity, virus maintenance and overall cell growth. These analyses suggest that the L11 P-site loop normally helps to optimize ribosome function by monitoring the occupancy status of the ribosomal P-site.

Over the past decade, atomic resolution ribosome structures have revealed the locations of critical elements. However, these static images do not reveal the dynamic movements within this complex macromolecule. The ribosome must coordinate multiple activities between spatially and functionally different sites in two subunits. These include three transfer RNA (tRNA)-binding sites, the peptidyltransferase and decoding centers and the elongation factor interacting regions. Events occurring in these regions must be carefully coordinated to assure rapid and accurate decoding of messenger RNAs (mRNAs). Current efforts in the field are focusing on determining the mechanisms by which these functional centers synchronize their actions and communicate with each other. The eukaryotic ribosome contains nearly 80 intrinsic proteins. The high degree of similarity across species, from the primary amino acid sequences to their tertiary structures, suggests conserved functional roles beyond serving as mere scaffolding for the rRNAs. Ribosomal protein L11 of Saccharomyces cerevisiae is an essential, highly conserved component of the 60S subunit (in bacteria and archaea, the homologous protein is named L5; the yeast nomenclature is used throughout this text to minimize confusion). At the primary amino acid sequence level, L11 is well conserved among eukaryotes ($67-97% identity), while bacterial and archaeal L5 proteins are less well conserved (27-55% identical) (Supplementary Figure S1A ). L11 is uniquely positioned at the interface between the large subunit central protuberance ( Figure 1A and 1B) and the head of the small subunit ( Figure 1A ) (1) (2) (3) (4) (5) . In the small subunit, the head region undergoes significant rotational movement relative to the central protuberance between the pre-and post-translocational states (6) , and the protein-protein interactions between L11 and S18 (S13 in bacteria and archaea) on the small subunit (the B1b and B1c intersubunit bridges) undergo the largest intersubunit structural rearrangements between these two states (7) (8) (9) . These observations suggest that L11 may play a central role as an informational conduit between the two subunits. Detailed analysis of X-ray crystallographic and cryo-EM structures ( Figure 1B ) reveals that the concave surface of the b-sheet portion of L11 interacts with specific nucleotides in the minor groove of 23S rRNA helix 84 (10) . L11 also makes contacts with the helix III and loop C regions of 5S rRNA; these connections have been hypothesized to help stabilize 5S rRNA interactions and may participate in an information signal transmission network linking functional centers within the ribosome (1, 11) . Importantly, the B1b and B1c intersubunit bridges with S18 are the only protein-protein interactions between the two subunits (2, 3, 5, 8) . Analyses of these structures indicate that contacts involving L11 and S18 through the B1b and B1c bridges break and rearrange after eEF-2 binding and ribosome ratcheting, controlled in part by differentially charged amino acid side chains between the two proteins (2, 4, 5, 7, 9, 12) . An internal loop of L11 that we denote the 'L11 P-site loop', which is roughly formed by amino acid residues 48-68, also directly contacts the T-loop of the peptidyl-tRNA in the P-site through tRNA nucleotide 56 (3) (4) (5) . At the level of primary amino acid sequence, the P-site loop is highly conserved among eukaryotes (85-100% identity), while it is less well conserved among bacteria and archaea (42-57% identity) (Supplementary Figure S1B) . At the biochemical level, however, the P-site loop is significantly more homogeneous, containing a large number of well-aligned charged and aromatic amino acids. In particular, A50, F57, R60 and I65 (yeast numbering) are universally conserved. An alignment of the P-site loop structures from yeast, Haloarcula marismortui, Thermus thermophilus and Escherichia coli reveals that the P-site loop is extremely well conserved at the structural level (Supplementary Figure S1C) . In yeast, L11 is encoded by the paralogous genes RPL11A and RPL11B located on chromosomes 16 and 7, respectively (13) . The 19-kDa proteins are 174 amino acids long and are identical except for an alanine (L11A) to threonine (L11B) difference at the third amino acid position. Analysis of L11 in the late 1980s (a.k.a. L16) showed that expression of either isoform was sufficient for cell viability (14) . However, when expressed as the sole form of L11, RPL11A mRNA transcripts accumulated to only 33-40% of wild-type levels as compared to cells expressing both isogenes, while RPL11B mRNAs accumulated to 60-66%. Expression of either isogene alone also affected 60S subunit assembly: a strain expressing only L11B grew at wild-type rates but synthesized fewer 60S subunits than wild-type cells (although apparently not below a threshold necessary for wild-type growth rates), while strains expressing only L11A grew more slowly than wild-type, and synthesized only 33-40% of wild-type levels of total L11 and 60S subunits (14) . A random mutagenesis screen of RPL11B for cold-sensitive mutants identified alleles that promoted 25S pre-rRNA processing and initiation defects (15) . The specific mutants identified in that study were S34P, S41P, S97F, A98V, S119C and G135D. In Arabidopsis, divergent 5 0 untranslated regions (UTRs) between the two isogenes were found to result in differential expression among plant tissues (16) . In addition to its function as a ribosomal protein, L11 has been implicated in p53 activation through its interactions with HDM2 in the nucleus of human fibroblast cells (17) , and mutant forms of L11 have been linked to Daimond-Blackfan anemia in humans (18) . Although the structural information suggests that L11 should play a significant role in translation, functional analyses of the protein in this role have not been performed. In this report, a series of mutants were generated using a reverse genetics approach to parse the role of the L11 P-site loop. Detailed biochemical and structural analyses focused on two multi-amino acid mutants with opposing effects on rRNA structure and tRNA binding. We propose that prior to peptidyltransfer, the presence of peptidyl-tRNA in the large subunit P-site positions the L11 P-site loop to interact with the Helix 84 of the large subunit rRNA. After peptidyltransfer, spontaneous translocation of the deacylated tRNA to the large subunit E-site allows the L11 P-site loop to extend into the P-site, breaking contact with Helix 84. By this model, we hypothesize that the L11 P-site loop functions locally as a sensor of the occupancy status of the ribosomal P-site. Restriction enzymes were obtained from Promega (Madison, WI, USA), MBI Fermentas (Vilnius, Lithuania) and Roche Applied Science (Indianapolis, IN, USA). The QuikChange XL II site-directed specific mutagenesis kit was purchased from Stratagene (La Jolla, CA, USA). DNA sequencing was performed by Genewiz (Germantown, MD, USA). Escherichia coli DH5a was used to amplify plasmid DNA. Transformation of yeast and E. coli and were performed as previously described (19) . YPAD, SD and 4.7 MB plates for testing the killer phenotype were as previously reported (19) . Plasmids for expression of dual luciferase reporters were described previously (20) . Saccharomyces cerevisiae strain PSY2088 (MAT rpl11a::HIS3 rpl11b::HIS3 ura3-52 leu2D1 trp1D63 his3D200 + YCpL11B URA3), an rpl11a/rpl11b gene deletion strain in which L11 is supplied by a URA3-CEN6 based RPL11B clone, was a generous gift from Dr. Pamela Silver (21) . The L-A and M 1 viruses were introduced into PSY2088 by cytoplasmic mixing (cytoduction) through nonproductive mating with JD758 [MATa kar1-1 arg1 (L-AHN M 1 )] to produce the Killer + strain JD1313 as previously described (19) . Wild-type RPL11B was isolated from yeast strain PSY2088 plasmid (pYCP50L11B URA3). Using flanking BamHI restriction sites, a 2.2-kb fragment of DNA containing both the 525-bp wild-type RPLL11B ORF plus the native 5 0 and 3 0 UTR regions (1228 bp and 485 bp, respectively) was purified by agarose gel electrophoresis. This 2238-bp fragment was ligated into BamHI digested pRS314, a low copy TRP1-selectable plasmid (purchased from ATCC, Manassas, VA, USA) (22) to create pRS314L11B-TRP1. This plasmid served as the template for generation of rpl11B mutants by site directed mutagenesis using the primers listed in Supplementary Table S1 . Wild-type and mutant pRS314L11B-TRP1 clones were transformed into JD1313, selected for growth on -trp medium, and cells having lost the URA3-based plasmid were identified by their ability to grow in the presence of 5-fluoroorotic acid (5-FOA) (23) . The effects of temperature and translational inhibitors were assessed by standard 10-fold dilution spot assays. Yeast were grown in H-tryptophan synthetic deletion (SD) media (-Trp) to mid log phase. OD 595 values were obtained, and cells were serially diluted 10-fold from 10 5 to 1 CFU per 2.2 ml and spotted on -Trp plates. Growth was monitored at 20 C, 30 C and 37 C, and pharmacogenetic assays utilized 2 mg/ml paromomycin, 40 mg/ml anisomycin or 30 mg/ml sparsomycin incubated at 30 C for 3-5 days. Killer virus assays were performed as previously described (19) . The dual luciferase reporter plasmids pYDL-control, pYDL-LA, pYDL-Ty1, pYDL-UAA (20) and pYDL-AGC 218 (24) were employed to quantitatively monitor programmed À1 ribosomal frameshifting, programmed +1 ribosomal frameshifting, suppression of a UAA codon and suppression of an AGC serine codon in place of an AGA argine codon in the firefly luciferase catalytic site respectively. In this study, the reporters were housed in LEU2-based reporters: the 0 frame dual luciferase reporter was pJD419, the L-A dsRNA virus À1 PRF containing reporter was pJD420 and the Ty1 containing +1 PRF reporter was pJD421. Cells were grown overnight in 5-ml volumes of -leu synthetic depletion media to mid log phase (A 595 = 0.8-1.5). Cells were washed, resuspended in lysis buffer (1X PBS pH 7.4, 1 mM PMSF) and lysed using 0.5-mm glass beads with a vortex mixer for 3-5 min at 4 C. Lysates were clarified by centrifugation for 5 min. at 8000 r.p.m. at 4 C. Samples were maintained on ice, and 5 ml of clarified lysate was added to 50 ml of pre-aliquoted Promega LARII reagent, mixed by pipetting, and read in a TD20/20 luminometer. Immediately upon completion of this read, 50 ml of Promega Stop and Glo buffer was added to the tube, pipetted to mix and read again. This was repeated 6-12 times per strain per reporter depending on the consistency of the data. Frameshifting rates were determined by taking the ratio of firefly to Renilla luciferases for each sample, and then taking the ratio of the average ratios of the 0 frame samples to that of test reporter ratios to obtain the rates for both À1 and+1 PRF. These results were then analyzed by t-test to determine statistical significance compared to wild-type levels as previously described (25) . Prior to determining rates of UAA readthrough (nonsense suppression), strains were cured of the endogenous yeast prion [PSI + ] by daily serial passage of cells in -trp liquid media containing 5 mM guanidine hydrochloride for 10 days. Rates of nonsense suppression were determined as previously described (20) using the LEU2selectable 0-frame control pJD419 and in-frame UAA containing reporter pJD702. Missense reporters were based on URA3 plasmids previously described for the sense reporter (20) and for the firefly luciferase 218 arginine codon (AGA) to serine (AGC) missense reporter plasmid pYDL-AGC (24) . Methodologies were the same as those for other dual luciferase assays described above. Cells were grown overnight in a 30 C shaker in 500 ml of YPAD media to mid-log phase (OD 595 0.8-1.5), cooled to 4 C for 1 h to allow ribosomes to run off of transcripts while remaining tightly coupled. Cells were harvested by centrifugation and washed three times with 40 ml 0.9% KCl solution. Cell pellets were stored at À80 C until needed, at which time they were thawed and resuspended in 1 ml binding buffer (10 mM Tris-HCl pH 7.5, 10 mM MgCl 2 , 60 mM NH 4 Cl, 2 mM DTT, 1 mM PMSF) per gram of cells. Cells were lysed with a 1:1 vol of Zirconian beads (BioSpec, Bartlesville, OK, USA) and disrupted using two 2-min pulses of a minibead beater. Lysates were clarified by centrifugation at 20 000 r.p.m. (50 000 g) using an MSL-50 rotor at 4 C for 25 min. Ribosomes were chromatographically purified using Sulfolink beads (Pierce, Rockford, IL, USA) as previously described (26) , and eluted from the resin in 8 ml of elution buffer (10 mM Tris-HCl pH 7.5, 10 mM MgCl 2 , 500 mM KCl, 2 mM DTT, 0.5 mg/ml heparin). Eluted ribosomes were treated with 2 mM puromycin and 1 mM GTP for 30 min at 30 C and were layered on top of a 22-ml glycerol cushion [50 mM HEPES-KOH pH 7.6, 10 mM Mg(CH 3 COO) 2 , 50mM NH 4 Cl, 1 mM DTT, 25% glycerol] and pelleted by centrifugation at 30 000 r.p.m. at 4 C for 18-20 h. Pellets were washed with 1 ml of storage buffer [50 mM HEPES-KOH pH 7.6, 10 mM Mg(CH 3 COO) 2 , 50mM NH 4 Cl, 1 mM DTT, 25% glycerol], and resuspended in 200-400 ml of storage buffer. Concentrations were determined spectrophotometrically (1 OD 260 = 20 pmol ribosomes). The salt-washed ribosomes were aliquoted and stored at À80 C for up to 3 months. Ribosomal rRNA quality was checked on 1.3% agarose gels and rRNA to protein ratios were monitored by determining OD 260 to OD 280 ratios. Polysome profiles were obtained by sucrose density gradient centrifugation as previously described (27) . Samples were split, and 10 ml of dimethyl sulfoxide (DMSO) was added to half of the samples, while 10 ml of 60 mM 1M7 was added to the other half. Samples were incubated at 30 C for 20 min. Ribosomes were precipitated by the addition of 275 ml of ice-cold 100% ethanol and stored at À20 C for 1-2 h. Ribosomes were pelleted by centrifugation at 10 000 r.p.m. for 2 min and resuspended in lysis buffer and rRNAs were isolated using an Ambion (Austin, TX, USA) RNAqueous Õ -Micro RNA isolation kit. Optical densities were taken at 260 nm and 280 nm to monitor the quantity and quality of RNA, and samples were resuspended at a concentration of 1 mg rRNA/7 ml in pure water. HPLC purified oligonucleotide primers purchased from IDT (Coralville, IA, USA) are listed in Supplementary Table 2. Oligonucleotides were resuspended to 25 pmol/ml, 3 0 end labeled with g[ 32 P]ATP with T4 polynucleotide kinase (Roche, Indianapolis, IN, USA), and purified from free radiolabeled nucleotide by passage through a MicroSpin G-25 column (GE Healthcare, Piscataway, NJ, USA). Annealing reactions utilized 1 mg of modified rRNAs and 3 ml of labeled oligonucleotide heated at 80 C for 1 min, followed by a 4-7-min incubation at 5-10 C below the T m of each oligonucleotide. Annealed rRNA/ primers (2 ml each) were added to 3 ml of cold enzyme mix [0.25 ml 10 mM dNTP, 0.25 ml 100 mM DTT, 1 ml 5X Superscript III Buffer, 0.25 ml Superscript III (Invitrogen Life Technologies, Carlsbad, CA, USA), 1.25 ml H 2 O]. For sequencing samples, an additional 1 ml of each ddNTP was added to each C, T, A, G, sample, respectively. Primer extension reactions were performed at 52 C for 25 min, with potential 2-min-long extensions preceding the 52 C at lower temperatures depending on the individual T m values of the primers. Denaturing RNA loading dye (2 ml) was added to each sample, heated to 94 C for 2.5 min, and samples were resolved through 6% urea-acrylamide denaturing gels. Gels were dried and radiolabeled samples were visualized by phosphorimagery. The published structures for the 70S ribosome from E. coli [PDB accession numbers: 2AVY, 2AW4; (2)], as well as yeast 80S structures from yeast (1S1I, 1S1H, 3JYV, 3JYW, 3JYX; (4, 12) ] were used in the analysis of this work and the generation of figures. Published T. thermophilus 70S subunits containing A-site, P-site and E-site Phe-tRNA were also employed (1G1X, (5) . All structures were visualized and manipulated using MacPyMol software (30) . The visualization of a single salient loop of L11 interacting with peptidyl-tRNA indicated that it might play a vital role in sensing peptidyl-tRNA occupancy status and transmitting this information to other functional centers of the ribosome. As cells expressing RPL11B alone were healthier than those solely expressing RPL11A, genetic manipulations began with the yeast rpl11AD rpl11BD double knockout strain JD1313 expressing wild-type (WT) RPL11B from a low-copy, URA3-selectable episomal plasmid (pRPL11B-URA3). Oligonucleotide site-directed mutagenesis was used to construct a series of mutants, each containing changes of 1, 4 or 10 sequential amino acids ( Figure 1C ). Stretches of amino acids from arginine 51 to arginine 61 in the L11 P-site loop were targeted for site-directed mutagenesis expressed from a low-copy, TRP1-selectable episomal plasmid under control of the endogenous RPL11B promoter (pRPL11B-TRP1). After transformation and selection on SD medium lacking tryptophan (-trp), cells expressing only mutant rpl11B alleles were identified by their ability to grow on SD-trp medium containing 5-flouroorotic acid (5-FOA). Three of the multiple substitution mutants were inviable as the sole forms of L11B. These were R 51 YTVRTFGIR 60 !alanine (i.e. 51-60A); deletion of residues 51-60 (51-60Á); and F 57 GIR 60 !alanine (57-60A). Viable mutants, R 51 YTV 54 !alanine (51-4A), V 54 RTF 57 !alanine (54-7A), R51A, Y52* (mutations including Á, A, R, E, S, I, Q, N, H and F), F57A and R61A were rescued from yeast into E. coli, and the mutations were confirmed by DNA sequencing. The L11 P-site loop mutants confer temperature-and drug-specific growth phenotypes displayed roughly wild-type growth. Cold sensitivity was assessed at 20 C and both mutants grew at wild-type rates. 51-4A showed enhanced growth at 37 C relative to itself at 30 C, while mutant 54-7A was similar to wild type. R51A grew at wild-type rates at 30 C and 37 C but showed enhanced growth at 20 C. The Y52* mutants displayed mutant-specific effects on growth rates at 30 C, but did not confer significant phenotypes at either 20 C or 37 C. F57A had wild-type growth rates at all temperatures, while R61A showed depressed growth at 30 C, which was rescued at 37 C. Small molecule inhibitors of protein translation are useful probes for identifying changes in ribosome function. This study utilized three such molecules: paromomycin, anisomycin and sparsomycin. The effects of all three drugs were monitored using dilution spot assays at 30 C on SD-trp media containing various drug concentrations. Paromomycin is an aminoglycoside antibiotic that increases translational error rates by artificially stabilizing codon:anticodon interactions at the decoding center in the small ribosomal subunit (31) . As compared to their intrinsic growth in the absence of drug, both the 51-4A and 54-7A mutants were slightly hypersensitive to 2 mg/ml paromomycin, as were Y52Á, Y52N and Y52H. In contrast, R51A, F57A and R61A were all paromomycin resistant ( Figure 2B ). Anisomycin competes with the 3 0 end of the aa-tRNA for binding to the A-site pocket of the ribosome (32, 33) . Both 51-4A and 54-7A showed anisomycin resistance at 40 mg/ml, as did several Y52* mutants, and R61A ( Figure 2B ). Sparsomycin binds to the P-site and interferes with peptidyl-tRNA binding and peptidyl transfer (33, 34) . 51-4A and 54-7A mutants were hypersensitive to 30 mg/ml sparsomycin, as were most of the Y52* mutants, with the exception of Y52F, which conferred slight resistance to this drug ( Figure 2B ). The yeast 'killer' system is composed of the L-A helper and M 1 satellite dsRNA viruses (35) . The L-A dsRNA viral genome encodes a capsid protein (Gag), and an RNA-dependent RNA polymerase (Pol) that is synthesized as a Gag-pol fusion protein consequent to a À1 Programmed Ribosomal Frameshifting (PRF) event (36) . The M 1 satellite dsRNA is encapsidated and replicated in L-A encoded viral particles, and the M 1 (+) strand encodes a secreted toxin that kills uninfected yeast through its interactions with the GPI-anchored Kre1p cell wall assembly protein (37) . Changes in À1 PRF efficiency alter the ratio of Gag to Gag-pol, and inhibit the ability of cells to maintain M 1 (19) . To monitor the effects of the mutants on Killer virus maintenance, colonies of JD1313 cells expressing either wild-type or mutant rpl11B alleles were spotted onto a lawn of diploid, Killer À indicator cells. Cells expressing wild-type RPL11B were Killer + as demonstrated by their ability to inhibit growth of the indicator cells ( Figure 2C ). In contrast, isogenic cells expressing the 51-4A, 54-7A and F57A mutants were Killer À . A weak killer phenotype, defined by decreased zones of growth inhibition, was observed in mutants Y52E, Y52N, Y52H and F57A. The rpL11B mutants affect translational fidelity 'Translational fidelity' is generically used to describe the accuracy of protein synthesis. A series of bicistronic reporter plasmids were used to quantitatively monitor the effects of the L11B mutants on four aspects of translational fidelity: À1 PRF, +1 PRF, suppression of a UAA nonsense codon and incorporation of a missense near-cognate amino acid. In JD1313 cells expressing wild-type RPL11B, À1 PRF directed by the L-A dsRNA viral signal was 6.07% ± 0.16%. This compares favorably with other 'wild-type' strains in our laboratory (normal range from 4% to 8% (20, 25) . The 51-4A mutant promoted increased À1 PRF (1.33 ± 0.06-fold relative to wild type), while 54-7A trended in the opposite direction (0.84 ± 0.03-fold relative to wild type) ( Figure 3 , and Table 1 ). Both these values were statistically significant and correlate well with the Killer À phenotypes. Y52Á, Y52N, Y52E and Y52H mutants also showed increased rates of -1 PRF, with statistically significant rates ranging from Y52H at 1.20-fold wild type to Y52E at 1.49-fold wild-type. Y52A, Y52S, Y52Q and Y52F all had wild-type rates of À1 PRF. While both À1 and +1 PRF are kinetically driven events, the substrates for the slippage are distinct: À1 PRF requires that both the ribosomal A-and P-sites are occupied by tRNAs, while+1 PRF occurs while the A-site is empty (38) . Rates of +1 PRF were monitored using a cis-acting signal derived from the Ty1 retrotransposable element using pYDL-Ty1. Baseline +1 PRF efficiencies in cells expressing wild-type RPL11B were 10.98% ± 0.30%. 51-4A had no effects on +1 PRF, while 54-7A promoted a small but statistically significant increase (1.18 ± 0.06-fold of wild type; Figure 3 ). Significant changes in +1 PRF were also observed in the Y52A, Y52S, Y52N, Y52E, Y52H and Y52F mutants. mRNA decoding occurs in the small subunit decoding center, and changes in termination codon recognition (nonsense suppression) is another indicator of altered translational fidelity. pYDL-UAA (39) , which contains an in-frame termination codon immediately 5 0 of the firefly luciferase gene, was used to monitor this parameter. The baseline rate of nonsense suppression in cells expressing RPL11B was 0.137% ± 0.003%. The 51-4A mutant slightly improved this aspect of translational fidelity, with nonsense suppression levels decreasing to 0.88 ± 0.04-fold of wild-type levels. 54-7A did not affect UAA recognition (Figure 3) . Y52Á, Y52A, Y52S, Y52N, Y52E and Y52H all promoted increased rates of nonsense suppression ranging from 1.31-to 1.78-fold wild type. pYDL-AGC 218 tests missense suppression levels by monitoring rates of incorporation of an arginine (AGA) near-cognate amino acid instead of a cognate serine (AGC) at the catalytic codon 218 within the firefly luciferase gene as previously described (24) . Thus, in this assay, mis-utilization of near-cognate tRNA Arg at the Ser AGC codon restores firefly luciferase activity. Wild-type missense levels were measured at 0.074% ± 0.002, comparable to previous studies (24) . Mutant 51-4A had significantly higher levels of missense suppression (measured at 1.21 ± 0.03-fold wild-type), while 54-7A did not significantly affect this phenomenon (1.07 ± 0.04 fold wild type) ( Figure 3 ). Missense suppression was not assayed for the single amino acid mutants. The mutant rpl11b alleles promote opposing effects on tRNA binding to the ribosomal A-and P-sites Sucrose gradient analyses were employed to fractionate cycloheximide arrested elongating ribosomes on mRNAs in lysates generated from JD1313 cells expressing wild-type L11B, 51-4A, and 54-7A. In all strains the 60S peak was smaller than that of the 40S fraction which can be attributed to the presence of only a single copy of RPL11B, which has previously been shown to effectively reduce the number of 60S subunits produced by the cell to 60-66% of true wild-type levels while having no visible phenotypic effect on growth (14) . No significant differences were observed among the samples (data not shown). Phenotypic variation in PRF and in the presence of anisomycin and sparsomycin are indicative of altered interactions between the ribosome and tRNAs. P-site tRNA K d values were determined in vitro by binding 2-fold serial dilutions of N-acetylated-[ 14 C]Phe-tRNA to ribosomes until saturation was achieved ( Figure 4A ), and the resulting data were used to determine steady-state single site binding K d values ( Figure 4B ). Wild-type ribosomes bound this P-site substrate with a K d of 72.3 ± 7.9 nM. The 51-4A mutants promoted a slight increase in affinity for P-site substrate (K d = 50.9 ± 11.2 nM), while 54-7A had the opposite effect (K d = 89.3 ± 10.4 nM). Given the physical interaction between the L11 P-site loop and peptidyl-tRNA, it was imperative to determine whether the observed small changes in P-site affinities promoted by the mutants were biochemically significant. To this end, multiple turnover puromycin reactions were performed. In these experiments, puromycin was added to ribosomes pre-incubated with excess P-site substrate, i.e. Ac-[ 14 C]Phe-tRNA Phe , and accumulation of the peptidylpuromycin product was monitored over time. In these reactions, the first round of peptidylpuromycin synthesis is very rapid. Next, in a slow step, the ribosome intrinsically translocates the deacylated tRNA Phe into the E-site (40), followed by the slow diffusion of Ac-[ 14 C]Phe-tRNA Phe into the P-site where it can react with puromycin. Repetition of this cycle results in slow multiple rounds of product synthesis ( Figure 4C ). Assuming that the L11 mutants do not affect either rates of intrinsic translocation or of Ac-[ 14 C]Phe-tRNA Phe diffusion into the P-site, changes in product accumulation, i.e. K obs , should be due to differences in binding affinities for the P-site substrate. Consistent with this model 51-4A promoted 1.46 ± 0.14-fold increased K obs relative to wild-type ribosomes, while 54-7A decreased K obs to Figure 3 . The L11B mutants promote defects in translational fidelity. Isogenic yeast cells expressing either wild-type or mutant forms of L11B were transformed with dual luciferase reporters and control plasmids and rates of translational recoding were determined. All results are graphed as fold wild type. À1 PRF was measured using the yeast L-A virus frameshift signal. +1 PRF was directed by the frameshift signal derived from the Ty1 retrotransposable element. Nonsense suppression denotes the percentage of ribosomes able to suppress an in-frame UAA termination codon positioned between the Renilla and firefly luciferase reporter genes. Missense suppression rates were evaluated by incorporation of an arginine (AGA) near-cognate amino acid instead of a cognate serine (AGC) at the catalytic codon 218 within the firefly luciferase gene. Error bars denote standard error. P-values are indicated above samples showing statistically significant changes. (Figure 4E and F). The P-site loop is flexible depending on the occupancy status of the P-site The highly basic nature of the P-site loop, its interaction with peptidyl-tRNA, and its proximity to 25S rRNA Helix 84 (H84) suggested that it might interact with either of these two RNA components depending on the occupancy status of the P-site. Changes in interactions between the P-site loop and local rRNA structures may in turn propagate outward to more distant regions of the ribosome. To test this, SHAPE (41) (42) (43) was employed to probe for structural alterations in selected regions of the 25S, 18S and 5S rRNAs due to either the L11B mutants or in wild-type ribosomes with occupied or unoccupied P-sites. Due to the large size and complex three-dimensional structure of the ribosome, the entire rRNA content was not examined. Rather, approximately one-third of the rRNA bases were interrogated, focusing on those bases closest to L11, the A-and P-sites, and the decoding center. In the first series of experiments, salt-washed wild-type and 51-4A, 54-7A, Y52Q and Y52F mutant ribosomes (chosen for structural analyses because they had the most pronounced genetic phenotypes) were treated with 1M7, an electrophile that adds an adduct onto the 2 0 OH groups of solvent exposed base sugars. Modifications were performed on salt-washed ribosomes because they represent the thermodynamic 'ground state' of the ribosome. Thus, the structural changes observed are indicative of changes in the full 'dynamic potential' of the ribosome as opposed to conformations locked in by e.g. occupation of binding sites by tRNAs or ribosome-associated factors. rRNAs were extracted, hybridized with 5 0 [ 32 P]-labeled oligonucleotide primers and reverse transcriptase primer extension reactions were performed. The products were separated through urea-acylamide denaturing gels, and visualized using a phosphorimager. 2 0 -OH ribose modification results in a strong stop 1-nt 3 0 of modified bases, and the intensity of the stops are proportional to the solvent accessibility and flexibility of riboses. Comparison of the protection patterns between wild-type and mutant ribosomes enables identification of specific bases which became protected or deprotected relative to WT. In all areas examined, rpl11b ribosomes Y52Q and Y52F matched the wild-type rRNA base modification profile (data not shown), while 51-4A and 54-7A ribosomes revealed consistently reproducible differences. The most significant changes in rRNA structure were observed in bases C2675-A2679 (E. coli numbering: C2306-2310) located in the terminal loop of 25S rRNA H84 ( Figure 5A and E). The two mutants promoted opposing patterns of base protection/deprotection in this structure. Specifically, as compared to wild-type ribosomes, 51-4A promoted enhanced protection of this loop, while the loop was deprotected in the 54-7A mutants. Analysis of the recent cryo-EM yeast ribosome structure (4) revealed that these H84 loop bases are located within 3 Å of the stretches of amino acids changed to alanines in both the 51-4A and 54-7A mutants ( Figure 5B ). These findings suggested that the two mutants had the effects of displacing the P-site loop into two opposing conformational states: extended toward the P-site (54-57A), or retracted into H84 (51-54A). To test whether these two states are naturally dependent on P-site occupancy, the experiments were repeated with wild-type and mutant ribosomes with or without tRNA Phe in their P-sites. Consistent with this model, addition of tRNA to the P-site of wild-type ribosomes resulted in slightly enhanced protection of the H84 terminal loop bases closest to the P-site loop (A2676-A2679). Interestingly, C2675 showed significant deprotection when the P-site was occupied by tRNA. This base is on the far side of the terminal end of H84 from the P-site loop, suggesting that H84 itself alters its conformation upon tRNA occupancy of the P-site ( Figure 5C ). 51-4A's H84 bases were unchanged between P-site bound and unbound ribosomes, consistent with the P-site loop positioned in the 'retracted' state in this mutant, although small differences in the protection patterns suggest that the P-site loop is in a slightly different orientation in this mutant. In contrast, while 54-7A ribosomes, i.e. the P-site loop 'extended' state, showed deprotection at all bases (C2675-A2679) for both P-site bound and salt washed ribosomes, bases A2676-A2679 were less deprotected when tRNA was in the P-site and C2675 was even more reactive, consistent with the notion that the P-site loop interacts with H84 when peptidyl-tRNA is in the P-site. Although no other SHAPE-specific changes were observed, several other phosphodiester bonds 3 0 of specific 25S rRNA bases were reproducibly more, or less, intrinsically labile as compared to wild type ( Figure 5D ). In both mutants, G2531 and G2534 located in expansion segment 31 (ES31) were more stable than in wild-type ribosomes as evidenced by reduced intensity of strong reverse transcriptase stops 1-nt 5 0 of these bases. Additionally, bases A2779-A2780 (E. coli A2407-U2408) located in the terminal loop of Helix 88 were hyper-labile in 51-4A mutant ribosomes as compared to WT, as shown by the presence of strong stops with increased intensity 1-nt 5 0 of these bases. These are mapped onto the two-dimensional structure of yeast 25S rRNA ( Figure 5E ). The L11 P-site loop is largely comprised of polar amino acids and carries a net positive charge, making it ideal for interactions with the phosphate backbones of nucleic acids, e.g. rRNA and tRNA. Positioned between H84 and the peptidyl-tRNA T-loop, several of its amino acids are within H-bonding distance of H84 ($3.3 Å ), while C56 of the peptidyl tRNA T-loop comes within 2.1 Å of G58 in the L11 P-site loop (4, 5) , suggesting that the L11 P-site loop can directly interact with both of the RNA-based structures. While currently available X-ray crystal structures are unavailable for ratchet-state ribosomes, a recently published examination of tRNA movement through the E. coli ribosome using large-scale analysis of cryo-EM images implicates the P-site loop as a dynamic arm interacting with and moving in relation to tRNAs passing across the P-site (44) . Although these studies were performed at resolutions of 9-20 Å , leaving considerable ambiguity regarding the precise residues involved, they clearly reveal highly dynamic interactions between the P-site loop and both P-site, and E-site tRNAs. Although death is not a phenotype per se, the inviable mutants are informative nonetheless in so far as they demonstrate that the amino acids F 57 GIR 60 are absolutely required for viability. While F57 is universally conserved, it does not appear to be essential on its own for viability, as witnessed in the mild phenotypes of the F57A mutant. Similarly, all single amino acid changes explored here resulted in viable cells, suggesting a certain degree of biochemical/biophysical redundancy within this essential loop. In support of this notion, the strongest growth phenotypes observed across a range of temperatures and small molecule translational inhibitors were concentrated in the multiple alanine substitutions, i.e. 51-4A and 54-7A, thus directing the bulk of the biochemical and structural analyses to these two mutants. Analysis of the results of the assays performed on the viable multiple alanine substitution mutants (summarized in Table 1 ) provoke the hypothesis that the L11 P-site loop may dynamically function to help the ribosome sense the occupancy status of the large ribosomal subunit P-site. This is modeled in Figure 6 . When the P-site is unoccupied, the P-site loop can extend into this space, moving away from the terminal loop of H84. Upon occupation of the P-site, the peptidyl-tRNA T-loop displaces the L11 P-site loop, causing its retraction into H84. By this model, the rRNA SHAPE analyses depicting increased protection of Helix 84 by the 51-4A mutant show that this mutant drives the L11 P-site loop equilibrium toward the 'retracted' state. Conversely, increased deprotection of Helix 84 in the 54-7A mutant suggests that this more mimics the P-site unoccupied state, i.e. the 'extended' P-site loop state. This analysis directly explains the P-site binding data. Retraction of the P-site loop from the P-site results in 51-4A ribosomes having higher intrinsic affinity for this substrate while extension of this structure into the P-site creates a steric clash with the peptidyl-tRNA T-loop, resulting in decreased affinity for this substrate. That neither mutant conferred optimal peptidyl-tRNA P-site occupancy may account for their hypersensitivity to sparsomycin, especially for 54-7A in which the P-site loop is already competing with the tRNA for the P-site. Mutants 57-60A, 51-60A and 51-60Á appear to disrupt the normal function of the P-site loop to a lethal level. In addition, the observation that tRNA binding to the P-site results in deprotection of C2675 implicates H84 itself as a structurally dynamic unit. The functional consequences of this are not clear, although it is tempting to speculate that this conformational change may play a role in the structural rearrangements of the B1b and B1c bridges between the pre-and post-translocational states. The lack of rRNA structural changes in the A-site or in the decoding center suggest that the biochemical and phenotypic effects observed are indirectly due to the changes described above. The reciprocal effects between Ac-aa-tRNA binding with the P-site and aa-tRNA interactions with the A-site are intriguing. In the aa-tRNA binding reactions, the ribosomal P-sites were occupied with daeacylated tRNA. We suggest that in the 51-4A mutant, the P-site ligand is more 'locked' into a suboptimal conformation, which in turn feeds back to the A-site, resulting in decreased affinity for its ligand. Conversely, the lessened ability of 54-7A mutant ribosomes to lock P-site ligand in a suboptimal conformation may account for the increased affinity of these ribosomes for A-site ligand. Anisomycin resistance by both mutants also followed the reciprocal P-site/A-site pattern, i.e. both mutants were sparsomycin hypersensitive. Paromomycin interacts with the decoding center in the small subunit, where it promotes misreading of near-cognate codons in the A-site by stabilizing codon-anticodon interactions (45) . This sensitivity may be attributable to an observed increase in missense incorporation of a near cognate arginine (AGA) over that of the sense serine codon (AGC) in mutant 51-4A. Intriguingly, 54-7A had wild-type levels of missense incorporation suggesting that its sensitivity to paromomycin was indirect. The reciprocal anisomycin/paromomycin phenotypes of the L11 mutants demonstrate the effects of this protein on A-site ligand based ribosomal functions over very long distances. Similar phenotypic patterns were previously observed with mutants of other large subunit components (46, 47) . The observed effects on -1 PRF are consistent with a recent kinetic analysis demonstrating that aa-tRNA slippage is the most highly weighted parameter in determining the rate at which this process occurs (Liao,P.Y. et al., submitted for publication). Here, increased affinity for aa-tRNA by the large subunit suggests that the 51-4A ribosomes stabilize the frameshifted (i.e. near-cognate) tRNAs, reducing their ability to be proofread, thus promoting increased rates of À1 PRF. This is consistent with the observed increased rates of missense decoding in this mutant. Conversely, post-slippage A-site tRNAs are even less stable in the 54-7A mutants, leading these to be more efficiently proofread, and thus promoting decreased À1 PRF efficiency. In both cases, altering À1 PRF from the optimum 'golden mean' precludes these cells from maintaining the yeast killer virus (19, 48) . Programmed +1 frameshifting is completely dependent on peptidyl-tRNA slippage. Increased +1 PRF in the 54-57A mutant is consistent with decreased affinity for this substrate. The failure to observe decreased +1 PRF in the 51-54A mutant, despite its increased affinity for peptidyl-tRNA, is not entirely clear, although this may be due to the inability of these ribosomes to achieve a threshold beyond which +1 PRF effects can be observed. The changes in rRNA stability observed in the terminal loop of Helix 88 and in ES31 are intriguing. Chemical protection experiments revealed the terminal loop of Helix 88 is involved in a kissing loop interaction with the terminal loop of Helix 22, and this interaction is apparent in the X-ray crystal and cryo-EM structures (4, 49) . Increased lability at A2779 and A2780 was tRNA H84 L11 S18 P-Site H84 L11 S18 C C Figure 6 . Model: the P-site loop acts as a sensor of the occupancy status of the P-site. (Left) When the large subunit P-site is unoccupied by tRNA, the L11 P-site loop is able to extend into this space leaving the distal loop of H84 partially deprotected from chemical attack. This conformation is favored by the 54-7A mutant of L11B. (Right panel) Occupation of the P-site by peptidyl-tRNA displaces the L11 P-site loop, causing it to tightly retract from the P-site and interact with H84, resulting in increased protection of the H84 terminal loop from chemical attack. H84 likely moves toward the P-site loop slightly, increasing the exposure of C2675 to the surrounding solvent. This conformation is favored by the L11B 51-4A mutant. previously observed in the Y11C mutant of ribosomal protein L10 (homolog of E. coli L16) located at the base of the aa-tRNA accommodation corridor, and in the É2922C (E. coli U2554) 25S rRNA mutant located in the peptidyltransferase center (50, 51) . The observation that mutations located in three very different and topologically distinct regions of the large subunit conferred similar structural effects suggest that this kissing loop interaction plays an important role in ribosome function. Its location on the cytoplasmic face of the ribosome where deacylated tRNA leaves the molecule implies that the interaction between the terminal loops of Helices 88 and 22 may be involved in gating this deacylated tRNA exit corridor open and closed. This is consistent with the model of allosteric coordination between the A-and E-sites (52, 53) , which would indicate that the defects conferred by all of these mutants on aa-tRNA binding might impair this E-site gating function. The decreased lability of C2531 and G2534 in ES31 is similarly intriguing, raising more questions than answers. No function is currently associated with this expansion segment, but recent cryo-EM analysis shows it to be located on a solvent accessible surface of the large subunit (4) . Perhaps this site is also involved in A-site/ E-site coordination. Alternatively, it may be a site for recognition of defective ribosomes by the nonfunctional ribosome decay apparatus.

Involvement of microRNAs in physiological and pathological processes in the lung

To date, at least 900 different microRNA (miRNA) genes have been discovered in the human genome. These short, single-stranded RNA molecules originate from larger precursor molecules that fold to produce hairpin structures, which are subsequently processed by ribonucleases Drosha/Pasha and Dicer to form mature miRNAs. MiRNAs play role in the posttranscriptional regulation of about one third of human genes, mainly via degradation of target mRNAs. Whereas the target mRNAs are often involved in the regulation of diverse physiological processes ranging from developmental timing to apoptosis, miRNAs have a strong potential to regulate fundamental biological processes also in the lung compartment. However, the knowledge of the role of miRNAs in physiological and pathological conditions in the lung is still limited. This review, therefore, summarizes current knowledge of the mechanism, function of miRNAs and their contribution to lung development and homeostasis. Besides the involvement of miRNAs in pulmonary physiological conditions, there is evidence that abnormal miRNA expression may lead to pathological processes and development of various pulmonary diseases. Next, the review describes current state-of-art on the miRNA expression profiles in smoking-related diseases including lung cancerogenesis, in immune system mediated pulmonary diseases and fibrotic processes in the lung. From the current research it is evident that miRNAs may play role in the posttranscriptional regulation of key genes in human pulmonary diseases. Further studies are, therefore, necessary to explore miRNA expression profiles and their association with target mRNAs in human pulmonary diseases.

A. miRNA definition, biology and function Discovery of microRNA (miRNA) lin-4 was the first short non-coding RNA discovered in 1993 as a regulator of developmental timing in Caenorhabditis elegans [1] . The first non-coding RNA identified in humans was let-7, which has been found involved in the control of developmental timing in humans and animals [2, 3] . Soon it became evident that these short non-coding RNAs are a part of much larger class of non-coding RNAs and the term microRNA (miRNA) was introduced [4] . To date, more than 900 miRNAs in Homo sapiens have been identified (940 in miRBase v15). MiRNAs are small non-coding RNAs~22 nucleotides (nt) long involved in the negative post-transcriptional gene regulation via RNA interference mechanism [5, 6] . The sequences of miRNAs are highly conserved among plants-microorganisms-animals, suggesting that miRNAs represent a relatively old and important regulatory pathway [7] . MiRNAs belong to the most abundant class of human gene regulators [8] : up to a third of the human genes are regulated by miRNAs [9] . MiRNAs are, therefore, key regulators of numerous genes in biological processes ranging from developmental timing to apoptosis [e.g. [10] [11] [12] [13] [14] ]. It has been speculated that miRNAs may be associated with the regulation of almost every aspect of cell physiology [8] . MiRNA genes are localized in the non-coding regions or in the introns of protein-coding genes in the genomic DNA. The miRNA genes are much longer than biologically active, mature miRNAs which originate through a multistep process [15] (Figure 1 ). Briefly, transcription by the RNA polymerase II leads to hundred or thousand nucleotides long primary miRNA transcripts (pri-miRNAs) [16] . A local stem-loop structure of pri-miRNAs is then cleaved in the nucleus by the dsRNA-specific ribonuclease Drosha/ Pasha to 70 nucleotides long precursor miRNA (pre-miRNA) [17] in a process known as "cropping" [18, 19] . Pre-miRNAs are then actively transported from the nucleus to the cytoplasm [20, 21] . In the cytoplasm, pre-miRNAs are subsequently cleaved by RNase III Dicer into~22-nt miRNA duplexes [17, 20] . One strand of the short-lived miRNA duplex is degraded ("passenger" strand, miR*), whereas the other ("guide", miR) strand is incorporated into the RNA-induced silencing complex (RISC) and serves as a functional, mature miRNA [8] . Selection of the "guide" strand is based on the base pairing stability of both dsRNA ends [22, 23] . Depending on the complementarity between miRNA and 3' untranslated region (UTR) of target mRNA there are two known mechanisms of miRNAs action on mRNAs: 1) target mRNA degradation and 2) translational inhibition with little or no influence on mRNA levels [24] (Figure 2) . Firstly, the deadenylation and subsequent degradation of the target mRNA occurs when miRNA is near-perfectly complementary with target mRNA [25, 26] . A recent study proved that mRNA degradation represents the major mechanism of miRNA regulation [27] . The authors showed that about 84% of all protein-coding mRNA targets undergo degradation while recognized by their cognate miRNA [27] . Secondly, the translational inhibition MiRNAs are transcribed by RNA polymerase II from the genomic DNA as long (hundred or thousand nucleotides) primary miRNA transcripts (pri-miRNAs). A local stem-loop structure of pri-miRNAs is then cleaved in the nucleus by the dsRNA-specific ribonuclease Drosha/Pasha to produce a 70 nucleotides long precursor miRNA (pre-miRNA). Pre-miRNAs in form of hairpins are then actively transported from the nucleus to the cytoplasm. In the cytoplasm, pre-miRNAs are subsequently cleaved by RNase III Dicer into~22-nt miRNA duplexes, consisting of the "guide" (miR) strand and the "passenger" (miR*) strand. The "passenger" strand is degraded, the "guide" strand is incorporated into the RNA-induced silencing complex (RISC) and serves as a functional, mature miRNA, acting by two different mechanisms according to the complementarity with the target mRNA. Adopted from Kim [15] . occurs when miRNA is only partially complementary to its target mRNA [28] [29] [30] . In light of the recent study by Guo et al [27] , this mechanism does not represent a predominant reason for reduced protein output. Besides the complementarity between miRNA and mRNA, several other factors may influence the miRNA action such as impaired processing, methylation, gene polymorphisms, gene amplification, deletion of Dicer, translocations and others [31] . It is evident that single miRNAs may regulate translation of numerous downstream mRNAs and each mRNA is likely to be regulated by several miRNAs simultaneously [30, 32] . Thus, identification of miRNA target genes has been a great challenge [33] . Numerous computational algorithms [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] were established which combined 5' seed matches, thermodynamic stability and conservation analysis in order to maximize specificity when predicting mRNA targets [44] (Table 1) . Nevertheless, various algorithms differ in the selection of mRNA targets and simultaneous application of several algorithms is, therefore, highly recommended. Nowadays, many web-based applications [45] [46] [47] [48] [49] [50] [51] [52] have been developed by combining existing prediction programs with functional annotations associated to many miRNA, gene, protein or biological pathway resources such as miR-Base, Ensembl, Swiss-Prot, UCSC genome browser, KEGG pathway and other databases [44] (Table 2) . However, because of high similarities in miRNA sequences, computational algorithms may predict a large number of putative miRNA binding sites on mRNA targets [33] . Thus, experimental validation in biological system is fundamental to complete the target prediction [44] ; the currently available methods [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] are listed in Table 3 . Of these, antagomir studies or immunoprecipitation of Ago-bound mRNAs have been specifically developed for miRNA-mRNA studies. Antagomirs represent a novel class of chemically engineered oligonucleotides used to silence endogenous microRNAs [64, 65] . Immunoprecipitation is then based on the observation that each member of the Argonaute (Ago) protein family (catalytic components of the RNA-induced silencing complex) can bind to miRNAs and to partially complementary sequences in the 3'-UTR of specific target mRNAs. Thus, using highly specific monoclonal antibodies against members of the Ago protein family, Ago-bound mRNAs can be co-immunoprecipitated [66, 67] . The lung has a very specific miRNA expression profile, highly conserved across mammalian species [68, 69] . However, the knowledge of the role of miRNAs in physiological and pathological conditions in the lung compartment is still limited and it is based mainly on the studies in animal models. MiRNAs have been shown to be involved in 1) the lung development and homeostasis, 2) in inflammation and viral infections and 3) miRNA deregulation may contribute to several pulmonary diseases ( Figure 3 ). Hereby, we summarize the knowledge of the involvement of miRNAs in the lung and current information on their posttranscriptional regulation ongoing in the lung compartment. Besides pathology we pay attention also to physiological lung because understanding miRNA function in normal condition is prerequisite to description of its involvement in disease. Several miRNAs such as miR-155, miR-26a, let-7, miR-29, miR-15/miR-16, miR-223, miR-146a/b and the miR-17-92 cluster have been shown to be involved in homeostasis and in the lung development ( Table 4 ). The pulmonary role of miR-155 was studied in murine lung, where it has been shown that miR-155 is crucially involved in the differentiation of naive T-cells into Th1 and Th2 cells [70, 71] . Mice deficient in bic/miR-155 became immunodeficient and displayed increased lung remodelling, higher bronchoalveolar leukocytes and impaired T-and B-cell responses to inflammatory stimuli [70] . Another member of miRNA family, miR-26a, has been shown to be selectively expressed in the bronchial and alveolar epithelial cells in murine lung [72] . Target mRNA of miR-26a is the transcription factor SMAD1, which is involved in the regulation of bone morphogenic protein signalling during lung development and pulmonary vascular remodelling [73, 74] . Thus, miR-26a might be important in controlling essential developmental and physiological events in the lung [75] . Also the miR-17-92 cluster is believed to regulate the lung development because its expression is high in embryonic development and steadily declines through development into adulthood [76] . Mice deficient in the miR-17-92 cluster died shortly after birth and lung hypoplasia/ventricular septal defects were demonstrated; moreover the absence of the miR-17-92 cluster let to upregulation of the pro-apoptotic protein Bim and inhibition of B-cell development [77] . On the other side, the overexpression of the miR-17-92 cluster in murine models resulted in an abnormal phenotype manifested by absence of terminal air sacs, which were replaced by highly proliferative, undifferentiated pulmonary epithelium [76] . Other miRNAs found to be involved in the pulmonary homeostasis are members of let-7 family [78] , miR-29 [79] , miR-15 and miR-16 [80, 81] , which Northern blot analysis [54] Quantitative real-time PCR [55] Ribonuclease protection assay [56] in situ hybridization [57] , [58] miRNA mimics [59] Western blot [60] Immunocytochemistry [61] Bead-based flow cytometry method [62] Suppression of miRNA expression in cells by antisense locked-nucleic acid oligonucleotides [63] Antagomir assays [64] , [65] Immunoprecipitation of Ago-bound mRNAs [66] , [67] function as tumor suppressors in lung cells. In addition, another miRNA, miR-223, has been shown to be crucial for normal granulocyte development and function in the lung [82] . MiR-223 mutant mice spontaneously developed neutrophilic lung inflammation with tissue destruction after endotoxin challenge [82] . Two miRNAs, miR-146a and miR-146b, have been shown to play central role in the negative feedback regulation of IL-1β-induced inflammation; the mechanism is down-regulation of two proteins IRAK1 and TRAF6 involved in Toll/interleukin-1 receptor (TIR) signalling [83, 84] . Also other miRNAs have been shown to regulate the inflammation in mouse lung exposed to aerosolized lipopolysaccharide (LPS): miR-21, -25, -27b, -100, -140, -142-3p, -181c, -187, -194, -214, -223 and -224 [72] . Increase in these miRNAs correlated with the downregulation of pro-inflammatory cytokine production such as TNFα [72] . The deregulation of miR-155, the miR-17-92 cluster and miR-223, miRNAs involved in lung development and homeostasis, resulted in the uncontrolled lung inflammation in murine models [70, 77, 82] . Based on the studies in murine models, there is evidence that miRNA expression may influence also the course of pulmonary viral infections [85, 86] . MiR-200a and miR-223 were detected in lethal influenza virus infection presumably contributing to the extreme miR-155 important for normal lung airway remodelling (A) [70] alteration of T-cell differentiation (A) [71] miR-26a highly expressed within bronchial and alveolar epithelial cells, important for lung development (H) [75] let-7 highly expressed in normal lung tissue, functions as a tumor suppressor in lung cells (H) [78] miR-29 functions as tumor suppressor in lung cells (H) [79] miR-15, miR-16 function as tumor suppressor genes (H) [80] , [81] miR-223 control of granulocyte development and function (A) [82] miR-146a/b central to the negative feedback regulation of IL-1β-induced inflammation (H) [83] , [84] miR-200a, miR-223 contribution to the extreme virulence of the r1918 influenza virus (A) [85] miR-17 family, miR-574-5p, miR-214 upregulated at the onset of SARS infection (A, H) [86] virulence of the r1918 influenza virus [85] . MiR-17 family, miR-574-5p and miR-214 were upregulated at the onset of SARS infection: these miRNAs may help the virus to evade the host immune system and are responsible for effective transmission at the initial stage of viral infection [86] . There is evidence that upregulation or downregulation of miRNAs is critical for the lung development/homeostasis and thus may contribute to development of pathological pulmonary conditions, namely to smokingrelated diseases including lung cancerogenesis, fibrosis, and other immune-mediated disorders including allergy (Table 5) . Recent studies have implicated the miRNAs in the pathogenesis of immune system mediated pulmonary diseases. Tan and colleagues [87] described that a single nucleotide polymorphism in the 3'UTR of HLA-G, a known asthma-susceptibility gene, disrupts the binding sites of three miRNAs (miR-148a, miR-148b, miR-152) targeting this gene. Thus, it is likely that the association of the HLA-G gene to asthma-susceptibility may be due to the allele-specific regulation of this gene by miRNAs [87] . MiR-21 is a further miRNA crucially involved in allergic lung inflammation. Its molecular target is IL-12p35, a cytokine contributing to polarization of Th cells toward Th2 cells [88] . MiR-126 is another miRNA found to be involved in the pathogenesis of allergic airways disease [89] . The blockade of miR-126 suppressed the asthmatic phenotype leading to diminished Th2 responses, suppression of inflammation, reduced airways hyperresponsiveness, inhibition of eosinophil recruitment, and lower mucus hypersecretion [89] . In bronchial epithelial cells stimulated with IL-4 and TNFα, let-7, miR-29a and miR-155 have been involved in the regulation of allergic inflammation [90] . Multiple members of let-7 family were also found upregulated in experimental asthma model and the pro-inflammatory role of let-7 miRNAs on the allergic cytokine expression was confirmed [91] . Another study showed that expression of RhoA in bronchial smooth muscle cells (BSMCs), a new target for asthma therapy, is negatively regulated by miR-133a [92] . The same group later revealed that IL-13 is capable of reducing the miR-133a expression in BSMCs and that the miR-133a downregulation causes an Table 5 MiRNAs involved in pathological processes in the lung miRNA Function (A animal studies, H human studies) References miR-155, miR-17-92 cluster deregulation results in uncontrolled inflammation (A) [70] , [71] , [77] miR-21, miR-27b, miR-100, miR-181c, miR-223, miR-224 increased following exposure to LPS (A) [72] miR-155 overexpressed in solid tumors, inhibition of tumor suppressor genes (A, H) [81] miR-223 impaired granulocyte function, regulator of granulocyte production and inflammatory response (A) [82] miR-148a/b, miR-152 allele-specific regulation of asthma susceptibility HLA-G gene (H) [87] miR-21 key role in asthma (A) [88] overexpressed in solid malignancies (A, H) [103] up-regulated in bleomycin-induced fibrosis and IPF (A, H) [110] miR-126 suppression of the asthmatic phenotype by blockade of miR-126 (A) [89] downregulated in cystic fibrosis airway epithelial cells (H) [111] let-7, miR-29a, miR-155 regulation of allergic inflammation in bronchial epithelial cells (A, H) [90] let-7 pro-inflammatory effect in experimental asthma (A) [91] role in lung cancer progression (H) [99] miR-133a regulator of expression of RhoA, target for asthma therapy (A, H) [92] , [93] miR-146a reduced expression in COPD fibroblasts (H) [95] miR-218, miR-15a, miR-199b, miR-125a/b, miR-294 deregulated due to smoking (A, H) [96, 97] miR-218 tumor suppressor in non-small cell lung cancer (H) [98] miR-17-92 cluster overexpressed in lung cancers (H) [102] miR-34 regulation of apoptosis in lung cancer cells (H) [105] [106] [107] miR-210 overexpressed in lung cancer (H) [108] let-7d pro-fibrotic effect in pulmonary fibrosis (A, H) [109] upregulation of RhoA, presumably resulting in an augmentation of the contraction [93] . Lung cancer and chronic obstructive pulmonary disease (COPD) share a common environmental risk factor in cigarette smoke exposure [94] . Although extensive studies of the involvement of miRNAs in lung cancer have been performed, there are only few reports focused on the role of miRNAs in COPD. Recent study on fibroblasts from COPD subjects stimulated in vitro with pro-inflammatory cytokines released less miR-146a than smokers without COPD [95] . The reduced miR-146a expression resulted in prolonged mRNA half-life of cyclooxygenase-2, thus increasing prostaglandin E2 in fibroblasts from COPD subjects [95] . There is evidence that smoking has influence also on other miRNAs. Expression profiling study in the rats exposed to environmental cigarette smoke revealed 24 downregulated miRNAs (especially let-7 family, miR-10, [96] . MiR-294, a known inhibitor of transcriptional repressor genes, was the only miRNA upregulated in smoke-exposed rats [96] . In another study, bronchial airway epithelial cells from current and never smokers differed in the expression of 28 miRNAs (especially miR-218, miR-15a, miR-199b, miR-125a/b, miR-294) in comparison to smokers, whereas the majority of deregulated miRNAs were downregulated in smokers [97] . Similar observation was observed in lung squamous cell carcinoma, where downregulation of miR-218 was associated with a history of cigarette smoking [98] . However, the majority of miRNA studies in smokingrelated diseases are focused on the role of miRNAs in lung cancer. Altered expression of miR-155 and let-7 has been reported in lung adenocarcinoma and expression of let-7 related to patient survival [99] . Moreover, it has been shown that let-7 may also play a role in lung cancer progression [99] [100] [101] . Further, increased expression of the miR-17-92 cluster has also been detected in lung cancer [102] . Another miRNAs involved in lung cancerogenesis are miR-21 and miR-34 families. MiR-21 was shown to regulate multiple tumor/metastasis suppressor genes in lung solid tumors [103] . MiR-34a/b/c have been identified to be a component of the p53 tumor suppressor network: p53 upregulates in response to DNA damage the members of miR-34 family [104] , thus regulating genes involved in the cell cycle and apoptosis [105] [106] [107] . Furthermore, miR-210 has been overexpressed in late stages of lung cancer, thus mediated mitochondrial alterations associated with modulation of hypoxia-inducible factor-1 activity [108] . Next, miR-218 was identified as a putative tumor suppressor in non-small cell lung cancer [98] . Recently, it was reported that miRNAs may play pivotal regulatory role also in the fibrotic processes ongoing in the lung: the downregulation of let-7 d in idiopathic pulmonary fibrosis (IPF) resulted in the pro-fibrotic effects [109] . Also, upregulation of miR-21 was reported in the lungs of IPF patients and in the murine lungs with bleomycin-induced fibrosis, whereas miR-21 expression was enhanced by pro-fibrotic TGF-β1 [110] . Another disease associated with miRNA change was cystic fibrosis. Downregulation of miR-126 was detected in cystic fibrosis bronchial epithelial cells and its expression correlated with upregulation of TOM1 mRNA both in vitro and in vivo [111] . TOM1, a miR-126 target, was reported to be involved in the regulation of innate immune responses through its involvement in the TLR2/4 and IL-1β and TNF-α-induced signalling pathways [111] . Small non-coding RNAs (miRNAs) play pivotal role in the posttranscriptional regulation of numerous human genes, mainly via degradation of target mRNAs. There is evidence that the lung has a very specific miRNA expression profile undergoing changes during the lung development. Studies namely in animal models have provided evidence that miRNAs participate in lung homeostasis and play pivotal role also in the control of pulmonary inflammation and viral infections. Recent studies showed evidence that upregulated or downregulated expression of various miRNAs play an active role in the pathogenesis of pulmonary diseases. Specific miRNA expression profiles were characterized for smoking related-diseases including COPD and lung cancer, immune-mediated pulmonary diseases and pulmonary fibrosis. Moreover, several miRNAs crucial for lung development and homeostasis such as let-7, miR-155 or miR-19-72 cluster have been identified to be deregulated in pulmonary allergy, asthma or lung cancer. The knowledge of altered miRNA expression profiles in diseased lung may thus offer new insights in the biology of pulmonary diseases. Moreover, miRNAs may represent attractive novel diagnostic biomarkers mainly due to their higher stability when compared to mRNAs [112] and could potentially provide possibilities for therapeutic intervention [31, 113, 114] .

Rapid detection of pandemic influenza in the presence of seasonal influenza

BACKGROUND: Key to the control of pandemic influenza are surveillance systems that raise alarms rapidly and sensitively. In addition, they must minimise false alarms during a normal influenza season. We develop a method that uses historical syndromic influenza data from the existing surveillance system 'SERVIS' (Scottish Enhanced Respiratory Virus Infection Surveillance) for influenza-like illness (ILI) in Scotland. METHODS: We develop an algorithm based on the weekly case ratio (WCR) of reported ILI cases to generate an alarm for pandemic influenza. From the seasonal influenza data from 13 Scottish health boards, we estimate the joint probability distribution of the country-level WCR and the number of health boards showing synchronous increases in reported influenza cases over the previous week. Pandemic cases are sampled with various case reporting rates from simulated pandemic influenza infections and overlaid with seasonal SERVIS data from 2001 to 2007. Using this combined time series we test our method for speed of detection, sensitivity and specificity. Also, the 2008-09 SERVIS ILI cases are used for testing detection performances of the three methods with a real pandemic data. RESULTS: We compare our method, based on our simulation study, to the moving-average Cumulative Sums (Mov-Avg Cusum) and ILI rate threshold methods and find it to be more sensitive and rapid. For 1% case reporting and detection specificity of 95%, our method is 100% sensitive and has median detection time (MDT) of 4 weeks while the Mov-Avg Cusum and ILI rate threshold methods are, respectively, 97% and 100% sensitive with MDT of 5 weeks. At 99% specificity, our method remains 100% sensitive with MDT of 5 weeks. Although the threshold method maintains its sensitivity of 100% with MDT of 5 weeks, sensitivity of Mov-Avg Cusum declines to 92% with increased MDT of 6 weeks. For a two-fold decrease in the case reporting rate (0.5%) and 99% specificity, the WCR and threshold methods, respectively, have MDT of 5 and 6 weeks with both having sensitivity close to 100% while the Mov-Avg Cusum method can only manage sensitivity of 77% with MDT of 6 weeks. However, the WCR and Mov-Avg Cusum methods outperform the ILI threshold method by 1 week in retrospective detection of the 2009 pandemic in Scotland. CONCLUSIONS: While computationally and statistically simple to implement, the WCR algorithm is capable of raising alarms, rapidly and sensitively, for influenza pandemics against a background of seasonal influenza. Although the algorithm was developed using the SERVIS data, it has the capacity to be used at other geographic scales and for different disease systems where buying some early extra time is critical.

Rapid detection of pandemic influenza at national or regional level is a public health issue of critical importance [1, 2] . Huge excess mortality and morbidity have been associated with the pandemics of influenza outbreaks in the past [3] . In the aftermath of the highly pathogenic H5N1 avian influenza outbreaks worldwide [4, 5] , the growing concern [3,4] of a virulent form of a possible human influenza pandemic has led to the setting up of influenza surveillance systems across the globe [6] . One of the main purposes of such worldwide expansion of influenza surveillance systems is the timely detection of influenza outbreaks of pandemic potential [7] . The importance of timely detection lies in buying some extra time for being prepared to deal with a pandemic [3, 8, 9] . This has also been corroborated by some recent mathematical modelling studies [10, 11] of pandemic influenza outbreaks: a key finding suggests that there would be a small window of opportunity for getting ahead of pandemic outbreak fronts and thus helping early pandemic mitigation efforts if it could be detected early on. Most developed countries as well as many from the developing world have some form of influenza surveillance in place [6] . These surveillance systems are based on the reporting of disease syndromes (e.g., reports of Influenza-like illnesses (ILI)) and are generally designed to monitor levels of seasonal influenza [12, 13] . Although the signature of pandemic influenza could be different from that of seasonal ones [14] , the traditional approach (patients presenting with clinical signs of ILI, collection of throat/nasal swab samples from some of these patients and, finally, laboratory confirmation of influenza) followed in influenza surveillance systems, in the absence of any detection algorithm applied to syndromic data, may not be able to pick it up early on. This is the reason that public health surveillance systems are being supplemented by the new state-of-the-art statistical tools [1, 2] . The development of these new statistical tools has demonstrated the potential to automate syndromic surveillance systems, to be able to raise specific and sensitive early alerts of adverse disease outbreaks. Indeed this is a fast growing and a very active area of scientific research at the moment [6] . At present, a number of methods [12, 13, 15] exist to establish the onset of peak activities in the epidemics of seasonal influenza. These methods are mostly based on regression [16, 17] or time-series [12, 13, 15] analysis of seasonal ILI data. One such method is the Moving-Average Cumulative sums (Mov-Avg Cusum) method [18] [19] [20] . Originally developed for the industrial quality control [21] , it is now frequently used for detecting the outbreaks of seasonal and pandemic influenza [12, 22] . Recently there has been a flurry of new detection methods based on sophisticated statistical approaches [1, 2] , including those aimed at real-time monitoring and projecting of influenza cases [23, 24] . However, challenges remain in terms of how to use the ILI surveillance data in a simple and efficient manner for timely detection of influenza pandemics. The basic reproduction ratio R 0 (i.e., the average number of new infections produced by a single infection in a totally naïve host population) plays a central role in our understanding of infectious disease dynamics. It determines whether a new infection will successfully invade the susceptible population [25, 26] . In the case of an ongoing epidemic, the effective reproduction ratio R replaces R 0 [25] . In the presence of disease tracing data, R can be estimated and the in-or out-of-control status of an ongoing epidemic can be established [27] . Where there is no availability of disease tracing data, as in influenza syndromic data, the weekly case ratio (WCR), defined as the ratio of the number of reported cases in a week to the number of cases reported in the previous week, may function as an indirect measure of R and may be suitable for raising public health alarms in the early stages of an emerging infection. Although pandemic influenza infections may grow exponentially in early invading stages (evident in the mortality data from the past pandemics [28] or in the mathematical modelling [10, 11, 29] of influenza pandemic), the detection algorithms so far employed in influenza surveillance systems largely ignore this natural behaviour as the basis for generating early warning of influenza outbreaks of pandemic potential. The aim of this paper is to develop a detection algorithm, based on the estimates of WCR for expected influenza pandemics, to facilitate sensitive, specific and rapid detection of a pandemic outbreak at a regional level based on existing surveillance systems. Using the influenza surveillance data from Scotland, we first sift through the spatiotemporal patterns in the historical data by calculating WCR and N HB , the number of health boards (HBs) that show increases in the weekly ILI cases. The joint probability distribution of WCR and N HB is then contrasted with expected and observed patterns in the presence of pandemic influenza. As described in the next section, the expected patterns for pandemic cases are obtained from a previously published mathematical model [29] . Observed patterns for pandemic cases are based on records from the 2008-09 season when influenza A(H1N1)v was circulating in Scotland. We compare the performance of our detection algorithm, using simulated influenza pandemics as well as data from the 2009 influenza A(H1N1v) epidemic in Scotland, with that of the Mov-Avg Cusum method and with the ILI rate threshold method, a slightly modified form of the baseline ILI activity indicator used by the Health Protection Agency (HPA) in the monitoring of seasonal influenza in the UK. In developing our pandemic detection algorithm, we used historical seasonal ILI data, collected and compiled under the Scottish Enhanced Respiratory Virus Infection Surveillance (SERVIS) system. The data set spans from the 2001-02 season through to the 2008-09 season. Seasonal ILI cases are normally reported weekly over a period of 33 weeks (from the first week of October to the third week of May) in different age-and sex-classes, by sentinel general practices (GPs) across Scotland. The SERVIS sentinel GPs are drawn from 13 Scottish health boards. (There are currently 14 HBs in Scotland; all HBs except the Western Isles HB have participated in the SERVIS network of the sentinel GPs.) The Scottish health boards widely vary in their population sizes from 20,000 to 1,360,000. The total number of the sentinel GPs varied from 20 to 44 across the influenza seasons considered, with a minimum of 1 GP in a HB to a maximum of 9 GPs in a HB. Altogether, the numbers of people registered with sentinel GPs represent 2.7% to 4.9% of the Scottish population of around 5.15 million. This system was designed as a national surveillance scheme with regional coverage. It was not designed to be used as a surveillance system in each health board separately. The weekly reported ILI cases at the national level are shown in Figure 1a . In our analysis we used the weekly ILI cases, aggregated at the HB level. An example of weekly HB-level ILI cases is shown for the 2004-05 influenza season in Figure 1b . Sentinel GPs are often recorded as reporting zero cases in a week in the SER-VIS data. It is unknown whether this represents true zeroes or non-reporting. However, we believe that certainly, in the data for the 2008-09 season, a blank means that no report was made by the practice and a zero means a report was made but no cases occurred. The historical influenza data from 6 influenza seasons from 2001-02 through 2006-07 were used in estimating the background pattern of the seasonal ILI cases. The ILI data for the season 2007-08 from 23 sentinel GPs recorded just 93 cases compared with over 300 cases for the other seasons. The whole of the UK reported influenza cases below the HPA baseline activity threshold in this season [30] . We therefore excluded the 2007-08 data from our analysis. (As we checked this and will be clear later, the inclusion of the historical ILI data of this season increases the detection efficiency of the WCR method. The exclusion of the data of this season, therefore, ensures conservative estimate for the WCR method in the performance testing.) The 2008-09 data-set contains the 2009 influenza A(H1N1)v pandemic cases, so it was used for performance testing of our detection algorithm with real pandemic data. The SERVIS ILI data used in the study is freely available from Health Protection Scotland on request (NSS.hpsflu@nhs.net) to anyone wishing to use them for any non-commercial research purposes. For our main analysis, we use simulated pandemic influenza data for Scotland. In brief, the pandemic model [29] is a stochastic, spatially structured, individual-based discrete time simulation. For the analyses carried out in this paper, the model pandemic outbreaks were run with a basic reproduction rate R 0 of 1.7 and a generation time of 3 days. (The R 0 value used here is slightly higher than what has been reported from various analyses [31] of the 2008-09 influenza A(H1N1)v outbreak data. But we have used a range of pandemic case reporting rates as discussed below.) Full model details are given in the Supplementary Information of [10, 29] . The pandemic model was simulated 10 times for the whole of Great Britain, starting on day 1 seeded with a single infection at a randomly chosen location. In our analysis, however, we define the pandemic start week as the week when the first influenza infections occur in Scotland, which may or may not be reported. From the simulated pandemic infections, we sample ILI cases at case reporting rate of α as if they would have been reported by sentinel GPs to SERVIS (see Additional file 1). In this work we have used three values of α: 0.5, 1 and 5%. The sampled sentinel pandemic cases were then converted to the HB-level daily reported influenza cases by summing across all participating sentinel GPs of the HBs, which in turn were converted into weekly reported ILI cases to match the temporal resolution of the SER-VIS data. The first wave of pandemic influenza cases could occur in the presence of seasonal cases at any time of the year. To take this into account, we add the simulated pandemic ILI cases to the seasonal ILI cases from each of the six seasons and use the resultant ILI time series for detecting pandemic, sliding the pandemic start week across the entire influenza season. We present our results using 300 (30 samples times 10 simulated pandemics) sampled time series of pandemic cases. Here we use 10 simulated pandemics each sampled 30 times. (Note that the results are invariant with other sampling schemes, e.g., 300 time series, each sampled from 300 different simulated pandemics. The sampling is carried out after the first Scottish cases are reported, during the exponential growth phase of the epidemic. All simulated pandemics are therefore observed to have very similar temporal dynamics.) Each of these 300 time series were, in turn, overlaid with seasonal ILI time series from each of the 6 seasons, making a total of 1800 time series to be analysed for 33 (from week 1 to week 33) pandemic start weeks in a typical influenza season. Pandemic detection algorithm Joint probability distribution of (WCR, N HB ) Our pandemic detection algorithm uses two metrics obtained from weekly reported ILI cases: the weekly case ratio (WCR) for cases aggregated across the region and N HB , the number of health boards reporting increases in the cases over the previous week. The weekly WCR is defined as W CR = total ILI cases reported to all SERV IS sentinel GPs in week w total ILI cases reported to all SERV IS sentinel GPs s in week w − 1 Note that WCR is not defined for week 1 and, also, not defined for any week where in the previous week all sentinel GPs did report zero ILI cases. (In the historical data we have one instance (out of the total 214 weeks from all the 6 seasons) that for a week, which was not a start week but well within the influenza season, there were no ILI cases reported. In this situation we simply replace zero in the denominator by 1.) Since WCR is a continuous variable, in order to create a joint distribution, it is binned with a bin size of 0.1. We then construct a joint probability distribution of the two metrics using the historical ILI data from all the 6 seasons (see Figure 2 ). The historical seasonal ILI data are temporally heterogeneous, with substantial week-to-week variability ( Figure 1 ). For making any useful and robust statistical inference the probability distribution requires smoothing. The smoothing was done assuming that the weekly reported ILI cases in a HB have a Poisson distribution with rate parameter equal to the reported total weekly ILI cases in the HB. The Poisson distribution is preferred to a binomial distribution as the numbers of weekly reported cases are very much smaller than the total population of a health board. We simulated the Poisson model to generate weekly ILI cases at the HB level (see Additional file 2). The model-generated ILI counts from individual runs for each season were then used, as described above, to calculate the joint distribution of WCR and N HB . A set of 10,000 model runs per season's data were used in the estimation of this joint distribution. A smaller (1,000) or larger (100,000) number of runs were also tested for the robustness of the results. We tested our algorithm for its specificity (i.e., not detecting a pandemic when no pandemic is occurring). To do this, we first calculate WCR and N HB for each week from a given seasonal HB-level ILI time series. We obtain the probability values of weekly pairs of (WCR, N HB ) from the joint probability distribution (the plots of the weekly probability values are given in Figure 3 ). In the second step, for a chosen threshold probability value δ, we count the number of weeks in which the probability of (WCR, N HB ) falls below the threshold. The total counts represent the number of false alarms (Fa s (δ)) in season s. The theoretical maximum number of false alarms ( FA s max ) in any given season is the total weeks minus 1 because WCR cannot be calculated for the first week. Finally, this leads us to calculate the specificity of the detection algorithm for a given threshold probability as follows: A detection threshold is defined as the threshold probability δ that gives rise to a pre-set specificity of our detection method. We adjust δ in order to achieve specificities of 95% and 99%. This allows us to compare our method to other algorithms which also have the same constant specificity. Detecting a pandemic Figure 4 shows an example of how our WCR detection algorithm works when applied to a simulated pandemic starting on week 1 and week 15. In the case of a pandemic starting on week 1 (week 15), the probability value of WCR and N HB falls below the threshold values for the first time on week 7 (week 21) triggering an alarm that week. We note here that the probability value is the probability mass function, i.e. P(WCR = xx and N HB = y), and not the more usual statistical significance level of P(WCR > = xx and N HB > = y). We compared the performance (in terms of median detection time and sensitivity) of our detection algorithm with that of the Mov-Avg Cusum method. This method was recently used by Cowling et al. [12] as a statistically robust automation tool for generating early alerts for the onset of peak activity of ILI cases. We also compare the performance of an ILI rate threshold method, similar to the HPA baseline influenza activity level. The detection thresholds used by these methods were adjusted so that all three methods had the same detection specificity described earlier. The d-week upper Cusum at time t is defined as follows: where Cusum t d ≤ + + = 7 0 . X t  and st  are the 7-week moving average and standard deviation of ILI cases in weeks t-d-1 to t-d-7 Note that d stands for a delay period and this method will only be informative from the (d+8) th week onwards. An alarm is triggered using this method when Cusum t + on a week t crosses a pre-set threshold [21] . Using the SERVIS ILI data for six seasons we preset the threshold values for this method for all 16 (Table in Additional file 3) . Therefore, this combination is used to make comparisons with other methods. The HPA set ILI consultation rates as proxies for influenza activity in the United Kingdom. The threshold rate for baseline-and epidemic-level ILI activities has recently been revised: the baseline threshold is lower from 50 to 30 consultations per 100,000 population while the epidemic threshold has been decreased from 400 to 200 consultations per 100,000 population [32] . These thresholds are derived from the time series analysis of historical seasonal data, and serve the purpose of establishing when and/or whether the community ILI activity warrants some intervention of the public health departments. In the SERVIS data set (2001 -2007) considered, the epidemic threshold rate was never crossed [30] . Here the ILI rate threshold is denoted by η cases per 100,000 population per week and an alarm is generated when the aggregated ILI cases in any week crosses this threshold. In order to compare this method with alternatives we adjust η to obtain the pre-set specificity of 95% and 99%. The two respective values of η are 24 and 34. Performances of the three methods are summarised in terms of sensitivity and median detection time (MDT) in Table 1 . Our algorithm is almost 100% sensitive (the lowest being 98% for a very low case reporting rate of 0.5%). Its MDT ranges from 3 to 5 weeks compared to 4 to 6 weeks for the Mov-Avg Cusum and threshold methods. While the threshold method is 100% sensitive, Mov-Avg Cusum is the worst performing method with sensitivity of 77% to 97%. Note that time to pandemic detection is counted from the start week of the first infections in simulated pandemics. There is a lag of about 3 to 5 weeks between the first infections arising in simulated pandemics and the first cases which get reported by sentinel GPs to SERVIS. If, therefore, the reference point is changed to the week of the first reported cases, then MDT of the WCR method will not be more than 2 weeks. The distribution of detection times ( Figure 5 ) is to show whether the temporal pattern in the background seasonal ILI cases will have any effect on pandemic detection. The detection specificity (Sp) and the pandemic case reporting rate (α) were set for all methods at the following values: Sp = 99% and α = 5%. Detection times are typically within 5-6 weeks for our method (Figure 5a ). This compares favourably with the ILI rate threshold method at η = 34 cases per 100000 population (Figure 5b ). The slightly longer detection times are present for the starting weeks falling in the period of late-November to mid-January (week 9 to week 15 in Figure 5a ). This is the time period when the seasonal ILI incidences show widespread and peak influenza activities ( Figure 1 ). In the case of simulated pandemics starting during this period, the above two factors together mask the probability of (WCR, N HB ) as seasonal one in the first few weeks of pandemic. This masking causes delay in generating an alarm and the delay is more pronounced at lower case reporting rates and a higher specificity and is, for example, responsible for reducing the method's sensitivity to 98% at α = 0.5% and Sp = 99%. Conversely, during the same time-period the threshold method produces its shortest detection times because the peak-level seasonal cases having been added to the pandemic cases help the weekly ILI rate quickly cross the rate threshold η. However, our method outperforms the ILI rate threshold method in the beginning and end of an influenza season. Mov-Avg Cusum performs poorly in terms of detection times in the first few weeks of the season (Figure 5c ). This is because it requires 9 weeks to calculate the first Mov-Avg Cusum statistic. This situation improves as we move well within the seasonal period. But even for the later starting weeks, the method takes comparatively longer time to detect a pandemic. However, this method also outperforms the ILI rate threshold method towards the end of a season. Detection of pandemics in the early weeks of its starting depends on the case reporting rate and specificity ( Figure 6 ). Our method outperforms the other two by rapidly detecting pandemics in a large fraction of model runs at specificity of 99% and case reporting rate α of 0.5%. It detects pandemics in >50% of total runs within the first 6 weeks of a pandemic starting while the Mov-Avg Cusum and the threshold methods detect pandemics, respectively, in <25% and <35% of total runs (Figure 6a) . The time to detection decreases when the specificity was lowered from 99% to 95% (Figure 6b ). In this case, about 25% and a slightly lower than 50% detection levels were achieved by the WCR method within the first 4 and 5 weeks while the Mov-Avg Cusum and the threshold methods still trailed below the 25% detection level. The same trend was observed when, for the fixed specificity of 99%, the case reporting rate α was raised from 0.5% to 1% (Figure 6c) to 5% (Figure 6e ). At the elevated reporting rates, decrease in specificity further increases the detection level for all methods. But the increase in pandemic detection within the first few weeks of pandemic is more pronounced for our method than the other two (Figs 6d & 6f) . As shown in Figure 7 , our algorithm, retrospectively, detects the 2008-09 pandemic outbreak 12 weeks (i.e., The performances of the three methods compared in terms of sensitivity (Sen) and median detection time (MDT) for different values of pandemic case reporting rates and detection specificities of 95% and 99%. Sen is given as percentage (%) of the model runs summed across all 33 weeks, i.e. calculated from a set of 1800 overlaid time series times 33 weeks. MDT is in weeks and calculated from those of (1800 × 33) runs in which a given method was able to generate a detection alarm. on week 41) after the first cases were reported on week 29 in Scotland (Figure 7a ). Clearly, here it does not perform as well as it does with the simulated pandemic data. In the next section we discuss possible reasons for this poor performance. The Mov-Avg Cusum method, which was the worst performer among the three methods using simulated pandemic data, also detects the pandemic in week 41 (Figure 7b ). Both methods outperform the ILI rate threshold method by 1 week (Figure 7c ). In this paper we compare three methods of detecting an influenza pandemic using an existing surveillance system in Scotland called SERVIS. The ILI rate threshold method uses current ILI case data to detect pandemics. This method is motivated by the current HPA's threshold levels [30, 32] to monitor influenza activity at the national scale. The HPA thresholds serve the purpose of establishing whether seasonal influenza activity warrants some intervention (e.g., the start of antiviral prescription) of the public health departments. Mov-Avg Cusum and other variants of Cusum are already being used in public health surveillance systems [1, 6, 12, 18] . The Mov-Avg Cusum method detects a pandemic when the cumulative number of current ILI cases is substantially higher than the expected cumulative number. The Mov-Avg Cusum statistic keeps accumulating the deviation between observed and expected values over time and when the accumulated value crosses a pre-set threshold, an alarm is triggered [21] . It has three adjustable parameters that require optimisation for specific surveillance systems. Finally, the WCR algorithm introduced in this paper is based upon a characteristic of epidemics, their exponential growth in the early stages before control measures and depletion of susceptibles have occurred. It also assumes that pandemic influenza would occur synchronously across spatial units of influenza surveillance system in a region (as is predicted by the mathematical models [29] for pandemic influenza in Scotland). It makes use of the joint probability distribution derived from the historical seasonal ILI data to detect a pandemic influenza. The other methods do not use any information from the data, other than to set thresholds to achieve required specificities. The WCR algorithm appears to provide a slightly more rapid and sensitive tool for detecting of pandemic influenza -median detection time for this method ranges from 3 to 5 weeks in comparison to 4 to 6 weeks for the other two methods. Although the WCR algorithm seems to do the job more efficiently with the simulated pandemics, it performed poorly with the 2008-09 pandemic data from SERVIS. There could be several possible reasons for this poor performance. First, the 2009 influenza A(H1N1)v epidemic happened outside the normal influenza season and, second, it was mild in severity [33, 34] . In addition, the number of SER-VIS sentinel GPs in the season 2008-09 was at its sparsest level -only 20 practices, as the SERVIS system was in the process of being phased out to be replaced by a system which automatically collected data from GP systems on a daily basis. These three factors will have contributed to poor reporting of the early pandemic cases, notwithstanding the huge media coverage given to the pandemic. This is consistent with the patchiness in the reported ILI cases through SERVIS sentinel GPs between weeks 29 and 41 (Additional file 4). No method will detect a pandemic in its early weeks if the early syndromic influenza data are not reported to the surveillance system. Finally, the 2009 pandemic influenza cases were more spatially heterogeneous than those predicted by the pandemic model. (It is interesting to note that during the period April to July 2009, when there was a sentinel practice within an outbreak area, Greenock and Govanhill in the GGC HB, there was no increased reporting of ILI.) This might have contributed to the observed patchiness in the sentinel reporting. An important aspect of our algorithm is that the detection threshold remains constant throughout. An implementation of a time-varying detection threshold could make this algorithm capable of using the seasonal ILI pattern more efficiently. In principle, this could be implemented by calculating the joint probability of (WCR, N HB ) either on a week-by-week basis, or on a slightly more coarse temporal scale of the time-windows of high/low ILI activities in the seasonal data. Clearly increasing the number of sentinel GPs and the frequency of ILI case reporting would improve the temporal resolution of the WCR algorithm. In future work we will explore how many sentinel GPs are required to achieve this aim. Furthermore, in outbreaks of a novel influenza strain, generally children and young adults of the population, who will have little or no prior immunity to the disease [4] , are disproportionately affected [28] . Implementing our detection algorithm using these data attributes will further improve the timeliness and specificity of the detection of pandemic influenza. SERVIS data contain age attributes which could be incorporated into our algorithm, but this requires more sentinel GPs to be of use. The WCR algorithm could be applied to any syndromic surveillance data structured by space and time. Syndromic data-sets include, but are not limited to, the triage nurse calls [35] (e.g., the NHS24/NHSdirect calls in the UK [36] [37] [38] ), the over-the-counter medicine sales data [39, 40] available in most of the developed countries, or online web search queries [41] . These data sets are highly useful in the early detection of unusual health events [1, 2] . Generally these data sets come with spatiotemporal attributes and, therefore, could potentially be integrated with the seasonal ILI data; this should enhance the detection process (in terms of timeliness, specificity and sensitivity) of pandemic influenza.

Responses of Human Endothelial Cells to Pathogenic and Non-Pathogenic Leptospira Species

Leptospirosis is a widespread zoonotic infection that primarily affects residents of tropical regions, but causes infections in animals and humans in temperate regions as well. The agents of leptospirosis comprise several members of the genus Leptospira, which also includes non-pathogenic, saprophytic species. Leptospirosis can vary in severity from a mild, non-specific illness to severe disease that includes multi-organ failure and widespread endothelial damage and hemorrhage. To begin to investigate how pathogenic leptospires affect endothelial cells, we compared the responses of two endothelial cell lines to infection by pathogenic versus non-pathogenic leptospires. Microarray analyses suggested that pathogenic L. interrogans and non-pathogenic L. biflexa triggered changes in expression of genes whose products are involved in cellular architecture and interactions with the matrix, but that the changes were in opposite directions, with infection by L. biflexa primarily predicted to increase or maintain cell layer integrity, while L. interrogans lead primarily to changes predicted to disrupt cell layer integrity. Neither bacterial strain caused necrosis or apoptosis of the cells even after prolonged incubation. The pathogenic L. interrogans, however, did result in significant disruption of endothelial cell layers as assessed by microscopy and the ability of the bacteria to cross the cell layers. This disruption of endothelial layer integrity was abrogated by addition of the endothelial protective drug lisinopril at physiologically relevant concentrations. These results suggest that, through adhesion of L. interrogans to endothelial cells, the bacteria may disrupt endothelial barrier function, promoting dissemination of the bacteria and contributing to severe disease manifestations. In addition, supplementing antibiotic therapy with lisinopril or derivatives with endothelial protective activities may decrease the severity of leptospirosis.

Leptospirosis is a geographically widespread zoonosis that has emerged as a significant public health problem in urban slums, particularly in the tropics. The infection is caused by species of spirochetes belonging to the genus Leptospira. There are more than 200 serovars of Leptospira distributed among both pathogenic and non-pathogenic species [1] . The pathogenicity of different strains can vary considerably depending on the host species and age, and on the infecting serovar [2] . The spirochete's mode of entry is through mucous membranes and cuts or abrasions on the skin [1] . Upon entry, the organisms travel through the bloodstream to multiple sites, and may cause liver and kidney damage, meningitis, and a variety of other inflammatory conditions. If the host survives the acute infection, leptospires can persist in the proximal renal tubules for weeks to months, protected from antibodies and causing little to no inflammation. The bacteria are then shed in the urine, and animal urine contamination of water is the primary source of human exposure. Although little is known about how Leptospira species establish infection in their hosts, adhesion to the host cell surface and extracellular matrix (ECM) by pathogens is often the first critical step in the initiation of infection. Several groups have investigated the adhesion of Leptospira interrogans to endothelial, fibroblast, kidney epithelial, and monocyte-macrophage cell lines cultured in vitro [3] [4] [5] [6] [7] [8] [9] . It is likely that pathogenic leptospires can attach to several different types of mammalian receptors to establish the infection. In fact, infectious strains of Leptospira have been shown to adhere to ECM components including collagen type IV, fibronectin and laminin, and also to the plasma protein fibrinogen [4, [10] [11] [12] . Adhesion to several ECM components is mediated at least in part by the LigA and LigB proteins [11] and a group of additional related proteins that were identified through homology to a laminin binding protein [10, 12] . Several studies have shown that the adhesion of pathogens to mammalian cells will provoke multiple changes in the physiology and/or gene expression of the host. The host-pathogen interactions that define a disease are clearly complex. Microarrays are a powerful tool to explore those host-pathogen interactions by analyzing the transcriptional profiles of host cells or pathogens. Although it has been documented that temperature and osmolarity alter leptospiral gene expression [13, 14] , no previously published research has focused on the mammalian cell responses to the bacteria. To understand how human endothelial cells alter gene expression in response to incubation with different strains of Leptospira, human gene arrays were probed with cDNA derived from the RNA purified from infected cells and uninfected controls. In this study, we discuss how global analysis of gene expression allows us to gain insights into host specific responses to infection with pathogenic Leptospira. The human microvascular endothelial cell line of dermal origin (HMEC-1) [15] was obtained from Dr. Ades (Centers for Disease Control and Prevention, Atlanta, Georgia) and cultured in endothelial basal medium (Clonetics, San Diego, CA) supplemented with 15% heat-inactivated fetal bovine serum (Hyclone, Logan, UT), 1 mg/ml hydrocortisone (Sigma-Aldrich, St. Louis, MO) and 10 ng/ml epidermal growth factor (Sigma-Aldrich). The immortalized human macrovascular endothelial cell line EA.hy926 [16] was kindly provided by Dr. C.-J. Edgell (University of North Carolina, Chapel Hill, NC) and grown in Dulbecco's modified Eagle medium with high glucose supplemented with 10% heat-inactivated fetal bovine serum (Gibco, Grand Island, NY) and HAT Media Supplement (Sigma-Aldrich). Both cell lines were cultured in the medium recommended by the supplier in a humidified atmosphere of 5% CO 2 and both cell media were supplemented with 1 U/mL penicillin, 1 mg/mL streptomycin, and 2 mM L-glutamine for routine propagation. Cells to be used for experimental infection with Leptospira strains were cultured without the antibiotics. The roles of proteoglycans in the endothelial cell response to L. interrogans were tested based on previously published protocols [17] . Briefly, chondroitin sulfate B was shown to bind L. interrogans and to competitively inhibit L. interrogans to mammalian cells, so it was tested for the ability to inhibit the endothelial cell responses to the bacteria described below. In addition, inhibition of proteoglycan synthesis by b-xyloside, which also decreases L. interrogans attachment to mammalian cells, was tested for any effect. Controls included chondroitin sulfate A, to which L. interrogans does not bind, and the sugar analog a-galactoside, which does not affect proteoglycan synthesis. The reference strain Leptospira biflexa serovar Patoc was obtained from the American Type Culture Collection (ATCC 23582, Manassas, VA), and is a non-pathogenic species. L. interrogans serovar Canicola (pathogenic, strain ATCC 23606 and strain 11203-32) were obtained from the ATCC and Dr. Richard Zuerner (USDA, Ames, IA), respectively. L. interrogans serovar Copenhageni (pathogenic, strain designated Fiocruz L1-130) was provided by Dr. David Haake (UCLA, Los Angeles, CA). Bacterial strains were maintained in ambient air at 30uC. Bacteria utilized for this study were at low passage from the suppliers (#passage 6) and cultured in EMJH medium [1] supplemented with 100 mg/ml of 5-fluorouracil and 1% rabbit serum (Sigma-Aldrich). For some experiments, the bacteria were radiolabeled by addition of 35 S cysteine plus methionine to the medium as described previously [17] . The bacteria were enumerated using a Petroff-Hausser counting chamber and dark field microscopy. Mammalian cells were plated in T-225 tissue culture flasks (BD Falcon, Bedford, MA) and grown up to 90% or higher confluence. When cells reached desired confluence, the monolayer was washed with PBS and the cells were lifted off the plastic culture flask with 5mM EDTA in PBS. This was done to allow access of the bacteria to endothelial cell surface receptors that are normally involved in attachment to the substratum, i.e. receptors that the bacteria may encounter when penetrating the vasculature. In addition, this approach minimizes degradation of mRNA that occurs during harvesting of adherent cells. After lifting, cells were spun for 10 minutes at 1,000 rpm, resuspended in the cell culture medium without antibiotics, and enumerated using a hemocytometer counting chamber. 2610 7 cells per sample were incubated in suspension with either L. biflexa serovar Patoc or L. interrogans serovar Canicola, or without any bacteria, for 1 h and 3 h at room temperature in the cell medium without antibiotics. The MOI (multiplicity of infection) used was 10 bacteria per mammalian cell. After incubation, cells were washed with phosphate buffered saline (PBS) and harvested for RNA isolation. The RNA was purified using RNeasy kit (Qiagen, Valencia, CA) with DNase digestion according to manufacturer's manual. The quality of RNA was checked using a Bioanalyzer (Agilent, Santa Clara, CA). Human HEEBO (Human Exonic Evidence Based Oligonucleotide) Arrays, consisting of 44,544 70mer probes representing 30,718 known genes, were purchased from Microarrays Inc. (Nashville, TN). 5 to 20 mg of total RNA from uninfected control and infected samples was used to generate cDNA labeled with aminoallyl (aa)-dUTP through a reverse transcription reaction using anchored oligo(dT) primers. The purified aa-dUTP-labeled cDNAs were coupled in 10 ml 0.1 M NaHCO 3 with either Cy3 or Cy5 NHS-ester dye. Cy-dye labeled cDNA was purified using a Cyscribe GFX column (Amersham Biosciences, Piscataway, NJ). The two differently labeled cDNAs were mixed and hybridized using Pronto Microarray Hybridization Kit in a hybridization chamber (Corning, Corning, NY), with the same array slide for 38 to 42 hr according to manufacturer's instruction. After a series of washes using the buffers provided in the kit, slides were spun dry and scanned under two laser channels in a Scanarray 4000 scanner (Packard Bioscience, Meriden, CT). Leptospirosis is a widespread zoonotic infection that primarily affects residents of tropical regions, but is seen occasionally in temperate regions as well. Leptospirosis can vary in severity from a mild, non-specific illness to severe disease that includes multi-organ failure and widespread endothelial damage and hemorrhage. To investigate how pathogenic leptospires affect endothelial cells, we compared the responses of two endothelial cell lines to infection by pathogenic versus non-pathogenic leptospires. Our analyses suggested that pathogenic L. interrogans and non-pathogenic L. biflexa caused changes in expression of genes whose products are involved in cellular architecture and interactions with the matrix, but that the changes were in opposite directions, with infection by L. biflexa primarily maintaining cell layer integrity, while L. interrogans disrupted cell layers. In fact, L. interrogans caused significant disruption of endothelial cell layers, but this damage could be abrogated by the endothelial protective drug lisinopril. Our results suggest that L. interrogans binds to endothelial cells and disrupts endothelial barrier function, which may promote dissemination of the bacteria and contribute to severe disease manifestations. This disruption may be slowed by endothelial-protective drugs to decrease damage in leptospirosis. Endothelial Cell Responses to Leptospira www.plosntds.org Images were overlaid and analyzed using Imagene (BioDiscovery, El Segundo, CA). Raw gene expression was imported from Imagene to GeneSifter (GeneSifter.Net, VizX Labs, Seattle, WA) for analysis. Data from 3 biological replicate experiments were normalized using Lowess normalization and by the median of the raw intensities for all spots in each sample for each array. The ratio of two fluorescence intensities of each spot reflected the ratio of each gene expressed in the infected and uninfected samples. Genes were considered to be induced or repressed when the ratio of infected/uninfected was at least 1.5 fold (increased or decreased), and the P value was ,0.05 by the Student's twotailed t test. For analysis involving more than one time point and/ or condition, the one way ANOVA test was performed. Microarray data are deposited in GEO archive under the accession numbers GSE23172 and GSE23173. EA.hy926 cells were seeded in tissue culture treated glass slides (BD Falcon) and grown at 37uC as described above. After cells reached 100% confluence, the monolayer was washed three times with PBS and medium without antibiotics was added. Four compartments of each slide were inoculated with 1610 7 bacteria (MOI = 10) of either L. biflexa serovar Patoc or L. interrogans serovar Canicola. The remaining four wells were left uninfected to serve as negative controls. In some cases, parallel experiments were performed using cells plated on coverslips in 24 well culture dishes, which allowed centrifugation to facilitate bacterialendothelial cell contact. At the end of the incubation (1 h, 3 h and 24 h) the slides were washed three times with PBS and fixed with 3% (wt/vol) paraformaldehyde in PBS at room temperature for 30 min. Cells were permeabilized with 0.1% Triton X-100 in PBS, washed three more times with PBS, and blocked overnight at 4uC with HEPES buffered saline (HBS) and 1% bovine serum albumin (BSA). On the next day the slides were washed again with PBS and incubated with fresh blocking solution for 1h at room temperature. After blocking, the layers were probed with either rabbit anti-L. interrogans (a gift from Dr. Richard Zuerner, USDA, AMES, IA) diluted 1:5000 or anti-L. biflexa antiserum (Biogenesis, Inc., Brentwood, NH) diluted 1:1000, followed by anti-rabbit IgG-TRITC conjugate (1:1000) plus phalloidin-FITC (200 U/mL) to stain filamentous actin. After repeated washing in PBS, chambers were removed from the slides and Prolong Anti-Fade (Invitrogen, Carlsbad, CA) was used to mount coverslips. Two different microscopes at two different institutions were used throughout the course of this work. At institution one, images were captured using a Zeiss Axioplan microscope with a digital charge-coupled device camera (Hamamatsu, Hamamatsu City, Japan) and co-localization of the fluorescent labels was done using Volocity software (Improvision Inc., Lexington, MA). At the second institution a Zeiss Axioimager Z1 with an Axiocam HrC camera and a Nuance Multi-Spectral Imaging System (software CRI Inc, Woburn, MA, v.2.6.0) was used. The endothelial cell lines EA.hy926 and HMEC were plated in 3.0 mm (2610 6 pores/cm 2 ) polyester transwell inserts (Corning) and cultured as described above. After reaching 100% confluence, as assessed by lack of penetration of the fluorescent dye FITCdextran 40,000 (and loss of penetration of the L. biflexa serovar Patoc), the monolayer was washed with PBS and cell medium without antibiotics was added to the inserts and wells. Inserts without cells were used as controls for these experiments. Bacteria were added to an MOI of 50 to allow reliable enumeration of bacteria crossing the cell layers or membranes without cells at early time points, and 10 mL from the insert and from the well were taken after 1 h, 3 h, 6 h, 24 h, 27 h, 48 h and 72 h. In addition to the non-pathogenic strain Patoc and the pathogenic Canicola, Leptospira interrogans serovar Copenhageni was also used to analyze the migration of leptospires through the cell monolayer. Motile leptospires were counted by dark-field microscopy using a Petroff-Hausser chamber. Data are shown for the time points through which the bacteria were motile; after 72 hr there was a progressive decrease in L. biflexa motility. To determine whether the bacteria were affecting the viability of the endothelial cells, four methods were used. First, adherent and EDTA-lifted endothelial cells infected at an MOI of 10 were washed, then incubated with the vital dye CellTracker Green (CT-CMFDA, 10 mM) plus DAPI (0.02 mg/ml) (Molecular Probes, now part of Invitrogen, Eugene, OR) for 1 hour at 37uC under 5% CO 2 . The samples were mounted and viewed using the Zeiss Axioplan microscope described above, and live cells (bright green cytoplasm) and dead cells (bright blue nuclei) were enumerated in at least three fields per sample in at least three independent experiments. Second, the cells were stained using the Vybrant Apoptosis Assay Kit 2 (Molecular Probes), which stains for annexin V and membrane permeability. Third, the APO-BrdU TUNEL kit, also from Molecular Probes, was used. A second TUNEL-based kit, Alert DNA Fragmentation kit (Clontech Laboratories, Inc., Mountain View, CA) was also used. For methods two and three, the cells were also assessed using fluorescence microscopy. Finally, cells were harvested, and DNA was purified and analyzed for fragmentation (an assessment of apoptosis) using conventional agarose gel electrophoresis. We identified statistically significant and reproducible changes in endothelial cell gene expression after incubation with each bacterial strain as compared to the uninfected controls and to each other. The data were analyzed using Webgestalt [18] to identify mammalian cell genes whose products comprise functional pathways in which multiple components showed alterations in gene expression (Table 1 ). Four pathways that show internally consistent changes in gene expression are the KEGG focal adhesion, regulation of actin cytoskeleton, leukocyte transendothelial migration, and ECM-receptor interaction pathways. They are considered together because a number of genes encode proteins whose functions participate in aspects of cell biology common to these pathways. Actin microfilaments are one of the three major components of the cellular cytoskeleton. The cytoskeleton participates in maintaining adhesion to and communicating with the extracellular matrix, cell migration, division, and signaling. b-Actin (ACTB) mRNA was decreased in response to L. interrogans but increased in response to L. biflexa, both as compared to the uninfected control cells (Table 2) . Guanine nucleotide-binding protein alpha-13 subunit (GNA13) mediates the activation of the small GTPase RhoA [19] which when activated controls the assembly of focal adhesions and actin in the formation of stress fibers [20] . Although RhoA was not differentially regulated in response to the bacteria, Rho GTPase activating protein 5 (RhoGAP5) was differentially expressed following the same pattern as GNA13, in which both genes were downregulated in response to the pathogenic leptospires in comparison to the uninfected controls, and upregulated in response to the non-pathogen. The effect of decreased GNA13 may be to decrease stimulation of Rho, while The changes in expression of several additional genes are consistent with changes in cellular architecture as a result of leptospiral infection of these endothelial cells. For example, decreases in the mRNAs for radixin (RDX, a protein that links the actin cytoskeleton to the plasma), caveolins 1 and 2 (CAV1 and CAV2, which couple integrins to the Ras-ERK pathway, titin, the ECM component laminin b1, and integrin subunits a v and b 3 ( Table 2) , were seen in cells infected with L. interrogans Canicola as compared to the uninfected controls. In contrast, the L. biflexa Patoc caused increases in mRNA levels for the same genes in infected cells vs. uninfected controls (Table 2) . Together, all of these gene expression patterns are consistent with the hypothesis that one effect of L. interrogans serovar Canicola is to promote actin remodeling and detachment of the cells from the ECM. A fundamental stage in the pathogenesis of Leptospira infections is the ability of the bacteria to cross mucous membranes and underlying epithelial barriers, as well as endothelial cell barriers, and disseminate to different organs. Although Leptospira species are Ea.hy926 endothelial cells were plated in tissue culture treated glass chamber slides and allowed to reach near confluence (assessed visually). The bacteria were added at MOI = 10 and incubated with the endothelial cells for 1 or 3 hours at 37uC, then were washed and fixed. The slides were stained with phalloidin-FITC, which illuminates F actin, plus anti-Leptospira antibodies followed by TRITC-conjugated secondary antibody. Retraction of the cell bodies in response to L. interrogans Canicola, but not L. biflexa Patoc, is evident, particularly at 3 hr infection. The brighter staining of rounded and retracted cells with FITC-phalloidin may be due to disorganization of cellular architecture without complete depolymerization of the actin, which in the increased depth and decreased area of the cytoplasm would appear more concentrated and therefore brighter. Changes in endothelial cell morphology were most evident, and at earlier time points, in cells with which the L. interrogans bacteria were associated. One higher magnification micrograph of L. interrogans Canicola infected cells is included because the bacteria are small when viewing fields of endothelial cells that provide information on integrity of the monolayer. Micrographs are representative of multiple (.12) independent experiments. L. interrogans Copenhageni caused essentially the same changes in endothelial cell morphology as L. interrogans Canicola (data not shown). doi:10.1371/journal.pntd.0000918.g001 Endothelial Cell Responses to Leptospira www.plosntds.org extracellular bacteria apparently devoid of actin modifying exotoxins [21] [22] [23] , and devoid of the specialized secretion systems utilized by many bacterial pathogens to deliver toxins that disrupt the host cell cytoskeleton (as reviewed in [24] [25] [26] [27] [28] ), pathogenic leptospires might be indirectly targeting the cytoskeleton via cell surface attachment mechanisms that co-opt the host cell signaling to achieve the same result. Decreased cellular adhesion to the ECM and rearrangement of the cytoskeleton may facilitate the migration of Leptospira through endothelial barriers as it disseminates from the site of inoculation. To further explore the possibility that actin rearrangements are triggered by Leptospira infection at the functional level, endothelial cells plated in chamber slides were infected at an MOI of 10 for 1 hour and 3 hours. As shown in Figure 1 , the bacteria were clearly more adherent to the cells than to the extracellular space, and the pathogenic bacteria caused dramatically more significant alterations in cellular morphology and integrity of the cell layer than did the non-pathogenic bacteria. The earliest change noted was a reduction in cortical actin (so the cell edges are less defined) and appearance of gaps in confluent cell layers, followed by loss of stress fibers and rounding of the cells. The images shown in Figure 1 are from cell layers that were just below confluence prior to infection, to allow better visualization of changes in individual cells. For example, while the cortical actin has largely disappeared in cells infected with L. interrogans Canicola by 1 hour postinfection, and stress fibers have disappeared and cell rounding is evident by 3 hours, the cells are largely unaffected at the same time points after infection with L. biflexa Patoc (Figure 1 ). L. biflexa Endothelial Cell Responses to Leptospira www.plosntds.org does adhere to mammalian cells in culture less efficiently than does L. interrogans (as shown and reviewed in [17] ), but even when bacterial contact with the cells was facilitated by centrifugation, the L. biflexa caused little disruption to cellular morphology and cell layer integrity (data not shown). Although these and subsequent experiments were performed using adherent cells, the morphologic changes are consistent with changes in mRNA levels seen using lifted cells in the microarray experiments. Despite the alterations in cellular architecture and monolayer integrity, no decrease in endothelial cell viability was found by any of several criteria (see Materials and Methods), even after infection times extended as long as 48 hours (Figure 2 ). The disruptions in the layers did, however, result in the ability of the pathogenic strain to cross the monolayers more efficiently than did the non-pathogenic bacteria (Figure 3) . After a brief period in which the endothelial layer did prevent significant transmigration of the bacteria, the layer rapidly became essentially irrelevant as a barrier to the penetration of the pathogenic bacteria, as the bacterial counts in the lower chamber were unaffected by whether or not cells had been plated on the membrane. Because Leptospira interrogans has been shown to bind to proteoglycans on the mammalian cell surface [17] , we tested a proteoglycan synthesis inhibitor, b-xyloside, for the ability to decrease damage to endothelial cell layers caused by L. interrogans Canicola. b-xyloside inhibits transfer of glycosaminoglycan chains to protein cores; a control sugar analog, a-galactoside, was tested in parallel. As shown in Figure 4 , inhibition of proteoglycan synthesis did not fully prevent the damage to the endothelial cell layers caused by L. interrogans. The inhibition of glycosaminoglycan chain attachment does not significantly affect the formation of holes in the cell layer caused by L. interrogans Canicola as assessed visually and by measurement of L. interrogans penetration of the cell layers (data not shown). b-xyloside does cause a reduction of L. interrogans Canicola and Copenhageni attachment to these cells ( [17] and data not shown), but does not abolish bacterial attachment, consistent with the hypothesis that additional nonproteoglycan molecules serve as substrates for L. interrogans attachment to cells. Direct bacterial attachment to the cells does appear to be required for the damage to the endothelial cell layers, as supernatants harvested from infected cell layers (infection times of 1-24 hr) and sterilized by centrifugation and filtration through 0.1 mm filters did not affect endothelial cell layer integrity (data not shown). Therefore, non-proteoglycan cell surface receptors are likely to be those primarily involved in the responses of the endothelial cells to L. interrogans attachment, and efforts to identify both the host cell and the bacterial cell molecules involved in these interactions are underway. As noted in the publication reporting the sequence of two L. biflexa Patoc strains [29] , there are a number of proteins predicted in the published L. interrogans genomes that are not present in the L. biflexa Patoc genome, including some that are postulated to have potential adhesin activities. These include proteins containing leucine-rich repeats, which are involved in many protein-protein interactions [29] . As stated in the publication of the L. biflexa genome, it is intriguing Ea.hy926 cell layers were treated with the proteoglycan synthesis inhibitor b-xyloside, or the control a-galactoside, as described in [17] prior to infection with L. interrogans sv. Copenhageni. After 3 hr, the cell layers were washed and fixed, then stained with phalloidin-FITC. Neither reagent significantly reduced disruption of the cell layers by L. interrogans, as alterations in cell morphology and significant gaps between cells were seen when L. interrogans was present, and trans-endothelial cell layer migration was not significantly affected (data not shown). Consistent with this result, chondroitin sulfates B and A, which do and do not inhibit L. interrogans attachment to mammalian cells, respectively [17] , also had no effect (not shown). doi:10.1371/journal.pntd.0000918.g004 Endothelial Cell Responses to Leptospira www.plosntds.org that a Treponema denticola leucine-rich repeat protein, LrrA, has been identified as an adhesion/tissue penetration factor [29, 30] . It is also possible that additional components of the surfaces of L. interrogans and L. biflexa might have different effects on host cells [31] [32] [33] . At this point, however, the determinants critical to the effects of L. interrogans-host cell interaction reported here remain to be identified, and neither bacterial adhesins nor host substrates can necessarily be predicted solely on the basis of the primary amino acid sequences. Several drugs currently in use in humans have been reported to have endothelial barrier protective function; all are in use as antihypertensive therapeutics, and some for other therapeutic purposes as well. We therefore tested four different drugs with different mechanisms of action for the ability to prevent the damage to endothelial layers in culture caused by L. interrogans. Lisinopril binds to and competitively inhibits angiotensin 1 binding to angiotensin converting enzyme (ACE), which is expressed by endothelial cells, while telmisartan competitively inhibits angio-tensin 2 binding to its receptor AT 1 . Dopamine is an antagonist of VEGF/VEGFR2-mediated cell layer permeability in treatment of human umbilical vein endothelial cells (HUVECs) in vitro at 10mM, as well as VEGF-mediated angiogenesis in vivo and proliferation of HUVECs at 1 mM in vitro [34, 35] . Furosemide is an anion transport blocker and is used as a diuretic but has antihypertensive activity as a consequence, and was used as a control not expected to preserve endothelial layer integrity. While telmisartan, furosemide, and dopamine did not protect the endothelial layers from the damage due to L. interrogans Copenhageni infection, lisinopril did at 100 nM, 1 mM and 10 mM ( Figure 5 , representing 3 independent experiments, and data not shown). There are several possible explanations for this, including: 1) lisinopril inhibits L. interrogans attachment to the cells, and 2) that attachment is unaffected but the interaction of the bacteria triggers activation of a signaling cascade or release of a mediator whose action or activation is inhibited by lisinopril. We therefore investigated the possibility that lisinopril might prevent Figure 5 . Effects of specific drugs that protect endothelial barrier function on damage caused by L. interrogans. Panels A and B: Ea.hy926 endothelial cell layers were infected with L. interrogans sv. Copenhageni or L. biflexa sv. Patoc as described in Materials and Methods, except that just prior to the addition of the bacteria the drugs lisinopril, telmisartan, dopamine, or furosemide were added to 1 mM. The micrographs shown in Panel A (representative of three experiments) were taken at the 6 hour time point; the graphs in Panel B show the transmigration of leptospires over the entire 72 hr. time course. Shown are the means and standard deviations of all data from three experiments. Statistical significance was determined using repeated measures ANOVA followed by Bonferroni's multiple comparison test. For wells with cells, L. interrogans vs. L. biflexa, p,0.001, L. interrogans with no additions vs. telmisartan p.0.05 (not significant), L. interrogans with no additions vs. lisinopril p,0.001, L. interrogans with telmisartan vs. lisinopril p,0.001. There were no significant differences in the absence of cells, and the drugs did not affect bacterial motility or attachment of 35 S-labeled leptospires to the cells (Panel C and data not shown). doi:10.1371/journal.pntd.0000918.g005 Figure 6 . Lisinopril concentrations effective in protection of endothelial cell layers from damage due to L. interrogans. Ea.hy926 endothelial cell layers were infected with L. interrogans sv. Copenhageni as described in Materials and Methods, except that just prior to the addition of the bacteria the drugs lisinopril or telmisartan were added to the concentrations indicated. The micrographs were taken at the 3 hour time point. Lisinopril at 10 mM, 1 mM and 100 nM blocked endothelial disruption by L. interrogans; lisinopril at 10 nM or below did not. doi:10.1371/journal.pntd.0000918.g006 Endothelial Cell Responses to Leptospira www.plosntds.org endothelial damage by blocking L. interrogans Copenhageni attachment to the cells, but no inhibition of adhesion of 35 Slabeled bacteria [17] was seen even at a concentration of lisinopril 10 fold over the concentration used for these experiments ( Figure 5 ). Although it was tempting to speculate that cell-surface-localized ACE could serve as a receptor for L. interrogans, as the enzyme is expressed by endothelial cells and proximal tubule epithelial cells [36] , and is therefore open to possible competition by the lisinopril, this is not consistent with our results to date. However, ACE2 is not inhibitable by lisinopril, but is a receptor for the SARS virus [37] , so there is precedent for ACE proteins serving as receptors for pathogens. It is also possible that the effect of lisinopril in our system is not related to ACE inhibition, but is instead due to additional effects of lisinopril, such as inhibition of isoprenoid synthesis, which is required for the post-translational modification of Rho GTPases, which in turn regulate the actin cytoskeleton [38] . In turn, this may lead to increased NO synthesis, which is protective of endothelial function in the face of a variety of insults. Given that doxycycline also has endothelial protective effects [39] , and that doxycycline is effective in treating leptospirosis [40] , our results may also provide a starting point for investigation into possible combinatorial therapeutic approaches to reduction of endothelial damage and consequent organ damage in human populations during leptospirosis outbreaks. Should this combinatorial approach prove useful in animal models, consideration as a focused approach to the treatment of human leptospirosis is warranted. The 1 mM dose shown in Figure 5 is at the high end of the physiologically relevant dosing range for humans, but administration of an antihypertensive to a patient with clinical manifestations of leptospirosis would be contraindicated, as further depression of blood pressure levels would be potentially lethal. However, in outbreak situations, this agent could potentially help to reduce endothelial damage if administered to affected populations as soon as an outbreak situation is recognized, prior to exposure of the majority of the population to pathogenic Leptospira species. In addition, protective effects of lisinopril were maintained even at a dose of 100 nM, which is well within the range routinely used in humans ( Figure 6 ). It will also be interesting to investigate the possibility that, on a population basis, patients on lisinopril fare better than patients not on this therapy during leptospirosis outbreaks. Reorganization of the actin cytoskeleton, as indicated by our microarray studies and by phalloidin staining of F actin, is essential to the pathogenesis of diverse bacterial infections, and pathogens use many different strategies to provoke changes in the cellular cytoskeleton in order to facilitate invasion of tissues, invasion of host cells, or evasion of phagocytosis (as reviewed in [24, 41, 42] ). A different spirochete, Treponema denticola, produces the protein Msp, which disrupts the actin cytoskeleton in neutrophils and fibro-blasts, preventing phagocytosis of the bacterium and inhibiting the cellular migration required to respond to and repair the damage caused by the pathogen and the host response at the site of infection [43, 44] . These activities are likely to facilitate invasion and colonization of periodontal tissues by T. denticola. Previous work by another laboratory demonstrated that L. interrogans Copenhageni crosses MDCK canine kidney epithelial cell layers in culture more rapidly than does L. biflexa Patoc [45] , but without significant disruption to the cell layers or the actin cytoskeleton. Consistent with these results, in experiments not shown here we also observed no significant damage to NRK (normal rat kidney) 293 (human kidney) or HEp-2 (human laryngeal) epithelial cell layers infected with L. interrogans Canicola or L. interrogans Copenhageni. The calculations of the proportions of bacteria crossing the cell layers differed between the two studies, but our protocol accounted for the replication of the L. interrogans Canicola and Copenhageni in the co-cultures, while the L. biflexa Patoc did not replicate (data not shown). Thus the endothelial cells tested here respond very differently to the bacteria than did the MDCK epithelial cells, and our results are the first to suggest a mechanism: disruption of actin dynamics by bacterial attachment to the cell surface. Thus, while L. interrogans has not been shown to secrete a toxin that modifies actin, the bacteria are able to manipulate the actin cytoskeleton indirectly. Even the pore forming toxin activity reported for Leptospira [46, 47] does not appear to have as large an effect, as the endothelial cells here were viable throughout the experiments. The leptospires may be able to establish disseminated infection in part due to the binding of the bacteria to one or more mammalian cell surface receptors that in turn, regulate the dynamics of the actin cytoskeleton in the mammalian cell. Deciphering the role of, and mechanisms behind, actin rearrangement in response to pathogenic Leptospira will provide insights into the mechanisms that leptospires uses to disseminate to different organs of the host to cause infection and disease, and provides a possible avenue for therapeutic intervention in conjunction with antimicrobial therapy.

A Porcine Adenovirus with Low Human Seroprevalence Is a Promising Alternative Vaccine Vector to Human Adenovirus 5 in an H5N1 Virus Disease Model

Human adenovirus 5 (AdHu5) vectors are robust vaccine platforms however the presence of naturally-acquired neutralizing antibodies may reduce vector efficacy and potential for re-administration. This study evaluates immune responses and protection following vaccination with a replication-incompetent porcine adenovirus 3 (PAV3) vector as an alternative vaccine to AdHu5 using an avian influenza H5N1 disease model. Vaccine efficacy was evaluated in BALB/c mice following vaccination with different doses of the PAV3 vector expressing an optimized A/Hanoi/30408/2005 H5N1 hemagglutinin antigen (PAV3-HA) and compared with an AdHu5-HA control. PAV3-HA rapidly generated antibody responses, with significant neutralizing antibody titers on day 21, and stronger cellular immune responses detected on day 8, compared to AdHu5-HA. The PAV3-HA vaccine, administered 8 days before challenge, demonstrated improved survival and lower virus load. Evaluation of long-term vaccine efficacy at 12 months post-vaccination showed better protection with the PAV3-HA than with the AdHu5-HA vaccine. Importantly, as opposed to AdHu5, PAV3 vector was not significantly neutralized by human antibodies pooled from over 10,000 individuals. Overall, PAV3-based vector is capable of mediating swift, strong immune responses and offer a promising alternative to AdHu5.

Experimental adenovirus-based vaccine vectors are promising alternatives to conventional vaccine platforms. In particular, human adenovirus serotype 5 (AdHu5) vectors are well-characterized and are being developed against several infectious disease models including influenza, hepatitis C, dengue and viral hemorrhagic fever viruses [1, 2, 3, 4] . Several candidates have demonstrated unique protective efficacy and can generate robust immune responses in both animal models and clinical trials [4, 5, 6, 7] . Pre-existing immunity against AdHu5 is, however, frequent in the human population and has been associated with undesirable clinical outcomes and the suspension of clinical trials [8, 9, 10] . One promising alternative is the development and evaluation of rare human, chimpanzee, or other mammalian adenovirus vectors with low seroprevalence in humans. A chimeric simian adenovirus 21 vector protected mice against lethal Ebolavirus challenge and generated robust T-cell responses against the glycoprotein in nonhuman primates [11] . A bovine adenovirus 3 (BAV3)-based vaccine previously demonstrated successful protection against avian influenza A virus H5N1 challenge in mice and was able to escape preexisting neutralizing antibodies against AdHu5 [12] . Similarly, a porcine adenovirus 3 (PAV3) vector was successful in several swine vaccination studies against classical swine fever and pseudorabies virus [13, 14, 15] . PAV3-based vaccines were able to evade pre-existing immunity and provide long-term protection in pigs [14] . The antigenic profile and reported in vivo efficacy as an animal vaccine makes the PAV3 vector a promising alternative adenovirus vector for human administration. Due to their genetic diversity and the availability of several significantly different isolates, avian influenza H5N1 viruses provide a valuable and challenging disease model for evaluating broad immune responses generated by potential adenovirus vectors. The external hemagglutinin (HA) glycoprotein mediates receptor binding, fusion, and can generate both strong antibody and cell-mediated immune responses [16] which can be directly assayed and provide useful comparison of adenovirus platforms. Highly pathogenic avian influenza H5N1 viruses have spread throughout domestic and aquatic bird populations in South East Asia and the World Health Organization (WHO) has confirmed 500 clinical cases of H5N1 cross-transmission into humans. Despite the limited incidence of human-to-human-transmission, high mortality rates (.60%) and continuous evolution of the virus represent a concern for future influenza pandemics [17, 18] . The emergence of pandemic swine-like H1N1 influenza A virus isolates in early 2009 highlights the need to generate cross-protective and lasting immune responses against diverging human and zoonotic influenza viruses. In addition to evaluation of different adenovirus platforms, the development of improved influenza vaccines would also help in better preparation against emerging pandemic viruses and could reduce the impact of infection on public health. This study evaluates the protective efficacy following lethal homologous challenge of a replication-incompetent porcine adenovirus 3 (PAV3) vector expressing the HA gene from the A/ Hanoi/30408/2005 H5N1 (H5N1-H05) influenza A isolate (PAV3-HA). The immunogenicity of HA and the success of previous AdHu5 H5N1-HA vaccines [1, 19] suggested that avian influenza H5N1 may be a good comparative model to evaluate the efficacy of a similar PAV3 vector. Previous studies showed that PAV3 does not exhibit crossreactivity with AdHu5 or BAV3 neutralizing antibodies [20, 21] . Additionally, PAV3 was not neutralized by the lowest dilution of 1:4 from 50 randomly selected human sera [20] In order to further address neutralization of PAV3 by an extended number of human sera, human Ig made of pooled sera from 10,000-60,000 individuals was evaluated. AdHu5, used as a control, was neutralized at the highest dilution of 1:160 (6.25610 23 mg/ml human Ig). In contrast, neutralization of PAV3 was not detected at 1:20 (5.061022 mg/ml), the lowest dilution tested. Previous studies showed little cross-reactivity between cell-mediated immune responses against AdHu5 and PAV3 [22] . An optimized expression cassette containing the codonoptimized H5N1-HA was inserted by homologous recombination into a replication-incompetent PAV3 vector containing deletions in the E1/E3 genes (described in [23] ). An AdHu5-HA vaccine was also developed in parallel as a control to compare levels of protection and immune responses. Expression of the PAV3-HA and AdHu5-HA vaccines was evaluated in VRIBL E1, HEK 293, and mouse AB12 cells (Figure 1a ). BALB/c mice were also vaccinated with 10 10 virus particles (vp) and in vivo expression of PAV3-HA or AdHu5-HA in muscle tissue was detected 4 days post-immunization ( Figure 1b) . A robust antibody response is important for the prevention of influenza virus infection. The ability of the PAV3-HA vaccine to generate humoral immune responses following immunization was assayed through detection of HA-specific antibodies by hemagglutination inhibition (HI) or neutralizing antibody (NAB) titres. BALB/c mice were vaccinated with 10 10 vp/mouse of PAV3-HA or AdHu5-HA and the development of HI and NAB antibody responses was monitored from serum collected at days 8, 10, 14, and 21 post-immunization. All samples were treated with receptor-destroying enzyme (RDE), followed by complement inactivation the next day. An HI assay was performed using serial dilutions of the treated serum combined with H5N1-H05 virus and horse red blood cells to detect inhibition of cell agglutination by serum antibody. The reciprocal of the highest dilution which did not agglutinate red blood cells was scored as the HI antibody titre. HI titres of 2060, 120669, 213692, 5336184 were detected for PAV3-HA and 27611, 67623, 213692, 4266184 for AdHu5-HA, respectively, with no statistical difference between the two vaccines ( Figure 2a ). The presence of neutralizing antibodies to inhibit active H5N1-H05 infection was also evaluated as a measure of the humoral response. Serial dilutions of treated serum were incubated with 100 pfu/well of H5N1-H05 virus and cells were monitored for the presence or absence of cytopathic effects (CPE) using a light microscope. The highest serum dilution which did not exhibit CPE was scored as positive for neutralizing antibody and titres were reported as the reciprocal of the dilution. NAB titres of 060, 7612, 7612 and 90650 or 060, 7612, 10610 and 4060 reciprocal dilutions were observed with the PAV3-HA or AdHu5-HA vaccines at 8, 10, 14, and 21 days, respectively (Figure 2b ). The increased neutralizing antibody response following PAV3-HA vaccination was statistically significant at day 21 (p = 0.045) relative to AdHu5-HA. In order to assess antibody titres immediately before challenge, mice were vaccinated with different doses of each vaccine and serum was obtained 25-days post-immunization. HI reciprocal titres of 9866326, 10936293, and 1280 for the PAV3-HA vaccine or 10666330, 10666330, and 11736261 with AdHu5-HA were detected for 10 8 , 10 9 or 10 10 vp/mouse doses respectively (Figure 2c ). Both vaccines had similar HI antibody titres at all doses, with no significant differences detected between the PAV3-HA and AdHu5-HA (p = 0.57). Based on clinical data, a reference HI antibody titre of 40 is recommended for influenza vaccine candidates as the minimum level to confer fifty percent protection in humans [24, 25] . Both vaccines met this requirement at all doses prior to challenge. Serum samples had an average NAB titre of 65620, 75614, or 106646 reciprocal dilution at 10 8 , 10 9 or 10 10 PAV3-HA vp/mouse respectively ( Figure 2d ). The AdHu5-HA vaccine generated NAB titres of 65620, 55620 or 133646 at 10 8 , 10 9 or 10 10 vp/mouse, respectively. Although a strong antibody response is important for immediate and long-term protection against influenza viruses, the induction of early cellular immune responses following vaccination may enhance clearance of virus-infected cells following H5N1 influenza virus infection. T-cell responses were assayed using an enzymelinked immunosorbent spot (ELISPOT) assay to detect secretion of interferon gamma (IFNc) by activated lymphocytes. Splenocytes were obtained from mice vaccinated at days 8, 10, 14, and 21 with 10 10 vp/mouse of PAV3-HA or AdHu5-HA and the cells were restimulated with pools of overlapping peptides corresponding to the entire H05-HA protein. The peptide corresponding to the immunodominant H5N1-H05 HA (IYSTVASSL, conserved influenza A virus) epitope was evaluated alongside an unrelated control peptide (TYQRTRALV, A/PuertoRico/8/34 H1N1 nucleoprotein). In mice vaccinated with PAV3-HA, 13,59862066, 14,4426 2541, 69546392, or 20566633 spot-forming cells (sfc) per million splenocytes were detected at days 8, 10, 14, or 21 respectively (Figure 3a) . Immunization with the AdHu5-HA vaccine generated 1,9656341, 12,85861749, 7332693, and 19026372 sfc/million, respectively. The number of sfc/million splenocytes detected at day 8 was statistically significant between PAV3-HA and AdHu5-HA vaccines (p,0.005). The H05-HA immunodominant epitope also stimulated the strongest T-cell responses on day 8 for the PAV3-HA vaccine compared to AdHu5-HA. PAV3-HA vaccinated mice generated an average of 8126138 sfc/million at day 8 whereas 2206133 sfc/million were detected from AdHu5-HA mice at the same time point (Figure 3b , p = 0.048). At day 10, average responses following stimulation with the H05-HA immunodominant epitope were 1,2166178 or 1,364636 sfc/ million from PAV3-HA or AdHu5-HA immunized mice respectively (p = 0.258). T-cell responses of 770674 and 453684 were detected from PAV3-HA or 791658 and 590684 from AdHu5-HA at days 14 and 21 post-immunization, respectively. Restimulation of splenocytes from PAV3-HA or AdHu5-HA immunized mice by the NP control peptide repeatedly generated less than 65 sfc/million. Cellular responses following vaccination with PAV3-HA were further characterized by detecting the relative frequency of peptide-specific CD8 + T cells expressing IFN-c by flow cytometry. Splenocytes were harvested ten days post-immunization and restimulated in vitro with the H05-HA immunodominant peptide and unrelated NP or ZGP (TELRTFSI, Zaire Ebolavirus glycoprotein) control peptides. Cells were stained with anti-mouse CD8-FITC (fluorescein isothiocyanate) and an anti-mouse IFNc-PE (phycoerythrin). The frequency of IFNc positive CD8 + T-cells was 3.460.3 and 2.860.3 percent for PAV3-HA or AdHu5-HA, respectively (Figure 3c ). Cells stimulated with control peptide NP or ZGP showed frequencies of CD8 + IFNc positive T-cells equal to 0.360.05 or 0.660.05 percent, respectively. These results suggest that the PAV3-based vaccine can stimulate a robust immune response in mice faster than AdHu5HA and comparable in strength. Protective efficacy of PAV3-HA following lethal challenge The early detection of the T-cell response at day 8 postvaccination with the PAV3-HA vaccine suggest that it may be a good candidate for immediate administration just prior or following a suspected exposure to H5N1 virus. The success of rapid vaccination regimen was evaluated in two parts: first, overall survival and second, total viral load in lung tissue, as determined by TCID50 assay. BALB/c mice were challenged with a lethal dose of H5N1-H05 virus at 5, 8 and 10 days post-vaccination and lungs were harvested at day 3 post-infection. Although mice challenged 5 days post-vaccination did not survive lethal challenge (Figure 4a Survival was also assessed in groups of 10 BALB/c mice vaccinated with 10 8 , 10 9 , or 10 10 virus particles of recombinant adenovirus PAV3-HA or AdHu5-HA by intramuscular (I.M.) administration, challenged 28 days later with lethal homologous H5N1-H05 virus. Protection against clinical signs of disease with minimal weight loss and full survival was observed at the 10 10 vp/ mouse dose for both Ad-based vaccines ( Figure 5 ). Complete survival was also observed for both vaccines at a dose of 10 9 vp/ mouse, with the PAV3-HA vaccinated mice either asymptomatic or having milder signs of disease (14% weight loss) compared to AdHu5-HA (21% weight loss). Another desirable characteristic that influenza vaccines should provide is long-term protection. Therefore, long term immunity was evaluated in mice challenged 12 months after vaccination in order to determine whether protective immune responses could be maintained. PAV3-HA afforded full protection in mice challenged with 100 LD50 of H5N1-H05 virus 12 months post-immunization whereas 50% of mice vaccinated with AdHu5-HA succumbed (Figure 6a ). Higher HI antibody titers for PAV3-HA compared to AdHu5-HA (186665 and 60623, respectively (p = 0.006)) may have translated directly to the improved survival observed with the PAV3-HA vaccine. NAB titers were 23615 and 10611 for AdHu5-HA (Figure 6b ). An ELISA assay was also performed to detect total IgG antibody titres against the H5N1-HA antigen. Serum was obtained from mice 25 days and 1 year post-vaccination, and unvaccinated control mice (Figure 6c ). Total antibody titers were significantly lower for both vaccines after 1 year. On average, higher levels of IgG antibodies were detected for the PAV3-HA vaccine, however the difference was not statistically significant compared to AdHu5-HA (p = 0.241). Currently, replication-deficient human adenovirus serotype 5 (AdHu5) vaccines are being evaluated against several pathogens. Several candidates have been shown to induce protective immune responses against emerging or re-emerging infectious pathogens such as Ebola and avian influenza (H5N1) viruses [4, 26, 27] Complete protection and long-term memory responses have also been reported in different animal models [7, 26, 28] ; however, the final development of AdHu5-based vector to approved human vaccines has been hampered by the presence of natural preexisting immunity to AdHu5 which is present in a large fraction of the human population. Neutralizing antibody to porcine adenovirus 3 (PAV3) was not detected from pooled immune globulin representing 10 to 60 thousand human sera suggesting that preexisting immunity to PAV3-based vaccines is not likely to be of concern for human applications. In addition, the documented compatibility of porcine and human tissues could translate into PAV3-based vaccines being efficacious while of low toxicity in humans [23] . The further development of a new adenovirus serotype also increases the spectrum of possible applications such as sequential vaccination with other adenovirus vectors (e.g. prime/boost or a second Ad-based vaccination against a different agent) [20, 29] . The present study compares in parallel the immune responses and protection generated by a PAV3-based vector and a similar AdHu5-based vector using an avian influenza H5N1 disease model of infection. PAV3-based vector expressing the H5N1 HA antigen was able to generate quick and robust immune responses against H5N1. This finding was supported by the rapid induction of improved immediate protection by the PAV3-HA vaccine compared to AdHu5-HA. Interestingly, both vaccines shared similar in vivo expression, suggesting that the increase in vaccine efficacy observed with PAV3-HA may be due to specific interactions with the immune system rather than enhanced expression of the antigen. A recent study suggests that PAV3-HA may activate different innate immune pathways to AdHu5-HA [30] . Additionally, the PAV3-HA offered improved long-term protection against lethal H5N1 virus challenge with higher levels of detectable antibodies by HI assay. Although full short-term (28 day) survival was observed following vaccination with PAV3-HA or AdHu5-HA at the 10 9 vp/mouse dose, greater weight loss was observed with the AdHu5-HA vaccine. Comparing HI and NAB antibody titres with observed clinical outcome, this suggests that there may have been incomplete neutralization of the virus following vaccination with AdHu5-HA, perhaps explaining why there were differences in the levels of protection observed following challenge 1-year post-vaccination. It has been shown that levels of neutralizing antibodies in serum decrease over time following H5N1 infection [31] . Following vaccination, the exact levels of neutralizing antibodies correlating with full protection from infection with homologous virus is still uncertain. In comparison with total IgG antibody levels, neutralizing antibody titres correlated better with early and long-term protection, similar to previously described reports [19, 32, 33] . Together the data suggests that PAV3 vector is an additional option to AdHu5 vector and could have several important applications including rapid or post-exposure protection against emerging influenza viruses or other infectious agents. A similar study described a BAV3-HA vaccine sharing comparable protective efficacy to the AdHu5-HA against an H5N1 virus. Two doses of the BAV3 vector generated similar humoral and cellmediated immune responses and was able to evade pre-existing neutralizing antibodies against AdHu5 [12] . Overall protective efficacy offered by PAV3-HA was similar to AdHu5-HA one month post-immunization and the immune response kinetics was also generally comparable at later time points. Although not explored in the current study, previous studies have evaluated the impact of pre-existing immunity to PAV3 and the potential reuse of PAV3-based vectors against different pathogens. Groups of outbred pigs with high PAV3 neutralizing antibody titres were vaccinated with a PAV3-based vaccine and vector re-administration did not result in hepatotoxicity or reduced transgene expression [14] . Doses of 10 13 particles/kg AdHu5 vector can also bypass pre-existing immunity, however, several pathologies including liver damage indicated by elevated transaminase, low platelets count, and lymphocytopenia were observed in nonhuman primates administered similar doses [9, 10, 34, 35] . Even though conventional vaccines have been relatively successful against influenza A infection, the ability of adenoviral vectors to rapidly generate a strong immune response may be useful to specific applications such as rapid immunization of health care workers in anticipation of a probable exposure to an emerging virulent pandemic virus such as H1N1-1918. In addition to strong antibody responses against the H5N1-H05 virus, both survival and cellular responses suggest that induction of an earlier T-cell response by the PAV3-may complement the developing antibody response to improve protection against H5N1 challenge and contribute directly towards lower viral load. Although the influenza glycoproteins have high antigenic variation, the generation of faster T-cell responses by a PAV3 vector against well-conserved influenza antigens may supplement a mismatched antibody response and may improve protection against a wider range of influenza viruses. Previous studies have shown that PAV3 vectors can transduce several human cell lines resulting in full transgene expression and adenoviral coat proteins [20, 21] . An additional safety feature is that wild-type PAV3 does not replicate in human cells [21] , Groups of 6 BALB/c were vaccinated with 10 10 vp/mouse of PAV3-HA () or AdHu5-HA (), or no vaccine (Control, ()) and challenged with 100LD50 of H5N1-H05 on days 5, 8, and 10 post-vaccination. Lungs were harvested from the mice on day 3 post-challenge. Virus titre was determined by TCID50 assay using serial dilution of lung homogenates on MDCK cells and monitoring for the presence of CPE over 48 hours. The TCID50 titre was calculated by the Reed & Muench method [39] and normalized/gram of lung tissue. Data is presented as log 10 TCID50/g of lung tissue. The data represent average values and standard deviations from one experiment performed with one vector preparation of each vaccine (* and ** represents p,0.05). doi:10.1371/journal.pone.0015301.g004 suggesting that a replication-competent PAV3 vector which would be easy to produce at high titres could also be a promising vaccine candidate. Nevertheless, performance of this new vaccine vector will need to be addressed in other animal species, including nonhuman primates, before its real utility as a human vaccine can be predicted more accurately. Overall, this study supports a complementary role for cellular immunity during early H5N1 infection and the further development of porcine adenoviruses as human vaccine candidates. All animal procedures and scoring sheets were first approved by the Animal Care Committee (Animal Use Document ID# H-08-010) at the Canadian Science Centre for Human and Animal Health, according to the guidelines set by the Canadian Council on Animal Care. Human embryonic kidney (HEK) 293 cells, Madin-Darby canine kidney cells (MDCK), and mouse AB12 cells were maintained in Dulbecco's modified eagle's medium (DMEM), supplemented with 10% fetal bovine serum (FBS), L-glutamate, sodium pyruvate (NaPyr), and antibiotics. Fetal porcine retina cells (VR1BL E1), expressing human adenovirus 5 (AdHu5) E1a and porcine adenovirus 3 (PAV3) E1b large genes, were maintained in minimum essential medium (MEM) alpha, supplemented with 10% FBS, L-glutamate, NaPyr, non-essential amino acids, HEPES buffered saline, penicillin/streptomycin, and 50 mg/ml Hygro- mycin B (BD Biosciences). Avian influenza H5N1 strain A/ Hanoi/30408/2005 (H5N1) was generously provided by Q. Mai Le and T. Hien Nguyen, National Institute of Hygiene and Epidemiology, Hanoi, Vietnam. Virus was propagated on MDCK cells cultured with virus diluent (MEM, 0.3% bovine serum albumin, and antibiotics) containing 3.0 mg/ml TPCK-treated trypsin (TPCK-trypsin) and titered by plaque assay. All infectious recombinant adenovirus constructs were propagated on HEK 293 or VR1BL E1 cells and purified by a cesium chloride density gradient. The full complementary DNA (cDNA) sequence from the A/ Hanoi/30408/2005 H5N1 hemagglutinin (H05-HA) gene was obtained, codon optimized, and synthesized from overlapping oligonucleotide primers, as previously described [36] . The H05-HA gene was first cloned into the pCAGa plasmid, containing a chicken-b-actin (CAG) promoter, to generate the pCAGa-HA construct. The pCAGa-HA expression cassette was then excised and inserted into two shuttle vector systems: pShuttle2 (Clontech) and pPAV227 (VIDO, University of Saskatchewan). Insertion of the expression cassette replaced the existing pShuttle2 cytomegalovirus (CMV) promoter with the CAG promoter, resulting in the pShuttle2-HA construct. Transfer plasmid pPAV227 was also modified to include the SV40 polyadenylation signal from pShuttle2, generating pPAV227-HA-PolyA. The transgene cassettes from pShuttle2-HA and pPAV227-HA-PolyA were cloned into replication-deficient DE1DE3 adenoviral vectors: pAdenoX (AdHu5, Clontech) or pFPAV227 (PAV3, VIDO, University of Saskatchewan). pAdenoX-HA was generated through digestion of both pShuttle2-HA and pAdenoX by homing endonucleases I-CeuI/PI-SceI and ligation of cohesive ends using T4 DNA ligase (Invitrogen). pFPAV227-HA was generated through homologous recombination in Escherichia coli strain BJ5183 (recBC, sbcBC) [37] of linearized pPAV227-HA-PolyA (Eco47III/TthIII1) and pFPAV227 (PacI). To obtain AdHu5-HA vaccine, HEK 293 cells were transfected with 10 mg of linearized pAdenoX-HA DNA in calcium phosphate (BD Biosciences) solution and cells were cultured until the appearance of cytopathic effects (CPE). Similarly, VR1BL E1 cells were transfected with 10 ug of linearized pFPAV227-HA DNA combined with Lipofectin (Invitrogen) and cells were cultured until CPE were apparent. Amplified adenoviruses containing cell lysates were harvested, freeze-thawed three times, and purified by CsCl gradients. The integrity of the H05-HA transgene cassette was confirmed through EcoRI restriction digests and by sequencing (DNA Core, National Microbiology Laboratory) with multiple primer sets. Total virus particles (vp) was determined by OD260 and total infectious particles was determined using anti-hexon antibodies against AdHu5 (AdenoX Rapid Titer kit, Clontech) or against PAV3 (VIDO, University of Saskatchewan). Four independent vector preparations were used for each vaccine. Total infectious particles and total viral particles were determined for both adenovirus vaccines, with ratios of 1:285, 1:333, 1:250, and 1:300 for PAV3-HA and 1:150, 1:250, 1:220, and 1:250 for AdHu5-HA. Protein expression of H05-HA by both vectors was confirmed in HEK 293, VRIBL E1, and AB12 cells using standard Western blotting techniques. To evaluate in vivo expression, groups of 6 BALB/c mice were vaccinated with 10 10 vp of PAV3-HA or AdHu5-HA and muscle tissue was harvested 4 days following immunization. Muscle tissues were homogenized and normalized per gram of muscle tissue in radioimmunoprecipitation (RIPA) buffer. Expression of each vaccine was detected from 75 mg of loaded muscle tissue. Groups of 10 BALB/c (Charles River Canada) were vaccinated with 10 8 , 10 9 , or 10 10 vp of recombinant adenovirus PAV3-HA or AdHu5-HA by intramuscular (I.M.) administration. Each vaccine was diluted in 100 ml and 50 ml was administered in each of the right and left hind limbs. All mice were anesthetized and challenged after 28 days through intranasal inoculation with 100 times the dose of A/Hanoi/30408/2005 virus required to obtain 50% survival (100 Lethal Dose 50 or 100LD50) in 50 ml virus diluent. The LD50 for the H5N1-H05 virus was 1.05 plaque forming units (pfu), therefore 100LD50 was 105 pfu. Mice were monitored for 15 to 20 days following challenge and signs of disease including weight loss, labored breathing, ruffled fur, and death were observed according to an approved scoring chart. All animal procedures were approved by the Institutional Animal Care Committee at the National Microbiology Laboratory (NML) at the Public Health Agency of Canada (PHAC), according to the guidelines of the Canadian Council on Animal Care. All infectious work was performed in the high biocontainment laboratory at NML/PHAC. Groups of 4 BALB/c mice were vaccinated with 10 10 vp of PAV3-HA or AdHu5-HA vaccines. The day before each experiment, ELISPOT-IFNc (BD Biosciences) plates were coated with purified mouse IFNc and incubated at 4uC, overnight. As well, overlapping 15mer peptides (Mimitopes, Australia) spanning the entire H05-HA protein were resuspended overnight in dimethyl sulfoxide (DMSO) and allocated in pools by matrix format. Spleens were harvested on days 8, 10, 14, and 21 postimmunization and splenocytes were plated at 5610 5 cells/well in RPMI 1640 (supplemented with 10% FBS, L-glutamine, NaPyr, HEPES, non-essential amino acids, 5610 23 M 2-b-mercaptoethanol, and antibiotics) and restimulated with each of the peptide pools (2.5 mg/ml per well). An individual 9mer peptide representing the immunodominant H5N1 H05-HA (IYSTVASSL, conserved influenza A viruses) epitope was also evaluated, along with negative controls PR8-NP (TYQRTRALV, A/PuertoRico/8/34 (H1N1) nucleoprotein) and ZGP (TELRTFSI, Zaire Ebolavirus glycoprotein). ELISPOT-IFNc plates were incubated at 37uC overnight (18-20 hours) and washed the following day according to the manufacturer's instructions. AEC Substrate Set (BD Biosciences) was used to develop spots formed by interferon gamma secreting cells. Spots were visualized and counted using an ELISPOT plate reader (AID ELISPOT reader, Cell Technology, Colombia, Maryland). Splenocytes obtained on day 10 post-vaccination, plated at 2610 6 cells/well, were restimulated with 9mer H05-HA, PR8-NP, or 8mer ZGP individual peptides in DMEM (supplemented with 10% FBS, L-glutamine, NaPyr, HEPES, non-essential amino acids, 5610 23 M b-mercaptoethanol, and antibiotics), IL2, and GolgiStop (Brefeldin A, BD Biosciences). Cells were stimulated for 5 hours and stained with anti-mouse CD8-FITC (fluorescein isothiocyanate) at 4uC for 30 minutes. BD Cytofix and permwash protocol was used to fix and permeabilized cells according to the manufacturer instructions. The following day, cells were stained

Liposome-Coupled Antigens Are Internalized by Antigen-Presenting Cells via Pinocytosis and Cross-Presented to CD8(+) T Cells

We have previously demonstrated that antigens chemically coupled to the surface of liposomes consisting of unsaturated fatty acids were cross-presented by antigen-presenting cells (APCs) to CD8(+) T cells, and that this process resulted in the induction of antigen-specific cytotoxic T lymphocytes. In the present study, the mechanism by which the liposome-coupled antigens were cross-presented to CD8(+) T cells by APCs was investigated. Confocal laser scanning microscopic analysis demonstrated that antigens coupled to the surface of unsaturated-fatty-acid-based liposomes received processing at both MHC class I and class II compartments, while most of the antigens coupled to the surface of saturated-fatty-acid-based liposomes received processing at the class II compartment. In addition, flow cytometric analysis demonstrated that antigens coupled to the surface of unsaturated-fatty-acid-liposomes were taken up by APCs even in a 4°C environment; this was not true of saturated-fatty-acid-liposomes. When two kinds of inhibitors, dimethylamiloride (DMA) and cytochalasin B, which inhibit pinocytosis and phagocytosis by APCs, respectively, were added to the culture of APCs prior to the antigen pulse, DMA but not cytochalasin B significantly reduced uptake of liposome-coupled antigens. Further analysis of intracellular trafficking of liposomal antigens using confocal laser scanning microscopy revealed that a portion of liposome-coupled antigens taken up by APCs were delivered to the lysosome compartment. In agreement with the reduction of antigen uptake by APCs, antigen presentation by APCs was significantly inhibited by DMA, and resulted in the reduction of IFN-γ production by antigen-specific CD8(+) T cells. These results suggest that antigens coupled to the surface of liposomes consisting of unsaturated fatty acids might be pinocytosed by APCs, loaded onto the class I MHC processing pathway, and presented to CD8(+) T cells. Thus, these liposome-coupled antigens are expected to be applicable for the development of vaccines that induce cellular immunity.

Vaccines have played an important role in disease prevention and have made a substantial contribution to public health. Upon natural infection, it is known that the host responds by inducing both humoral and cellular immunity against the pathogen. However, most of the currently approved vaccines work by inducing humoral immunity [1] [2] [3] . For protection against viruses that are highly mutable and frequently escape from antibodymediated immunity, such as influenza A viruses, HIV, and HCV, humoral immunity is insufficient [4] [5] [6] [7] . Consequently, the development of vaccines that induce cellular immunity is critical to novel vaccine strategies. T lymphocytes respond to peptide fragments of protein antigens that are displayed by MHC molecules on antigen-presenting cells (APCs). In general, extracellular antigens are presented via MHC class II molecules to CD4 + T cells while intracellular antigens are presented via MHC class I molecules to CD8 + T cells [8, 9] . However, a number of reports have demonstrated that a significant level of crossover, so-called 'cross-presentation', occurs in APCs [10] [11] [12] [13] [14] . Using this phenomenon, novel vaccine preparation inducing antigen-specific CTLs that effectively eliminate virus-infected cells is expected. The mechanisms of cross-presentation have been studied intensively [15] [16] [17] while the details have been left unclear. Part of the antigens taken via phagocytosis by APCs are known to be translocated into the cytosol and degraded by local proteases [18, 19] . In another pathway, some antigens internalized into endocytic compartments are loaded onto MHC class I molecules [20] . We previously reported that antigens chemically coupled to the surface of liposomes induced antigen-specific IgG but not IgE antibody production [21, 22] . In addition, antigens chemically coupled to the surface of liposomes consisting of unsaturated fatty acids were presented not only to CD4 + -but also to CD8 + T cells by APCs [23] . Since liposome-coupled antigens induce antiviral immunity [24, 25] , they are expected to be applicable for the development of viral vaccines without inducing antigen-specific IgEs, which cause allergic reactions. In the present study, we investigated the mechanism by which the liposome-coupled antigens were cross-presented by APCs to CD8 + T cells. Confocal laser scanning microscopic analysis of macrophages co-cultured with DQ-OVA-liposome conjugates MHC class I of macrophages were stained with red fluoresceinlabeled anti-mouse H-2D d mAb (Fig. 1A : left column), and MHC class II of macrophages were labeled with DM-DsRed ( Fig. 1A : right column) as described in Materials and Methods. DQ-OVA, which exhibits green fluorescein upon proteolytic degradation, was coupled to liposomes consisting of unsaturated (oleoyl) or saturated (stearoyl) fatty acid, and added to the culture of macrophages. After incubation for 2 hr, the recovered macrophages were analyzed using confocal laser scanning microscopy. The results shown in Fig. 1 demonstrate that DQ-OVA coupled to oleoyl liposomes was processed at both MHC class I and class II compartments, while most of the DQ-OVA coupled to stearoyl liposomes was processed at the MHC class II compartment. Alexa 488 -labeled OVA were coupled to liposomes and were added to the cultures of macrophages. As shown in Fig. 2 , OVA coupled to oleoyl liposomes were internalized by APCs more efficiently than those coupled to stearoyl liposomes at 37uC. Interestingly, OVA coupled to oleoyl liposomes but not stearoyl liposomes were internalized significantly by APCs even in a 4uC environment. One of two kinds of inhibitors, cytochalasin B and DMA, which inhibit APC phagocytosis and pinocytosis of antigens, respectively, was added to the culture of macrophages 1 hr prior to the addition of Alexa 488 -OVA-or DQ-OVA-coupled oleoyl liposomes. One hour later, flow cytometric analysis was performed. As shown in Fig. 3 , the effect of cytochalasin B on the antigen uptake and digestion of liposome-coupled OVA by APCs was limited. On the other hand, DMA significantly reduced both antigen uptake and digestion of antigens by macrophages. DQ-OVA-coupled oleoyl liposomes were added to the culture of macrophages in which either EEA1 or LAMP-1 were costained. The co-localization of the liposome-coupled antigens and intracellular organelles in the APCs was analyzed using confocal laser scanning microscopy. As shown in Figure 4 , although most of the DQ-OVA coupled to oleoyl liposomes was processed beyond LAMP-1-expressing compartments (green spots), a portion of DQ-OVA was processed at compartments expressing LAMP-1 (yellow spots). Co-localization of EEA1-expressing compartments with liposome-coupled-DQ-OVA was significantly less than that of LAMP-1-expressing compartments with DQ-OVA (Fig. 4B ). In agreement with the results shown in Fig. 3 , antigen presentation by APCs pulsed with liposomal antigen was significantly inhibited by DMA but not by cytochalasin B in both CD4 + -and CD8 + T cell responses (Fig. 5 ). In general, extracellular antigens are presented via MHC class II molecules to CD4 + T cells, whereas intracellular antigens are presented via MHC class I molecules to CD8 + T cells. Consequently, most APCs do not present exogenous antigens via MHC class I since exogenous antigens do not gain access to the cytosolic compartment. Therefore, exogenous antigens usually do not prime CTL responses in vivo. This segregation of exogenous antigens from the class I pathway is important to prevent CTL from killing healthy cells that have been exposed to foreign antigens but are not infected [26] . However, there are several exceptions to this rule, reflecting the ability of the exogenous antigens to be delivered into the cytosolic compartments [13] [14] [15] [16] [17] . We have previously reported that antigens coupled to the surface of liposomes comprised of unsaturated fatty acid are presented to both CD4 + -and CD8 + T cells [23] . Confocal laser scanning microscopic analysis demonstrated that a portion of the liposome-coupled antigens were taken up and processed beyond the MHC class II compartment. In the present study, we confirmed that OVA coupled to oleoyl liposomes was processed at both the MHC class I and class II compartments ( Fig. 1 ). Flow cytometric analysis demonstrated that OVA coupled to oleoyl liposomes was incorporated more efficiently by macrophages than OVA coupled to stearoyl liposomes (Fig. 2) . Furthermore, OVA coupled to oleoyl liposomes was taken up by macrophages even in a 4uC environment, in which antigen entry could only occur via plasma membrane translocation. In general, antigen processing pathways largely depend on the route of antigen uptake, and liposomes with a certain lipid component are known to fuse with the plasma membrane [27] . The uptake of OVA coupled to oleoyl liposomes in a 4uC environment observed in the present study suggested that oleoyl liposome might fuse with the plasma membrane and thereby allow the liposome-coupled antigen direct access to the cytosol. The role of endocytosis in the uptake of the liposomal antigen was further examined by using specific inhibitors for antigen uptake (Fig. 3) . Cytochalasin B treatment of APCs prior to the addition of liposomal antigen in the culture had little effect. However, treatment of APCs with DMA significantly reduced the uptake of liposome-coupled OVA. Consequently, it was suggested that antigens coupled to oleoyl liposomes might be taken up by APCs via at least two pathways, penetration and pinocytosis. The analysis of intracellular pathways of antigens coupled to oleoyl liposomes using confocal laser scanning microscopy demonstrated that a portion of liposomal antigens taken up by APC were translocated to the lysosomal compartments expressing LAMP-1 (Fig. 4) , suggesting that the liposomal antigens processed at lysosomal compartment and beyond lysosomal compartment might be presented to CD4/ CD8 + T cells via MHC class II and class I, respectively. In agreement with the results of antigen uptake shown in Fig. 3 , the treatment of splenic CD11c + cells with DMA significantly reduced antigen presentation of liposomal antigens to both CD4 + -and CD8 + T cells as evaluated by T-cell activation (Fig. 5) . It was reported that pinocytosis and scavenger receptor-mediated endocytosis by APC facilitate antigen presentation to CD4 + T cells; by contrast, mannose receptor-mediated endocytosis by APC has been shown to facilitate antigen presentation to CD8 + T cells [28] . However, as described in Materials and Methods, the oleoyl liposomes used in the present study do not contain mannose. Thus, the data in the present study demonstrated that antigens coupled to oleoyl liposomes were internalized by APCs through both penetration and pinocytosis. The antigens coupled to the surface of oleoyl liposomes were processed at both MHC class I and class II compartments and presented to CD4 + -and CD8 + T cells. Although the detailed pathway leading to presentation to both CD4 + -and CD8 + T cells remains unclear, the observed behavior of antigens coupled to oleoyl liposome in APCs seems quite unique. Taken together, coupling of antigens to oleoyl liposome might potentially serve as a novel method to induce both humoral and cellular immunity. Mice CBF1 mice (8 weeks of age, female) were purchased from SLC (Shizuoka, Japan). All experiments were approved (No. 208021 and 209082) by an independent animal ethics committee at National Institute of Infectious Diseases, Tokyo, Japan. All phospholipids were provided by NOF Co. (Tokyo, Japan). Reagent grades of cholesterol were purchased from Wako Pure Chemical (Osaka, Japan). Ovalbumin (OVA, Grade VII) was purchased from Sigma-Aldrich. For the analysis of the processing of liposome-coupled OVA by macrophages, DQ-OVA, which exhibits green fluores- Figure 2 . Uptake of liposome-coupled OVA by macrophages. Alexa-labeled OVA was coupled to either stearoyl or oleoyl liposomes and added to the culture of cloned macrophages as described in Materials and Methods. Thirty minutes after the onset of the culture, macrophages were recovered and analyzed using flow cytometry. doi:10.1371/journal.pone.0015225.g002 Figure 1 . Confocal laser scanning microscopic analysis of macrophages co-cultured with DQ-OVA-liposome conjugates. A, DQ-OVA was coupled to either stearoyl or oleoyl liposomes and added to the culture of cloned macrophages expressing DM-DsRed (class II) or labeled with red fluorescein (class I), as described in Materials and Methods. Two hours after the onset of the culture, macrophages were recovered and analyzed using confocal laser scanning microscopy. These optically merged images are representative of most cells examined by confocal microscopy. Yellow, co-localization of green (DQ-OVA after proteolytic degradation) and red (macrophage DM or class I); cell only, macrophages without co-culture with DQ-OVA-coupled liposomes. B, the green-and yellow-color compartments in the immunofluorescent pictures were quantified by the image analysis software MetaMorph, as described in Materials and Methods. Ratios of the yellow to green compartments are shown. Data represent the mean values 6 SD of the images shown in Fig. 1A . Asterisk, significant (p,0.01) difference of samples. doi:10.1371/journal.pone.0015225.g001 cence upon proteolytic degradation, was purchased from Molecular Probes, Inc. Synthetic CpG ODN (5002: TCCAT-GACGTTCTTGATGTT) was purchased from Invitrogen and was phosphorothioate-protected to avoid nuclease-dependent degradation. OVA was labeled with fluorescence using an AlexaFluor 488 protein labeling kit (Invitrogen) according to the manufacturer's protocol. Liposomes consisting of two different kinds of lipid were used in this study. Liposomes consisting of saturated fatty acids were composed of distearoyl phosphatidylcholine, distearoyl phosphatidyl ethanolamine, distearoyl phosphatidyl glycerol acid, and cholesterol in a 4:3:2:7 molar ratio (stearoyl liposomes), and liposomes consisting of unsaturated fatty acids were composed of dioleoyl phosphatidylcholine, dioleoyl phosphatidyl ethanolamine, dioleoyl phosphatidyl glycerol acid, and cholesterol in a 4:3:2:7 molar ratio (oleoyl liposomes). The crude liposome solution was passed through a membrane filter (nucleopore polycarbonate filter, Coster) with a pore size of 0.2 mm. Liposomal conjugates with plain OVA, Alexa-labeled OVA, or DQ-OVA were prepared essentially in the same way as described previously [22] . Briefly, to a mixture of 90 mg of liposomes and 6 mg of OVA in 2.5 ml phosphate buffer (pH 7.2), 0.5 ml of 2.5% glutaraldehyde solution was added in dropwise fashion. The mixture was stirred gently for 1 h at 37uC, and then 0.5 ml of 3 M glycine-NaOH (pH 7.2) was added to block excess aldehyde groups. This was followed by incubation overnight at 4uC. The liposome-coupled OVA and uncoupled OVA in the resulting solution were separated using CL-4B column chromatography (Pharmacia). The amount of lipid in the liposomal fraction was measured using a phospholipid content assay kit (Wako Pure Chemical). The OVA-liposome solution was adjusted to 10 mg lipid/ml in PBS, sterile-filtered using a Millex-HA syringe filter unit (0.45 mm, Millipore), and kept at 4uC until use. For the measurement of OVA coupled to liposome, radiolabeled OVA (methyl-14 C; purchased from New England Nuclear) was mixed with cold OVA and used for coupling with liposome and for determining the calibration curve. The Mice were immunized subcutaneously (s.c.) with the OVAliposome conjugate at a dose of 1 mg lipid/100 ml/mouse in the presence of 5 mg/mouse CpG. Macrophage hybridoma clone 39, obtained from the fusion of splenic adherent cells from CKB mice and P388D1 [29] , was used. The DNA fragment coding the full-length H2-DMb2 [30] was amplified by PCR with two primers (59-ATGGCTGCACT-CTGGCTGCTGCTGCTGGT-39 and 59-GATGCCGTCCT- TCTGGGTAGGTGGATCC-39). The PCR product was cloned into the CMV promoter-driven expression plasmid pDsRedN1 (BD Clontech). This construct omitted the stop codon of H2-DMb2 and encoded the H2-DMb2 fused with DsRed. The cloned plasmid DNA was transfected to macrophage hybridoma clone 39 with Effectene transfection reagent (Qiagen) according to the manufacturer's protocol. During the transfection to clone 39, the medium containing cDNA and the transfection reagent was replaced with fresh medium after an 8-h transfection, and then clone 39 was cultured for 40 h. To obtain stable cell lines, clone 39 was passaged at 1:5 into RPMI 1640 containing 10% FCS with 50 mg/ml geneticin (G-418; Sigma-Aldrich). Cells showing the best fluorescence were selected using a FACS Vantage cell sorter (BD Bioscience). After cell sorting, clone 39 expressing DM-DsRed was cultured in RPMI 1640 containing 10% FCS with 200 mg/ml geneticin. To investigate the capture of OVA-liposome conjugates by macrophages, macrophage clone 39 was incubated for 30 min at 4uC or 37uC in the presence of fluorescence-labeled OVAliposome conjugates that contained a final concentration of 4 mg/ ml OVA. After the incubation, cells were washed with ice-cold PBS. In the case of using Alexa-labeled OVA-liposome conjugates, cells were then incubated with 1.2 mg/ml trypan-blue for 5 min at 4uC to block the fluorescence of Alexa-OVA attached to the cell surface. After the cells were washed, they were analyzed on a FACS Caliber flow cytometer (BD Bioscience). The histograms of fluorescence distribution were plotted as the number of cells versus fluorescence intensity on a logarithmic scale. To investigate the localization of OVA-liposome conjugates by macrophages, macrophage clone 39 or DM-DsRed-expressing cloned macrophage 39 was cultured for 18 h at 37uC on 8-hole heavy Teflon-coated slides (Bokusui Brown) and was then incubated with DQ-OVA-liposome conjugates, prepared using oleoyl or stearoyl liposomes, for 2 h at 37uC. The slides were then washed with MEM and fixed with 4% paraformaldehyde in PBS for 10 min at room temperature. After fixation, they were incubated for 10 min in 0.1 M glycine-HCl (pH 7.0) to block the remaining aldehyde residue. They were then washed two times in PBS. After washing, the slides were sealed with PBS:glycerin (1:9) and analyzed under an LSM510 confocal laser scanning microscope system (Zeiss). For analysis of co-localization of OVA and MHC class I, early endosomal antigen 1 (EEA1) or lysosomalassociated membrane protein-1 (LAMP-1) after blocking of the remaining aldehyde residue, cloned macrophage 39 was subsequently permeabilized with 0.05% saponin-TBS for 10 min at room temperature. After being washed twice with PBS, they were reacted with biotin-conjugated mouse anti-mouse H-2D d mAb (34-2-12, 10 mg/ml; BD Biosciences), goat anti-mouse EEA1 Figure 5 . IFN-c production by splenic CD4/CD8 + T cells of mice immunized with OVA after co-culture with CD11c + cells pulsed with OVA coupled to oleoyl liposomes. Splenic CD4/CD8 + T cells were taken from mice immunized with OVA and were cultured with CD11c + cells pulsed with OVA coupled to oleoyl liposomes with or without inhibitors as described in Materials and Methods. IFN-c production of T cells in the supernatants in the absence of inhibitors was normalized to 100%. Data represent the mean values 6 SD of triplicate culture. Asterisk, significant (p,0.01) difference as compared with the 'no inhibitor' group. doi:10.1371/journal.pone.0015225.g005 polyclonal antibody (N19, 1 mg/ml; Santa Cruz Biotechnology) or rat anti-mouse LAMP-1 monoclonal antibody (1D4B, 1 mg/ml; Santa Cruz Biotechnology) for 18 h at 4uC. After being washed three times with TBS, they were reacted with Alexa 546conjugated streptavidin (1:200 diluted; Invitorogen) to detect MHC class I, Alexa Fluor 568-labeled Ab (rabbit anti-goat IgG, 10 mg/ml; Invitrogen) to detect EEA1 or Alexa Fluor 568-labeled Ab (goat anti-rat IgG, 10 mg/ml; Invitrogen) to detect LAMP-1 for 4 h at room temperature. They were then washed two times in TBS. After the washing, the slides were sealed with PBS:glycerin (1:9) and analyzed under an LSM510 confocal laser scanning microscope system (Zeiss). Quantification of confocal image analysis was done by single cell identification using the image analysis software MetaMorph (Molecular Devices Co., Tokyo, Japan), and the relative fluorescence intensity of green, red, and yellow pixels was assessed. The relative fluorescence intensity of all individual colors was then expressed as percent of the total fluorescence intensity. p values were calculated by the Student's t test with two-tailed distribution and two-sample unequal variance parameters. In the case of inhibition studies, cloned macrophage 39 or CD11c + cells were incubated with indicated inhibitors 60 min before and throughout the antigen pulse. Cytochalasin B [31] and DMA [28] were purchased from Sigma. Preparation of CD11c + cells and CD4 + -and CD8 + T cells CD11c + spleen cells of naïve mice and CD4 + T and CD8 + T spleen cells of mice immunized with OVA-liposome conjugates were prepared with the magnetic cell sorter system MACS, according to the manufacturer's protocol using anti-CD11c, anti-CD4 and anti-CD8 antibody-coated microbeads (Miltenyi Biotec). Culture of CD4 + -and CD8 + T cells with CD11c + cells pulsed with OVA CD11c + cells were incubated with or without the indicated inhibitors for 60 min in a 24-well plate prior to the addition of OVA-liposome conjugates made using oleoyl liposomes. The final concentration of OVA-liposome added to the macrophage culture was 500 mg lipid/ml, which included 24 mg OVA. After 60 minutes' incubation, CD11c + cells were washed 3 times in ice-cold medium and 2610 5 cells were co-cultured with 5610 5 CD4 + T cells or CD8 + T cells, in a 48-well plate. A preliminary experiment showed that the optimal culture period in the above culture condition was 2 days for IFN-c production by CD4 + T cells and 5 days for IFN-c production by CD8 + T cells. After incubation in a CO 2 incubator for 2 or 5 days, the culture supernatants were collected and assayed for IFN-c. IFN-c in the culture supernatants was measured using the Biotrak mouse ELISA system (GE Healthcare). All test samples were assayed in duplicate, and the SD in each test was always ,5% of the mean value.

Therapeutic Vaccination in Chronic Hepatitis B: Preclinical Studies in the Woodchuck

Recommended treatment of chronic hepatitis B with interferon-α and/or nucleos(t)ide analogues does not lead to a satisfactory result. Induction of HBV-specific T cells by therapeutic vaccination or immunotherapies may be an innovative strategy to overcome virus persistence. Vaccination with commercially available HBV vaccines in patients did not result in effective control of HBV infection, suggesting that new formulations of therapeutic vaccines are needed. The woodchuck (Marmota monax) is a useful preclinical model for developing the new therapeutic approaches in chronic hepadnaviral infections. Several innovative approaches combining antiviral treatments with nucleos(t)ide analogues, DNA vaccines, and protein vaccines were tested in the woodchuck model. In this paper we summarize the available data concerning therapeutic immunization and gene therapy using recombinant viral vectors approaches in woodchucks, which show encouraging results. In addition, we present potential innovations in immunomodulatory strategies to be evaluated in this animal model.

World Health Organization estimates that about 2 billion people worldwide have been infected with hepatitis B virus (HBV). Since the introduction of preventive vaccination programs against hepatitis B in over 170 countries, the number of new infections is continuously decreasing. Despite the success of prophylactic vaccines, chronic HBV infection is still a global health problem. Over 360 million people are persistently infected with HBV, of whom 1 million die each year from HBV-associated liver cirrhosis or hepatocellular carcinoma (HCC). The outcome of HBV infection varies greatly from person to person. In most of the cases the infection is cleared spontaneously, however, 5%-10% of adults develop chronic infection. By contrast, 40%-90% of children which are born to HBV-infected mothers will progress to develop a persistent liver disease [1] . In the recent, years a marked progress has been made in the treatment of chronic hepatitis B. Currently, the two types of antiviral therapies are approved: treatment with pegylated interferon alpha 2a (PEG-IFNα) or nucleos(t)ide analogues, such as adefovir, entecavir (ETV), lamivudine, telbivudine, and tenofovir [2] [3] [4] [5] . However, the efficacy of those therapies in preventing liver cirrhosis and HCC is still limited. Treatment with PEG-IFNα leads to a sustained antiviral response in only one third of patients, regardless of combining the therapy with polymerase inhibitors. On the other hand, the treatment with nucleos(t)ide analogues significantly suppresses HBV replication that leads to a decrease of necroinflammation in the liver. However, those antivirals cannot completely eradicate the virus. After withdrawal of the drug, the rebound of viremia is observed in the majority of patients. Furthermore, the long-term treatment is subsequently associated with the appearance of drugresistant HBV strains that is often the cause of the therapy failure [6, 7] . Therefore, the new approaches in treating chronic hepatitis B are urgently needed. It is well documented that an appropriate adaptive immune response is required to efficiently control the HBV infection. T cell-mediated immune response directed against 2 Hepatitis Research and Treatment hepatitis B virus antigens is crucial for resolution of the infection [8] [9] [10] [11] [12] . HBV-specific CD8 + T cells are able to clear HBV-infected hepatocytes by secretion of Th1 antiviral cytokines, such as interferons (IFNs) and tumor necrosis factor alpha (TNFα), and direct cytotoxic mechanisms (perforin/granzyme, ligand-ligand induced cell death, e.g., Fas-Fas-L) [12] [13] [14] [15] [16] . An early, vigorous, polyclonal, and multispecific cellular immune response against the viral proteins is associated with the clearance of hepatitis B in acutely-infected patients. In contrast, chronic HBV carriers demonstrate weak, transient, or often undetectable CD8 + T cell response that correlates with HBV persistence [17] [18] [19] [20] [21] . Humoral immune response, especially neutralizing antienvelope antibodies, play a key role in preventing HBV spread to noninfected hepatocytes [20, 22] . Recent studies indicate that several mechanisms may be involved in the loss of the function of HBV-specific T cells during chronic hepatitis B. It was shown that high-level viremia negatively influences the virus-specific immune responses. High viral replication in the liver with viral load higher than 10 7 copies/mL is correlating with hyporesponsiveness of virus-specific CD8 + T cells in patients with chronic hepatitis B [23] . Moreover, the prolonged exposure to viral antigens occurring during the chronic viral infections can trigger the T cells to become tolerant and prone to apoptosis. The interaction between programmed death 1 (PD-1) receptor and its ligand PD-L1 (also known as B7-H1) plays an important role to prevent an overreaction of the immune system [24] . Recent studies revealed that inhibitory molecules such as PD-1 and CTLA-4 are markedly upregulated on virus-specific T cells, resulting in exhaustion (e.g., lack of IFNγ production and proliferation) [25] . Simultaneously, this mechanism can contribute to the development of the chronic infection by impairment of the effective antiviral response. This hypothesis was previously proven for hepatitis C virus (HCV) [26, 27] and human immunodeficiency virus (HIV) infection in humans [28] [29] [30] , as well as lymphocytic choriomeningitis virus (LCMV) infection in mice [31, 32] , and more recently for HBV [33, 34] . Furthermore, several studies imply that functional defects of antigen presenting cells (APCs), mainly dendritic cells (DCs), may contribute to the impaired T cell response in chronic hepatitis B patients [35] [36] [37] [38] [39] [40] [41] . In vitro studies showed that DCs isolated from HBV chronic carriers produce lower amount of antiviral cytokines, such as type I interferons and TNFα, in comparison to healthy controls [35, 36] . In addition, those DCs are less efficient in T cell activation and stimulation of T cell proliferation [35, [39] [40] [41] . The novel report demonstrated that myeloid DCs from chronic HBV patients express increased level of inhibitory PD-L1 molecule and therefore may down regulate functions of HBV-specific T cells [39] . Several investigations underline the significance of CD4 + CD25 + regulatory T cells in pathogenesis of persistent viral infections [42] . In HCVand HIV-infected patients, it was shown that regulatory T cells may downregulate HCV-and HIV-specific CD8 + and therefore influence the disease progression [43] [44] [45] . The role of regulatory T cells in HBV infection is still not clear. Nevertheless, the increased numbers of CD4 + CD25 + regulatory T cells were detected in the blood and the liver of patients with chronic severe hepatitis B [46] . In addition, the liver itself is an organ with tolerogenic properties that might contribute to the immunological tolerance during chronic HBV infection [47, 48] . Finally, viruses developed the strategies to efficiently evade the host immune response resulting in persistent infections. Viral immune escape due to the mutation of CD4 + , CD8 + , and B cell epitopes in a given HLA background have been observed in patients infected with HIV, HCV, and HBV [49] [50] [51] [52] [53] [54] . Several studies demonstrate that the treatment with lamivudine alone, or in combination with interleukin-12 (IL-12), result in the restoration of the HBV-specific CD4 + and CD8 + immune response in chronic HBV-infected individuals. However, the therapeutic effect was not sustained in those patients [55] [56] [57] . Over 20 years, continuous efforts have been undertaken to develop a therapeutic vaccine for chronic hepatitis B to enhance the virus-specific immune responses and overcome persistent HBV infection [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] . Numerous clinical trials of therapeutic immunization exploited the conventional prophylactic hepatitis B surface antigen-(HBsAg-) based protein vaccines. These studies demonstrated reductions in viremia, HBeAg/anti-HBe seroconversion, and HBV-specific T cell responses in some patients. However, the anti-viral effect was only transient and did not lead to an effective control of the HBV [58] [59] [60] [61] [62] [63] [64] [65] . Combination of the HBsAg protein vaccines with antiviral treatment with lamivudine did not lead to a satisfactory improvement of the therapies [66] [67] [68] . The strategies designed to specifically stimulate HBVspecific T cell responses were also not successful [69] [70] [71] . The lipopeptide-based vaccine containing a single cytotoxic T lymphocyte (CTL) epitope derived from HBV nucleocapsid was able to induce a vigorous primary HBV-specific T cell response in naïve subjects [76] . However, in HBV chronic carriers, the vaccine initiated only poor CTL activity and had no effect on viremia or HBeAg/anti-HBe seroconversion [69] . The DNA vaccine expressing small and middle envelope proteins proved to elicit the HBV-specific cellular immune response in chronic HBV carriers, however, this effect was only transient [70] . Yang et al. presented the novel DNA vaccine for treatment of chronic hepatitis and combined the immunizations with lamivudine treatment [71] . The multigene vaccine contains five different plasmids encoding most of HBV antigens and human IL-12 gene as a genetic adjuvant. The combination therapy led to sustained antiviral response in 6 out of 12 HBV chronically infected patients. The responders were able to clear HBeAg and had undetectable viral load at the end of a 52-week follow-up. Those effects were correlating with a detectable T cell response to at least one of the HBV antigens. [71] . Nevertheless, further studies are needed to evaluate this strategy on a larger cohort of HBV chronic carriers. The therapeutic vaccine-based HBsAg complexed with human anti-HBs was proposed by the group of Wen et al. 3 [77] . Immunogenic complexes (ICs) stimulate robust T cell responses by increasing uptake of HBsAg through Fc receptors on APCs and, therefore, modulate HBsAg processing and presentation. It was demonstrated that this vaccine administered to HBeAg-positive patients led to decrease of HBV DNA in serum, HBeAg seroconversion, and development of anti-HBs in part of the subjects [78] . Currently, the IC-based vaccine is the only one that entered phase III of clinical trials in chronic hepatitis B patients [79] . Even though the IC-based vaccine led to antiviral effect, clearance of HBV was not observed in treated patients. It seems that the vaccine alone is not sufficient to achieve the full control over HBV. Therefore, some steps have been undertaken to combine the IC-based vaccine with nucleos(t)ide analogues treatment, (Wen et al., personal communication) . The ongoing clinical trial will show whether IC are effective as a therapeutic vaccine in chronic hepatitis B. Over the years, various animal models, including chimpanzees, woodchucks, ducks, and HBV transgenic mice, were established for development and evaluation of novel therapeutic strategies. Considering the cost, ethical reasons, and available amount, HBV transgenic mice are the most widely used models. Studies using HBV transgenic mouse models demonstrated that DNA immunization with the expression plasmids encoding different HBV proteins could induce HBV-specific antibodies and stimulate CTL responses. However, the functionality of HBV-specific CTLs induced in transgenic mice may be not fully developed [80] [81] [82] . Improvement of DNA vaccination regimen [83] and blockade of PD-1/PD-L1 interaction [34, 84] could enhance functional T cell responses and lead to inhibition of viral replication in vivo without causing hepatitis. Apart from the DNA immunizations, the other therapeutic approaches including administration of Toll-like receptor (TLR) ligands, HBV-specific siRNA, and direct activation of APCs were evaluated in HBV transgenic mice [85] [86] [87] . Those strategies were able to effectively reduce the HBV replication, and are currently under investigation as combined therapies. Nevertheless, this model has a significant limitation. As the HBV genome is inserted into the mouse chromosome, full HBV life cycle does not take place in the transgenic mice and no liver inflammation can be observed [88] . Thus, the animal models with naturally occurring hepadnaviral infection are required for the longterm evaluation of the therapeutic effect. In comparison to chimpanzees, woodchucks are easily available and affordable. In this paper we would like to introduce woodchucks as a useful preclinical model for designing of the new therapeutic vaccines in chronic hepadnaviral infections. We will summarize the available data concerning therapeutic immunization approaches in woodchucks and present potential innovations in immunomodulatory strategies that yet to be evaluated on this animal model. The Eastern woodchuck (Marmota monax) is naturally infected by woodchuck hepatitis virus (WHV). WHV was discovered in 1978 as a virus closely related to HBV [89] and classified as a member of Hepadnaviridae family. WHV and HBV show a marked similarity in the virion structure, genomic organization, and the mechanism of replication, but differ in several aspects, for example, regulation of transcription (Table 1 ) [90] . WHV causes acute self-limiting and chronic infection similar to HBV infection in the pathogenesis and profiles of the virus-specific immune response [91] . This feature of the woodchuck model makes it so significant for investigation of the new therapeutic approaches in chronic hepatitis B. Experimental infection of neonates or adult woodchucks with WHV reflects the outcome of HBV infection in humans. In adult woodchucks infection with WHV usually leads to the resolution of infection and only 5%-10% of animals will develop the chronic hepatitis. The exposure of woodchuck, neonates to WHV results in development of chronic WHV infection in 60%-75% of the cases [92] . The continuous replication of WHV in the liver during the chronic infection is nearly always associated with development of HCC in the woodchucks [93, 94] . After diagnosis of HCC the survival prognosis of the animals is estimated on about 6 months, like in humans. The common features of HBV-and WHVinduced carcinogenesis give the opportunity to examine the new anti-HCC therapies in the woodchucks [95] . For many years, the studies on immunopathogenesis of WHV infection in woodchucks were restricted to determination of humoral immune responses [96] . The lack of appropriate methods to evaluate antigen-specific T cell responses was the serious limitation of this model. Proliferation assay for peripheral blood mononuclear cells (PBMCs) based on incorporation of [ 3 H]-thymidine by cellular DNA, routinely used for human and mouse system, has been ineffective in the woodchuck PBMCs [97, 98] . The failure of this approach is consistent with the fact that woodchuck lymphocytes do not express the thymidine kinase gene (Menne et al., unpublished results). This obstacle had been overcome by usage of the alternative radioactively labeled nucleotide 2[ 3 H]-adenine [72] . Development of 2[ 3 H]-adenine-based proliferation assay enabled to detect the T-helper lymphocyte responses after stimulation of woodchuck PBMCs with WHV core, surface and X antigens (WHcAg, WHsAg, and WHxAg, resp.) [72, 99] . In addition, using the 2[ 3 H]-adenine-based proliferation assay in PBMCs from acutely infected animals, several T-helper epitopes within WHcAg [72] and WHsAg were identified [Menne et al., unpublished results] . Recently established, a novel CD107a degranulation assay for woodchuck PBMCs and splenocytes made a significant breakthrough in studying pathogenesis of hapadnaviral infections in the woodchuck model [73] . Several studies demonstrated that detection of CD107a, as a degranulation marker, is a suitable method for determination of antigen 4 Hepatitis Research and Treatment Surface glycoproteins (large-L, medium-M, small-S), core protein, "x" protein, "e" antigen, DNA polymerase with reverse-transcriptase activity [22, 105] The corresponding proteins [91] Replication strategy Replication of HBV DNA occurs by reverse transcription of an RNA intermediate within cytoplasmic nucleocapsids [22] The same mechanism [97] Genetic diversity 8 major genotypes [105] 1 major genotype (minor sequence differences) [91] Integration into host chromosome Yes [22] Yes, often close to N-myc oncogene region [106] Clinical course of infection were characterized ( Figure 1) . In contrast to self-limiting infection, WHV chronic carriers demonstrate weak or no virus-specific T cell responses against the identified epitopes [72, 73, 99] . The establishment of the assays for monitoring of cellular immune response in woodchucks is of great importance for a reliable evaluation of therapeutic and immunomodulatory strategies for treatment of chronic hepatitis B in the woodchuck model [96, 102, 103 ]. Recently described advancements in the characterization and monitoring of the woodchuck immune system during the WHV infection, made this animal model particularly useful for development of the immunomodulatory approaches in chronic hepatitis B. The natural occurrence of chronic WHV infection in woodchucks, that is closely related to HBV infection in humans, allows to evaluate the potentially new therapeutic strategies directly in chronic WHV carriers. Up to date, several studies of diverse therapeutic vaccinations have been carried out in woodchucks ( Table 2) . The pioneer investigations based on therapeutic vaccines based on WHV core [96] or surface antigens in combination with a helper peptide FIS [120] , or with potent Th1 adjuvants like monophosphoryl lipid A [121] did not lead to satisfactory results. Those experiments proved that vaccinations could induce specific B-and/or T cell responses in chronic WHV carriers. However, this alone was not sufficient to achieve the control of virus replication. It is assumed that high level viremia, during the chronic hepatitis B, can inhibit the therapeutic effect of the vaccination. Treatment of chronic HBV patients with lamivudine could transiently restore HBV-specific T cell immune response [55, 56] . Therefore, reduction of viral load by the nucleos(t)ide analogues pretreatment might support the efficacy of immunization to enhance the virus-specific immune responses. This hypothesis was tested in three experimental trials of the combination therapies in chronic WHV carriers. The first study performed by Hervás-Stubbs et al. was based on lamivudine therapy [108] . Five chronically WHVinfected woodchucks were treated orally with the drug for 23 weeks. At week 10, after decline of WHV DNA by 3-5 logs, three animals were vaccinated with 3 doses of serum-purified WHsAg combined with T-helper FIS peptide derived from sperm whale myoglobin. The vaccination induced T-helper responses against WHV antigens, shifting the cytokine profile from Th2 to Th0/Th1. However, no beneficial effect on WHV viral load and WHsAg levels was observed in comparison to nonimmunized animals. After withdrawal of the lamivudine treatment the values of viremia returned to the pre-treatment levels. The second trial evaluated the therapy with a very potent antiviral drug: clevudine (previously called L-FMAU) combined with a WHsAg-based immunization [74, 109, 110] . A large cohort of thirty 1-2-year-old chronically WHVinfected woodchucks was enrolled in the study. Half of the animals were orally treated with clevudine (10 mg/kg/day) for 32 weeks; the other 15 woodchucks received placebo. After withdrawal of clevudine treatment, 8 animals from each group were vaccinated with the four doses of formalin inactivated alum-adsorbed WHsAg and 7 were injected with the saline as a control. Combination of the drug and vaccine therapy resulted in marked reductions WHV DNA (6-8 logs) and WHsAg in serum during the 60-week monitoring period, in contrast to the vaccine only and placebo groups, where both markers remained at high levels. Combination therapy did not enhanced anti-WHs responses beyond those measured for vaccine alone. However, treatment with clevudine and vaccine together led to more sustained and robust lymphoproliferative responses to WHsAg and additionally to WHcAg, WHeAg, and WHxAg. Moreover, combination therapy delayed the onset of the liver disease and prevented HCC development in up to 38% of treated chronic WHV carriers in the long-term follow-up study [111] . Recently, a novel therapeutic approach for treatment of chronic hepatitis B in a woodchuck model was described. The therapy combined the antiviral treatment with immunization with plasmid DNA and antigen-antibody immunogenic complex vaccines together [112] . DNA vaccines are considered to stimulate both humoral and cellular immune response, polarizing T cells in the direction of Th1 response [122] . Immunization of the naïve woodchucks with the plasmids encoding WHV core and preS2/S genes (pWHcIm and pWHsIm, resp.) induced the lymphoproliferative responses against the antigens and provided a protection against WHV challenge [123] . In addition, the DNA vaccine expressing HBsAg proved to elicit the vigorous T cell responses in chronic HBV carriers, however, this effect was only transient [70] . The HBsAg/anti-HBs IC vaccine is currently under the investigation in chronic HBV patients [77] [78] [79] . To evaluate the efficacy of previously mentioned immunotherapy in woodchucks, firstly 10 chronic WHV carriers were treated with 15 mg of lamivudine, daily for 21 weeks. At week 10, four animals were pretreated with cardiotoxin and then received three immunizations with DNA vaccine containing three plasmids expressing WHsAg, WHcAg, and woodchuck IFNγ (pWHsIm, pWHcIm and pWIFN, resp.). Simultaneously, the other four woodchucks received three doses of the combination of DNA vaccine and WHsAg/anti-WHs immunogenic complex. Two chronic WHV carriers served as lamivudine monotherapy control. Lamivudine treatment resulted in only a slight decrease of WHV DNA levels in the woodchucks serum (0,7 and 0,32 log, resp.). Surprisingly, the DNA vaccination did not lead to any additional therapeutic effect beyond that observed for lamivudine treatment alone. In contrast, the triple combination of antiviral treatment, plasmid DNA encoding WHcAg, WHsAg, and wIFNγ and IC vaccines was able to decrease WHV viral load up to 2,9 log and the serum WHsAg up to 92%. Moreover, three of the four treated animals developed anti-WHs antibodies. Nevertheless, these effects were not sustained and all parameters reached the baseline levels shortly after withdrawal of lamivudine treatment. In addition, the vaccination did not induce WHVspecific T cell responses in the majority of woodchucks, even in animals that exhibited virological responses. Significant Hepatitis Research and Treatment 7 lymphoproliferative responses against WHV antigens were detected only in one animal after three immunizations with DNA vaccine [112] . The study demonstrated the benefit of using the combinatory therapy in chronically WHVinfected woodchucks. However, the transient therapeutic effects, suggest that this strategy needs further optimization. The results from the previous studies clearly confirm the poor efficacy of the lamivudine therapy in woodchucks [108, 112, 124] . A new strategy evaluated the potency of an entecavir treatment and increased number of immunizations [Lu et al., unpublished results] . Chronically WHV-infected woodchucks were pretreated with the entecavir for 21 weeks; 10 weeks in a daily and 11 weeks in a weekly manner. During the weekly administration of the drug, one group of animals received 6 immunizations with two-plasmid DNA vaccine (pWHsIm and pWHcIm),the second group received combination of DNA vaccine together with purified WHV core and surface antigens, and the third group remained untreated. The entecavir therapy resulted in rapid and significant decrease of the viral load and WHsAg levels in serum of the animals. The effect was especially pronounced in animals that additionally received vaccines. In woodchucks treated only with entecavir, the increase of viremia was observed already during the weekly administration or immediately after withdrawal of the drug. By contrast, in both groups of animals, that were immunized with DNA or DNA/proteins vaccines, the delay before the rebound of WHV replication was significantly prolonged. In addition, entecavir treatment was effective to suppress WHV replication and enhanced the induction of WHV-specific T cell responses. An increased CTL activity was detected in individual woodchucks after DNA or DNA/proteins vaccinations. Moreover, two animals completely eliminated the virus from the blood and were WHV DNA negative in the liver [Lu et al., unpublished results] . Altogether, the results obtained in the woodchuck model concerning combination of nucleot(s)ide therapy and immunization proved the synergistic effect of both therapeutical approaches. The therapeutic effects observed during such therapies were significantly increased and prolonged in comparison to the monotherapy alone. In addition, those therapeutic approaches could stimulate the WHV-specific T cell responses, usually impaired in WHV chronic carriers [72, 73] . A combination of antiviral treatment and vaccination is required for the improvement of virus specific T cell responses. Designing of the future therapeutic approaches should include pretreatment with the potent antiviral drugs, such as entecavir or clevudine, that proved their efficacy in the woodchuck model. Previous results from therapeutic immunization trials on woodchucks, chimpanzees, and humans indicate that the licensed vaccines are not able to boost a functional antiviral T cell response. There is a need to use more potent strategies. Vaccines based on recombinant viruses have gained a great interest because of their ability to stimulate robust humoral and cellular immune responses. Viral vectors were investigated as prophylactic and therapeutic vaccines against many human pathogens such as measles virus, herpes simplex virus (HSV), human papillomavirus (HPV), HIV, and rabies [126] [127] [128] [129] [130] . However, the utility of those recombinant vaccines in the treatment of chronic hepatitis B was not yet evaluated. Preliminary results obtained from the study in chronically HBV-infected chimpanzees immunized with retroviral vector, based on Moloney murine leukemia virus, encoding HBcAg suggest that further investigation of viral-vector based vaccines should be taken into consideration [131] . In the experiment, one of the three therapeutically immunized chronic carrier chimpanzees cleared the virus and showed HBeAg seroconversion. Significant ALT elevations observed in this animal implicate restoration of HBV-specific cytotoxic and humoral responses without causing fulminant hepatitis. Moreover, the other two chimpanzees demonstrated high anti-HBe titers after the therapy and one of them HBcAgspecific CTLs [131] . This study demonstrates not only the benefit of using the recombinant viral-vectors for treatment of chronic HBV infection in primate model, but also the possible advantage of using core antigen-based therapeutic vaccines. Even though the retroviral vector vaccination was well tolerated in the chimpanzees, several clinical trials suggest that gene therapy with traditional retroviral vectors can lead to oncogenesis [132, 133] . Therefore, the usage of another recombinant virus as a carrier of the proteins could be beneficial. Recombinant adenoviruses have been one of the intensively investigated viral vectors for therapeutic purposes. Development of the novel methods for manipulating of the viral genome resulted in the three generations of the recombinant adenoviruses and with increasing capacity [125] (Figure 2 ). Several trials imply the usefulness of those vectors in gene therapy of genetic diseases and cancer [134] [135] [136] [137] . For many years, the first generation replication-deficient E1 or E1/E3-deleted adenoviral vectors have been explored as the vaccine carriers in prevention of the infectious diseases [138] . Adenoviral vectors have several advantages that can be beneficial for potent therapeutic vaccines. First of all, adenoviruses are relatively susceptible for genetic modifications and can be easily produced in high titers. After transduction of the cells, adenoviral genome is not integrated into the host DNA and stays in the episomal form. As a result, the risk of the possible activation of the cellular oncogenes is minimal. Adenovirusbased vaccines proved to elicit a vigorous and sustained humoral and T cell responses to the incorporated antigen that is considered to be crucial in clearance of persistent viral diseases [127, [139] [140] [141] . The benefit of adenoviral vectors as a vaccine carrier is not only limited to stable delivery of proteins of interest. Several findings on additional immunostimulatory effects, for example, induction of the innate immune response, that originate from the nature of adenoviruses itself, may enhance the vaccine efficacy. Capsid of adenoviruses demonstrates immunostimulatory properties, that is why the coadministration of the adjuvant is usually unnecessary. Those vectors can directly transduce DCs causing their maturation and upregulation of MHC and costimulatory molecules on their surface, thus lead to enhanced antigen presentation. Moreover, it was shown that AdV-transduced DCs are secreting antiviral cytokines, such as IFNα, TNFα, and IL-6 [142] . Interleukin-6 is one of the most important factors that suppress the function of the regulatory T cells [143, 144] . Nevertheless, modified adenoviruses apart from the abovementioned advantages have one serious limitation. Thus far, vectors that were comprehensively examined as the vaccines have been based on the human adenovirus serotype 5 (Ad5) [127] . This serotype is the most common in the human population. Anti-Ad5 neutralizing antibodies are detectable in 45%-90% of adults [145] . The preexisting immunity directed against Ad5 is considered as a main reason of failure in the phase I clinical trial of a protective HIV-1 vaccine. STEP study guided by Merck pharmaceutical concern, based on 3-dose regimen of a trivalent Ad5 vaccine, suggested that the immunization might increase the risk of HIV-1 infection in the subjects with high neutralizing anti-Ad5 titers [146] [147] [148] . Moreover, even single immunization may induce immunity to the vector in seronegative individuals. The negative effect of the pre-existing or Ad5-induced immunity against the vaccine, mostly when the therapy requires multiple dosages, may be overcome by heterologous prime-boost regimen. The utility of the rare human serotypes (e.g., serotype 35) [149, 150] or recombinant adenoviruses of nonhuman origin has been recently tested [151] . In particular, subsequent priming immunizations with plasmid DNA vaccine followed by a booster vaccination with AdV seem to be a very promising strategy. DNA primeadenovirus boost regimen proved to induce more robust and potent immune response in comparison to plasmid DNA alone and provided protection against the pathogen challenge in several animal models of infectious diseases [149, [152] [153] [154] . Furthermore, a clinical trial of multiclade HIV-1 DNA plasmid-Ad5 boost vaccine, HIVuninfected individuals demonstrated high immunogenicity even in the presence of high anti-Ad5 antibody titer. In addition, the vaccine proved to be well tolerated in the participants of the study [155] . Several studies indicate that the transgene expression level can be increased from adenoviral vectors by the presence or insertion of an intron sequences [156] [157] [158] . Therefore, we constructed the new recombinant adenoviruses serotype 5 and 35 encoding WHV core protein and containing an intron between promoter and WHcAg gene sequences. Preliminary experiments showed that vaccination with the AdVs containing the intron sequences led to induction of robust cellular and humoral immune responses in mice. Moreover, immunization of the mice in DNA prime-AdV boost manner, using improved vectors, resulted in more vigorous and multispecific T cell responses in comparison to immunization with plasmid DNA alone [Kosinska et al., unpublished results]. Immunization of chronically WHV-infected woodchucks with plasmid DNA vaccine in combination with entecavir treatment showed a marked therapeutic effect. Addition of the recombinant adenoviruses to this regimen could be a new, more potent approach in treatment of chronic hepatitis B. We will apply DNA prime-AdV boost approach in WHV chronically infected woodchucks in combination with nucleos(t)ide analogs and evaluate its therapeutic potential. Over the last 20 years, modified adenoviruses have been extensively studied as a vehicle for gene delivery to the liver, because of their high transfection efficiency and their natural tropism for hepatocytes [159, 160] . Moreover, the development of the third generation of adenoviral vectors that lack all viral coding sequences (e.g., helper-dependent adenoviral vectors), resulted in their increased capacity and minimized immunogenicity of the vector allowing longterm transgene expression [161] . High cloning capacity of those vectors enables usage of inducible or tissue-specific promoters and coexpression of multiple therapeutic or immunomodulatory genes [162] . So far, several trials of virus-mediated gene therapy for treatment of chronic hepatitis and HCC were performed in chronically WHV-infected woodchucks and in cell culture systems. Those strategies were mainly based on delivery of antiviral cytokines, such as IFNα, IFNγ, IL-12 by recombinant adenoviruses, to reduce viral replication or modulate the immune response ( Table 3) . Transduction of primary woodchuck hepatocytes from chronic WHV carriers with helper-dependent AdV encoding woodchuck IFNα (wIFNα) resulted in the reduction of WHV proteins expression in vitro [169] . In vivo studies on chronically WHV-infected woodchucks, demonstrated that a single injection of 1 × 10 12 vp of this vector into the liver's portal vein could inhibit WHV replication by 1 log up to 11 weeks after the treatment [163] . The same approach with helper-dependent AdV expressing woodchuck IFNγ (wIFNγ) did not show any antiviral effect, even though the transduction led to the production of biologically active interferon [163] . Another study combined intravenous delivery wIFNγ by recombinant adenoviral vector with nucleos(t)ide analogues therapy. Chronic WHV carriers were treated with clevudine and emtricitabine (FTC), together, for 8 weeks and after the initial drop in viral load one group of animals received additionally two i.v. injections of 3 × 10 10 PFU of Ad-IFNγ. Delivery of wIFNγ induced inflammation, caused by T cell infiltration, and increased hepatocyte turnover. However, this effect did not induce additional antiviral outcome in comparison clevudine/emtricitabine biotherapy alone [164] . Similarly, poor therapeutic effect was observed for gene therapy based on both wIFNγ and wTNFα. Intravenous injection of those recombinant adenoviruses during clevudine treatment led to decrease of replicative intermediates of WHV DNA in the liver, beyond what could be achieved by clevudine alone. Nevertheless, 6 weeks after injection there was no significant difference between the groups of WHV carriers receiving AdV expressing the cytokines or beta-galactosidase as a control [165] . The benefits of using the immunomodulatory genes in this study are difficult to assess, since it was reported that adenovirus infection alone is sufficient to transiently suppress the WHV replication in chronically infected woodchucks [170] . The lack of therapeutic effect by direct delivery of IFNγ is consistent with in vitro data obtained from persistentlyinfected woodchuck primary hepatocytes. Treatment of the cells with wIFNγ, even in the presence of wTNFα, was not able to inhibit the WHV replication. Moreover, high concentration of those cytokines resulted in the loss of the cells during the culture [171] . This observation underlines the cytotoxic effect of Th1 cytokines on the woodchuck hepatocytes. Rapid downregulation of the IFNγ expression, after transduction of the liver cells with viral vector, could be one of the mechanisms to protect the organism from the potential toxicity of this cytokine in vivo [163] . In addition, several reports indicates that the level of wIFNγ and wTNFα is higher in the liver of chronic WHV carriers in comparison to naïve animals [172, 173] . Therefore, continuous presence of inflammatory cytokines in the liver during the chronic WHV infection could result in hyporesponsiveness of hepatocytes to such a therapy. The novel strategy to treat chronic WHV hepatitis is based on adenovirus-mediated delivery of murine IL-12 (mIL-12) gene into hepatocytes [166] . Interleukin-12 is a proinflammatory cytokine produced naturally by antigen presenting cells. IL-12 stimulates production of IFNγ and TNFα by T and natural killer (NK) cells and enhances their cytotoxic activity [174] . In the study, mIL-12 gene expression could be regulated by inducible promoter that was responding to progesterone antagonist RU486. Eight chronic WHV carriers received single dose of 2 × 10 10 i.u. of AdV expressing mIL-12 (HC-Ad/RUmIL-12) by intrahepatic injection at laparotomy. Two weeks after, the expression of mIL-12 was induced by the administration of RU486. The IL-12 treatment resulted in intense and sustained suppression of WHV replication in the liver as well as decreased viral loads in the serum. This effect, however, was visible only in the animals with basal viremia lower than 10 10 WHV copies per milliliter of serum. Animals, which responded to the therapy, developed a vigorous T cell response to WHcAg, measured by woodchuck IL-2 production, and demonstrated WHeAg and WHsAg seroconversion. Moreover, the FoxP3 levels in the livers of those animals were decreased, while in nonresponder woodchucks FoxP3 values were significantly upregulated [166] . This finding suggests that the intrahepatic expression of IL-12 may inhibit the regulatory T cells in the liver during the chronic WHV infection. Indirect induction of inflammatory cytokines, such as IFNγ and TNFα by IL-12, seems to be a more efficient strategy in breaking the tolerance to virus antigens than direct delivery of those cytokines. It suggests that probably additional events occur in the liver after AdV-mediated IL-12 transfer that supports the antiviral effects of this therapy. Adenoviral delivery of genes for cytokines and other immunomodulators is widely used in cancer therapy in the animal tumor models as well as in patients [137, [175] [176] [177] [178] . The T cells play an important role not only in defense against the pathogens, but also in antitumor immunity and inhibition of the tumor growth. Interleukin-12 inhibits the angiogenesis and induces a potent antitumoral immune response by stimulation of IFNγ secretion. Therefore, IL-12 is a promising candidate for cancer gene therapy [179] [180] [181] [182] [183] . Strategy based on recombinant adenoviruses expressing IL-12 demonstrated antitumor effect in the murine models with transplantable HCC [184, 185] and was also evaluated in woodchucks [168] . In the study, large (2-5 cm) intrahepatic tumors of 5 woodchucks were injected with a single dose of 1 × 10 9 PFU AdV expressing IL-12 and B7.1 molecule (AdIL-12/B7.1). The B7.1 molecule (also known as a CD80) is naturally expressed on the professional APCs and provides the synergistic effect in the tumor regression [181, 186, 187] . In 4 out of 5 animals, AdIL-12/B7.1 was delivered by laparotomy into the three HCC nodules and three nodules were injected with a vector expressing GFP as a control. Animals were sacrificed 7-14 days later and the tumor volumes were assessed. On average, treated tumors showed an 80% reduction in the volume whereas the size of the AdGFP-injected nodules increased. Remission of the tumors was associated with CD4 + and CD8 + T cell infiltration into the tumor tissue and increased local IFNγ levels after AdIL-12/B7.1 injection. One of the treated woodchucks received the intratumoral injection by magnetic resonance imaging (MRI) guidance and was monitored for 7 weeks. During this period the tumor size decreased from 8,6 cm 3 to 0,5 cm 3 [168] . This observation shows that administration of AdIL-12/B7.1 during MRI guidance, with therapeutic effect similar to laparotomy, could prevent the animals from harmful consequences of the surgery. The study proved that the gene therapy based on IL-12 leads may be a promising strategy to treat HCC. By contrast, treatment with AdV encoding herpes simplex thymidine kinase combined with gancclovir administration did not lead to reduction in the tumor size [167] . Nevertheless, the short time of monitoring during the study makes it difficult to evaluate the prolonged antitumoral effect of this approach. A recent study presents gene therapy with semliki forest viral vector expressing high levels of murine IL-12 (SFV-enhIL-12) on remission of HCC in chronically WHV-infected woodchucks. In the research, the vector was delivered by surgery into multiple sites of HCC tumors in the liver [75] . A total of nine woodchucks were enrolled in the experiment. Six of the woodchucks, two animals each, received different doses of SFV-enhIL-12: 3 × 10 9 vp, 6 × 10 9 vp, and 1, 2 × 10 10 vp, and three animals served as a control and received saline injections. The tumor size was monitored by ultrasound examination for 23 to 24 weeks. In all woodchucks, reduction in tumor volume was observed, however, this effect was transient and dose dependent. Animals treated with the highest dose of SFV-enhIL-12 showed the most spectacular reduction of the tumor size 71% and 80%. Nevertheless, the tumors started to grow between 6 and 14 weeks after the treatment. The antitumoral effect was associated with the induction of the immune response towards the tumor antigens, demonstrated by T cell proliferation assay, upregulation of leukocyte markers expression, and cytokine production, such as IFNγ, TNFα, IL-6, and IL-12. In addition, the therapy resulted in transient induction of lymphoproliferative responses against WHcAg and WHsAg and led to short-term reduction in WHV viral load [75] . The results presented here indicate that viral-mediated gene therapy in treatment of chronic hepatitis B and HCC needs further optimization. However, treatment of the woodchucks with viral vectors allowed to achieve a long-lasting expression of the cytokines and their higher concentration preferably in the liver. Therefore, this strategy is proven to be more effective than an approach based on using of the soluble cytokines. In addition, adenovirusmediated gene transfer is proven to be a safe and a welltolerated strategy in the woodchucks. The current progress indicates the feasibility of therapeutic approaches for treatment of chronic HBV infection. There is a general agreement that a combination of antiviral treatment and immunomodulation is essential to achieve a sustained control of HBV infection. However, many scientific questions are still not answered. The question how HBV infection leads to defective immune responses to HBV proteins remains to be investigated. This issue is the key to a more rational design of new therapeutic approaches. Recently, HBV proteins were found to suppress host innate responses [188] . It has to be clarified whether an early blockage of innate immune responses may further negatively influence the priming of adaptive immune responses. In addition, different groups reported consistently that TLR2 and TLR4 signalling may be impaired in chronic HBV infection patients [189, 190] . Thus, it is worthy to test whether an enhancement of innate immune responses in chronic carriers is necessary for restoration of specific immune responses. With the increasing number of available vaccine formulation, a more crucial question raised recently: what is the optimal combination of these vaccines. Obviously, it is necessary to test the mutual influences of different types of vaccines to maximize their effects and avoid the negative interference between the vaccines. Finally, the future design of therapeutic vaccines needs to be considered in nonnaïve hosts since patients have undergone other infections. It is yet not possible to foresee how the pre-existing infections and immunological backgrounds will influence the effect of therapeutic vaccines. Understanding these issues will be helpful for the translation of recent progresses for clinical use of therapeutic vaccines.

Human immunome, bioinformatic analyses using HLA supermotifs and the parasite genome, binding assays, studies of human T cell responses, and immunization of HLA-A*1101 transgenic mice including novel adjuvants provide a foundation for HLA-A03 restricted CD8(+)T cell epitope based, adjuvanted vaccine protective against Toxoplasma gondii

BACKGROUND: Toxoplasmosis causes loss of life, cognitive and motor function, and sight. A vaccine is greatly needed to prevent this disease. The purpose of this study was to use an immmunosense approach to develop a foundation for development of vaccines to protect humans with the HLA-A03 supertype. Three peptides had been identified with high binding scores for HLA-A03 supertypes using bioinformatic algorhythms, high measured binding affinity for HLA-A03 supertype molecules, and ability to elicit IFN-γ production by human HLA-A03 supertype peripheral blood CD8(+ )T cells from seropositive but not seronegative persons. RESULTS: Herein, when these peptides were administered with the universal CD4(+)T cell epitope PADRE (AKFVAAWTLKAAA) and formulated as lipopeptides, or administered with GLA-SE either alone, or with Pam(2)Cys added, we found we successfully created preparations that induced IFN-γ and reduced parasite burden in HLA-A*1101(an HLA-A03 supertype allele) transgenic mice. GLA-SE is a novel emulsified synthetic TLR4 ligand that is known to facilitate development of T Helper 1 cell (TH1) responses. Then, so our peptides would include those expressed in tachyzoites, bradyzoites and sporozoites from both Type I and II parasites, we used our approaches which had identified the initial peptides. We identified additional peptides using bioinformatics, binding affinity assays, and study of responses of HLA-A03 human cells. Lastly, we found that immunization of HLA-A*1101 transgenic mice with all the pooled peptides administered with PADRE, GLA-SE, and Pam(2)Cys is an effective way to elicit IFN-γ producing CD8(+ )splenic T cells and protection. Immunizations included the following peptides together: KSFKDILPK (SAG1(224-232)); AMLTAFFLR (GRA6(164-172)); RSFKDLLKK (GRA7(134-142)); STFWPCLLR (SAG2C(13-21)); SSAYVFSVK((SPA250-258)); and AVVSLLRLLK(SPA(89-98)). This immunization elicited robust protection, measured as reduced parasite burden using a luciferase transfected parasite, luciferin, this novel, HLA transgenic mouse model, and imaging with a Xenogen camera. CONCLUSIONS: Toxoplasma gondii peptides elicit HLA-A03 restricted, IFN-γ producing, CD8(+ )T cells in humans and mice. These peptides administered with adjuvants reduce parasite burden in HLA-A*1101 transgenic mice. This work provides a foundation for immunosense based vaccines. It also defines novel adjuvants for newly identified peptides for vaccines to prevent toxoplasmosis in those with HLA-A03 supertype alleles.

parasite burden using a luciferase transfected parasite, luciferin, this novel, HLA transgenic mouse model, and imaging with a Xenogen camera. Conclusions: Toxoplasma gondii peptides elicit HLA-A03 restricted, IFN-γ producing, CD8 + T cells in humans and mice. These peptides administered with adjuvants reduce parasite burden in HLA-A*1101 transgenic mice. This work provides a foundation for immunosense based vaccines. It also defines novel adjuvants for newly identified peptides for vaccines to prevent toxoplasmosis in those with HLA-A03 supertype alleles. Toxoplasmosis is a disease of major medical importance. Toxoplasma gondii causes congenital infections responsible for stillbirths and spontaneous abortions [1] [2] [3] [4] [5] . In addition, it causes neurologic disorders, uveitis, and systemic infections in immune-compromised patients. Toxoplasmic encephalitis is a cause of morbidity and mortality in those with congenital disease and persons with AIDS [6] . T. gondii is acquired by consumption of lightly cooked meat, especially lamb and pork contaminated with bradyzoites [7, 8] or by ingestion of food or water contaminated with oocysts containing sporozoites, which are the product of a sexual cycle in the intestine of the cat [8, 9] . Development of an effective vaccine would prevent this disease. Attenuated T. gondii tachyzoites (e.g., S48, TS-4, T-203, ΔRPS13)have been employed for live vaccinations of non-human animals [10] [11] [12] [13] [14] [15] . Parasites attenuated by knockout of gene transcription recently have been proposed as a new type of attenuated vaccine candidate [13] [14] [15] . However, safety considerations may limit vaccination of humans with live organisms. Development of peptide-based vaccines created with epitopes which elicit IFN-γ production by CD8 + T cells [16] is a promising strategy to mobilize the immune system against T. gondii in humans [16] [17] [18] [19] [20] . Thus, an effort to create an immunosense epitope-based vaccine was initiated. Recently, three peptide epitopes, KSFKDILPK (SAG1 224-232 ), AMLTAFFLR (GRA6 164-172 ), and RSFKD LLKK (GRA7 134-142 ), were found by our group to elicit IFN-γ from peripheral blood mononuclear leukocytes (PBMCs) from T. gondii seropositive HLA-A03 supertype humans but not from PBMCs of T. gondii seronegative HLA-A03 supertype humans [18] . Herein, initially we designed 4 CD8 + T cell epitope containing vaccine formulations comprised of lipopeptides that incorporate PADRE as well as a novel oil-in-water emulsion that includes a specially formulated synthetic MLA derivative called GLA-SE [21] [22] [23] [24] . First, efficacy of various vaccine formulations consisting of the three previously identified CD8 + T cell peptides eliciting HLA-A*1101-restricted, CD8 + T cell-mediated IFN-γ production in vitro from HLA-A*1101 mice was examined. Then, in order to identify additional peptides from T. gondii tachyzoites, bradyzoites and sporozoites of Type I and Type II strains that are effective in eliciting IFN-γ from HLA-A03 supertype restricted CD8 + T cells, the first step was to select proteins with biologic properties, such as being secreted, compatible with MHC Class I processing. Then, bioinformatic algorithms to identify HLA-A03 supertype bound peptides were utilized to find additional, novel T. gondii-derived, potential CD8 + T cell eliciting epitopes restricted by the HLA-A03 supertype. We screened peptides from tachyzoite, bradyzoite and sporozoite proteins (GRA10, GRA15, SAG2C, SAG2D, SAG2X, SAG3, SRS9, BSR4, SPA, MIC) of the type II T. gondii strain, ME49, searching for those with high binding scores in a bioinformatic analysis (IC 50 < 50 nM) and then in binding assays. In addition, peripheral blood mononuclear cells from seropositive and seronegative persons were tested for response to these peptides using an IFN-γ ELISpot assay. These latter studies were performed to attempt to identify other peptides that would be promising candidates for inclusion in a multi-epitope, next generation immunosense vaccine. Then, information obtained from testing the first three peptides studied was used to guide formulation and administration of a peptide pool. This pool included the first three peptides identified earlier and tested in the initial experiments combined with the newly identified peptides that elicited IFN-γ from human HLA-A03 supertype restricted CD8 + T cells. All the peptides in this pool were tested with a universal T helper epitope called PADRE and a new, promising adjuvant called GLA-SE with Pam 2 Cys in HLA-A*1101 transgenic mice. Capacity to induce IFN-γ production by spleen CD8 + T cells, and to protect against parasite burden following subsequent challenge were determined. For parasite challenge, a luciferase transfected Type II Prugneaud parasite was administered, followed by luciferin administration and imaging with a Xenogen camera system a week later. This allows detection and quantization of bioluminescent parasites as a biomarker to assess efficacy of immunizations in protection. Three CD8 + T cell epitopes had been identified from T. gondii proteins, based on their significant recognition by T cells from T. gondii seropositive HLA-A03 individuals [18] . A universal CD4 + T cell epitope, PADRE (AKFVAAWTLKAAA), was linked in sequence with the N-terminal end of each of the three different T. gondii CD8 + T cell epitopes: KSFKDILPK (SAG1 224-232 ), AML-TAFFLR (GRA6 164-172 ), RSFKDLLKK (GRA7 134-142 ). Also, these three epitopes were linked together with three alanines as the linker. The N-terminal end of each resulting CD4-CD8 peptide or polypeptide was extended by a lysine covalently linked to two molecules of the palmitic acid moiety. The lipopeptides (Lp) were named as LpKS9, LpAM9, LpRS9 and LpKS9-AM9-RS9. They are shown in Figure 1 . Immunogenicity of lipopeptides in HLA-A*1101 transgenic mice HLA-A*1101 transgenic mice were immunized twice at intervals of three weeks with lipopeptides which were administered in PBS. Two weeks after the last immunization, the spleens were removed from immunized mice and the ability of splenocytes to produce IFN-γ upon stimulation with peptides was analyzed. Transgenic mice immunized with the three single peptide lipopeptide vaccines had T cells that produced IFN-γ ( Figure 2 ). The lipopeptide vaccines LpKS9 and LpAM9 stimulated higher IFN-γ production than LpRS9 ( Figure 2 ). However, LpKS9-AM9-RS9 with three peptide epitopes linked together did not stimulate strong IFN-γ responses when splenocytes from these mice were exposed to each of the individual peptides in vitro. Only results with LpKS9(KS9) and LpAM9(AM9) achieved statistical significance ( Figure 2 ). In order to determine which formulation was most immunogenic, vaccination with a single peptide or a mixture of the peptides was compared with linked lipopeptide vaccines. Results of a representative experiment are presented in Figure 3 . Mice immunized with a vaccine formulated with a single peptide SAG1 224-232 , KS9-PADRE with the adjuvancy of GLA-SE and Pam 2 Cys, elicited IFN-γ production (SFC:248 ± 65; mean ± SEM) four-fold higher (p < 0.01) than mice that received LpKS9 GLA-SE vaccination (63 ± 17) (Figure 3a ). Similar results, with substantial IFN-γ production, were found when splenocytes from mice immunized with a mixture of peptides and adjuvants. These results were compared with IFN-γ production by splenocytes from mice immunized with a lipopeptide vaccine constructed with the three peptides linked together with alanine spacers LpKS9-AM9-RS9. Significant IFN-γ production was only present with KS9 and AM9 stimulation of spleen cells, and was much less robust and not significant when the spleen cells were stimulated by RS9 peptide. To determine whether, and if so how, adjuvants effected immunogenicity of these peptides, HLA-A*1101 transgenic mice were immunized with pools of peptides that included all of the three peptides (KS9, AM9, RS9) alone or with varying adjuvants. HLA-A*1101 transgenic mice were immunized with: (1) CD8 + epitope peptide pool, (2) peptide pool plus PADRE, (3) peptide pool Figure 1 Schematic representation of the synthetic lipopeptide immunogens used in this study. The C-terminal end of a promiscuous CD4 + T cell peptide epitope (PADRE) was joined in sequence with the N-terminal end of one of three different T. gondii CD8 T cell epitopes: SAG1 224-232 (A), GRA6 164-172 (B); GRA7 134-142 (C) or three epitopes linked together (D) with a three alanine linker. The N-terminal end of each resulting CD4-CD8 peptide was extended by a lysine covalently linked to one molecule of palmitic acid. This results in a four lipopeptides construct. The abbreviation Lp is used throughout the manuscript whenever the lipopeptide (Lp) has been studied. When there is a mixture of components or undivided components they are named individually. Sequence of PADRE is AKFVAAWTLKAAA. Structure of Pam 2 Cys is PAM 2 KSS. a = abbreviation for name of protein from which peptide is derived. Figure 1 . The mice were immunized twice at intervals of three weeks. Ten to fourteen days after the last immunization, spleen cells were separated from immunized mice and stimulated by appropriate peptides in an ex vivo IFN-γ ELISpot assay. Data presented are averages of three independent replicate experiments. *, P < 0.05; **, P < 0.01. (Figure 3b ) were used as immunogens. For the lipopeptide immunizations, the mice were vaccinated twice at intervals of three weeks. For the immunizations containing individual components including the same peptides, the mice were vaccinated three times at intervals of two weeks. Ten to fourteen days after the last immunization, spleen cells were separated from immunized mice and stimulated by the appropriate peptide in an ex vivo IFN-γ ELISpot assay. Data presented are a representative example from three independent experiments. *, P < 0.05; **, P < 0.01. plus PADRE emulsified with GLA-SE, and (4) peptide pool plus PADRE and Pam 2 Cys emulsified with GLA-SE. Mice were inoculated three times at intervals of two weeks. Eleven to fourteen days post immunization, spleen cells were isolated and exposed to each individual peptide. A peptide was considered immunogenic if it induced IFN-γ spot formation that was significantly higher in the immunization group compared with the group inoculated with PBS. After the immunizations, only KS9 and AM9 were found to be immunogenic in HLA-A*1101 transgenic mice, and only when the iniversal CD4 + helper T cell peptide epitope PADRE was included. Robust responses were observed when GLA-SE was added in the vaccine. Greater responses were elicited when Pam 2 Cys was used as an adjuvant for some peptides but did not enhance responses to all of them, and in fact reduced the effect of some peptides. Figure 4 shows the representative data of IFN-γ spot formation from the four immunization groups which were stimulated by individual peptides. Vaccination with peptide pools and adjuvants protects mice against type II parasite challenge HLA-A*1101 transgenic mice were immunized with peptide pools plus PADRE and Pam 2 Cys in GLA-SE three times at intervals of two weeks. Mice were challenged 2 weeks after the last immunization. They were imaged 7 days after they had been challenged with 10,000 Pru (Fluc) using the Xenogen in vivo imaging system. As shown in Figure 5 , numbers of luciferase expressing parasites in immunized HLA-A*1101 transgenic mice were significantly less compared to numbers of parasites in unimmunized mice. Mean 8 ] for immunized mice. Differences were significant (e.g., p < 0.0064, for the pooled experiments, using natural log transformed data and two-sample t test). We initially identified 3 epitopes that provided protection against parasite challenge. As a next generation vaccine that might be even more robust, elicit CD8 + T cells that would be effective against all three life cycle stages and Type I and II genetic types of the parasite, we sought to identify additional HLA-A11 epitopes to be used for vaccine development. In order to identify additional peptides from T. gondii that were present in tachyzoites, bradyzoites, and sporozoites of Type II strains for HLA-A03 supertype restricted CD8 + T cells, we screened candidate peptides from tachyzoite, bradyzoite and sporozoite proteins (GRA15, GRA10, SAG2C, SAG2X, SAG3, SRS9, SPA and MIC) of the Type II T. gondii strain (Tables 1 and 2 ). For parsimony in numbers of peptides utilized, we attempted to include peptides present in Type I as well as Type II parasite genetic types. Peripheral blood mononuclear cells from seropositive T. gondii donors were tested for response to these peptides by using the IFN-γ ELISpot assay. Pooled peptides were tested initially. Pools 2 and 5 were significantly different (p < 0.05) from the control (data not shown). Then the individual peptides were tested ( Figure 6 ). Three out of 34 epitopes elicited responses greater than 50 IFN-γ SFC from seropositive donors PBMC, but not from seronegative donors PBMC ( Figure 6 ). These peptides and the proteins from which they are derived are: STFWPCLLR (SAG2C 13-21 ); SSAYVFSVK (SPA 250-258 ); and AVVSLLRLLK (SPA 89-98 ) adding proteins expressed in sporozoites and bradyzoites to the peptides selected. All peptides identified herein then were found to show high binding affinity to three to five HLA-A03 supertype alleles in the MHC-peptide binding assay ( Table 3 ). The numbers in Table 3 which indicate high binding affinity are those less than 500 IC 50 nM. They are indicated by bolded font numbers in Table 3 . Vaccination with peptide pools including newly identified peptides and adjuvants elicits IFN-g and provides more protection to mice against Type II parasite challenge Then, HLA-A*1101 transgenic mice were immunized with peptide pools which included these newly identified peptides: KSFKDILPK (SAG1 224-232 ); AMLTAFFLR (GRA6 164-172 ); RSFKDLLKK (GRA7 134-142 ); STFWPCLLR (SAG2C 13-21 ); SSAYVFSVK (SPA 250-258 ); AVVSLLRLLK (SPA 89-98 ) plus PADRE and Pam 2 Cys in GLA-SE, They were immunized three times at intervals of two weeks. Significant IFN-γ spot formation responses were observed in vitro by the cells from immunized mice exposed to all the peptides except GRA7 134-142 . Figure 7a shows the representative data of IFN-γ spot formation from the four immunization groups which were stimulated by individual peptides. The inclusion of the newly identified peptides thus had potential to enhance immunogenicity in transgenic HLA-A*1101 mice. Mice were challenged 2 weeks after the last immunization. They were imaged five days after they had been challenged with 10,000 Pru (Fluc) using a Xenogen in vivo imaging system. As shown in the initial experiment in Figure 7b , the numbers of luciferase expressing parasites in immunized HLA-A*1101 mice were significantly reduced compared to the numbers of parasites in unimmunized mice. Results An algorithm was developed to calculate projected population coverage of a T cell epitope-based vaccine using MHC binding or T cell restriction data and HLA gene frequencies [17] . We used this web-based tool http://www.iedb.org/ to predict population coverage of these HLA-A03 supertype peptide epitopes-based vaccine. The population coverage calculation results in Table 4 indicate that such coverage is varied in different geographic regions. The HLA-A03 supertype molecules that present these peptides would be expected to be present in 28 In this study, we first evaluated immunization of HLA-A*1101 transgenic mice with either mixtures of peptides or lipopeptides derived from three identified T. gondii specific HLA-A*1101 restricted CD8 + T cell epitopes emulsified in 3-deacylated monophosphoryl lipid A (GLA-SE) adjuvant. Immunizations of transgenic mice with a mixture of CD8 + epitope peptide pools plus PADRE and adjuvants were able to induce splenocyte to produce IFN-γ and to protect against challenge with high numbers of Type II parasites. Conjugation of CD8 + T cell determinants to lipid groups is known to enhance specific cell-mediated responses to target antigens in experimental animals and humans [25] [26] [27] [28] [29] , although mechanisms whereby immunity is achieved remains poorly understood. Lipopeptides hold several advantages over other conventional vaccine formulations; for instance, they are self-adjuvanting and display none of the toxicity-associated side effects of other Th1-inducing adjuvant systems. In our work, transgenic mice that were immunized with three short lipopeptide vaccines had T cells that produced IFN-γ. Among them the lipopeptide vaccine formulated with KS9 or AM9 In vivo protection demonstrated with imaging using Xenogen camera. HLA-A*1101 transgenic mice immunized with peptide pool and adjuvants were protected compared to control mice inoculated with PBS when they were challenged with 10,000 Prugneaud strain (Fluc)-T. gondii luciferase expressing parasites. There were a total of 5-9(usually 4-5 per group) mice tested in each control or immunization group. Differences between control and immunization groups were significant(p < 0.0064 using natural log transformed data and two-sample t test). Peptides derived from these proteins and the position within the proteins 2 Binding affinity was performed by MHC binding assay. 3 PBMC from four T. gondii-seropositive HLA-A03 supertype persons and four seronegative persons were stimulated with peptides, the T cell that produce IFN-γ were tested by ELISpot assay. 4 Splenic T cell were isolated from HLA-A*03 supertype(which includes the HLA-A*1101 haplotype) mice 10 to 14 days after peptide immunization and tested for their ability to generate IFN-γ in response to peptide. stimulated higher IFN-γ production than the lipopeptide vaccine formulated with RS9. Unexplained and variable responses have been observed to high affinity binding peptides in other models, e.g., studies of Livingstone, Alexander, Sette et al with Lassa fever virus. It will be of interest to better understand possible mechanisms for such lack of response in future studies. However, the lipopeptides with three epitope peptides linked together with alanine spacers did not stimulate an IFN-γ response by splenocytes from immunized mice when the splenocytes subsequently were exposed to each of the peptides in vitro. The reason why the lipopeptides with the three linked peptides did not work well in the transgenic mice might be related to a frame shift caused by the linkers that altered the response to the original peptides rather than the alanines functioning for the intended purpose of introducing a cleavage motif. The three linker "AAA" between the peptides had previously been demonstrated in other systems to result in sensitization to each linked peptide. However, surprisingly, it did not appear to work well herein. Because the three linked peptides in the lipopeptide formulation were not effective and we had found that a mixture of the components with a single peptide was as or more robust than the lipopeptide, we tried this approach with the three peptides that had been included in the linked lipopeptide with the universal helper CD4 + T cell peptide, PADRE, and adjuvants as described below. The response was robust both in vitro and in vivo (Figures 4 and 5) . Some studies have shown palmitoylated lipopeptide constructs to elicit long-lived, protective cellular responses against a variety of pathogens, including Hepatitis B virus (HBV), influenza virus, and Plasmodium falciparum [25] [26] [27] [28] [29] . Our work herein shows that mice immunized with mixture of CD8 + and CD4 + eliciting peptides and lipid Pam 2 Cys emulsified in GLA-SE elicited higher IFN-γ production than mice immunized with lipopeptides constructed with the same components of CD4 + and CD8 + eliciting peptides, and Pam 2 Cys. The approach using cocktails of non-covalently linked lipid mixed to helper T lymphocytes(HTL) and CD8 + T cell (cytolytic T lymphcyte[CTL] and IFN-γ eliciting) epitopes for simultaneous induction of multiple CD8 + T specificities would have significant advantages in terms of ease of vaccine development. HTL responses are crucial for the development of CD8 + T responses, at least in the case of lipidated covalently or non-covalently linked HTL-CTL epitope constructs formulated in PBS. Several previous studies have illustrated a role for CD4 + responses for development of CD8 + CTL responses, both in humans and in experimental animals [30] [31] [32] [33] [34] [35] . The inclusion of PADRE, a synthetic peptide that binds promiscuously to variants of the Figure 6 New peptides tested with PBMCs from HLA-A03 seropositive and seronegative donors. Then Peripheral blood mononuclear cells from seropositive T. gondii donors were tested for response to these predicted HLA-A03 supertype restricted CD8 + T cell epitope individual peptides by using IFN-γ ELISpot assay. Peptides that induced significant IFN-γ spot formation compared to DMSO are denoted by an asterisk. *, P < 0.05. Figure 7 Addition of peptides to pool robustly protects HLA-A*1101 mice against Type II parasite challenge. HLA-A*1101 transgenic mice were immunized with PBS (controls) or peptide pool with PADRE and Pam 2 Cys in GLA-SE. Splenic T cells were isolated 10-14 days post immunization and exposed to each peptide in an ex vivo IFN-γ ELISpot assay (Figure 7a ). HLA-A*1101 transgenic mice immunized with peptide pool and adjuvants were protected compared with control mice inoculated with PBS when they were challenged with 10,000 Pru (Fluc)-T. gondii luciferase expressing parasites (Figure 7b ). Mice were immunized and in a subgroup immune function was studied at the same time as the challenge shown was performed. Differences between control and immunized mice were significant (p < 0.0064 using natural log transformed data and two-sample t test). human MHC class II molecule DR and is effective in mice, also augmented CD8 + T cell effector functions by inducing CD4 + T helper cells [30] [31] [32] [33] [34] [35] . Both CD4 + and CD8 + epitopes were targeted in order to drive a protective immune response [34, 35] . Adjuvanting antigens contributes to the success of vaccination. An example herein is that 3-deacylated monophosphoryl lipid A(GLA-SE), a detoxified derivative of the lipopolysaccharide (LPS) from Salmonella minnesota R595 was a potent adjuvant. This GLA-SE is a novel adjuvant which was formulated in an emulsion [21] [22] [23] [24] . This is a Toll-like receptor 4 (TLR4) agonist that is a potent activator of Th1 responses [21] [22] [23] [24] . It has been used as an adjuvant in human vaccine trials for several infectious disease and malignancy indications. It has been very effective as an adjuvant providing CD4 + T cell help for immunizations against other protozoan infections such as leishmaniasis [21] [22] [23] [24] . In our study, a robust response was observed when GLA-SE was included in preparation for immunization of mice. Pam 2 Cys (S-[2,3-bis(palmitoyloxy)propyl] cysteine) is a lipid component of macrophage-activating lipopeptide. Pam 2 Cys binds to and activates dendritic cells by engagement of Toll-like receptor 2 (TLR-2) [24] . Tolllike receptors (TLRs) function as pattern-recognition receptors in mammals [36] . We have found that both TLR2 and TLR4 receptors participate in human host defense against T. gondii infection through their activation by GPIs and GIPLs(Melo, Hargrave, Miller, Blackwell, Gazinelli, McLeod et al, in preparation, 2010). TLR2 and TLR4 likely work together with other MyD88-dependent receptors, including other TLRs, to elicit an effective host response against T. gondii infection [36] . In our study, there was a slightly more robust response observed when Pam 2 Cys was co-administered for some peptides, but not all of them. The goal of the present study was to identify HLArestricted epitopes from T. gondii and evaluate whether they could provide protection against parasite challenge measured as protection against a luciferase producing Type ll parasite using a Xenogen camera system. In the future, additional more detailed studies involving analyses over longer times, other strains of the parasite and challenge with life cycle stages, evaluation of multiple organs including eye and brain, studies of protection in congenital infections, comparisons of delivery of these peptides as DNA encoding them versus other formulations. This future work will follow up and extend these intitial studies of reduction of parasite burden seen in Figures 5 and 7 . Various peptide-based approaches to induction of IFN-γ responses were evaluated as part of ongoing efforts to develop immunosense vaccines for use in humans with each of the supermotifs which would in total include more that 99% of the human population worldwide. Robust protection was achieved in the HLA-A*1101 transgenic mice challenged with Type II parasites following immunizations. In order to identify additional peptides from T. gondii that were present in tachyzoites or bradyzoites [37, 38] or sporozoites of Type I and II strains and elicited IFN-γ from HLA-A03 + supertype (which includes the HLA*1011 allele) restricted CD8 + T cells, bioinformatic algorithms were utilized to identify novel, T. gondii-derived, epitopes restricted by the HLA-A03 supertype. Then PBMC cells were tested to determine whether the peptides elicited IFN-γ from human CD8 + T cells from seropositive persons. This was intended to collectively provide broad coverage for the human population with HLA-A03 supertype worldwide. The additional peptides we identified as immunogenic for human peripheral blood cells were also robust in eliciting IFN-γ from splenocytes of HLA-A*1101 mice and protection when used to immunize these mice. These findings will facilitate development of an immunosense epitope-based vaccine for human use. A human immunome-based and parasite genome based bioinformatices approach was used to define candidate HLA-A03 supertype restricted peptides. Immunogenicity of a group of T. gondii HLA-A03 supertype restricted peptides, and therefore the proteins from which they are derived, for immune humans and HLA-A*1101 transgenic mice was demonstrated. These peptides elicit interferon-γ production by human CD8 + T cells. They also elicit interferon γ production by mouse splenocytes when utilized to immunize HLA-A*1101 transgenic mice with a lipopeptide with a universal CD4 + T cell eliciting epitope, PADRE, or in peptide pools with PADRE, Pam 2 Cys and GLA-SE, a novel adjuvant. Immunization studies demonstrate the need for and the efficacy of adjuvants in immunization of these HLA transgenic mice. Immunogenic peptides included KSFKDILPK (SAG1 224-232 ); AML-TAFFLR (GRA6 164-172 ); and RSFKDLLKK (GRA7 134-142 ); STFWBCLLR (SAG2C [13] [14] [15] [16] [17] [18] [19] [20] [21] ; SSAYVFSVK (SPA 250-258 ); and AVVSLLRLLK (SPA 89-98 ). The studies herein provide a foundation for immunosense based vaccines to prevent toxoplasmosis in those with the HLA-A03 supertype and information about how they can be adjuvanted. HLA-A03 supertype CD8 + T cell epitopes included: KSFKDILPK (SAG1 224-232 ), AMLTAFFLR (GRA6 164-172 ), and RSFKDLLKK (GRA7 134-142 ). PADRE (AKF-VAAWTLKAAA) was the universal CD4 helper peptide used in vaccine constructs. Pam 2 Cys (Pam 2 -KSS) also was included. Lipopeptide constructs used in this study are shown in Figure 1 . Peptides and lipopeptides were synthesized by Synthetic Biomolecules, San Diego at > 90% purity. Additional HLA-A03 supertype bound peptides and their initial grouping into pools for in vitro studies are shown in Tables 1 and 2 . A TLR4 agonist, a GLA-SE adjuvant, was synthesized by the Infectious Diseases Research Institute (Seattle, Washington) as a stable oil-in-water emulsion. AMLTAFFLR (GRA6 164-172 ) and additional new peptides were first dissolved in DMSO and then diluted in PBS. HLA-A*1101/K b transgenic mice were produced at Pharmexa-Epimmune (San Diego, CA) and bred at the University of Chicago. These HLA-A*1101/K b transgenic mice express a chimeric gene consisting of the 1 and 2 domains of HLA-A*1101 and the 3 domain of H-2K b , and were created on a C57BL/6 background. For each test, we used 4-5 mice for each group. Each experiment was repeated 2 to 3 times. Experiments in Figure 7b were performed including a subgroup analyzed for immune response in parallel with a subgroup in the challenge shown. All studies were conducted with approval of the Institutional Animal Care and Use Committee at the University of Chicago. Transgenic T. gondii used for in vivo challenges was derived from Type II Prugniaud (Pru) strain and expresses the firefly luciferase (FLUC) gene constitutively by tachyzoites and bradyzoites. It was created, and kindly provided by S. Kim, J. Boothroyd and J. Saeij (Stanford University) and was maintained and utilized as previously described [18, 37, 39] . To evaluate peptide immunogenicity, HLA-A*1101 transgenic mice were inoculated subcutaneously (s.c.) at the base of the tail using a 30-gauge needle with single peptides or a mixture of CD8 + T cell peptides (50 μg of each peptide per mouse) and PADRE (AKFVAAWTLKAAA) emulsified in 20 μg of GLA-SE (TLR4 agonist) with or without Pam 2 Cys. Pam 2 Cys concentration was 5 mg/ml. For immunization with lipopeptides, HLA-A*1101 mice received 20 nmol lipopeptide dissolved in PBS or emulsified in GLA-SE. As controls, mice were injected with PBS or PBS/GLA-SE. For the lipopeptide immunizations, the mice were vaccinated twice at intervals of three weeks. For the peptide immunizations, mice were immunized three times at intervals of two weeks. For challenge studies, mice were immunized with peptide emulsions and challenged intraperitoneally (i.p.) 14 days post-immunization using 10,000 Type II parasites. Mice infected with 10,000 Pru-FLUC tachyzoites were imaged 7 days post-challenge using the in vivo imaging system (IVIS; Xenogen, Alameda, CA). Mice were injected i.p. with 200 μl of D-luciferin, anesthetized in an O 2 -rich induction chamber with 2% isoflurane, and imaged after 12 minutes. Photonic emissions were assessed using Living image® 2.20.1 software (Xenogen). Data are presented as pseudocolor representations of light intensity and mean photons/region of interest (ROI). All mouse experiments were repeated at least twice. There were 4-5 mice for each group. In the experiment in Figure 7b a subgroup of mice was used for studying immune response in parallel with the subgroup in the challenge shown. Mice were euthanized 7 to 14 days after immunization. Spleens were harvested, pressed through a 70 μm screen to form a single-cell suspension, and depleted of erythrocytes with AKC lysis buffer (160 mM NH 4 Cl, 10 mM KHCO 3 , 100 M EDTA). Splenocytes were washed twice with Hank's Balanced Salt Solution (HBSS) and resuspended in complete RPMI medium (RPMI-1640 supplemented with 2 mM L-GlutaMax [Invitrogen], 100 U/ml penicillin, 100 μg/ml streptomycin, 1 mM sodium pyruvate, 50 M -mercaptoethanol, and 10% FCS) before they were used in subsequent in vitro assays. Peptide concentration was 20 mg/ml and 0.20 μl was used per well. ELISpot assays with murine splenocytes were performed using α-mouse IFN-γ mAb (AN18) and biotinylated α-mouse IFN-γ mAb (R4-6A2) as the cytokine-specific capture antibodies. Antibodies were monoclonal antibodies. 5 × 10 5 splenocytes were plated per well. PBMC were obtained, HLA haplotype was determined, and they were processed and cryopreserved as described [18] . ELISpot assays with human PBMCs were similar to those with murine splenocytes but used α-human IFN-γ mAb (1-D1K) with biotinylated α-human IFN-γ mAb (7B6-1) with 2 × 10 5 PBMCs per well. All antibodies and reagents used for ELISpot assays were from Mabtech (Cincinnati, OH). Antibodies were monoclonal antibodies. Both murine and human cells were plated in at least 3 replicate wells for each condition. Results were expressed as number of spot forming cells (SFCs) per 10 6 PBMCs or per 10 6 murine splenocytes. Protein sequences derived from GRA10, GRA15, SAG2C, SAG2D, SAG2X, SAG3, SRS9, BSR4, SPA, and MIC were analyzed for CD8 + T cell epitopes based on predicted binding affinity to HLA-A03 supertype molecules using ARB algorithms from immunoepitope database (IEDB) http://www.immuneepitope.org [40, 41] . A total of 34 unique peptides IC 50 < 50 nM of all ranked nonameric peptides were selected. All protein sequences were from ToxoDB 5.1. Quantitative assays to measure binding of peptides to HLA class I molecules are based on inhibition of binding of radiolabeled standard peptide. Assays were as described [42] . Concentration of peptide yielding 50% inhibition of binding of radiolabeled probe peptide (IC 50 [43, 44] . Statistical analyses for all in vitro assays were performed using 2-tailed student's T test. Natural log transformed data and two-sample t test were used to analyze data shown in Figures 5 and 7b . Two-tailed P values < 0.05 were considered statistically significant. Peptides were considered immunogenic in mice if they induced IFN-γ spot formation from immunized mice that were significant (P < 0.05) compared with spot formation from control mice. All mouse experiments were repeated at least twice. There were 4-5 mice for each group. The experiment in Figure 7b determined immune response and imaged mice in parallel.

Conserved epitopes of influenza A virus inducing protective immunity and their prospects for universal vaccine development

Influenza A viruses belong to the best studied viruses, however no effective prevention against influenza infection has been developed. The emerging of still new escape variants of influenza A viruses causing epidemics and periodic worldwide pandemics represents a threat for human population. Therefore, current, hot task of influenza virus research is to look for a way how to get us closer to a universal vaccine. Combination of chosen conserved antigens inducing cross-protective antibody response with epitopes activating also cross-protective cytotoxic T-cells would offer an attractive strategy for improving protection against drift variants of seasonal influenza viruses and reduces the impact of future pandemic strains. Antigenically conserved fusion-active subunit of hemagglutinin (HA2 gp) and ectodomain of matrix protein 2 (eM2) are promising candidates for preparation of broadly protective HA2- or eM2-based vaccine that may aid in pandemic preparedness. Overall protective effect could be achieved by contribution of epitopes recognized by cytotoxic T-lymphocytes (CTL) that have been studied extensively to reach much broader control of influenza infection. In this review we present the state-of-art in this field. We describe known adaptive immune mechanisms mediated by influenza specific B- and T-cells involved in the anti-influenza immune defense together with the contribution of innate immunity. We discuss the mechanisms of neutralization of influenza infection mediated by antibodies, the role of CTL in viral elimination and new approaches to develop epitope based vaccine inducing cross-protective influenza virus-specific immune response.

Influenza remains a serious respiratory disease in spite of the availability of antivirals and inactivated trivalent vaccines, which are effective for most recipients. Influenza viruses are RNA viruses with strongly immunogenic surface proteins, especially the hemagglutinin. Error-prone RNA-dependent RNA polymerase and segmented genome enable influenza viruses to undergo minor (antigenic drift) as well as major (antigenic shift) antigenic changes, which permit the virus to evade adaptive immune response in a variety of mammalian and avian species, including humans. The unpredictable variability of influenza A viruses, which cause yearly epidemics in human population, is the main reason why no effective prevention against influenza infection exists up to date. Currently available vaccines induce antibodies against seasonal and closely related antigenic viral strains, but do not protect against antibody-escape variants of seasonal or novel influenza A viruses. Therefore, there is a call for development of a vaccine, which would be protective against virus strains of different HA subtypes and would not need to be updated every year. New approach to prepare a universal vaccine lies in the selection of conserved epitopes or proteins of influenza A virus, which induce cross-protective immune response, particularly M2, HA2, M1, NP [1] [2] [3] . Influenza infection induces specific humoral immunity represented by systemic and local antibody response, as well as cellular immunity, represented by specific T-cell response ( Figure 1 ). Both of them are important in the host defense against influenza infection, because of their close cooperation mediated by various immune mechanisms. Dendritic cells and macrophages (antigen presenting cells, APCs) play an important role in initiating and driving of adaptive immune response [4] . Exogenous viral antigens, including inactive viral particles, intact viruses or infected cells, are taken up by APCs through endocytosis or phagocytosis. Their further processing results in generation of peptides that are presented via MHC I or MHC II molecules to CD8+ precursor T-cell and CD4+ helper T-cell precursors (Th0), respectively. Th0 cells are subdivided to Th1and Th2-type helper cells, based on the cytokine profiles they produce. Following influenza infection, APCs secrete IL-12 that contributes to the differentiation of Th0 into Th1 cells, which secrete IFN-γ and help to produce IgG2a antibodies [5, 6] . Th1 cells also produce IL-2, required for the proliferation of the virus-specific CD8+ CTLs. In contrast, when IL-10 is present early in The cellular immune response (right) is initiated after recognition of viral antigens presented via MHCI and MHC II molecules by antigen presenting cells (APC), which then leads to activation, proliferation and differentiation of antigen-specific CD8+ T or CD4+ cells. These cells gain effector cell function and either they help directly (Th1 or Th2 cell) to produce antibodies or, CTL effector cells recognize antigen peptides presented by MHCI on APC and kill the virus infected cells by exocytosis of cytolytic granules. The humoral immune response (left)is mediated by specific antibodies (e.g IgG, IgA) produced by antibody secreting plasma cells (ASC) which are the final stage of B cell development. This process is aided by CD4+ T helper and T cell-derived cytokines essential for the activation and differentiation of both B-cell responses and CD8+ T cell responses. the immune response , Th0 cells differentiate to Th2 cells, which secrete IL-4, IL-5, IL-6 and help preferentially drive IgG1, IgA and IgE Ab production by anti- body-secreting plasma cells (ASCs) [6] [7] [8] [9] . CD8+ precursor T-cells, which maturate into CTLs (cytotoxic T lymphocytes), release antiviral cytokines (IFN-γ) upon recognition of short viral peptides presented by MHC I molecules on virus-infected epithelial cells, and destroy the virus infected cells by exocytosis of cytolytic granules. The granules contain cytolytic protein perforin and granzymes. Perforin is a protein that creates pores in membranes of infected cells. Granzymes are members of serine protease family. In the presence of perforin, granzymes enter into the cytoplasm of infected cells and initiate proteolysis, which triggers destruction of the target cell [10, 11] . CTLs could mediate killing of infected cells also by perforin-independent mechanisms of cytotoxicity. This involves binding of Fas receptor in the infected target cell membranes with the Fas ligand (FasL) expressed on activated CTLs. Interaction of FasL with corresponding Fas receptor leads to the activation of caspases, which induce apoptosis in influenza infected cells [12] [13] [14] . Unlike T-cells, which recognize linear epitopes presented by MHC molecules, B cells can recognize antigen in its native form. Antibody response against influenza infection is mediated by secretory IgA antibodies and serum IgG antibodies. IgA are transported across the mucosal epithelium of the upper respiratory tract, where they represent the first immunobarrier to influenza viruses. IgG transude from the serum to the mucus by diffusion and are primarily responsible for the protection of the lower respiratory tract [15] . Specific antibodies induced by influenza virus infection can neutralize infection by several different mechanisms ( Figure 2 ). They can directly block virus attachment to the target cells by interfering with virus-receptor interaction and thus prevent influenza infection ( Figure 2A ). These antibodies are directed to the globular domain of the surface antigen, hemagglutinin [16] . However, because of high variability of influenza A viruses, neutralization activity of these Abs is limited to viral strains, which are antigenically similar to the inducers of Ab Figure 2 Mechanisms of antibody-mediated neutralization during influenza infection. A. Serum IgG or B. mucosal IgA antibodies specific to hemagglutinin prevent influenza infection by blocking attachment to host cell receptors. C. After binding, the virus is internalized by receptor mediated endocytosis. The low pH in the endosome triggers conformational changes in hemagglutinin that expose fusion peptide located in HA2 required for membrane fusion. In this step, antibodies bound to HA2 block the fusion of viral and endosomal membranes and prevent release of ribonucleoprotein complex into the cytoplasm of target cell. D. Intracellular neutralization of influenza virus through transcytotic pathway of IgA that complex with viral proteins and inhibit assembly of progeny virions. E. Antibodies specific to neuraminidase inhibit release of budding viral particles and further spread of influenza infection by inhibition of neuraminidase activity. production. By contrast, it was shown that mucosal immunity mediated by secreted form of IgA Abs in the upper respiratory tract is more cross-protective against heterologous virus infection than systemic immunity mediated by IgG Abs [17, 18] . The strong cross-protective potential of IgA Abs appears to be the consequence of their polymeric nature, resulting in higher avidity of Abs for the influenza virus compared to the monomeric serum IgG Abs [18] . After synthesis by ASC, dimeric IgA (dIgA) Abs bind to the polymeric immunoglobulin receptor expressed on the basolateral surface of the epithelial cells and are transcytosed to the apical surface, where the poly-Ig receptor is cleaved, secretory IgA are released and prevent infection by blocking attachment to the epithelial cells ( Figure 2B ). Moreover, dIgA Abs are able to bind to the newly synthesized viral proteins within infected cells, thus preventing virion assembly ( Figure 2D ) [19] . After attachment to the receptor on the target cell, influenza virus is internalized by receptor-mediated endocytosis. Conformational changes of hemagglutinin triggered by the low pH in the endosome activate viral and endosomal membrane fusion. In this step, antibodies, which bind to the non-receptor binding region of HA, could interfere with the low-pH induced conformational change in the HA molecule required for the fusion. Inhibition of the fusion between viral and endosomal membrane proteins mediated by such antibodies prevents release of the ribonucleoprotein complex (RNP complex) into the cytoplasm of the target cell, resulting in the inhibition of viral replication ( Figure 2C The interaction of opsonic complement proteins with complement receptor on macrophages (CR) increases the rate of phagocytosis of macrophages, causing direct virolysis or improvement of antibody-mediated inhibition of virus attachment to host cells [26, 27] . However, contribution of complement to the protective capacity of antibodies is contradictory, since it was shown that passive transfer of murine polyclonal anti-eM2 serum into C3-negative mice had protective effect [24], while human monoclonal anti-M2 antibodies could not protect complement-depleted mice [25] . It should be noticed that though some antibodies directed to conserved antigens such as M2 do not prevent infection by direct binding to virus, they can contribute to an earlier recovery from the infection by indirect antibodymediated mechanisms after binding to Fc-receptors on macrophages or NK cells. It is possible that the same mechanism of protection is mediated by antibodies to HA2 glycopolypeptide (HA2 gp), a conserved part of HA. They also do not prevent infection, but their strong protective potential has been proved in vivo [28-31]. For this reason understanding the role of the Fc effector function of antibodies in the clearance of influenza infection is required. Both, ectodomain of M2 and HA2 gp are conserved antigens inducing antibodies protecting against influenza infection. Therefore, various studies are focused on these two antigens as inductors of heterosubtypic antibody response. M2 protein is a single-pass type III membrane protein forming homotetramers representing pH-gated proton channel incorporated into the viral lipid envelope. This proton channel is essential for efficient release of viral genome during viral entry [32]. M2 protein is abundantly expressed at the apical plasma membrane of infected epithelial cells, but only a small number (16-20 molecules/virion) of M2 molecules are incorporated into virions [33, 34] . Great attention is paid to the extracellular N-terminal domain of M2 protein (eM2), a 23 amino acid peptide, which is highly conserved in all human influenza A strains. It is therefore an attractive target for preparation of a universal influenza A vaccine. In contrast to hemagglutinin and neuraminidase, eM2 is a weak immunogen [35] . Therefore, various approaches to increase its immunogenicity were used. All of them are based on increasing the immunogenicity of small antigen molecules by insertion of their multiple copies into a suitable immunogen Neirynck et al [36] prepared a fusion protein composed of eM2 and hepatitis B virus core (HBc) protein. This fusion protein has the ability to aggregate into the highly immunogenic virus-like particles inducing a long-lasting protection against lethal influenza A infection. High in vivo protective effect of described virus-like particles was proven after intraperitoneal or intranasal immunization of mice and subsequent infection with lethal dose of influenza viruses of various HA subtypes [37] . Efficacy of these particles has been increased by application of new adjuvant CTA1-DD. Combination of the eM2-HBc construct with the new adjuvant led to the protection of mice against lethal infection and a remarkably lower morbidity [38] . Various constructs of eM2 peptide engineered by conjugation to carrier proteins were evaluated as a vaccine, which successfully protected animals against infection with homologous but also heterologous human strains [24,36,37,39-42]. A different approach to increase immunogenicity of eM2 was described by other groups. Constructs composed of four tandem copies of the eM2 peptide fused to flagellin, a ligand of TLR5 (Toll-like receptor 5) [41], or glutathione-S-transferase fusion protein bearing various numbers of eM2 epitope copies [42], were used as immunogens. These studies showed that high eM2 epitope densities in a single recombinant protein molecule resulted in enhanced eM2-specific humoral response and higher survival rates of infected animals. Another way to stimulate the immune system by small peptide was described by Ernst et al. Immunization of mice with these eM2-HD liposomes was protective against influenza virus strains of various subtypes and stimulated the production of specific IgG1 antibodies in mouse sera. Moreover, mice passively immunized with these antibodies were protected against lethal infection. M2 protein in its native state forms a homotetramer, comprising also conformational epitopes, which might play important role in eM2 immunogenicity. It was shown that oligomer-specific antibodies were induced by recombinant eM2 protein mimicking the natural quaternary structure of M2 ectodomain in viral particle [43] . For this purpose, a modified version of leucine zipper from yeast transcription factor GCN4 was bound to eM2. High titers of antibodies recognizing M2 protein in the native conformation were obtained after intraperitoneal or intranasal immunization with this recombinant protein, and immunized mice were fully protected against lethal dose of influenza A virus [43] . Such vaccine could improve quality of humoral immune response with antibodies elicited not only against linear epitopes but also against conformational epitopes. Above described results indicate that eM2 is a valid and versatile vaccine candidate to induce protective immunity against any strain of human influenza A viruses, and give a promise for finding new "universal" vaccine against flu. Hemagglutinin (HA) is the major influenza virus target antigen recognized by neutralizing antibodies. It is a surface glycoprotein, synthesized as a single polypeptide, which is trimerized. Each monomer of HA is synthesized as a precursor molecule HA0 post-translationally cleaved by host proteases into two subunits, HA1 and HA2 linked by a single disulfide bond [16] . Cleavage into HA1 and HA2 gp is essential for the infectivity of the virus particle and spread of the infection in the host organism [44] . The HA1 of influenza A virus forms a membranedistal globular domain that contains the receptorbinding site and most antigenic sites recognized by virus-neutralizing antibodies preventing attachment of virus to the host cell. Escape variants with mutation in the antigenic site easily avoid neutralization by existing host antibodies, leading to seasonal influenza outbreaks [45] . In spite of continual antigenic changes of hemagglutinin, common epitopes shared by various strains were identified. Although the degree of sequence diversity between HA subtypes is great, particularly in the HA1 glycopolypeptides, HA2 is its rather conserved part. According to documented results, HA2 has the prerequisite to be one of the potential inductors of protective heterosubtypic immunity [1, 28, 29, [46] [47] [48] . HA2 represents the smaller C-terminal portion of hemagglutinin, which forms a stem-like structure that mediates the anchoring of the globular domain to the cellular or viral membrane. N-terminal part of HA2 gp, termed the fusion peptide, plays a substantial role in the fusion activity of influenza virus. It was demonstrated that the rearrangements of HA as well as the fusion process is temperature-and pH-dependent [49, 50] . At neutral pH, the N-terminus of the fusion peptide is inserted into the inter-space of HA trimer. At low pH, which triggers the fusion process, N-terminus of the fusion peptide is exposed and inserted into the target membrane, allowing the release of the ribonucleoprotein complex into the cytoplasm [51, 16] . Although the epitopes of the [62, 63] . Moreover, it was shown that passive immunization with monoclonal antibodies against HA2 gp, as well as active immunization with recombinant vaccinia virus expressing chimeric molecules of HA, improve the recovery from influenza infection and contribute to a milder course of infection [28, 29] . A recent study showed that increased immunogenicity of HA2 gp could be achieved by unmasking of HA2 gp after removing the highly immunogenic globular head domain of HA1 gp. Headless HA trimers form the conserved HA stalk domain, on which HA2 epitopes are more accessible for B cells than in the native HA. Vaccination of mice with this headless HA immunogen elicited antibodies cross-reactive with multiple subtypes of hemagglutinin and provide protection against lethal influenza virus infection [31] . Hemagglutinin HA1-HA2 connecting region, as well as N-terminal fusion peptide of HA2, are the broadly conserved parts of HA, the latter conserved even among all 16 subtypes of influenza A viruses [1, 47, 61, 64] . Protective potential of the fusion peptide or HA1-HA2 cleavage site of influenza A viruses were investigated by several groups. They found that mice vaccinated with a peptide spanning the HA1-HA2 connecting region exhibited milder illness and fewer deaths upon virus challenge [64, 65] . Generation of monoclonal antibodies against universally conserved fusion peptide has attracted interest in the recent past, as such antibodies are known to inhibit the HA fusion activity and to reduce virus replication in vitro and also in vivo [28, 30, 54, 62, 63] . Additionally, passive immunotherapy with Abs reactive with all strains of influenza A could be an alternative for some populations at high risk of infection, like infants, the elderly and the immunocompromised patients, who may not benefit from active vaccination. Several groups described the potential of human monoclonal antibodies against HA2 subunit and its fusion peptide with broadspectrum protection as a universal passive immunotherapeutic agent against seasonal and pandemic influenza viruses [66] [67] [68] [69] . Sui et al. [70] obtained a panel of highaffinity human antibodies that bind to the highly conserved pocket in the stem region of hemagglutinin, comprising part of the fusion peptide and several residues of the HA1 subunit. These antibodies showed a broad degree of cross-reactivity. Moreover, it was suggested that the conformational epitope on HA recognized by one of these neutralizing antibodies (F10) is recalcitrant to the generation of escape mutants [70] . Thus, identification of antibodies against conserved epitopes of hemagglutinin shows the way for their use in passive immunotherapy, designing of antivirals and represents an important step towards development of cross-protective universal vaccine against influenza virus that potentially does not require annual adjustment. Nucleoprotein (NP) and matrix protein (M1) of influenza virus are conserved structural influenza antigens, to which antibody response is induced after natural infection. These antibodies, however, do not display a considerable effect on protection against influenza infection [22] . On the other hand, NP, M1 and other inner influenza antigens play important role in the cellular immune response. It was demonstrated that NP-or M1-specific Th cells could augment protective antibody response, aiding the B cells to produce antibodies specific to hemagglutinin [71] . CTL play an important role in the control of influenza virus infections. They eliminate virus-infected cells, on which surface they recognize foreign antigens derived from endogenously expressed viral antigens presented by MHC class I molecules. Thus, they contribute to the clearance of the virus from the infected tissue and prevent the spread of viral infection. Although CTL do not prevent influenza infection, their beneficial effect on the course of infection was observed after the adoptive transfer of virus-specific CTL clones to mice, resulting in direct lysis of infected cells [72] [73] [74] . In addition, depletion of CTL in infected mice led to higher titers of the virus in lungs, increased mortality and more severe disease [75] . Depending on their antigen specificity, CTLs may be subtype-specific or, in case they recognize the internal antigens, they are broadly cross-reactive with various influenza A viral strains. Early studies in mice showed that the majority of influenza-specific CTLs were reactive across subtypes [76, 77] , what underlines their important role in heterosubtypic immunity. This high crossreactivity is explained by the antigenically conserved targets of CTL represented mostly by inner influenza antigens (e.g. NP, M1 and PB1, PB2) [78] [79] [80] [81] . However, some conserved T-cell epitopes were identified also on variable surface influenza antigens [82] [83] [84] . Recent data support the beneficial role of T-cell response in reducing the severity of infection also in humans [85] [86] [87] [88] . Additionally, cross-reactive CTLs recognized different subtypes of influenza A virus and their protective effect was shown also in individuals, who did not have specific antibodies against a given influenza virus they were exposed to [89] . Therefore, vaccination strategies focused on generating T-cellmediated immune responses directed towards conserved epitopes of influenza virus are also considered. T-cell epitopes are intensively studied as an alternative to the current vaccine strategy based mainly on the induction of the strain specific virus-neutralizing antibodies. Identification of conserved CTL epitopes shared by many influenza strains could represent the basis of vaccination strategies. This approach would be beneficial in the case of annual influenza epidemic and a potential pandemic, when humoral immunity is poorly or not protective due to the absence of pre-existing antibodies against emerging strains in the population [90, 91] . While CTL mediated immunity is considered to be weak, epidemiological data indicate induction of crossprotective immunity in humans, who overcame influenza infection in the past [85] . It was shown that memory T-cells against the conserved epitopes confer protection from the infection with the virus strains of different subtypes in humans [82, 85, 86, 88, 89, 92, 93] . Studies in mice demonstrated that, similar to the live influenza vaccine, adenovirus-based vaccine and DNA immunization induced CTL cross-protective immune response against infection with multiple influenza A subtypes [94] [95] [96] [97] [98] . The variable rate of cross-reactive CTL response was achieved also by using adjuvants, or various formulations and delivery systems with experimental influenza vaccines in preclinical animal studies [reviewed in [99] ]. It was shown that application of virus-like particles or virosomal vaccines could be successfully used for efficient delivery of multiple CTL epitopes to the target cells resulting in induction of CTL response [100, 101] . Heterosubtypic immunity mediated by CTLs was described in naturally infected humans [88, 89, 102] . It is developed mainly against conserved epitopes of NP, M1 and NS1 [82, [103] [104] [105] [106] . Kreijtz et al. showed that virusspecific CTL developed in humans as a response to previous exposition to seasonal influenza A viruses of the H3N2 and H1N1 subtypes displayed considerable cross-reactivity also with avian influenza viruses (e.g. A/H5N1) [86] . Thus, it could be supposed that obtained pre-existing T-cell immunity in humans may help to decrease the severity of infection during a pandemic outbreak in comparison to those individuals, who lack cross-reactive influenza specific CTL populations [86, 88, 107] . Therefore, vaccines based on conserved CTL epitopes represent a reasonable approach to generate effective broadly protective cellular immunity against influenza viruses of various subtypes. To develop vaccines capable of stimulating effective T-cell response, it is necessary to understand the factors contributing to the immunodominance of CTL epitopes. During viral infection, a large number of peptides are generated by processing of viral proteins in the proteasomes of infected cells. Only a small fraction of these peptides are presented by MHC class I molecules and subsequently recognized by specific CTL. This hierarchy of CTL response proved in animals [108] and in humans [104] is called immunodominance. There are several factors, which contribute to this phenomenon: HLA haplotype and its binding affinity to individual epitopes, repertoire of T-cell receptors, processing and presentation of viral peptides and interaction of CTL with antigen-presenting cells [109, 110] . It was shown that efficiency of epitope processing is one of the dominant factors affecting immunogenicity of multi-epitope vaccine [111, 112] . The most frequently used models for such immunological studies are inbred mice, like B57BL6 (H-2 b ) or BALB/c (H-2 d ) mice. Therefore, T-cell influenza specific epitopes in inbred mice were studied by many authors. Comprehensive analysis regarding existing influenza A epitopes in mice among avian and human influenza strains was done by Bui et al. [113] . However, not all Tcell epitopes are equally immunogenic. In inbred mice B57BL6 (H-2 b ), peptides from nucleoprotein D b NP 366-374 and from a subunit of viral RNA polymerase D b PA 224-233 are immunodominant, while nucleoprotein epitope K d NP 147-155 is immunodominant in BALB/ c (H-2 d ) mice [84, [114] [115] [116] . In contrast to inbred mice, the search for CTL epitopes suitable for development of CTL epitope-based vaccine in humans is more complicated [113] . The main reason is that HLA genes in humans are extremely polymorphic. Therefore, the knowledge of HLA restriction in population, which will be vaccinated, is necessary. The complexity of HLA molecules could be reduced by clustering them into sets of molecules that bind largely overlapping peptides. Such clustering was introduced by Sette and Sidney in 1999. They defined HLA supertypes as a set of HLA molecules that have similar peptide binding motifs and overlapping peptide binding repertoires [117] . Nine different supertypes (A1, A2, A3, A24, B7, B27, B44, B58, B62) were defined on the basis of their specifity for the main anchor positions of presented peptides. Later, other three HLA I supertypes (A26, B8 and B39) were described by Lund et al. [118] . Recent analysis provided an update of HLA I alleles classification into supertypes and is expected to facilitate epitope identification and vaccine design studies [119] . An example of most frequently recognized conserved epitopes of influenza antigens in humans represents M1 58-66 CTL epitop, which is restricted by the high prevalence allele HLA-A*0201 and could be a promising vaccine candidate [120] . Computer programs available today can predict binding epitopes of a given protein for the most common HLA allele [121, 122] . In silico analysis supports the proposition that the T-cell response to cross-reactive T-cell epitopes induced by vaccination or seasonal viral exposition may have the capacity to attenuate the course of influenza infection in the absence of cross-reactive antibody response [123, 124] . The ability to predict the CTL epitope immunogenicity and recognition patterns of variant epitopes enhances the probability of the optimal selection of potential targets of immune response and can be utilized for vaccine design [93, 113, 125] . In spite of the differences in various classification schemes, the concept of HLA supertypes has been effectively used to characterize and identify promiscuously recognized T-cell epitopes from a variety of different disease targets, as are those of hepatitis C virus [126, 127] , SARS [128] or HIV [129, 130] but also influenza virus [131] . A critical requirement for CTL epitope-based strategy is to identify and select promiscuous CTL epitopes that bind to several alleles of HLA supertypes to reach maximal population coverage. The utilization of supertyperestricted epitopes, which bind with significant affinity to multiple related HLA alleles, provides solution to this problem [117] . As described before, 80-90% population coverage can be achieved in most prominent ethnicities by focusing on only three major HLA class I supertypes -A1, -A3 and -B7 [132, 133] . By including two additional supertypes (A1, A24), 100% population coverage in all major ethnicities could be reached [117, 132] . Recently, HLA class I -A2, -A3 or -B7 supertype-restricted epitopes conserved among different viral subtypes of influenza virus were identified, what could be of relevance for the development of a potential supertype-restricted, multiepitope CTL-based vaccine protective against any subtype of influenza virus [82, 103, 113, 134] . One of the drawbacks of currently available inactivated vaccines is the lack of broad cross-protective humoral and cell-mediated immune response against any influenza virus. Their efficacy is limited due to the genetic variation of influenza viruses. Therefore, their annual reformulation is necessary in an attempt to antigenically match the currently circulating strain for each of the three vaccine strains or their subunits (HA and NA of H1N1 and H3N2 of influenza A virus as well as of influenza B virus) from which they are composed. Increasing amount of information about conserved epitopes of influenza viruses brings us closer to the development of the universal vaccine. Such vaccine should contain both, conserved B-cell epitopes that are important for induction of cross-protective antibodies and CTL epitopes for the involvement of the cellular arm of the immune response to the overall protective effect [90] . It was shown that the pre-existing memory T-cell immunity as defense against seasonal influenza strains may have the capacity to moderate the course of disease in the case of newly emerging flu viruses in the absence of cross-reactive antibody response [86, 93, 123, 124] . It was also shown that it would be possible to elicit the CTL response simultaneously directed against multiple supertype-restricted conserved CTL epitopes [135] [136] [137] [138] [139] . This could be relevant for the development of a potential supertype-restricted multiepitope CTL based vaccine, with the effort to reach wide population coverage. Even though recent reports support a beneficial role of T-cell response in reducing human infections [86] [87] [88] 124] , there are still many questions regarding the feasibility of designing an effective supertype-restricted CTL epitope based vaccine in humans. In addition to CTL epitopes, B-cell epitopes from conserved influenza antigens that can elicit cross-protective humoral response should also be considered as a component of novel vaccines. Recently, highly cross-reactive monoclonal antibodies directed against conserved epitopes of HA2 subunit, including fusion peptide, were identified [28, 30, 66, [68] [69] [70] . HA2 subunit region as well as M2 protein are promising candidates for design of vaccine constructs aimed at providing broad-spectrum immunity to influenza viruses [1, 28, 31, 37, 45] . Cross-protective potential of HA2 and eM2 could be increased by optimization of their delivery and immunogenicity using vaccine vectors that target multiple Toll-like receptors for efficient stimulation of innate immunity and subsequent enhancement of the adaptive immune response [41, 140] . Conserved B-and T-cell epitopes, thus, could represent the basis for preparation of universal vaccine and bring new hope for development of pandemic or universal influenza vaccine.

The intracellular dynamic of protein palmitoylation

S-palmitoylation describes the reversible attachment of fatty acids (predominantly palmitate) onto cysteine residues via a labile thioester bond. This posttranslational modification impacts protein functionality by regulating membrane interactions, intracellular sorting, stability, and membrane micropatterning. Several recent findings have provided a tantalizing insight into the regulation and spatiotemporal dynamics of protein palmitoylation. In mammalian cells, the Golgi has emerged as a possible super-reaction center for the palmitoylation of peripheral membrane proteins, whereas palmitoylation reactions on post-Golgi compartments contribute to the regulation of specific substrates. In addition to palmitoylating and depalmitoylating enzymes, intracellular palmitoylation dynamics may also be controlled through interplay with distinct posttranslational modifications, such as phosphorylation and nitrosylation.

Cellular proteins undergo a vast array of dynamic and static modifications to their amino acid backbone, which can dramatically extend and regulate functional output. Some of these modifications, phosphorylation in particular, have been the subject of intensive research, and their regulation and intracellular functions are extensively characterized. In contrast, our understanding of S-palmitoylation (hereafter referred to as palmitoylation), the attachment of palmitate (C16:0) or other long chain fatty acids to cysteine residues, has lagged behind some of its more popular cousins. Indeed, enzymes that catalyze palmitoylation reactions have only recently been identified after landmark studies in yeast, settling a long-running debate as to whether this process was predominantly enzyme mediated or spontaneous (Lobo et al., 2002; Roth et al., 2002; Fukata et al., 2004; Keller et al., 2004) . Palmitoylation is one of a group of lipid modifications (collectively termed lipidation) that appears on eukaryotic proteins and includes the common N-myristoyl and isoprenyl modifications. N-myristoylation describes the addition of myristic acid (C14:0) to a glycine residue with an exposed NH 2 group after cleavage of the immediately adjacent initiating methionine (Zha et al., 2000; Resh, 2006a) . This process is predominantly cotranslational, mediated by soluble enzymes, and has a strict consensus sequence (MGXXXS/T). N-myristoylation can also occur posttranslationally, notably after caspase-mediated protein cleavage during programmed cell death (Zha et al., 2000) . Prenylation is a posttranslational process also catalyzed by soluble enzymes, involving the attachment of farnesyl or geranylgeranyl isoprenoids to a C-terminal cysteine present within a defined consensus sequence (Wright and Philips, 2006) . Unlike the catalysts of N-myristoylation and prenylation reactions, the enzymes that mediate palmitoylation are polytopic membrane proteins (Fukata et al., 2004; Mitchell et al., 2006) , implying that cellular palmitoylation reactions occur at the cytosol-membrane interface. There are 23 putative S-palmitoyl transferases in mammals, characterized by the presence of a DHHC (aspartate-histidine-histidine-cysteine) motif within an 50 amino acid cysteine-rich domain. The large number of these DHHC proteins coupled with their localization to distinct membrane compartments (Ohno et al., 2006) implies that the cellular palmitoylation machinery is a highly regulated and coordinated system. There is no strict consensus sequence for palmitoylation, however, palmitoylated cysteines do share some common characteristics: (a) they are often adjacent to myristoylation and prenylation sites, (b) the surrounding amino acids tend to be basic or hydrophobic, and (c) they are frequently located in the cytoplasmic regions flanking transmembrane domains (or within transmembrane domains). An elegant combined proteomic and genetic analysis in yeast revealed that some DHHC proteins appear to exhibit a preference for a particular class of substrate, e.g., transmembrane proteins or myristoylated/ prenylated proteins (Roth et al., 2006) . Similarly, some palmitoylated yeast proteins displayed a marked dependence on a specific DHHC protein (of which seven are expressed in yeast). However, this study also revealed that DHHC proteins can have overlapping substrate specificities, which is consistent with previous studies in mammalian systems showing that specific substrates can be palmitoylated by more than one DHHC protein (Fukata et al., 2004; Fang et al., 2006; Fernández-Hernando et al., 2006; Greaves et al., 2008 Greaves et al., , 2010 Tsutsumi et al., 2009; Shmueli et al., 2010) . S-palmitoylation describes the reversible attachment of fatty acids (predominantly palmitate) onto cysteine residues via a labile thioester bond. This posttranslational modification impacts protein functionality by regulating membrane interactions, intracellular sorting, stability, and membrane micropatterning. Several recent findings have provided a tantalizing insight into the regulation and spatiotemporal dynamics of protein palmitoylation. In mammalian cells, the Golgi has emerged as a possible super-reaction center for the palmitoylation of peripheral membrane proteins, whereas palmitoylation reactions on post-Golgi compartments contribute to the regulation of specific substrates. In addition to palmitoylating and depalmitoylating enzymes, intracellular palmitoylation dynamics may also be controlled through interplay with distinct posttranslational modifications, such as phosphorylation and nitrosylation. hydrophobicity/membrane affinity and preference for ordered cholesterol-rich membrane domains or rafts (Melkonian et al., 1999; Brown, 2006) . A central function of palmitoylation is the regulation of protein sorting (Greaves et al., 2009b) , and it might be predicted that this simple lipid modification would have a very limited repertoire of effects on this process. However, this is not the case, and palmitoylation has been shown to act as a highly versatile sorting signal, which regulates protein trafficking to many distinct intracellular compartments. Several studies have highlighted regulatory effects of palmitoylation on either retention or anterograde trafficking of proteins at the ER-Golgi or protein cycling within the endosomal/lysosomal system (Linder and Deschenes, 2007; Greaves et al., 2009b) . Recent studies have extended our knowledge of the functional effects of palmitoylation on protein sorting by highlighting novel roles for this modification in targeting to cilia and flagella (Emmer et al., 2009; Cevik et al., 2010; Follit et al., 2010) and to the cleavage furrow of dividing cells (Hannoush and Arenas-Ramirez, 2009) . What is the underlying mechanistic basis for the effects of palmitoylation on protein sorting? For some soluble proteins, palmitoylation may be a passive sorting signal, acting only as an essential membrane anchor that allows other domains of the protein to regulate subsequent trafficking. However, palmitoylation also has active effects on protein sorting, achieved by partitioning of proteins into cholesterol-rich membrane subdomains or rafts (Levental et al., 2010) , changing protein orientation at the membrane and thus affecting protein-protein interactions (Hayashi et al., 2005; Lin et al., 2009) , regulating the ubiquitination status of proteins and thereby modulating ubiquitinationdependent protein sorting (Valdez-Taubas and Pelham, 2005; Linder and Deschenes, 2007; Abrami et al., 2008) , or affecting the transmembrane orientation of palmitoylated proteins in the membrane bilayer (Abrami et al., 2008) . Palmitoylation often couples with either N-myristoylation or prenylation to regulate membrane interactions of soluble proteins. N-myristoylation or prenylation of proteins in the cell cytosol provides a degree of hydrophobicity, although single lipid modifications are only sufficient for transient membrane interaction (Shahinian and Silvius, 1995) . In contrast, two closely positioned lipid modifications promote stable membrane attachment (Shahinian and Silvius, 1995) . Thus, at the cellular level, single myristoyl or prenyl chains facilitate transient membrane association, which is sufficient to allow access to membranebound DHHC proteins; subsequent palmitoylation by DHHC proteins will then promote stable membrane binding by inhibiting membrane dissociation (kinetic trapping; Fig. 1 A) . In this way, palmitoylation is essential for stable membrane association of many proteins, including farnesylated Ras and myristoylated G subunits (Hancock et al., 1990; Linder et al., 1993; Parenti et al., 1993) . Several soluble lipidated proteins are modified exclusively by S-palmitoylation, and these proteins may use an intrinsic weak membrane affinity for transient membrane interaction before palmitoylation (Greaves et al., , 2009a . Therefore, it is clear that a major function of palmitoylation is to mediate stable membrane attachment of soluble proteins. However, the effects of this posttranslational modification are more complex and diverse than that of a simple membrane anchor, and indeed, many transmembrane proteins are also palmitoylated. Analyses of different proteins and systems have revealed that palmitoylation has many distinct effects on modified proteins, regulating protein trafficking, protein stability, membrane microlocalization, and protein-protein interactions (Resh, 2006a,b; Greaves and Chamberlain, 2007; Linder and Deschenes, 2007; Greaves et al., 2009b; Noritake et al., 2009) . Although these effects of palmitoylation appear diverse, they are likely determined by two particular properties of palmitate: Figure 1 . Regulation of membrane binding and trafficking of peripheral proteins by palmitoylation. (A) Proteins modified with single lipid groups (prenylation or N-myristoylation; green circles) have a weak membrane affinity that allows transient membrane interaction. Palmitoylation by membrane-bound DHHC proteins promotes stable membrane association by kinetic trapping (Shahinian and Silvius, 1995) . Note that some peripheral proteins are exclusively palmitoylated, and these proteins were suggested to interact with membranes before palmitoylation by way of an intrinsic weak membrane affinity (Greaves et al., , 2009a . (B) Palmitoylation of Ras-farnesyl by Golgi-localized DHHC proteins leads to a dramatic increase in membrane affinity by kinetic trapping. This increased membrane residency facilitates entry of palmitoylated Ras (red circles) into transport vesicles that deliver it to the plasma membrane. It is possible that palmitoylation also serves to move Ras into cholesterol-rich domains from which Golgi exit vesicles are formed . Depalmitoylation of Ras can occur anywhere in the cell, perhaps modulated by Apt1, resulting in membrane release and cytosolic diffusion before repalmitoylation at the Golgi. For simplicity, the figure only depicts depalmitoylation occurring at the plasma membrane. This palmitoylation/depalmitoylation regulation of protein sorting is not specific for Ras proteins and may be a common mechanism underlying the sorting of many peripheral palmitoylated proteins (Kanaani et al., 2008; Tsutsumi et al., 2009; Rocks et al., 2010) . this approach indicated a half-life for N-Ras palmitoylation of 20 min (Magee et al., 1987) and showed that palmitoylation of G subunits of heterotrimeric G proteins can be modulated by agonist stimulation (Degtyarev et al., 1993; Mumby et al., 1994; Wedegaertner and Bourne, 1994) . Recently, the intracellular dynamics of palmitoylation and palmitoylation-dependent protein trafficking have been investigated using fluorescence imaging. The power of biochemical radiolabeling techniques comes from the ability to selectively follow the fate of a labeled pool of protein by pulse chase. In contrast, conventional confocal microscopy analysis of fluorescent proteins is not well suited to allow older proteins to be distinguished from newly synthesized proteins, static protein localizations to be differentiated from dynamic trafficking, and resolution of palmitoylated and unpalmitoylated proteins. However, these issues can be addressed by synchronizing the protein under investigation or by selective photoactivation/ photobleaching of a specific pool of the protein or by using visible membrane accumulation of soluble proteins as an indicator of palmitoylation. These techniques were successfully used in the landmark publications that detailed the palmitoylationdependent cycling pathway of Ras proteins (Goodwin et al., 2005; Rocks et al., 2005) . Recent studies used microinjection of semisynthetic N-Ras as a means to study real-time spatiotemporal dynamics of palmitoylation and membrane targeting (Rocks et al., 2005 ; microinjection of fluorescent protein allows the analysis of a synchronized (i.e., chemically identical and with the same initial localization) pool of protein. This elegant approach involves coupling chemically synthesized farnesylated peptides encompassing the C-terminal membrane targeting domain of N-Ras to a Cy3-labeled protein consisting of the remainder of N-Ras (amino acids 1-181) via a maleimidocaproyl linker. By using these semisynthetic farnesylated proteins with intact or mutated palmitoylation sites, high-resolution temporal dynamics together with spatial information on cellular palmitoylation and depalmitoylation reactions were open to investigation. Immediately after microinjection, farnesylated N-Ras displayed a dispersed localization, which is consistent with its presence in the cytosol and transient association with intracellular membranes . However, a rapid enrichment of the Cy3-labeled construct became apparent at the Golgi region (t/2 14 s) with plasma membrane staining visible at later time points. The simplest interpretation of these observations is that palmitoylation of the farnesylated N-Ras occurs at the Golgi, promoting early accumulation at this compartment, and is followed by anterograde transport to the plasma membrane. This notion agrees well with previous analyses of Ras protein cycling (Goodwin et al., 2005; Rocks et al., 2005) . Nevertheless, it should be noted that palmitoylation of the microinjected semisynthetic Ras protein was not directly assayed in this study. Although technically challenging, it will therefore be of major interest in follow-up studies to perform correlative measurements of palmitoylation and intracellular localization of microinjected Ras. In addition, a recently published study reported that palmitoylated Ras proteins traffic from Golgi to recycling endosomes en route to the plasma membrane (Misaki et al., 2010) . In contrast to other static lipid modifications, the versatility of palmitoylation as a membrane interaction and protein sorting module is greatly enhanced by its reversibility. It has long been appreciated that the palmitoylation of many (but not all) proteins is dynamic (Magee et al., 1987) and can be modulated in response to cell stimulation (Degtyarev et al., 1993; Mumby et al., 1994; Wedegaertner and Bourne, 1994) . Although palmitoylation dynamics of transmembrane proteins can impact sorting to distinct membrane compartments, depalmitoylation of soluble proteins can also mediate membrane release and cytosolic diffusion. Thus, the rapid palmitoylation and depalmitoylation dynamics of many proteins add an extra level of complexity to the effects of this posttranslational modification on protein sorting. The effects of palmitoylation-depalmitoylation dynamics have been extensively analyzed for palmitoylated Ras isoforms (Goodwin et al., 2005; Rocks et al., 2005; Roy et al., 2005) . Farnesylated H-and N-Ras exhibit a weak membrane affinity that mediates interaction with Golgi membranes. Palmitoylation of Ras at this compartment promotes stable membrane association and trafficking from the Golgi to the plasma membrane. Subsequent depalmitoylation releases Ras from the plasma membrane into the cytosol, and the process of palmitoylation at the Golgi and trafficking to the plasma membrane is repeated ( Fig. 1 B) . This palmitoylation cycle achieves a constant flux of Ras proteins from Golgi to plasma membrane and was suggested to prevent spillover of Ras onto other cellular membranes by constantly resetting the intracellular localization (Goodwin et al., 2005; Rocks et al., 2005) . The Ras cycling pathway is an example of how constitutive palmitoylation dynamics coordinate protein sorting. For some proteins, palmitoylation turnover and corresponding changes in intracellular localization can also be regulated by cell activity. Protein palmitoylation dynamics that are subject to regulation have been particularly well characterized for postsynaptic proteins; palmitoylation of the molecular scaffold PSD95 and AMPA and NMDA glutamate receptors are all regulated by synaptic activity (El-Husseini et al., 2002; Hayashi et al., 2005 Hayashi et al., , 2009 ). PSD95 is a peripheral palmitoylated protein that coordinates protein clustering at postsynaptic sites. This protein was suggested to display a decreased palmitoylation in response to glutamate receptor activation, which corresponded with a reduced synaptic clustering of both PSD95 and AMPA receptors. Palmitoylation of GluR1/GluR2 subunits of AMPA receptors and NR2A/2B subunits of NMDA receptors is also modulated by synaptic activity. The palmitoylation status of GluR1 affects association with the 4.1N protein, regulating activity-dependent receptor internalization and plasma membrane insertion dynamics (Hayashi et al., 2005; Lin et al., 2009) . Dynamic palmitoylation of NR2 subunits also modulates cell surface expression of the NMDA receptor ). Thus, neurons use dynamic palmitoylation as a response mode to couple changes in synaptic activity to changes in protein localization. Classically, palmitoylation dynamics have been studied using radiolabeled palmitate and pulse-chase protocols; for example, proteins or could instead reflect a different rate of palmitate turnover and subsequent Golgi recycling (Fig. 2) . Interestingly, proteins in this category, including Fyn, TC10, R-Ras, RhoB, and Rap2C all displayed colocalization with a Golgi marker when vesicular traffic through the secretory pathway was blocked by low temperature . This observation implies that after their synthesis, these proteins traffic via the Golgi and is consistent with the notion that the Golgi may be a specialized reaction center for the palmitoylation of all newly synthesized peripheral proteins and for a subset that undergo rapid depalmitoylation dynamics. The suggestion that palmitoylation of newly synthesized peripheral proteins is a Golgi-specific event is largely consistent with previously published data. Although DHHC palmitoyl transferases associate with a range of organelles, including the ER, Golgi, and plasma membrane (Ohno et al., 2006; Greaves et al., 2010) , independent analyses of DHHC substrate pairs have returned a high percentage of Golgi DHHC proteins as positive hits (Fukata et al., 2004 Huang et al., 2004; Fernández-Hernando et al., 2006; Greaves et al., 2008 Greaves et al., , 2010 Tsutsumi et al., 2009) . Indeed, DHHC9 displays specificity toward H-and N-Ras (Swarthout et al., 2005) and is Golgi localized (Swarthout et al., 2005) . However, Rocks et al. (2010) argued against any rigid requirements for DHHC substrate specificity and reported that DHHC9 knockdown (at the mRNA level at least) did not affect the intracellular localization of H-Ras or recovery kinetics after photobleaching of Golgi-localized H-Ras. To examine more closely whether the palmitoylation domain of N-Ras contained a recognition motif for interaction with specific DHHC proteins, the amino acids around the Rocks et al. (2010) did not report association of microinjected Ras proteins with recycling endosomes, and it will be important to clarify this apparent discrepancy between these studies. The progressive pattern of localization of microinjected Ras (i.e., dispersed → Golgi → plasma membrane) was also observed for semisynthetic constructs in which the N terminus of N-Ras (1-181) was ligated to either myristoylated peptides derived from G i1 or Fyn or to the N-terminal 20 amino acids from GAP43, all of which contain palmitoylation sites. This result suggests that palmitoylation of N-Ras at the Golgi is not dependent on farnesylation per se but may instead be a common feature of peripheral palmitoylated proteins with an underlying membrane affinity. Note that GAP43 lacks any other lipid modifications and had the slowest rate of Golgi accumulation (t/2 = 29 s); this protein may bind to membranes before palmitoylation via an intrinsic weak membrane affinity (Greaves et al., , 2009a . It will be particularly interesting in follow-up studies to examine whether microinjected full-length G i1 , Fyn, and GAP43 display identical intracellular localization dynamics to the peptide sequences that were fused to N-Ras. This issue is relevant because previous work has shown that regions distant from palmitoylation sites can have a marked influence on the specificity of interaction with DHHC proteins; for example, the yeast vacuolar protein Vac8 was palmitoylated specifically by Pfa3 in vitro, but the isolated palmitoylation domain from this protein was palmitoylated by all five yeast DHHC proteins that were examined (Nadolski and Linder, 2009) . Many other palmitoylated peripheral proteins do not display obvious steady-state localization at the Golgi but instead associate with the plasma membrane and endosomal membranes (Adamson et al., 1992; Kasahara et al., 2007; Sandilands et al., 2007) . The absence of such proteins from the Golgi might indicate that they are not modified by Golgi-localized DHHC Figure 2 . Regulation of protein localization by palmitoylation dynamics. The illustration depicts three palmitoylated proteins that have different rates of depalmitoylation. In this context, the term depalmitoylation refers to the complete absence of palmitoyl groups on the protein. Rapid depalmitoylation is associated with an enriched steady-state localization on Golgi membranes. This is achieved by depalmitoylation promoting membrane release and subsequent palmitoylation by Golgi-specific DHHC proteins leading to an accumulation at this compartment. Rapid depalmitoylation prevents excessive accumulation on endosomes via vesicular trafficking from the plasma membrane. In contrast, proteins that have a slower rate of depalmitoylation are maintained on membranes for longer and reach endosomal membranes via the plasma membrane. Note that a slower depalmitoylation rate may be achieved by a relative resistance to thioesterases, and/or the presence of many palmitoylated cysteines, and/or palmitoylation by DHHC proteins beyond the Golgi. All of these situations would limit the amount of the protein in a completely depalmitoylated state. This slower rate of depalmitoylation and membrane release limits the steady-state distribution on Golgi membranes. Ras palmitoyl transferase (Erf2) is localized to the ER (Bartels et al., 1999) . Thus, there currently appears to be marked differences in the palmitoylation pathways of mammalian and yeast Ras proteins, which might reflect reported differences in the intracellular trafficking itineraries of these proteins (Dong et al., 2003; Wang and Deschenes, 2006) . A key aspect of membrane targeting and cycling of dually lipidated proteins such as H/N-Ras, and the compartment where palmitoylation occurs, is the membrane interaction dynamics of the monofarnesylated protein. It is often implied that peripheral proteins bind to all or any intracellular membrane before palmitoylation; but is this the case? Does farnesylated Ras have an equal affinity for all intracellular membranes and, therefore, an unbiased membrane sampling before palmitoylation? In vitro experiments suggest that this may not be the case, as membrane interactions of farnesylated peptides are affected by membrane lipid composition (Gohlke et al., 2010) . Binding of farnesylated N-Ras to liquidordered membranes containing saturated phospholipids and cholesterol was reduced compared with liquid-disordered membranes made from an unsaturated phospholipid. These findings are particularly relevant given the high concentration of cholesterol in the plasma and endosomal membranes compared with membranes such as the ER (Mondal et al., 2009) . Semisynthetic proteins containing a single farnesyl or myristoyl group were suggested to interact with the plasma membrane based on total internal fluorescence microscopy analysis ; however, there was no indication of whether this interaction occurred with the same efficiency as interaction with ER-Golgi membranes. Visual inspection of the localization of farnesylated Ras proteins and peptides expressed in cells via plasmid transfection reveals clear ER and Golgi staining, whereas interaction with other membrane compartments (including the plasma membrane) is less obvious (Choy et al., 1999; Rocks et al., 2005 Rocks et al., , 2010 . Therefore, peripheral proteins may display a preference for interaction with specific membrane compartments rather than binding randomly to any intracellular membrane. This would clearly provide palmitoylation-dependent protein cycling with more specificity and directionality. The precise membrane interaction dynamics of peripheral proteins before palmitoylation therefore merits further high-resolution analysis. Although the Golgi appears to function as a hub for palmitoylation of newly synthesized and cycling peripheral proteins, certain proteins undergo dynamic palmitoylation remodeling without accessing the Golgi. This suggests that active DHHC proteins are localized in post-Golgi compartments. Indeed, DHHC2 and DHHC5 associate with the plasma membrane in neuroendocrine cells (Greaves et al., 2010) , and these proteins are present on post-Golgi membranes in neuronal dendrites and at the postsynaptic density Li et al., 2010) . SNAP25, a multiply palmitoylated peripheral protein, is modified by Golgi-localized DHHC proteins (Fukata et al., 2004 Huang et al., 2004; Greaves et al., 2009a Greaves et al., , 2010 , which palmitoylation site were changed to their stereoisomeric (i.e., D-amino acids) counterparts. The insertion of these D-amino acids, which was predicted to disrupt any specific protein-binding site, had no major effect on Golgi accumulation of the microinjected proteins . Further analysis suggested that there was also not an essential DHHC recognition domain in the remainder of the Ras protein. These observations suggest that the DHHC proteins that modify Ras do not require a specific signature of the palmitoylated domain or upstream region and implies that palmitoylation of peripheral proteins may not be restrained by tight enzyme substrate specificities; the major requirement for palmitoylation presumably being a suitable cysteine in close membrane proximity. However, this idea is not readily consistent with other studies that have highlighted features of both DHHC proteins and substrate proteins that contribute to specificity of interaction (Greaves et al., 2009a (Greaves et al., , 2010 Huang et al., 2009; Nadolski and Linder, 2009) or that identified a requirement for a specific DHHC protein for palmitoylation of a specific substrate (Roth et al., 2006; Ohyama et al., 2007; Stowers and Isacoff, 2007; Emmer et al., 2009; Huang et al., 2009; Noritake et al., 2009; Tian et al., 2010) . Indeed, depletion of a single DHHC protein, Erf2, had a marked impact on the palmitoylation of yeast Ras (Bartels et al., 1999; Roth et al., 2006) . Further work is clearly required to delineate the reasons for these apparent inconsistencies. If the Golgi really is a super-reaction center for palmitoylation of peripheral proteins in mammalian cells, it will be of great interest to determine how this is achieved. Are Golgilocalized enzymes highly efficient? Are cofactors required for palmitoylation enriched at the Golgi? Do peripheral proteins interact in a slightly different way with Golgi membranes, making them more susceptible to palmitoylation? Most peripheral palmitoylated proteins associate with post-Golgi membranes (particularly the plasma membrane) after palmitoylation. Thus, specific palmitoylation at the Golgi might be an important prerequisite to ensure plasma membrane trafficking. A particular feature of palmitate that might facilitate Golgi to plasma membrane trafficking is its affinity for cholesterol-rich lipid raft domains (Melkonian et al., 1999; Levental et al., 2010) . It was recently proposed that raft domains might act as platforms for vesicle budding from the Golgi , and thus, palmitoylation-dependent association with these domains would be predicted to promote protein traffic to the plasma membrane. In support of the idea that palmitoylation at the Golgi is important for correct trafficking of peripheral proteins, we recently identified a connection between the palmitoylation of cysteine-string protein (CSP) at the Golgi and its subsequent sorting . A mutant form of CSP with an enhanced membrane affinity associates with ER membranes and is only palmitoylated when ER and Golgi membranes are mixed by applying brefeldin A . Interestingly, however, after washout of brefeldin A, the Golgi recovers but the CSP mutant remains trapped at the ER , perhaps reflecting a disconnect between forward transport of peripheral palmitoylated proteins and the lipid environment of ER membranes (cholesterol poor). Although the Golgi has been highlighted as a possible hub for palmitoylation of peripheral mammalian proteins, the yeast In contrast to the wealth of information available on palmitoylating enzymes, our understanding of the proteins that regulate protein depalmitoylation is poor. Two main candidate thioesterases have been identified. Protein palmitoyl thioesterase 1 (Ppt1) depalmitoylates H-Ras and different G subunits in vitro (Camp and Hofmann, 1993) . Although there are reports that a cytosolic pool of Ppt1 may be present in cells (Kim et al., 2008) , this protein is thought to be predominantly localized to the lysosomal lumen (Hellsten et al., 1996) , where it is believed to function in depalmitoylation reactions occurring during protein degradation. Acyl protein thioesterase 1 (Apt1) reportedly displays thioesterase activity toward G i  1 , H-Ras, eNOS, and certain viral proteins (Duncan and Gilman, 1998; Yeh et al., 1999; Veit and Schmidt, 2001) but is inactive against other proteins such as caveolin (Yeh et al., 1999; Veit and Schmidt, 2001) . Importantly, Apt1 has a cytosolic localization, suggesting that it can regulate cellular palmitoylation dynamics. In support of this idea, overexpression of Apt1 into HEK293 cells was reported to increase the rate of removal of radiolabeled palmitate from G s  in pulse-chase experiments (Duncan and Gilman, 1998) . Despite Apt1 being identified many years ago, the physiological importance of this protein as a thioesterase is not clear. However, a recent study reported an important function for Apt1 in controlling dendritic spine volume, possibly by regulating palmitoylation and membrane localization of G 13 (Siegel et al., 2009 ). The recent description of a novel Apt1 inhibitor (palmostatin B) should provide an important tool to more finely dissect the function of this protein in cellular palmitoylation dynamics (Dekker et al., 2010) . Initial analysis with palmostatin B suggests that it promotes a moderate increase in Ras palmitoylation and disrupts the intracellular localization of this protein. Although our understanding of the mechanisms of depalmitoylation is limited, the spatiotemporal dynamics of this process were investigated by microinjection of semisynthetic N-Ras containing both farnesyl and palmitoyl chains . Previous analysis of an N-Ras protein in which the palmitoyl group was attached by a noncleavable thioether linkage revealed a dispersed intracellular localization without Golgi enrichment (Rocks et al., 2005) . This localization was suggested to reflect a requirement for active palmitoylation/ depalmitoylation cycling to achieve the correct localization of Ras. The palmitoylated protein with a cleavable thioester bond was therefore not expected to display any initial membranetargeting specificity. Despite this, the construct rapidly accumulated at the Golgi (t/2 = 27 s). This Golgi accumulation was suggested to follow on from binding of the farnesylated/ palmitoylated protein to any membrane, depalmitoylation, cytosolic diffusion, and subsequent repalmitoylation at the Golgi. There were two main interpretations made from this behavior of farnesylated and palmitoylated N-Ras: (1) depalmitoylation must be very rapid to account for the speed of Golgi accumulation and (2) depalmitoylation must occur throughout the cell because if it was confined to a specific location, association of the farnesylated and palmitoylated protein with some membranes would be irreversible. promote stable membrane attachment of SNAP25 (Greaves et al., 2009a) . Recent work also reported that SNAP25 can be palmitoylated by DHHC2 but that this enzyme is unable to promote stable membrane attachment (Greaves et al., 2010) . This observation is consistent with the idea that palmitoylation of newly synthesized SNAP25 (which promotes membrane association) is restricted to the Golgi and that modification by post-Golgi DHHCs is only relevant once SNAP25 has been trafficked from Golgi to plasma membrane. This may reflect a weaker association between SNAP25 and DHHC2, such that productive interaction can only occur when SNAP25 is stably membrane associated. However, it is also consistent with the idea that unpalmitoylated SNAP25 may have a higher affinity for Golgi membranes than the plasma membrane, again reinforcing the importance of membrane affinities and preferences of peripheral proteins before palmitoylation (see previous paragraph). In hippocampal neurons, DHHC2 is associated with mobile dendritic vesicles of unknown origin, and total internal reflection microscopy suggested that inhibition of synaptic activity promotes an increase in DHHC2 levels either at or just beneath the plasma membrane . Interestingly, this movement of DHHC2 correlates with enhanced palmitoylation and synaptic clustering of PSD95 (El-Husseini et al., 2002; Noritake et al., 2009) , and depletion of DHHC2 was reported to inhibit the increase in synaptic clustering of PSD95 after synaptic blockade. This work illustrates that dynamic palmitoylation can be achieved without peripheral proteins (such as PSD95) visiting the Golgi. It is interesting to note that depletion of DHHC3, which is localized to the somatic Golgi, also inhibited synaptic accumulation of PSD95. However, in contrast with DHHC2 depletion, knockdown of DHHC3 had no effect on the activity-dependent increase in synaptic clustering of PSD95 . This suggests that Golgi-localized DHHC3 is involved in the initial palmitoylation of newly synthesized PSD95, before dendritic targeting. CSP, an important neuroprotective DnaJ chaperone, is palmitoylated by Golgi-localized DHHC enzymes (DHHC3, DHHC7, DHHC15, and DHHC17; Greaves et al., 2008) . In this regard, CSP is similar to most other peripheral palmitoylated proteins. Consistent with the analyses of CSP palmitoylation in mammalian cells, disruption of DHHC17 in Drosophila melanogaster resulted in a loss of palmitoylation and mislocalization of CSP (Ohyama et al., 2007; Stowers and Isacoff, 2007) . Surprisingly, however, DHHC17 does not exhibit a Golgi localization in Drosophila neurons but, instead, has a presynaptic distribution on synaptic vesicles or at the presynaptic plasma membrane (Ohyama et al., 2007; Stowers and Isacoff, 2007) . Although palmitoylation cycles have not been reported for CSP, it is possible that DHHC17 is important for regulating local palmitoylation dynamics of CSP in Drosophila presynaptic terminals. Finally, there is also strong evidence to show that peripheral membrane proteins can undergo palmitoylation beyond the confines of the Golgi in yeast cells. As discussed earlier, Ras is modified by ER-localized ERF2 (Bartels et al., 1999) , and palmitoylation and membrane association of the yeast vacuolar fusion protein Vac8 are also markedly reduced after depletion of the DHHC protein Pfa3, which is localized to the vacuole membrane (Hou et al., 2005; Smotrys et al., 2005) . modulated by distinct posttranslational modifications present on the target protein. Interplay between palmitoylation and phosphorylation was recognized many years ago for the  2 -adrenergic G proteincoupled receptor (Moffet et al., 1993) . Mutation of the palmitoylated cysteine in the C-terminal tail of this receptor led to an increased level of basal phosphorylation and a loss of coupling to Gs (O'Dowd et al., 1989; Moffet et al., 1993) . The effects of mutating the palmitoylation site were not caused by a loss of palmitoylation per se but rather by the increased phosphorylation of the palmitoylation-deficient mutant (Moffett et al., 1996) . Palmitoylation-phosphorylation interplay has also been reported to regulate trafficking of AMPA and NMDA receptor subunits Lin et al., 2009) . For GluR1 subunits of AMPA receptors, depalmitoylation was suggested to enable the more-efficient phosphorylation of neighboring serine residues. This phosphorylation in turn increased the interaction of GluR1 with 4.1N protein, which modulated plasma membrane internalization and insertion dynamics of AMPA receptors (Lin et al., 2009) . How does palmitoylation regulate phosphorylation? One potential mechanism is membrane insertion of palmitoylated cysteines restricting access of protein kinases to the adjacent phosphorylation sites (Fig. 3 A) . If palmitoylation can regulate phosphorylation, can phosphorylation regulate palmitoylation? The STREX variant of BK potassium channels contains a PKA phosphorylation site within the cytoplasmic C-terminal tail that mediates channel inhibition (Tian et al., 2001) . A recent study reported that the STREX variant also contains palmitoylated cysteines adjacent to the PKA If depalmitoylation is not restricted to a specific membrane compartment, this suggests (a) a common depalmitoylase with access to many membranes (consistent with the cytosolic localization of Apt1), (b) a group of depalmitoylating enzymes with wide membrane compartment coverage, or (c) nonenzymatic depalmitoylation. The same spatiotemporal pattern of localization was observed for a farnesylated/palmitoylated protein containing D-amino acids at the palmitoylation site , suggesting that depalmitoylation does not require a specific recognition sequence around this region. How would a cytosolic thioesterase like Apt1 recognize membrane-embedded thioester linkages without the presence of a defined consensus sequence? It is clear that much more research on Apt1 is required and that the identity of novel inhibitors will greatly facilitate this. One interesting angle is to determine whether the turnover of palmitate on cellular proteins correlates with the ability of Apt1 to promote depalmitoylation in vitro. For example, caveolin is not a substrate of Apt1 in vitro and does not undergo rapid dynamic palmitoylation in cells (Parat and Fox, 2001) , and the reverse is obviously true for proteins such as eNOS and Ras. Interplay between palmitoylation and other posttranslational modifications DHHC proteins are clearly master regulators of intracellular palmitoylation reactions, and thioesterases (such as Apt1) may be equally important. However, are these enzymes the only means of regulating palmitoylation? In fact, there is evidence that palmitoylation/depalmitoylation dynamics can also be (1) Negatively charged phosphate group prevents palmitoylation of an adjacent cysteine by blocking membrane interaction. (2) Palmitoylation-mediated membrane association prevents access of protein kinases to an adjacent phosphorylation site. (3) Phosphorylation could alter the depalmitoylation rate of a neighboring cysteine, e.g., by increasing access to a thioesterase enzyme. (B) Phosphorylation of a soluble protein prevents palmitoylation by inhibiting transient membrane interaction. (C, 1) Possible regulatory effects of nitrosylation on palmitoylation. Nitrosylation may prevent palmitoylation by direct competition for cysteine residues. (2) It is also possible that nitrosylation could directly displace palmitate. Note that the examples shown do not illustrate the full range of effects that phosphorylation might have on palmitoylation and vice versa. will be more physiologically relevant for palmitoylated proteins that interact either directly or indirectly with NOS enzymes. Landmark studies in yeast highlighted the DHHC protein family as the catalysts of intracellular membrane fusion reactions and provided an essential spark to the palmitoylation field. This, together with the continuing development of new methodologies and reagents, has served as a platform for rapid expansion in our understanding of the mechanisms, spatiotemporal dynamics, and outcomes of protein palmitoylation. Techniques such as acyl-biotin exchange and click chemistry have offered highly sensitive alternatives to radiolabeling for the study of cellular palmitoylation (Drisdel and Green, 2004; Martin and Cravatt, 2009; Yap et al., 2010) . These approaches also permit the palmitoylation status of proteins to be studied in situ without cell labeling (acyl-biotin exchange) or the analysis of spatial patterns of palmitoylated proteins by fluorescence imaging (Hannoush and Arenas-Ramirez, 2009 ). Furthermore, both techniques have facilitated analysis of the cellular palmitoylome in both yeast and mammalian cells (Roth et al., 2006; Kang et al., 2008; Martin and Cravatt, 2009 ). The recent development of Apt1 inhibitors is an important step toward further delineating the function of this protein and developing an enhanced understanding of cellular depalmitoylation dynamics (Dekker et al., 2010) . It is expected that further technological developments will maintain the rapid pace of palmitoylation research. In particular, the development of more specific and selective inhibitors against the DHHC protein family (Resh, 2006b) will be a key to delineating the individual functions of these proteins and their contribution to the spatiotemporal dynamics of cellular palmitoylation reactions. site that regulate plasma membrane binding of the cytosolic C terminus (Tian et al., 2008) . Introduction of a phosphomimetic mutation at the PKA phosphorylation site or activation of cellular PKA perturbed palmitoylation-dependent membrane association of the C-terminal tail of STREX. Importantly, the full-length channel lacking the palmitoylated cysteines in the STREX domain was no longer subject to PKA-mediated inhibition, implying that phospho-regulation of the STREX channel is achieved via changes in palmitoylation and membrane association of the cytoplasmic C terminus. Phosphorylation also appears to regulate palmitoylation of the cyclic nucleotide phosphodiesterase (PDE) isoform PDE10A2 (Charych et al., 2010) . The N terminus of PDE10A2 contains a palmitoylated cysteine residue (Cys11) that is essential for membrane anchoring and efficient dendritic transport in striatal neurons. Palmitoylation was perturbed when phosphomimetic mutations were introduced into a downstream PKA phosphorylation site . Does phosphorylation at Thr-16 actively promote depalmitoylation of Cys-11, or does it block palmitoylation of this cysteine? In fact, PKA activation markedly enhanced PDE10A2 phosphorylation but had no acute effect on membrane expression levels. This observation suggests that phosphorylation does not promote depalmitoylation and membrane release of PDE10A2 but, instead, likely inhibits membrane binding by blocking palmitoylation. This phosphoregulation of palmitoylation might be relevant to many palmitoylated peripheral proteins and could represent a mechanism to promote a shift toward the depalmitoylated state of a protein in the absence of active (thioesterase driven) depalmitoylation. Negatively charged phosphate groups are likely to inhibit palmitoylation of neighboring cysteines by interfering with membrane interactions before palmitoylation (Fig. 3) . Another posttranslational modification that may impact palmitoylation dynamics is nitrosylation. Nitric oxide (NO) is produced from l-arginine by NO synthase enzymes (NOS) and can directly modify cysteines by S-nitrosylation (Stamler et al., 1992) ; this modification might therefore regulate palmitoylation dynamics by direct competition. The NO donor SIN-1 inhibited the basal level and the isoproterenol-stimulated increase in palmitate incorporation into 2 adrenergic receptor (Adam et al., 1999) . Palmitate incorporation into H-Ras, caveolin, SNAP25, and certain viral proteins has also been reported to be modified by NO donors (Hess et al., 1993; Baker et al., 2000; Akerström et al., 2009) . Indeed, the NO donor S-nitrosocysteine accelerated removal of radiolabeled palmitate from H-Ras in pulsechase experiments (Baker et al., 2000) , raising the intriguing possibility that NO may directly displace palmitate from modified proteins (Fig. 3 C) . Overall, there is sufficient published data to suggest that NO may be an important regulator of palmitoylation dynamics. The development of more sensitive techniques to directly study nitrosylation is required to more rigorously delineate interplay between this modification and palmitoylation. It will be particularly important in future studies to determine how palmitoylation dynamics are affected by endogenously produced NO. The intracellular diffusion range of NO from its site of production is limited, and thus, interplay between palmitoylation and nitrosylation

The impact of sex, gender and pregnancy on 2009 H1N1 disease

Children and young adults of reproductive age have emerged as groups that are highly vulnerable to the current 2009 H1N1 pandemic. The sex of an individual is a fundamental factor that can influence exposure, susceptibility and immune responses to influenza. Worldwide, the incidence, disease burden, morbidity and mortality rates following exposure to the 2009 H1N1 influenza virus differ between males and females and are often age-dependent. Pregnancy and differences in the presentation of various risk factors contribute to the worse outcome of infection in women. Vaccination and antiviral treatment efficacy also vary in a sex-dependent manner. Finally, sex-specific genetic and hormonal differences may contribute to the severity of influenza and the clearance of viral infection. The contribution of sex and gender to influenza can only be determined by a greater consideration of these factors in clinical and epidemiological studies and increased research into the biological basis underlying these differences.

Sex and gender differences can affect exposure to pathogens, vulnerability to infectious diseases, health seeking behaviours and immune responses to pathogens, resulting in differences between males and females in the incidence, duration, severity and case fatality rates following an infection [1, 2] . Sex refers to the biological and physiological characteristics that define males and females, whereas gender refers to the roles, behaviours, activities and attributes that individual societies consider appropriate for men and women. The impact of sex and gender on infection is tied to the age of the individual, as both biological and cultural factors can change dramatically with age. Consideration of these factors can result in a more effective public health response to infectious diseases, including influenza, and yet they are often inadequately addressed in clinical and basic research studies. A systematic review of the literature regarding sex, gender, pregnancy and the 2009 H1N1 pandemic indicates these are important factors which alter the severity of the disease as well as the prevention and treatment measures. A greater awareness of how sex and gender impact upon the biology of 2009 H1N1 infection could provide important insights into the unique morbidity and mortality patterns associated with this pandemic. Virus An influenza pandemic was declared by the World Health Organization (WHO) in June 2009 and the virus, 2009 H1N1, became the primary influenza virus strain isolated from humans by the end of the winter influenza season in the southern hemisphere [3] . It was the dominant influenza A virus strain circulating in the northern hemisphere for the entire influenza season, effectively outcompeting both seasonal influenza A virus strains [3] . The pandemic has been termed mild due to the relatively low mortality. Confirmed influenza virus infections, however, have increased substantially compared to recent years and the US Center for Disease Control (CDC) estimates of the number of people infected with 2009 H1N1 are greater than what would be expected in a standard influenza season [3] . Most cases of severe disease and mortality after infection with seasonal influenza A virus occur in the ≥65 years population. In contrast, 2009 H1N1 has not been associated with a large number of infections in this age group but has the highest attack and hospitalization rates in individuals between the ages of 0-40. The reduced number of cases in those aged ≥65 stems in part from the fact that antibodies generated to pre-1950 H1N1 viruses cross react with 2009 H1N1, resulting in limited protection from 2009 H1N1 infection [3] . Presence of co-morbidities or risk factors for severe disease Several populations are at risk for severe disease from seasonal as well as 2009 H1N1 infection [4] , including individuals who have pre-existing illnesses or medical conditions, pregnant women, immunosuppressed individuals (either through treatment, HIV infection, or as a result of a pre-existing immunosuppressive disorder) and children aged 0-4 years. Medical conditions associated with an increased risk of severe disease include chronic respiratory disorders (for example, asthma, bronchitis, chronic obstructive pulmonary disease [COPD] and cystic fibrosis), neuromuscular disorders (for example, cerebral palsy, myasthenia gravis and muscular dystrophy), metabolic diseases (for example, diabetes) and chronic renal, heart or liver disorders [3] . Factors such as obesity and hypertension are not normally associated with severe disease from seasonal influenza but have been suggested as risk factors for severe disease from 2009 H1N1 in some studies [5] [6] [7] . The protective immunity induced by influenza vaccinations is mediated primarily by antibodies that recognize the viral haemagglutinin protein and neutralize virus infectivity. After virus infection, host innate immune responses, including production of cytokines and chemokines, are activated which initiate a cascade of immunological events that lead to the development of specific immune responses to the virus. Controlling and clearing influenza virus infection requires neutralizing antibodies and cell-mediated immunity (for example, activation of T cells) [3] . The influx of immune cells into an influenza-infected lung can lead to the overproduction of various cytokines and chemokines -often called a 'cytokine storm' -which can enhance the virus-induced lung damage resulting in severe illness. A limited number of studies suggest that an altered cytokine and chemokine response is contributing to severe 2009 H1N1 disease [8, 9] . Therefore, immunity to influenza viruses represents a balance between immune responses inducing protection and clearance of virus versus causing pathology. Utilizing published observational reports of patients with confirmed 2009 H1N1 infection and those admitted into intensive care units worldwide, the incidence, severity and case fatality rates following infection appear to differ between males and females, but often are age-dependent and vary between countries. The outcome of infection with 2009 H1N1 is generally worse for females, but the magnitude of this difference varies across geographical regions. Assessments of male-female differences in reported incidences of infection is confounded by two factors: (1) many countries do not disaggregate data by both sex and age which may mask sex differences among the age groups that are most likely to be exposed -children and young adults; and (2) the profound differences in health seeking behaviours between males and females [10] . Household transmission studies of children and adults reveal that being female (female relative risk [RR]: 1.87, 95% confidence interval [CI]: 1.17-2.73) is a significant factor associated with higher secondary attack rates of influenza-like illness, with attack rates being higher among children and young adults than older adults (>55 years of age) [11] . Reported male-female differences in the incidence of infection vary with age in several countries, with a higher incidence of infection with 2009 H1N1 in young women than young men of comparable age [12] [13] [14] [15] . While pregnancy has been clearly linked with increased disease severity, the vast majority of infected females of reproductive ages are not pregnant, suggesting that additional factors are contributing to the increased incidence of infection. In contrast, in Asia, the majority of reported H1N1 cases have been male (57.1%) [16, 17] . In China, males (male odds ratio [OR]: 1.94, 95% CI: 1.07-2.66) also shed the 2009 H1N1 virus in pharyngeal and nasopharyngeal samples for a longer duration than females [18] suggesting that the transmission potential may be higher in males. Other countries reported no male-female differences in the number of cases of 2009 H1N1, but did not analyse the data stratified by both age and sex [19] [20] [21] [22] [23] [24] [25] . One trend that appears consistent across more than 60% of the datasets evaluated is that more females are hospitalized with critical illness than males ( Figure 1 ). The first cases in the USA were in California (April-May 2009), where the a majority of hospitalized cases (21/26) were women, five of whom were pregnant [26] . Initial analyses of data from critically ill patients in the USA during the first wave reported no male-female difference [27] , but subsequent state-specific reports from the first and second waves illustrated differences between the sexes [28] [29] [30] . In Canada, a significant majority of critically ill patients have been young women (female RR: 1.3, 95% CI: 1.0-1.6) [5, 31] . Other countries also report that rates of hospitalization have been higher among females than males, with a majority of the females being of reproductive age (15-49 years of age) [14, [32] [33] [34] [35] [36] . Analyses of cohorts of patients in Mexico and Australia/ New Zealand revealed a trend for more females than males being hospitalized [6, 37] . Evaluation of these differences in some countries is confounded by age, as many studies do not report male-female differences according to age group [6, 27, 37] . An examination of sex differences disaggregated by age is needed in larger, more complete datasets. The reason for the greater proportion of hospitalized women is not known, but many cases involve co-morbid conditions, including chronic respiratory diseases (for example. asthma and COPD), which are often more severe in females [38] [39] [40] . Mortality from 2009 H1N1 is not common but data from South Africa, where the incidence of co-infection with HIV and tuberculosis is high, reveal that 65% of fatal cases were females of reproductive age, of whom almost half were pregnant [41] . RR of death is higher for young adult women (female RR: 1.5, 95% CI: 0.9-2.3) than men in Canada [31] . In Australia, 58% of fatal cases were male [35] and in Brazil and Peru case fatality rates have been equal between males and females [14, 20] . No consistent pattern of male-female differences in mortality from the 2009 H1N1 has emerged. Increased morbidity and mortality in pregnant women has been documented during influenza pandemics and influenza seasons where virus infection rates are particularly high [42] . Pregnant women represent a disproportionately higher percentage of severe cases with the increased risk ranging from four-to 10-fold greater compared with the general population ( Figure 2 ). Increased morbidity and mortality in pregnant women has been reported in many datasets [5, 14, 27, 28, 36, 37, 41, [43] [44] [45] . The disease course and clinical presentation [46] [47] [48] has been studied and comparisons of disease in pregnant women to age-matched non-pregnant women [37, 49, 50] or to the general population [51] have been made. Disease severity is increased during the second and third trimester. However, no clear clinical parameter has been associated with pregnancy-associated increased morbidity and mortality. There are no significant differences in the general symptoms of disease, the progression to viral pneumonia, acute respiratory distress syndrome (ARDS) or secondary bacterial pneumonia in pregnant women compared to the control populations. Severe disease was also associated with a greater than sixfold increase in adverse neonatal outcomes when compared with pregnant women suffering mild disease [49] . An increased risk of severe disease may be present during the early postpartum period but the reported number of cases is limited and requires additional investigation [37, 50] . Female to male ratios of hospitalization with confirmed 2009 H1N1 were calculated using published datasets [5, 6, 14, 16, [27] [28] [29] [30] [32] [33] [34] [35] [36] [37] 43, [134] [135] [136] . Pink bars = higher rates of hospitalization in females; blue bars = higher rates of hospitalization in males; grey bars = similar hospitalization rates in males and females. Details about sample sizes, time of data collection, and criteria for hospitalization are contained within each individual reference. Pregnancy itself is considered a risk factor for severe disease [3] but the biological basis for this has not been established. Pregnancy-associated changes in immune function, hormone levels, cardiopulmonary stress and difficulties in treatment for respiratory disease are often cited as important factors [52] . The presence of other risk factors may increase the risk of severe disease in a pregnant woman (Figure 3) . The presence of a known co-morbidity in pregnant women with severe 2009 H1N1 disease can vary greatly and has been documented as 16% to 56% [37, [46] [47] [48] [49] [50] 53] . There is no one or cluster of co-morbidities associated with increased disease severity in pregnant women. Data indicate that when no additional co-morbidities were present, pregnant women still had a seven-to tenfold higher rate of severe disease when compared to age-matched, nonpregnant women [49, 50] . Certain risk factors predispose patients to increased morbidity and mortality following exposure to influenza viruses [54] and the severity and prevalence of these underlying conditions often differ between males and females ( Figure 4 ). The 2009 H1N1 virus causes disproportionate disease among young adults, a population that has a distinct repertoire of risk factors associated with exposure and worse outcome following infection compared with very young or old. Healthcare workers, as well as those in frequent contact with young children, are at a higher risk of exposure to influenza viruses than the general public [55] . Women represent over 50% of the healthcare workforce in many countries and nurses, teachers of young children and day-care workers are predominantly female [10] which potentially leads to a gender-specific occupational risk for influenza acquisition. Hand hygiene compliance, one of the most effective ways to prevent transmission of influenza, is significantly better among female than male (male OR: 0.6, 95% CI: 0.4-0.98) healthcare workers [56] . Among healthcare workers in the USA, self-reported rates of use and knowledge about appropriate personal protective equipment in response to influenza are similar between the sexes [57] . Differences in health seeking behaviour or healthcare access may impact both the acquisition and manifestation of influenza. A WHO survey in 59 countries from 2002-2004 revealed that adult women are more likely to seek healthcare in both higher and lower income countries [10] . The quality of care for women in some parts of the developing world is not equal to that received by men [58] . In some developing countries, knowledge of the pandemic was higher among men than women, which might reflect the fact that there are greater educational opportunities and greater chances of socialization for men [59] . Chronic medical conditions predispose patients to increased influenza-related morbidity [54, 60] and malefemale, as well as pregnancy-associated, differences in disease prevalence have been reported. (1) Respiratory disease Asthma has been a significant underlying condition in children and adults hospitalized with critical illness [27, 61] . Data from the USA and Canada illustrate that, prior to puberty, boys have more asthma exacerbations than girls. However, this trend is reversed in adulthood [39] . Rates of asthma attacks, numbers of asthma-related emergency room visits, numbers of asthma-related hospitalizations and duration of hospitalization are higher in women than men in the USA [39, 62] . Rates of asthma, as well as incidence of asthma attacks, appear to be the same in pregnant and age-matched non-pregnant women [63] . Cystic fibrosis and COPD have been identified as risk factors for severe illness with 2009 H1N1 [3] and the [27] , Chicago, IL, USA [28] , California, USA [43] , New York, USA [49] , Australia and New Zealand [37] , Canada [5] and Brazil [14] . Estimates of the general population and pregnant woman are based on data from the US Census Bureau or the World Health Organization. progression of cystic fibrosis and long-term survival is significantly worse for females than males, especially among individuals diagnosed in childhood [64] . Females with COPD report worse symptoms, lower exercise capacity, more airway hyper-responsiveness and worse health-related quality of life than males [65, 66] . Although morbidity from these conditions may be worse in females, mortality -both from all causes and from respiratory-related disease alone -is still higher in males with COPD [65] , illustrating the complexities involved in assessing the significance of sex and gender for a particular co-morbidity. (2) Hepatic disease Chronic hepatic disease is a risk factor for severe 2009 H1N1 disease [3] . The development of hepatocellular carcinoma occurs at a 2:1 to 4:1 ratio for males to females [67] . The prevalence of serum hepatitis B virus (HBV) is consistently higher in men than women [68] . Males are more than twice as likely to die from liver cancer, which suggests that men may be more sensitive to the effect of HBV infection on the development of liver cancer [69] . Men also are twice as likely to develop cirrhosis [70] . (3) Cardiovascular disease The rates and severity of cardiovascular disease differ between the sexes and these differences have been evaluated in the elderly [71] . As they have not been identified as an at-risk population for severe disease from 2009 H1N1 influenza, sex differences in cardiovascular disease may not be a critical factor. (4) Metabolic disorders Diabetes and morbid obesity have emerged as novel risk factors for severe 2009 H1N1 disease [3] . The lifetime risk of diabetes is higher in women than men, at least in the USA where approximately 55% of all diabetic-related deaths are women which may be a reflection of the fact that women tend to live longer than men [72] . In the USA, gestational diabetes and rates of diabetes in obese adolescent girls have been increasing [73, 74] . Gestational diabetes occurs in up to 14% of all pregnancies [75] . Women, particularly those of lower socioeconomic status, also receive less adequate diabetes care than men of the same socioeconomic status [76] . Females, particularly in developing countries, tend to have higher rates of obesity [77] . According to the WHO, in 138 of 195 countries, females are over 50% more likely to be obese than males [78] . In some countries, the body mass index for women is 5-8 points higher than for men [78] . The higher rates of obesity and diabetes in females may be significant factors contributing to higher 2009 H1N1-related morbidity in women. It has not yet been determined whether prepregnancy obesity or excess weight gain during pregnancy represent equivalent risks. Precise parameters for documenting obesity in pregnant women have not been established [79] . Influenza in immunocompromised individuals is associated with an increased severity of disease [54] and HIV is recognized as a co-morbidity for 2009 H1N1 influenza [54, 80] . The rate of HIV in females are approaching that of males worldwide [81] . HIV RNA levels are consistently lower in women than men [82] . However, women have a 1.6-fold higher risk of progression to AIDS than men with equal viral loads [83, 84] . There also are gender disparities in access to care for women with HIV, with women traditionally having greater difficulty accessing treatment [85, 86] . Whether infection with HIV and progression to AIDS differentially affects the outcome of influenza virus infections in males and females has not been evaluated. The precise impact of sex, gender and pregnancy on responses to the 2009 H1N1 vaccines is not known [87] [88] [89] [90] . Data from clinical trials of seasonal influenza vaccines reveal pronounced sex differences in the rates of vaccination, antibody responses to the vaccines and adverse reactions to the vaccines and illustrate that these differences must be considered in response to the 2009 H1N1 vaccine. Seasonal influenza vaccination data further reveal that pregnant and non-pregnant women generate comparable immune responses and experience similar adverse side effects [54] . Available data on rates of 2009 H1N1 vaccination have not been analysed by sex [91] but rates of seasonal influenza vaccination vary significantly with respect to sex and age [92] [93] [94] . Rates of vaccination among women are lower than men in some European countries [92] and may reflect greater negative beliefs about the risks associated with vaccination [95] , differences in physician recommendations regarding vaccination or occupational differences. Among healthcare workers in China, 73% of women reported intentions to decline both the H5N1 and 2009 H1N1 vaccines compared to 64% of men [96] . In France, acceptance (either receipt or intention to receive) of the 2009 H1N1 vaccine was higher among men and was higher among pregnant women and other groups with co-morbid conditions [97] . In the USA and Canada, vaccination against seasonal and 2009 H1N1 influenza during pregnancy is recommended irrespective of trimester [54, 98, 99] . The vaccination rate of pregnant women against 2009 H1N1 virus has been estimated to be only 38%, which is still higher than that normally seen with seasonal influenza vaccine [91] . Numerous studies reveal that haemagglutination inhibition (HAI) titres following seasonal influenza vaccination are consistently higher in women than men of comparable ages [100] [101] [102] [103] [104] , which suggests that women may be better protected against influenza disease following vaccination than are men. Women aged 18-64 years generate a more robust neutralizing antibody response following vaccination than men [102] . Pregnant women appear to have similar responses to seasonal influenza vaccines compared to non-pregnant women. The National Institutes of Health reports that 47 out of 50 (94%) pregnant women immunized with 2009 H1N1 vaccine achieved antibodies levels considered to be protective within 21 days of inoculation [105] . Women report more severe local and systemic reactions to influenza virus vaccines [100, 102, 104, [106] [107] [108] . Women also experience worse reactions to vaccine adjuvants [109] , which should be considered for 2009 H1N1 vaccines that are administered with adjuvant [88, 89] . The extent to which adverse reactions to the 2009 H1N1 vaccine differ in either frequency or severity between males and females has not been reported [60] . Seasonal, H5N1 and MF-59-adjuvanted influenza vaccines are reported to be safe for pregnant women [110] [111] [112] . A study of 50 pregnant women who received the 2009 H1N1 vaccine reported it was well-tolerated with no significant adverse side effects documented [105]. Antivirals are an effective treatment following infection with influenza viruses when administered early during the course of disease. The 2009 H1N1 viruses analysed, to date, are all resistant to the adamantadine class of antivirals but remain sensitive to neuraminidase inhibitors [3] . Available data indicate that the rate of prescribing antivirals to seasonal influenza virus-infected individuals, ranging in age from infants to adults, is similar between males and females in the USA [113] [114] [115] . In contrast, inappropriate prescription of antibiotics for seasonal influenza is greater for women [114] . A meta-analysis of data from randomized, doubleblind clinical trials illustrates that, following treatment with oseltamivir, men return to their baseline wellness faster than women, suggesting that antiviral treatment for seasonal influenza may be more effective in men [116] . Whether this observation reflects patient reporting biases, need for differential drug doses or other confounding factors is not clear. These data do, however, indicate that sex and gender should be considered when evaluating the efficacy of antiviral treatment for 2009 H1N1. Prompt administration of neuraminidase inhibitors is recommended for any pregnant woman with influenzalike symptoms [3] . Administration of antivirals within 48 h of symptom onset correlates with a mild or uneventful disease course in pregnant women [47, 48, 51] . Pregnant women who do not take antivirals, or begin treatment >72 h after symptom onset, have significantly higher morbidity and mortality rates compared to those who have early antiviral treatment [48, 49] . Sex differences in the immune responses to influenza viruses have not been systematically examined [117] . Using data from other virus-host systems, several immunological, hormonal and genetic mechanisms have been identified as being differentially expressed between the sexes and altered during the course of pregnancy, which may account for male-female differences and pregnancyassociated increases in the severity of 2009 H1N1. Generally, women mount higher immune responses to viral infections [118] . Heightened antiviral immunity in women is beneficial for virus clearance, but may be detrimental if it becomes excessively high or prolonged, leading to pathology and even death. Over the course of pregnancy, inflammatory and antiviral immune responses are suppressed which can alter responses to viruses, such as influenza. Women are at a greater risk of progressing to AIDS than men, despite having significantly less HIV RNA in circulation and host-mediated pathology is hypothesized to contribute to this sex bias [82] . Plasmacytoid dentritic cells (pDCs) are significant producers of type I interferons (IFN-α), which signal the activation of cytotoxic T cells for the elimination of virally infected cells. pDCs from women react more strongly to HIV-1 encoded toll-like receptor 7 (TLR7) ligands than pDCs derived from men, resulting in higher levels of immune cell activation [83] . Women with higher progesterone (P4) concentrations have greater numbers of activated pDCs in response to the HIV TLR7 ligand than women with lower P4 concentrations [83] . Several genes (for example, the Tlr7 gene that encodes a receptor that recognizes RNA viruses, including influenza viruses) that encode for immunological proteins are on the X chromosome and may escape X inactivation, resulting in higher amounts of expression in women [117] . X chromosomal variation also alters the course of progression of AIDS differently in women than men [84] . Whether female-biased immunopathology contributes to the severity of 2009 H1N1 disease in women requires consideration. The prevalence of HBV, titres of HBV DNA and development of hepatocellular carcinoma are higher in males than females and involve the effects of hormones on viral and host gene expression [67, 68, 119] . Among HBV positive males, elevated concentrations of testosterone and expression of certain androgen receptor gene alleles correlate with an increased risk of hepatocellular carcinoma [120, 121] . In HBV transgenic mice, castration of males reduces, whereas replacement of testosterone in castrated males increases, serum HBsAg concentrations [122] . Chemically-induced hepatocellular carcinoma is more severe in male than female mice, which is mediated by increased inflammatory cytokine production by liver cells in males and can be reversed with oestradiol (E2) treatment [123] . Sex steroids modulate sex differences in the prevalence of HBV and development of liver cancer through effects on immune responses to HBV. Whether sex steroids affect the pathogenesis of influenza virus infection should be examined. The impact of sex steroids, including androgens, oestrogens and progesterone (P4), on the activity of immune cells may contribute to sex differences and the effects of pregnancy on responses to 2009 H1N1. Generally, androgens, including dihydrotestoesterone and testosterone, suppress the activity of immune cells [124] . The immunosuppressive effects of androgens may reflect the inhibitory effects of androgen receptor signalling mechanisms on transcriptional factors that mediate the production of pro-inflammatory and antiviral cytokines [125] . Oestrogens affect both innate and adaptive immune function. Oestradiol can have bipotential effects with low doses enhancing and high doses reducing proinflammatory cytokine production [126] . Low E2 concentrations promote helper T cell type 1 (Th1) responses and cell-mediated immunity and high concentrations of E2 augment helper T cell type 2 (Th2) responses and humoral immunity which may be responsible for some female as well as pregnancy-associated changes in immune responses [126] . Another oestrogen that affects the functioning of the immune system is oestriol (E3), which is produced during pregnancy by the placenta. When E3 levels are high, inflammatory responses and the symptoms of Th1mediated autoimmune diseases -including multiple sclerosis -are reduced [127, 128] . Whether the effects of pregnancy on responses to 2009 H1N1 reflect the effects of E3 on immune responses requires investigation. Progesterone suppresses innate immune responses [125, 129] . Elevated concentrations of P4 during pregnancy inhibit the development of Th1 immune responses that can lead to fetal rejection and promote production of Th2 immune responses [130, 131] . Progesterone also suppresses antibody production [132] . Recent data illustrate that pregnant women with severe 2009 H1N1 have lower levels of total IgG2 than healthy pregnant women or women with only moderate H1N1 disease [133] . As IgG2 levels are enhanced in a Th1dependent manner, this reduction in total IgG2 may be related to pregnancy-associated modulation of the immune response. As data from the pandemic continue to be analysed, a number of factors should be considered by clinicians, epidemiologists and scientists in order to better understand the role of sex, gender and pregnancy on 2009 H1N1 disease. • Age-and sex-associated differences in exposure and severity of infection must be documented, as many biological and behavioural differences occur over the course of the lifespan. • The outcome of infection is worse for females, but the magnitude of this difference varies across countries and the differential contribution of gender and sex in different regions of the world must be considered. • Excessively high innate and cell-mediated immune responses, including the production of cytokines and chemokines, may contribute to increased severity of influenza in females. • Higher antibody responses to influenza vaccines in females may lead to an increased protection from disease. • Sex should be considered when effective vaccine and antiviral dosages are determined in order to maximize efficacy while limiting adverse side effects. • As the outcome of influenza infection can be worse for females, efforts should be made to increase acceptance of vaccines in both pregnant and nonpregnant females. • The 2009 H1N1 infection of pregnant women needs to be studied carefully in order to determine the factors that are driving the increased morbidity and mortality rates. • Sex hormones have profound effects on the immune responses to vaccines and infection and should be examined in clinical samples and animal models. • Animal models of infection can provide important insights into the role of sex, pregnancy, and hormones on the immune response to vaccination, infection, and antiviral treatment. Abbreviations CI: confidence interval; COPD: chronic obstructive pulmonary disease; DC: dendritic cell; E2: 17β-oestradiol; E3: oestriol; HBV: hepatitis B virus; OR: odds ratio; P4: progesterone; pDC: plasmacytoid DC; RR: relative risk; Th1: helper T cell type 1; Th2: helper T cell type 2.

PhEVER: a database for the global exploration of virus–host evolutionary relationships

Fast viral adaptation and the implication of this rapid evolution in the emergence of several new infectious diseases have turned this issue into a major challenge for various research domains. Indeed, viruses are involved in the development of a wide range of pathologies and understanding how viruses and host cells interact in the context of adaptation remains an open question. In order to provide insights into the complex interactions between viruses and their host organisms and namely in the acquisition of novel functions through exchanges of genetic material, we developed the PhEVER database. This database aims at providing accurate evolutionary and phylogenetic information to analyse the nature of virus–virus and virus–host lateral gene transfers. PhEVER (http://pbil.univ-lyon1.fr/databases/phever) is a unique database of homologous families both (i) between sequences from different viruses and (ii) between viral sequences and sequences from cellular organisms. PhEVER integrates extensive data from up-to-date completely sequenced genomes (2426 non-redundant viral genomes, 1007 non-redundant prokaryotic genomes, 43 eukaryotic genomes ranging from plants to vertebrates) and offers a clustering of proteins into homologous families containing at least one viral sequences, as well as alignments and phylogenies for each of these families. Public access to PhEVER is available through its webpage and through all dedicated ACNUC retrieval systems.

Viruses are responsible for a large number of infectious diseases and cancers. Recently, new viral diseases have emerged leading to severe consequences on human activities. The emergence of many of these new viruses can be attributed to recombining viruses as well as to host species jump (1) (2) (3) . Therefore, understanding how viruses interact with their hosts and more specifically how the complex interactions between viruses and their host organisms are acquired and maintained throughout evolution, remains a major challenge (4) (5) (6) (7) . In order to assess this question, it is of prime importance to be able to detect and quantify the occurrence of lateral gene transfer events, and the impact of these events on viralhost co-evolution. Indeed, the mechanisms behind fast viral adaptation are far from being elucidated. Thus, we developed a global approach aimed at providing accurate evolutionary and phylogenetic information to tackle these questions. The major drawback of currently available databases of homologous families to the study of viral homologies and lateral gene transfer in viruses is their taxonomic compartimentalization. Indeed, current databases present families of homologies either restricted to viruses only [Protein Clusters (8) , GeneTree (9) ] or to viral taxonomic groups [Viral Orthologous Cluster (10) ], some also not presenting viral information [HomoloGene (11) ]. The few databases that do present viral and non-viral sequences, such as Pfam (12) or the Conserved Domain Database (13) do not provide complete phylogenetic trees. This translates into the fact that it is not currently possible to have a global view on viral-host lateral gene transfers due to the difficulty of obtaining global information on cross-taxa transfers at the viral level. We present the first public release of PhEVER, a unique database of homologous gene families containing sequences (i) of all completely sequenced viruses and (ii) from fully sequenced cellular organisms. The protein sequences are clustered-without a priori and according to similarity criteria-into families containing either only viral sequences or both viral and cellular sequences. PhEVER integrates extensive data from up-to-date completely sequenced genomes spanning a wide taxonomic range (2426 non-redundant viral genomes, 1007 non-redundant prokaryotic genomes, 43 eukaryotic genomes ranging from plants to vertebrates). To our knowledge, this is the most complete database of families of homologous viral sequences. Indeed, it not only spans all known viral groups but it also has the unique feature of presenting homologies with eukaryotic and prokaryotic sequences. The database offers a clustering of proteins into homologous families containing at least one viral sequence, as well as pre-computed alignments and phylogenies for each of these families. Alignments and phylogenies are built according to state-of-the-art phylogeny procedures and we provide tools to edit them and recompute them on the fly (14) . We also provide the possibility for users to assign their sequence of interest to a family and to re-build the phylogeny accordingly through the HoSeqI tool (15) . PhEVER thus constitutes a comprehensive working tool to detect sequence homologies and possible gene transfer events. Public access and documentation is available through the database webpage and through all dedicated ACNUC retrieval systems (16) . We developed a genome-wide cross-taxa approach to build a database of families of homologous sequences and to provide accurate alignments and phylogenies for the constructed families. The layout of the implementation of the PhEVER database is represented in Figure 1 and the details concerning its sequence content are available in Table 1 . In order to avoid redundancy due to the availability of numerous genomes of similar bacterial and viral strains in public databases, we collected all completely sequenced viral and bacterial genomes from RefSeq Viral (17) and Genome Reviews (18) , two non-redundant and curated databases of completely sequenced genomes. To this high-quality curated data composed of Archaea, Bacteria and Eukarya, we added nine eukaryotic genomes from Ensembl (Aedes aegypti, Anopheles gambiae, Bos taurus, Caenorhabditis elegans, Danio rerio, Drosophila melanogaster, Gallus gallus, Homo sapiens, Mus musculus) (19) to allow for a large representation of species from the different domains of life. The PhEVER database was structured under the ACNUC system (20) allowing it to be queried using a web interface and a large number of tools specifically developed for this database management system (14) . From the flat files containing the genomes and their annotations, two databases were built under the ACNUC database management system (20) , which is specifically aimed at building, storing and querying biological sequence data. One of them contains the nucleic sequences, the other contains the proteins generated by translating all CDS of the complete genomes-using the Figure 1 . Flow chart of the PhEVER building process. Complete genomes and their annotations were retrieved from three external public databases (Ensembl, Genome Reviews and RefSeq Viral) to provide high-quality non-redundant data for Eukarya, Archaea, Bacteria and Viruses. Two databases (nucleic acids, proteins) were constructed from this data to form PhEVER. All annotated CDS and mature peptides were translated and used for the clustering procedure. The homologous families thus produced were annotated in PhEVER, and alignments and phylogenies were built for each family and incorporated in the databases. appropriate genetic codes. For viral genomes presenting polypeptides which further maturate in vivo into mature peptides, the mature peptides were added to the set of translated proteins according to the annotations specified in the given genome ( Figure 1 ). Table 1 lists the global content of the databases as well as their original sources. Annotations were extracted from UniProtKB via the cross-references found in the CDS (21) . Subsequently, sequences in the nucleic and protein PhEVER database were clustered into families and were assigned a family accession number. Alignments and phylogenies were built for each of these families according to state-of-the-art procedures ( Figure 1 ). The classification of all organisms present in the database was retrieved from the taxonomy database at National Centre for Biotechnology Information (11) and is available on the PhEVER web interface. The clustering of proteins into homologous families was constructed using an automated procedure similar to the one described in (14) and implemented within a parallel framework in a software package called SiLiX. Briefly, sequences were assigned to protein families by simple transitive link using the following criteria. A similarity search of all translated proteins and mature peptides against all was performed using BLASTP2 similarity search with the BLOSUM62 substitution matrix, a 10 À4 e-value threshold and the 'm S' filter option (22, 23) . For each pair of sequences, HSPs which were not compatible with a global alignment were removed. Two sequences were included in the same family if the sum of the remaining HSPs covered >80% of the proteins length (and at least 100 amino acids) and if their identity was !35%. These two criteria were previously shown to provide a good trade-off between the ability of clustering sequences from divergent organisms and the quality of resulting alignments for subsequent phylogenetic analyses (14) . Finally, only families containing at least one viral sequence were integrated in the database. For each of the families, a small description built from the gene annotations ordered by frequency is available on the web interface. Figure 2 presents the distribution of viral species, proteins and families according to each viral group (A) as well as a Venn diagram representing the families content (B). Figure 2A presents PhEVER's broad taxonomical distribution covering all Baltimore groups (24). This distribution is naturally biased towards dsDNA, ssDNA and positive-sense ssRNA viruses reflecting the bias in genomic sequencing efforts. Indeed these groups contain long-studied viral families-either for their medical interest or for their economical impacts-such as Caudovirales, Poxviridae, Herpesvirales, Flaviviridae, Picornavirales or Parvoviridae. Figure 2 (B) shows a large number of orphan families indicating that a significant proportion of viral proteins (32%) do not contain any homologs with proteins from known genomes. These proteins, among which some might possibly be caused by annotation errors, are unfit for comparative functional analysis and should be the focus of future experimental studies to validate them and to provide with crucial information on viral mechanisms. Figure 2B also shows the small number of families sharing sequences from both viruses and eukaryotes compared to the relatively high number of families sharing sequences from both viruses and bacteria. This observation may be due to different underlying biological mechanisms but might also be an indicator of a still low coverage of the Eukarya domain. One of the applications of PhEVER is the detection of horizontal gene transfer events by comparing a gene family tree with the expected species tree. The discrepancies between the gene family history and the species phylogeny can then be an indication of possible events including a gene duplication, a gene loss or a horizontal gene transfer. The quality of the gene phylogenies is therefore essential and in this perspective, we implemented a procedure based on a rigorous methodology. First, the clustering into families was built with criteria leading to conservative families. Second, maximum likelihood phylogenetic trees were inferred for all families based on conserved aligned blocks. In our databases, for all families containing at least three sequences, pre-computed alignments and phylogenies are therefore already available. This allows for a simple and accurate overview of any family without the need of heavy computations and Number of proteins associated to a family, followed by the proportion of proteins associated to a family in the taxonomic group. can be a useful tool to search for lateral gene transfers in viral genes. For each family containing at least three sequences, alignments were estimated using MUSCLE with default parameters (25) . All alignments were treated with Gblocks (26) to select conserved blocks. Phylogenetic trees were inferred by maximum likelihood using PhyML (27) with a JTT evolutionary model (28) . Branch support was inferred using the Shimodaira-Hasegawa-like non-parametric procedure implemented in PhyML (27, 29) . To accommodate for weak phylogenetic signal, a thorough exploration of the tree space was made through topological rearrangements using the Nearest Neighbor Interchange topology search method. Finally, for visualization purposes, the trees were then rooted using midpoint rooting. This procedure allowed us to build accurate phylogenies sustained by branches of high support values. Indeed, Figure 3 shows that $80% of all branch supports have a value higher than 75 and one-third are above 95. Global statistics are biased by the presence of low branch supports. These are mostly due to few small families of less than 10 leaves, indicating that most families present robust phylogenies ( Figure 3B ). Finally, Figure 3C indicates that the low branch support values are attributable to very similar sequences with small branch lengths and might be linked to unresolved topologies. By contrast, all branch lengths longer than 0.15 subst/site display a high branch support which reveals the accuracy of our phylogenetic inference procedure. PhEVER is structured under the ACNUC sequence database management system and a large number of tools have been developed around this database management system [see (16) for an overview]. PhEVER can therefore be queried using (i) web applications, (ii) standalone software (iii) or embedded within Python, R or C code. The PhEVER web interface is available at http://pbil.univ-lyon1.fr/databases/phever and allows to search for sequences or families by combining several criteria (including species, gene names, annotation terms) as well as by crossing taxa. The graphical user interface QUERY_WIN and the terminal-based interface Raa_query (16) implement more features than the web interface and allow for remote ACNUC access and query as well as for the automatization of querying processes through standalone software. Finally the C language API, Python language API (16) and the seqinR package for R (30) implement tools for integrating queries in user designed code. Note that the PhEVER database can also be installed on a local machine or server for fastest data access. All files necessary for the PhEVER For all figures, only trees with more than four leaves are presented here. Branches with length smaller than 10 À5 were considered unresolved multifurcations and were discarded. installation are provided through our ftp website or by simple request. The PhEVER web interface allows for two query forms represented in Figure 4A and B. On the one hand ( Figure 4A ), the HoSeqI tool allows to search in PhEVER families with a user provided query (15) . This query is used to BLAST the proteins present in PhEVER and to match the most related family. Alignments and phylogenies for this new family containing the user provided sequence can be recomputed on the fly, visualized and manipulated with Java applets ( Figure 4F ) (31, 32) . On the other hand ( Figure 4B ), the query tool allows to directly query the database with a very diverse range of terms including gene name, annotation term, species name, protein accession number, genome accession number and family accession number. The species and sequences represented in each family detected by the query can then be viewed ( Figure 4D ) as well as the alignments and phylogenies which can be edited online via Java applets ( Figure 4E ) (31, 32) . More details on how to query PhEVER are presented in the Supplementary Data. PhEVER is the first open access database to provide information at the cross-taxa scale for the analysis of virus-virus and virus-host protein transfers. It compiles information from all kingdoms of life, and handles data from the genomes of all completely sequenced viruses and prokaryotes and of a large range of eukaryotes. It is the largest database of viral homologous families and offers highly accurate pre-computed alignments and phylogenies, making it a powerful tool for the analysis of horizontal gene transfer and more widely for the analysis of gene history. Our objective is to continue the development of PhEVER around the analysis of protein evolution in the context of virus-virus and virus-host interactions. More specifically, the next step we have under development is the detection of evolutionary conserved modules in the proteins present in PhEVER. Indeed, there is strong evidence that proteins evolve in a modular way, where modules are defined as parts of proteins sharing a common evolutionary history. These modules act as small interchangeable blocks of sequences that may be combined into proteins and form novel functions (33) (34) (35) . We are interested in providing a global tool allowing to analyse the weight of modular evolution in viral adaptation. We will therefore implement the detection of modules in PhEVER proteins to provide information concerning the exchanges of genetic information at the sub-protein level. In conclusion, PhEVER aims at being a comprehensive tool for the analysis of virus-virus and virus-host relationships from an evolutionary point of view, namely through the analysis of genomic interchanges. It should become a valuable tool for anyone working on viral evolution, but also to understand the general mechanisms behind protein evolution and functional innovation. Public access and documentation is freely available through the database webpage (http://pbil.univ-lyon1.fr/ databases/phever/) and through all dedicated ACNUC retrieval systems. More information on dedicated ACNUC retrieval systems, such as standalone query software (16) , the seqinR package for R (30) or the C and Python APIs, can be obtained in the Supplementary Data or on the PhEVER webpage. The PhEVER flat files for local installation are available through our ftp server (ftp://pbil.univ lyon1.fr/pub/phever) and instructions are available on the database webpage. The PhEVER database is updated every 6 months. This update frequency allows to follow the fast pace of viral and prokaryotic genome sequencing as well as to obtain updated genomic annotations for large eukaryotic genomes. Previous versions of the database remain available upon request.

Reporting errors in infectious disease outbreaks, with an application to Pandemic Influenza A/H1N1

BACKGROUND: Effectively responding to infectious disease outbreaks requires a well-informed response. Quantitative methods for analyzing outbreak data and estimating key parameters to characterize the spread of the outbreak, including the reproductive number and the serial interval, often assume that the data collected is complete. In reality reporting delays, undetected cases or lack of sensitive and specific tests to diagnose disease lead to reporting errors in the case counts. Here we provide insight on the impact that such reporting errors might have on the estimation of these key parameters. RESULTS: We show that when the proportion of cases reported is changing through the study period, the estimates of key epidemiological parameters are biased. Using data from the Influenza A/H1N1 outbreak in La Gloria, Mexico, we provide estimates of these parameters, accounting for possible reporting errors, and show that they can be biased by as much as 33%, if reporting issues are not accounted for. CONCLUSIONS: Failure to account for missing data can lead to misleading and inaccurate estimates of epidemic parameters.

The recent outbreak of pandemic strain Influenza A H1N1 (pH1N1), as well as other infectious disease outbreaks that have taken place recently illustrate the need for a rapid public health response and the ability to collect and analyze data efficiently. Unnecessary panic and disruption to society is more likely to be avoided and appropriately measured public health responses are more likely to take place when we have accurate information on the virulence and pathogenicity of an emerging disease. For these reasons, quantitative methods have been developed and continue to be developed to facilitate the assimilation of emerging data. Important quantities to estimate include the basic reproductive number, R 0 , defined as the average number of secondary infections created by a single infected individual in an entirely susceptible population. Numerous methods have been proposed for the estimation of this quantity, including deterministic and stochastic compartmental models [1] , branching processes [2] , networks [3, 4] , and, more recently, a likelihood based method [5, 6] . Another parameter of interest is the serial interval, or the distribution of the interval in time between an infector and infectee presenting with symptoms [7, 8] . It has been recently shown that this quantity may be time dependent; for example it can contract during the course of an epidemic as prevalence of the disease increases [9] . Other quantities of interest include the case fatality rate [10] and the attack rate in subpopulations, such as age groups. Among methods that can be implemented with relatively straightforward data, we typically assume complete observations. Clearly this assumption is more often than not violated in practice, especially when dealing with national or even regional data. For instance the scare surrounding the anthrax attacks in the fall of 2001 in the United States led to a large number of individuals reporting to medical care facilities with suspected anthrax. Should analysis have focused on suspected cases in that situation, the magnitude and threat of the event would have been greatly exaggerated. During the recent H1N1 outbreak, the number of suspected cases of disease likely is composed of several cases that will not be confirmed, however there are undoubtedly an even larger number of undetected cases, at least in the initial stages of the epidemic before the large public health response was launched and later on as the growth of the epidemic in many locations rendered it impossible to continue to track a large portion of the cases. Further, as the pandemic progressed sick individuals were cautioned to stay at home, rather than seek medical attention unless they were acutely ill [11] , driving up the number of unreported cases. Several of the unreported initial cases could arise from individuals who are asymptomatic, but still carrying and transmitting the virus or from others whose illness was not sufficiently acute to warrant seeking medical attention. To our knowledge, the issue of the impact of this misspecification of the number of cases on the estimation of epidemic parameters has not been well-studied. In what follows we use the likelihood based methodology described in [5] to broach the subject and investigate the impact of underreporting on estimates of both R 0 and the serial interval. First we provide an overview of the methodology we employ and introduce notation to describe the occurrence of misspecification of cases. Second we provide some theoretical results describing the impact on the estimation of the reproductive number. We illustrate this through a simulation study. Finally we investigate the impact that various plausible underreporting schemes would have on estimates obtained from the recent H1N1 outbreak in La Gloria, Mexico and compare these to estimates of R 0 obtained using the method proposed by Wallinga and Teunis (WT method) [3] and a simple exponential growth model [12] . In what follows, we assume that the outbreak is in its initial phase and that there is an unlimited supply of susceptible individuals. This implies that all contacts that an infected individual has are with susceptible individuals. Additionally, we follow standard methods and assume homogenous mixing among individuals in the system being studied. Following [5] , we assume that N t = {N 1 , ..., N T } are the number of new cases each day of the epidemic, with T being the total number of days of data analyzed, and that the serial interval is given by p = {p 1 , ..., p k } where k is the maximal length of the serial interval and p j is the probability of an infectee presenting with symptoms j days after the infector. In practice we can model the p j with a multinomial distribution or a truncated continuous distribution, such as the Gamma or Weibull, and estimate the parameters of that distribution so that the dimensionality of the estimation is independent of k. Then, we show in [5] that the likelihood of the case counts N t = {N 1 , ..., N T } is given by a thinned Poisson: . Consider that on a given day M t = q t N t of the cases are observed, where q t ≥ 0. Therefore little changes in (1) and it can be shown that the likelihood becomes . Given that q t is known, estimation proceeds as described in [5] . In reality we seldom known q t and our intent here is to quantify the effects of ignorance of q t on estimation of R 0 and the serial interval. We first consider the case where the serial interval (the p j ) is well known and specified, perhaps from contact tracing data or historical information. Then the MLE for R 0 in the complete data case is given bŷ which is comparable to the branching process estimator described by [2] , as illustrated in [5] . If p j is incorrectly specified we know that the estimates of R 0 are impacted [5] . Our interest here is the study of the impact of missingness therefore we assume that p j is correctly specified so as to avoid confounding the effect of these two issues when estimating R 0 . In the case where data is incorrectly reported, the estimator for R 0 obtained from [2] is We consider two simple missingness patterns in this scenario. First let the missingness be constant, i.e. q t = q for all t. Second let q t = q 1 for t ≤ t c and q t = q 2 for t > t c , where t c might correspond to a public health announcement or certain number of cases occurring so as to raise alarm of an epidemic. In these cases it is likely that q 1 < q 2 . This is the likely scenario initially in the current H1N1 outbreak, where cases were accumulating for some time before public announcements were made and increased surveillance was implemented. Numbers of confirmed cases available early on in the investigation likely underestimate the true number of cases dramatically. One can also imagine cases where q 1 > q 2 . This might occur in instances of overreporting such as occur in times of panic, for instance following the scare after the anthrax attacks in 2001 in the US [13] when more than 20,000 individuals started antibiotics until it was determined that they did not have anthrax. Additionally if all suspected cases of H1N1 were considered early in the epidemic, this could possibly overestimate the true number of cases. Further, as time has progressed in the H1N1 pandemic, it has become virtually impossible to ascertain all cases. Therefore it is likely that reporting initially increased and then began to decrease again as case counts escalated. We now provide results to illustrate the impact of these reporting schemes on the estimation of R 0 . In the following scenarios we use the branching process estimator to avoid the complication of the serial interval. In essence this implies that we assume that the serial interval is one day long in all scenarios or that the data is grouped into generations, rather than days or some other time unit. We now compare the two branching process estimators of the reproductive number,R 0 and  R 0 that make use of the observed numbers of cases denoted by M t = {M 1 , ...., In other words,R 0 is the naïve estimator that does not account for reporting issues and  R 0 is the true estimator. We prove the following two propositions in the Appendix. In summary, if the reporting fraction does not change through time, then the estimate of R 0 is unaffected. However if the reporting fraction increases (decreases) then we will overestimate (underestimate) R 0 if we ignore misreporting (labeled naïve, above). In order to quantify the impact of missing data on the estimation of R 0 and the serial interval, we consider a simulation study. We use multinomial and gamma distributed serial intervals. The multinomial represents a recent estimate for the current Influenza A/H1N1 outbreak in the USA and has a mean, μ of 2.21 days and variance, s 2 of 0.89 with k = 4 [14] . The Gamma distributed serial interval represents a disease with a mean of 8 days and variance of 16 days with k set to 20. We consider three values for R 0 : 1.5, 2, and 2.5, making six simulation scenarios. We then apply four missingness schemes to each dataset. The first two scenarios assume that the reporting fraction is constant through time. The second two schemes have the reporting fraction increase once 30 cases are accumulated. Following are the schemes that we consider: 1. q t = 0.1, for all t, 2. q t = 0.5, for all t, 3. q 1 = 0.05, q 2 = 0.5, 4. q 1 = 0.4, q 2 = 0.9. Thus we have six sets of complete data and 24 sets of incomplete data, where each set of data has 10,000 simulated epidemics. We show results for data simulated with two initial cases (i.e. N 0 = 2). Epidemics are simulated to stop when 500 cases are created or they die out. We only consider those that have at least ten cases for analysis, since fewer than ten cases would likely not be detected and considered an epidemic. Thus all simulated epidemics that die out before ten cases are accumulated, as well as those which have fewer than ten cases after the missingness pattern is applied, are not analyzed. Additionally those simulations with more than k zeroes in a row after the missingness operation is applied are not analyzed. This would be comparable to an undetectable epidemic since cases are so sparse in time that they are likely not connected to the same source. The results of the simulations are given in Figure 1 and 2. Consistent with our theoretical results we observe that when the reporting fraction is constant, the estimates of R0 are unaffected by a failure to control for missingness. However if the reporting fraction increases, then the estimates are smaller when we adjust for the missingness. We also note that [15] has recently described a tendency of this method to overestimate the mean of the serial interval when the serial interval is short, such as in cases of influenza. Thus part of the effect seen could be attributed to this phenomena, but likely will be uniformly so. In [16] the authors report initial findings on the current H1N1 Influenza pandemic, including results from data collected on a localized outbreak in La Gloria, Mexico. Data for this analysis was obtained by surveying 1575 Figure 1 Simulation results for the estimate of the reproductive number. The first boxplot (All) in each frame shows results when all the data is used. Each subsequent couplet of boxplots first shows the estimates when missingness is correctly accounted for and second, when missingness is ignored when estimating. The numbers on the × axis denote the missingness scheme applied to the data (see text of manuscript for a detailed description). For example the first couplet corresponds to constant missingness of 0.05 with the left boxplot giving the results when missingness is accounted for and the right one for estimates when missingness is ignored. The horizontal line indicates the true value of the parameter. villagers out of 2243 villagers recorded in 2005 [17] . Of those surveyed, 615 cases were reported. It was later discovered that some of these reported cases were from seasonal flu. Figure 3 shows the observed data. In addition to employing our likelihood based method (hereafter MLE method) to obtain estimates for the reproductive number and serial interval from the observed data, we also consider various schemes of underreporting among both those surveyed (due to asymptomatic cases or misclassification) and those not surveyed. Additionally we consider the possibility of overreporting, given that it was later noted that some of the cases reported were actually seasonal flu strains [18] . These reporting patterns are informed using the following pieces of information. Estimates of the attack rate for Influenza vary greatly. In [19] the authors report an attack rate of 68% among servicemen during an H1N1 outbreak in Finland during the winter of 1977-78. Among the 1575 surveyed in La Gloria an attack rate of 39% (615/1575) was observed. Finally, a recent report in Peru [20] indicates that 33% of cases were asymptomatic, meaning that the attack rate in La Gloria could actually be as high as 58% if we Figure 2 Simulation results for the estimate of the mean of the serial interval. The first boxplot in each frame shows results when all the data is used. Each couplet thereafter shows estimates when missingness is correctly accounted for on the left and on the right, when missingness is ignored in estimation. The numbers on the × axis denote the missingness scheme applied to the data(see text of manuscript for a detailed description). The horizontal line indicates the true value of the parameter. For several of the gamma distribution scenarios the actual estimates are extremely large and thus the value of the median is included on the plot where necessary. can extrapolate from Peru. The number of missing cases was calculated as a Poisson random variable with mean given by AR*(2243*AR-615). We generate 1000 epidemic sizes for each attack rate. We first assume that there is a constant reporting fraction. In other words we attempt to study the impact of missing information on the individuals that were not surveyed, assuming that they would have followed the same trends as those who were surveyed. Using the simulated epidemic sizes, we superimpose the missed cases on the observed cases using a multinomial distri- The data simulated from these assumptions are shown in Figure 3 . We consider three additional scenarios where we allow the reporting fraction to vary through time. In this case we let the reporting fraction follow a logistic function given by: where r describes the growth rate of the epidemic and is calculated at 0.19 in this scenario and q 1 and q 2 are set to 0.4 and 0.9, respectively and represent the reporting fractions at the start and end of the epidemic. Again the multinomial distribution with q t , as calculated from the logistic function, is used to assign the missing cases to days of the epidemic. Figure 4 illustrates the simulated outbreaks. Finally we consider the possibility that a significant fraction of the cases were not actually pandemic influenza strain and, in fact, cases were overreported. We still assume that there was an overall underreporting of influenza like illness (ILI) according to the schemes described above. However, we additionally assume that only a fraction of those cases are indeed of the pandemic strain. This is done by randomly sampling from all of the ILI cases using a binomial distribution with probability 0.25, 0.50 or 0.75, indicating that 25%, 50% or 75% of cases, respectively, are pH1N1. For the MLE analyses, we calculate the estimates of R 0 and the serial interval, assuming the serial interval follows a multinomial with a maximal length of four days [14] . We estimate the parameters under two scenarios. First we consider the first 16 days when the epidemic curve is in exponential growth and 326 cases had been observed. We perform estimation assuming an infinite number of susceptible individuals. Second, we consider the entire epidemic curve and estimate the effective reproductive number R t , by allowing R t to following a four parameter logistic distribution, as shown in [6] . We report the value for R 0 from this analysis. The median of the estimates over the 1000 simulated datasets is shown for each scenario along with the interquartile range of estimates obtained. Additionally we provide estimates obtained on the same data where all cases are assumed to be pH1N1 using the method described by Wallinga and Teunis [3] for the entire dataset with either the serial interval estimate obtained from the MLE based method or that obtained by [16] . R 0 is reported as the average R t over the first 16 days of the epidemic. We further use a simple exponential growth model to estimate the exponential growth parameter and R 0 , using both serial interval estimates [12] . Table 1 shows the results from the analysis for the MLE based method assuming that all the cases are pH1N1. The results that assume overreporting of pH1N1 are given in the appendix (Tables 2 and 3 ). We first consider the results obtained by considering the exponential growth phase of the epidemic (the first 16 days). Without any adjustments for underreporting, we estimate the basic reproductive number to be 1.41 and the mean of the serial interval to be 2.09 days with a variance of 1.15 days. When we assume constant missingness, the estimates for the reproductive number and the mean of the serial interval vary slightly from these estimates with their IQR containing the original values. Allowing the reporting fraction to vary according to a logistic function yields estimates of the reproductive number and the serial interval that are consistently less than the estimates from the original data. When all of the data are considered, the results are similar. Without adjustment for missingness R 0 is estimated to be 1.42 and the mean of the serial interval is 1.96 days with a variance of 1.14. Under the constant missingness scheme the reproductive number estimates increase with the attack rate (between 1.49 and 1.54), as well as the mean of the serial interval (2.15-2.25 days). When the reporting fraction increases through time, the reproductive number is smaller and decreases as the AR increases (1.29 to 1.05). The mean of the serial interval follows a similar pattern ranging between 2.01 for AR = 39% and 1.45 for AR = 68%. The results from the overreporting scenarios follow a similar pattern, assuming that the amount of overreporting is consistent throughout the epidemic. The results indicate that if reporting were in fact increasing throughout the epidemic, then the original results could be overstating the magnitude of both the reproductive number and the mean of the serial interval. Table 4 provides the results using the method of Wallinga and Teunis [3] and the exponential growth model to estimate R 0 . The estimates using the MLE estimator of the serial interval (i.e. with mean of 1.96 for WT or 2.09 for exponential) are comparable to those obtained with the MLE method and are 1.48 for the WT method and 1.40 for the exponential method compared to 1.41 for the MLE method. Those with the Fraser et al [16] estimate (mean SI = 1.91 days) tend to differ as a direct function of the mean SI (WT R 0 = 1.57, exponential R 0 = 1.36). Overall the impact of Key: The first block of results consider the entire outbreak and fit the reproductive number using a four parameter logistic function. The second block consider only the first 16 days where the epidemic is in exponential growth. For each attack rate shown (39%, 58% and 68%) estimates are shown when data missingness is assumed to be constant and when it is assumed to follow a logistic pattern with the initial reporting fraction at 0.4 increasing to 0.9. missing data is the same, regardless of the estimation method used. We have shown the impact of reporting issues on estimates of the reproductive number and the serial interval. Using a MLE based method, we show that the estimate of the reproductive number is unaffected if the reporting fraction is constant through time. However, if the amount of reporting increases, then we will overestimate the reproductive number if no adjustment is made. The converse is true should reporting tend to decrease over time. The simulation results and the work from La Gloria tend to support this theoretical result. We note that using other methods of estimation of the reproductive number (Wallinga and Tuenis [3] and exponential growth) yield the same trends. From our simulation work, we notice that the mean of the serial interval appears to follow the trend of the reproductive number. For instance, if the reproductive number is overestimated, then the mean of the serial interval tends to be too large, as well. Thus missing data not only impacts the estimate of the reproductive number, but also the estimation of the serial interval. We additionally note several caveats to the work presented here. First, we have assumed homogenous mixing by not accounting for any variability in the disease parameters among subgroups. Clearly disease outbreaks are dynamic and impacted by multiple factors in the effected population. We follow the precedent commonly used and assume homogeneity in the population. It is not clear how much results might change if this assumption is relaxed. Second, we take a frequentist approach and do not allow for variability in the parameters that are assumed known [15] recently published work allowing for a Bayesian approach to estimation where prior information on the serial interval can be incorporated into estimation. The values for the end results are likely to not vary significantly, if the prior information is not informative, there is sufficient data for estimation, or the serial interval is longer than one day. However there are cases where this prior information can improve the estimates and might impact the conclusions drawn here. In our setting we assume that there is no prior information, or if there is, such as the serial interval being known, that it is known with certainty. In this case, this assumption might lead to overestimating of the mean of the serial interval. Failure to account for ascertainment of cases can have a substantial impact on the estimation of key epidemiological parameters. If the fraction of cases reported does not change dramatically throughout the course of the epidemic, then estimates may not be impacted substantially. When the reporting fraction varies through the course of an epidemic, as it likely will, estimates can be substantially impacted. It is important that epidemiological studies of infectious disease outbreak seek to account and better understand the nature of reporting and make appropriate adjustments in the methods used to obtain results. We have shown how this can be done for the MLE method and illustrated its use for recent Influenza A/H1N1 outbreak data from La Gloria, Mexico. In that case results were off by as much as 34% when underreporting was not considered. Data do exist, however, to obtain an idea of the level of underreporting [13] use information on hospital admissions in the current H1N1 pandemic to obtain an estimate of the degree of underreporting. Further [15] has recently shown that incorporating contact tracing data into the estimation of the serial interval in a Bayesian framework can improve estimates. A similar approach could be used to improve estimates to account for suspected levels of misreporting. Clearly the need exists to use innovative methods to ascertain reporting issues in data using existing data, or by the collection of additional data or well-planned studies that can be rapidly initiated in the event of an outbreak. For instance, one might consider carefully studying a smaller population to determine the rate of reporting there. This could inform the overall rate of reporting. At the least, estimates should be reported as ranges, rather than as a single point estimate to indicate plausible values for the estimate. It is straightforward to observe from the definition of  R 0 that when q t = q, the correct estimator simplifies to is not impacted by the missingness in the data. Without loss of generality, we consider the simple branching process estimator, where p 1 = 1 and p j = 0 if j > 1. Additionally let q 2 = rq 1 , where r > 1. Then the estimator with missingness taken into consideration becomes . We now illustrate the impact of reporting issues on the standard error of the estimators,R 0 and  R 0 using the formula provided in [21] and given by Proposition A2. If q t = q 1 , t ≤ t c and q t = q 2 , t > t c then i. If q 1 < q 2 < 1 then SE R R SE ( ) ( ) 0 0 >  . ii. If q 1 > q 2 > 1 then SE R R SE ( ) ( ) 0 0 <  . iii. If q 2 > q 1 > 1 or q 2 <q 1 < 1 then the results for the SE are not clearly defined. Following are the results when we consider that only a fraction of the total cases in La Gloria were actually of the pH1N1 strain. For this scenario, we simulate 1000 datasets, as before. These datasets are interpreted as providing the total number of ILI cases. Of these, we assume that only a proportion is actually of the pandemic strain. We allow this to be 0.25, 0.50, or 0.75. Cases are randomly selected as pH1N1 cases and included in the analysis. Results are given in Tables 2 and 3 .

Pathological and ultrastructural analysis of surgical lung biopsies in patients with swine‐origin influenza type A/H1N1 and acute respiratory failure

BACKGROUND: Cases of H1N1 and other pulmonary infections evolve to acute respiratory failure and death when co‐infections or lung injury predominate over the immune response, thus requiring early diagnosis to improve treatment. OBJECTIVE: To perform a detailed histopathological analysis of the open lung biopsy specimens from five patients with ARDS with confirmed H1N1. METHODS: Lung specimens underwent microbiologic analysis, and examination by optical and electron microscopy. Immunophenotyping was used to characterize macrophages, natural killer, T and B cells, and expression of cytokines and iNOS. RESULTS: The pathological features observed were necrotizing bronchiolitis, diffuse alveolar damage, alveolar hemorrhage and abnormal immune response. Ultrastructural analysis showed viral‐like particles in all cases. CONCLUSIONS: Viral‐like particles can be successfully demonstrated in lung tissue by ultrastructural examination, without confirmation of the virus by RT‐PCR on nasopharyngeal aspirates. Bronchioles and epithelium, rather than endothelium, are probably the primary target of infection, and diffuse alveolar damage the consequence of the effect of airways obliteration and dysfunction on innate immunity, suggesting that treatment should be focused on epithelial repair.

Recently, a novel swine-origin influenza A (H1N1) virus with molecular features of North American and Eurasian swine, avian, and human influenza viruses [1] [2] [3] [4] has been associated with an outbreak of respiratory disease. According to the World Health Organization (WHO), between 25 April and 11 October 2009, 399,232 confirmed cases of H1N1 influenza virus and 4,735 deaths occurred throughout the world. 5 Brazil reported 1,528 deaths up to 10 November 2009. 6 Swine-origin influenza A (H1N1) virus infection can cause severe acute respiratory failure (ARF), requiring admission to an intensive care unit (ICU) in 15-30% of previously healthy young to middle-aged people. 3, 4, 7, 8 Death may occur when co-infections or lung injury prevail over the immune response, resulting in a progressive worsening of lung function (low compliance and oxygenation). Early diagnosis and a complete understanding of the pathological features of the H1N1 virus are important to help to improve treatment and the prognosis of this lethal disease. Analysis of the lung tissue from an open lung biopsy (OLB) of these severe cases can help in understanding the pathogenesis of this severe and sometimes fatal development. Until now, no reports of OLB findings used to guide the treatment of patients with H1N1 pneumonitis have been published, although according to many authors OLB is safe and diagnostically useful in patients with ARF, enabling appropriate therapy. [9] [10] [11] [12] The pathogenesis of ARF associated with swine-origin influenza virus (S-OIV) infection in humans is unknown. The influenza virus triggers pulmonary inflammation owing to an infiltration of inflammatory cells and an immune response. Bronchial epithelial cells are the primary target and the principal host for the virus. 13, 14 Normally, influenza viruses are recognized and destroyed by innate immune mechanisms which involve macrophages, interferon (IFN) a, b and other cytokines, natural killer (NK) cells and complement. When influenza viruses escape from these early defense mechanisms, they are captured and eliminated by adaptive immune mechanisms, where T and B cells and their antigen-specific effectors (cytotoxic T lymphocytes, cytokines such as IFNc and antibodies) target the virus. Additionally, antigen-specific memory cells (T and B cells) are involved in the prevention of the subsequent viral infection. 14 Thus, pathological findings obtained by an OLB, coupled to ultrastructural and immunologic analysis, may have an impact on decisions about changes in treatment strategies employed for these critically ill patients, and also provide a greater understanding of the pathophysiology of S-OIV infection. The objective of this study was to analyze pathologically and ultrastructurally S-OIV lung infection and the pulmonary immune response in a series of five cases with OLB. We studied pathologically and ultrastructurally five patients suspected of having a pandemic S-OIV virus who developed ARF requiring ventilatory support. Nasal swabs for RT-PCR for H1N1 were collected from all patients. The OLBs indicated by the clinicians were carried out after receiving consent from the families. These patients had a severe evolution of the virus and more information about the physiopathology of the disease was required in order to provide adequate treatment. If no improvement of the respiratory status was seen in the patients with ARF after $5 days (defined as no decrease of the Lung Injury Score) an OLB was indicated. 15 Lung tissue sections (4 mm thick), prepared from 10% formalin-fixed, routinely processed, paraffin-embedded blocks, were stained with hematoxylin-eosin. The following methods of histochemical staining were carried out: Grocott's methenamine silver stain, Brown-Brenn, and Ziehl-Neelsen. The following pathological changes were analyzed: a) necrotizing bronchiolitis, b) alveolar collapse, c) dilatation of the airspaces, d) hyaline membrane, e) fibroplasia, f) squamous metaplasia, g) multinucleated cells, h) alveolar hemorrhage, i) acute inflammatory exudates, j) atypical pneumocytes. Pathological changes were graded, using two sections, according to a five-point semiquantitative severity-based scoring system as: 0 = normal lung parenchyma, 1 = changes in 1-25%, 2 = changes in 26-50%, 3 = changes in 51-75%, and 4 = changes in 76-100% of examined tissue. This semiquantitative analysis is currently routinely used in most studies of the department of pathology of the University of Sã o Paulo Medical School. 16, 17 For immunohistochemistry, the avidin-biotin-peroxidase complex and streptavidin-biotin enzyme complex immunostaining methods were used with antibodies against: lymphocytes CD4 (clone: MO834, dilution 1:1000), CD8 (clone: M7103, dilution 120), CD20 (clone: M755, dilution 140), macrophages-histiocytes CD68 (clone: M814, dilution 130), mouse monoclonal antibodies from DAKO, Carpinteria, CA, USA; S100 (clone: Z311, dilution 11000) rabbit polyclonal antibodies from DAKO; CD1a (clone: MCA1657, dilution 1: 200) mouse monoclonal antibodies from Serotec, Oxford, UK; natural killer, NK (clone: MS136P, dilution 11000) mouse monoclonal antibodies from Neomarkers, Fremont, CA, USA; interleukin 4 (IL-4) (dilution 140), IL-10 (dilution 140) goat polyclonal antibodies from R&D Systems, Minneapolis, MN, USA; IFNc (clone: MAB285, dilution 130), mouse monoclonal antibodies from R&D Systems; tumor necrosis factor alpha (TNFa) (clone: AF210NA, dilution 140) all mouse monoclonal antibodies from R&D Systems; inducible nitric oxide synthase (iNOS) (dilution 1500) polyclonal rabbit from Calbiochem, La Jolla, CA, USA. Immunohistochemical reactions were carried out in accordance with the manufacturer's instruction. Diaminobenzidine was used as the color substrate, and Meyer's hematoxylin was used for counterstaining. Cell immunophenotypes and immune expression of cells using the different methods of immunohistochemical staining were identified and graded according to a five-point semiquantitative intensity-based scoring system as: 0 = negative, 1 = positive in 1-25%, 2 = positive in 26-50%, 3 = positive in 51-75%, and 4 = positive in 76-100% of examined tissue. 18 Small blocks (1 mm 3 ) of lungs were fixed in 2% glutaraldehyde/2% paraformaldehyde in cacodylate buffer overnight, then fixed in 1% osmium tetroxide, dehydrated, and embedded in araldite. Ultrathin sections obtained from selected areas were double-stained and examined in a Philips TECNAI 10 electron microscope at 80 kV. For each electron microscopy image (15/case), the following structural changes were analyzed: a) cytoplasmic swelling, b) degenerative changes, c) sloughing of necrotizing alveolar epithelial cell type I (AECI) and II (AECII), d) denudation of the epithelial basement membrane, e) hyaline membranes, f) alveolar septal collapse, g) viral particles such as tubuloreticular structures (TRS) and cylindrical confronting cisternae (CCC), h) multinucleated AECII. Ultrastructural findings were graded according to a five-point semiquantitative severity-based scoring system as: 0 = normal lung parenchyma, 1 = changes in 1-25%, 2 = changes in 26-50%, 3 = changes in 51-75%, and 4 = changes in 76-100% of examined tissue. 16, 17 Five patients (two male, three female) mean age 48 years (range 35-81) were studied; only patient No 4 had preexisting medical illnesses (Table 1) and chest x-ray abnormality at disease onset. All the patients presented with a 4-10 days' (median 5 days) history of shortness of breath and flu-like symptoms and rapid clinical deterioration. They were transferred to the ICU for tracheal intubation and ventilation (range 8-25 days; median 17) and diagnosed as having ARF. 15 All the patients received 75 mg twice a day by nasal enteral tube of olsetamivir (range 4-14 days; median 10) and intravenous steroids (range 9-20 days; median 12). After obtaining these results the dose was changed from 75 mg twice a day to 150 mg twice a day through a nasal enteral tube, in accordance with the Brazilian guidelines. The presence of the H1N1 virus was confirmed in all five patients (Table 1) by nasal swab or lung tissue positivity of RT-PCR according to guidelines from the Centers for Disease Control and Prevention. 19 Other microbiological investigations, including the isolation of other viruses, were negative. During the evolution of disease in the patients in the ICU, Staphylococcus aureus was isolated from a blood culture (patients 2 and 3) and Klebsiella spp were identified in tracheal aspirate specimens (patient 1). Patients 1, 2 and 4 are alive, but patients 3 and 5 died of respiratory failure, with concurrent congestive heart failure, hepatic encephalopathy, and acute renal failure. Table 2) . Pulmonary specimens from patients 3 and 5 presented more intense changes at optical microscopy. The membranous and respiratory bronchioles were extensively compromised by epithelial necrosis, squamous metaplasia, and obliteration by fibroplasia ( Figure 1A -F). The parenchyma was modified by extensive alveolar collapse, dilatation of the airspaces, alveolar hemorrhage, and sparse hyaline membrane formation ( Figure 1G -I). There was interstitial thickening, with mild to moderate fibroplasia ( Figure 1I ), but a disproportionately sparse infiltrate of inflammatory cells, mainly histiocytes, including multinucleated forms, lymphocytes and megakaryocytes ( Figure 1J-K) . Atypical bronchiolar and alveolar epithelial cells (AECs) were seen in all five patients, although the distribution was focal ( Figure 1J ). These atypical forms included multinucleated giant cells with irregularly distributed nuclei ( Figure 1K , L) or bronchiolar and AECs with large atypical nuclei, prominent eosinophilic nucleoli, and granular amphophilic cytoplasm ( Figure 1M ). However, distinct viral inclusions were not apparent. The ultrastructural features were represented by bronchial and alveolar epithelium necrosis, a destroyed alveolar epithelium/basement membrane unity and the presence of viral-like particles (Table 3) . Patients 3 and 5 presented more prominent changes at submicroscopic level. Cytoplasmic swelling, necrosis, and degenerative changes of the endoplasmic reticulum and other organelles were present in bronchial and AECs (Figure 2A-C) . A large number of bronchiolar and AECs were detached from the basement membrane and were showing apoptosis (Figure 2A, B) . Lymphocytes also exhibited apoptosis. Sloughing of apoptotic bronchiolar cells and AECs causing denudation of the epithelial basement membrane was followed by deposition of hyaline membranes ( Figure 2D ). Ultrastructural evidence of alveolar collapse was also present by the apposition of the alveolar septa ( Figure 2E -G). The regenerating bronchiolar epithelium extended along the adjacent alveolar septa showing features of cells with prominent surface microvilli with decreased or absent lamellar bodies and considerable cytologic atypia ( Figure 2H -L). Increased myofibroblasts and collagen fibers were also present ( Figure 2I ). Multinucleated epithelial cells with prominent nucleoli were noted in most cases, although such cells were sparse ( Figure 2K ). The proliferating bronchiolar and AECs containing TRS and CCC, probably representing residual viral-like particles, were distinguished in all cases ( Figure 2M -R). TRS appeared as reticular aggregates of branching membranous tubules located within the cisternae of the endoplasmic reticulum ( Figure 2M -O) or were compact ( Figure 2Q , R). CCC were identified as elongated, slightly curved cylindrical structures ( Figure 2P , Q), ring shaped ( Figure 2R ) or fused membranous lamellae, representing cisternae of endoplasmic reticulum. (Table 4) . Patients 3 and 5 presented with immunologic impairment. In all patients small aggregates of macrophages, CD4+ Thelper cells, CD8+ T-cytotoxic cells, CD20+ B-cells, CD1a+ dendritic cells, S100+ dendritic cells, natural killer lymphocytes were present around vessels and bronchioles. Dendritic cells and TNFa were expressed sparsely in macrophages, AECs and endothelial cells, whereas IFNc was expressed in small mononucleated cells in lungs from patients with S-OIV. There was a very strong expression of IL-4, IL-10 and iNOS in small mononucleated cells. This case series documents for the first time the pathological and ultrastructural findings of lung tissue from five patients admitted to the ICU with ARF and S-OIV infection who were submitted to OLB. S-OIV (H1N1) virus and the pulmonary syndrome is an acute respiratory illness, first identified in Mexico with at present, 399,232 cases registered, 4,735 deaths, affecting more than 179 countries. 2, 5 Our patients, most of them previously healthy, had an atypical influenza-like illness that progressed during a period of 5-7 days. The two patients who died showed a higher degree of pathological commitment of the disease at the OLB. Most of our patients were young to middle-aged and had previously been healthy. Increased risk for severe S-OIV illness is found in young children, 10-19 age groups, patients older than 65 years, pregnant women, obese people and those with comorbidities. 1, 7, 20 Fifteen to thirty per cent of patients with H1N1 infection required ICU admission. Mortality among the patients who required mechanical ventilation was around 58%. 7 In our case series the OLB findings showed that the lung damage was most likely due to infection by the influenza virus. The main pathological finding revealed necrotizing bronchiolitis and DAD, respiratory epithelial cells probably being the primary target of the infection. The extensive destruction of the respiratory and AECs and dysfunction in the immune and adaptative immune response led to DAD. As previously reported, possible mechanisms of damage include direct injury to the respiratory and alveolar epithelium exposing the basement membrane and leading to alveolar collapse by loss of surfactant, 13,14,21 with a secondary cytokine storm. 22 This is followed by exudation of macromolecules from the circulation, which finally form hyaline membranes. Activation of the cytokines is part of the immune reaction aiming to eradicate the virus. In this study, the systemic IFNc and TNFa cytokine activation probably resulted in reactive hemophagocytic syndrome in the bronchiole-associated lymphoid tissue and possibly also mediated the epithelial necrosis. 23 A mild inflammatory infiltration is most often seen in viral pneumonias; this has been explained by a cytokine-mediated blockade of lymphocytopoiesis and also by blockade of release from the bone marrow. 24 In our cases, expression in the lung of IFNc by small mononucleated cells and TNFa by macrophages and AECs was low. This finding may be supported by Kim and colleagues, 25 Conversely, we found a very strong expression of IL-4, IL-10 and iNOS by macrophages. The sparse inflammatory and immune reaction found in our samples, which involves targetting of the virus by NK cells, lymphocytes T and B cells, CD8+ cytotoxic T-lymphocyte cells, as well as CD1a and S100 cells, may be due to a combination of lymphoid tissue necrosis and apoptosis and exhaustion of lymphoid proliferation in response to the cytokine overdrive. In addition, the high IL-10 expression associated with its anti-inflammatory action may explain the low degree of inflammation observed in our cases. Taken together, our results suggest that in S-OIV infection, altered innate and adaptative immune responses may lead to incomplete virus eradication in the primary target of the infection and, consequently, imbalance between inflammation and reparation, resulting in bronchiolar obliteration and DAD. DAD is likely to be a consequence of bronchiolar obstruction and consequent hypoxia rather than direct invasion of the viruses. It is a severe pattern of lung injury and could be secondary to various pulmonary and extrapulmonary insults. 26 In this series of cases we found DAD in which alveolar collapse was prominent, differing from classic DAD found in ARF or secondary to other pulmonary and extrapulmonary insults. This finding may have important implications in the ventilation strategy of the patients. 27 In addition, the presence of intra-alveolar hemorrhage may suggest virus-associated hemophagocytic syndrome. 23 The lung tissue score was obtained independently by two different investigators. The pathologic findings were graded according to a five-point semiquantitative severity-based scoring system: 0 = normal lung parenchyma; 1 = changes in 1-25%; 2 = 26-50%; 3 = 51-75%; and 4 = 76-100% of the examined tissue. Table 3 -Semiquantitative analysis of electron microscopy. In our current series, pulmonary ultrastructural analysis was important to obtain an understanding of the pathophysiology of this new disease. First, we demonstrated apoptosis and necrosis in the bronchiolar epithelium together with viral-like particles, thus suggesting the bronchiolar epithelium as the primary target of the virus infection. Second, we documented the submicroscopic pattern of a clastogenic DAD in S-OIV infection. Third, we found indirect evidence of virus infection in alveolar and bronchiolar epithelial cells represented by the TRS and CCC. These submicroscopic structures were demonstrated ultrastructurally in the lung tissue of all the patients and their presence suggests an inactivation of the virus by oseltamivir treatment or an altered innate immune response of these patients. They appeared mainly in respiratory cells and AECs and have previously been described in a variety of cell types. 28, 29 Usually, they occur in endothelial cells and lymphocytes from patients with autoimmune diseases and viral infections. 30 Patients with acquired immunodeficiency syndrome present TRS and CCC in these same cells. 31 The mechanism of TRS and CCC production in vivo is not definitely established. Nevertheless, clinical and experimental studies have shown that the presence of both structures in these diseases is directly associated with the increase of IFNa and IFNb but not with IFNc. 29 One theory to explain the nature and pathogenesis of TRS and CCC suggests that these structures are incomplete viral particles. 30 In our study, these viral-like particles were noted mainly in the respiratory epithelial cells, but not in the other cell types within the lung. These observations reinforce the hypothesis that the primary target cells for S-IOV infection are probably the bronchiolar epithelium. The atypical morphology of the bronchiolar and alveolar epithelial cells was probably related to viral cytopathic effects or reactive changes. In fact, the presence of multinucleated epithelial cells is not exclusive to S-IOV, and is seen in pneumonia caused by the family of Paramyxoviridae, including parainfluenza viruses, measles, mumps, respiratory syncytial virus and, perhaps, metapneumovirus. 31 Although multinucleated cells were seen in our cases, these probably reflect non-specific secondary changes. We describe a case series of five patients with influenzalike illness with pneumonia and ensuing ARF who underwent OLB with subsequently confirmed diagnosis by RT-PCR testing for S-IOV infections. This report has some limitations. First, this study may not validate the importance of OLB in this population; however, it did provide information about this new disease. Second, it is difficult to compare our findings with those of others because to our knowledge no studies reporting an OLB in patients with S- The lung tissue score was obtained independently by two different investigators. The pathologic findings were graded according to a five-point semiquantitative severity-based scoring system: 0 = normal lung parenchyma; 1 = changes in 1-25%; 2 = 26-50%; 3 = 51-75%; and 4 = 76-100% of the examined tissue. IFN, interferon; TNFa, tumor necrosis factor a. IOV have been published. Although there are already many autopsy series with patients with H1N1 that can be used for comparison, the pathological findings at autopsy are modified mainly by the presence of associated co-infections and mechanical ventilation. [32] [33] [34] [35] [36] [37] [38] [39] In summary, we have presented the pulmonary pathology in a confirmed and well-defined series of cases of S-IOV infection associated with ARF. The pathological features, in addition to necrotizing bronchiolitis and DAD, included the presence of multinucleated cells and intra-alveolar fibrin exudates (organizing pneumonia-like lesions). Although each of these features is non-specific, their combined occurrence, together with positive serologic, microbiologic, and immunologic investigations and/or ultrastructural tissue examination enables the diagnosis of S-IOV infection to be confirmed, and is particularly useful in clinically suspicious cases that do not fulfill the WHO criteria or in clinically inapparent cases. We have shown that viral-like particles can be successfully demonstrated in lung tissue by ultrastructural examination, highlighting the importance of OLB, particularly in those patients without confirmation of the virus. We also showed that bronchioles and epithelium, rather than endothelium, are probably the primary target of infection, and that DAD is the consequence of airways obliteration and dysfunction on innate immunity, suggesting that the treatment should be focused on epithelium repair.

Ventilatory and ECMO treatment of H1N1-induced severe respiratory failure: results of an Italian referral ECMO center

BACKGROUND: Since the first outbreak of a respiratory illness caused by H1N1 virus in Mexico, several reports have described the need of intensive care or extracorporeal membrane oxygenation (ECMO) assistance in young and often healthy patients. Here we describe our experience in H1N1-induced ARDS using both ventilation strategy and ECMO assistance. METHODS: Following Italian Ministry of Health instructions, an Emergency Service was established at the Careggi Teaching Hospital (Florence, Italy) for the novel pandemic influenza. From Sept 09 to Jan 10, all patients admitted to our Intensive Care Unit (ICU) of the Emergency Department with ARDS due to H1N1 infection were studied. All ECMO treatments were veno-venous. H1N1 infection was confirmed by PCR assayed on pharyngeal swab, subglottic aspiration and bronchoalveolar lavage. Lung pathology was evaluated daily by lung ultrasound (LUS) examination. RESULTS: A total of 12 patients were studied: 7 underwent ECMO treatment, and 5 responded to protective mechanical ventilation. Two patients had co-infection by Legionella Pneumophila. One woman was pregnant. In our series, PCR from bronchoalveolar lavage had a 100% sensitivity compared to 75% from pharyngeal swab samples. The routine use of LUS limited the number of chest X-ray examinations and decreased transportation to radiology for CT-scan, increasing patient safety and avoiding the transitory disconnection from ventilator. No major complications occurred during ECMO treatments. In three cases, bleeding from vascular access sites due to heparin infusion required blood transfusions. Overall mortality rate was 8.3%. CONCLUSIONS: In our experience, early ECMO assistance resulted safe and feasible, considering the life threatening condition, in H1N1-induced ARDS. Lung ultrasound is an effective mean for daily assessment of ARDS patients.

Since the first outbreak of a respiratory illness caused by Influenza A (H1N1) virus in Mexico [1] , several reports have described the need of intensive care [2] [3] [4] or extracorporeal membrane oxygenation (ECMO) assistance [5] in young and often healthy patients. Beginning August 2009, the Italian Ministry of Health and the Tuscany Ministry of Health issued instructions to identify and establish referral centers able to care for the more severely ill influenza patients. Therefore, several referral centers were identified throughout the national territory among the hospitals already experienced in extracorporeal respiratory support techniques. The referral ECMO centres, in addition to being capable of guaranteeing the most advanced treatment in influenza related respiratory failure, were also entrusted with providing support to the nearby hospitals and assuring safe transportation. In the present investigation we report our experience, as an ECMO referral center, in H1N1-induced acute respiratory distress syndrome (ARDS) and we present the critical care service planning in response to the H1N1 pandemic. Following the instructions of the Italian Ministry of Health and Tuscany Regional Ministry of Health, an Emergency Medical Service was established in the Careggi Hospital in Florence Italy for the novel pandemic influenza. The Careggi Hospital ECMO Team is composed of: an intensivist, a cardiac surgeon, a cardiologist, a nurse, and a perfusionist. All of the members of the team are properly trained in ECMO treatment. An ambulance and a car are equipped with an ECMO circuit, a transport ventilator and all of the materials needed to initiate extracorporeal support in the peripheral hospitals, and permit safe transportation while on extracorporeal circulation to our referral hospital. The requirement of ECMO was decided based on the Italian Ministry of Health criteria (Table 1) . From September 2009 to January 2010, all patients admitted to our ICU with severe respiratory failure due to H1N1 infection were included in this study. Patient demographics and clinical characteristics were collected from institutional ICU database (FileMaker Pro, File-Maker, Inc, USA), from Italian Group for the Evaluation of Interventions in Intensive Care Medicine database (GiViTI Margherita Project, Istituto Mario Negri, Bergamo, Italy) and from ECMO national network database. Discrete variables are expressed as counts and percentages, whereas continuous variables are reported as medians with 25th to 75th interquartile range (IQR). The Internal Review Board approved this retrospective study and informed consent for data publication was obtained from the patients or relatives. Pressure volume curves were calculated with ventilator's built in application (Draeger Evita XL, Draeger Medical AG, Lubeck Germany) starting from a PEEP level of 5 cm H 2 O, with a pressure limit of 40. Ventilation parameters were set on the basis of this calculation, with a PEEP of 2 cmH 2 O above the lower inflection point of the pressure-volume curve, and a peek pressure below the upper inflection point. In all cases, pressure plateau was limited to 30 cmH 2 O and the tidal volume was kept below 6 ml/Kg [6] . Recruitment manoeuvres (40 sec at 40 cmH 2 O) were performed twice a day, if needed, to improve pulmonary gas exchange. Cannulation was conducted percutaneously with Seldinger technique in all cases, and cannulas position was confirmed by transesophageal echocardiography. Heparin infusion during extracorporeal lung assistance was monitored every two hours by bedside aPTT measurement (Hemochron Jr. Signature plus, ITC Europe, Milan, IT), which was maintained between 50 and 80 seconds. In case of renal replacement therapy requirement in ECMO patients, a continuous veno-venous hemodiafiltration circuit was assembled on the ECMO circuit (aspiration on pre-pump line, restitution on preoxygenation line). ECMO patients were ventilated with protective parameters, and respiratory rate and ECMO flow were adjusted to achieve normocarbia and oxygen saturation above 92%. assayed on pharyngeal swab, subglottic aspiration and bronchoalveolar lavage in accordance with published guidelines [7] . Bronchoalveolar specimens were obtained with a mini-invasive system (Kimberly-Clark BAL Cath, Kimberly-Klark N.V. Zaventem -Belgium), or by bronchoscopy. Patients were isolated in negative pressure atmosphere rooms, and staff wore full protective garments (including FFP3 respirators, 3 M Italia SpA, Segrate, Italy), until 2 consecutive tests were confirmed negative. During the study period only one case of suspected transmission of influenza to a nurse occurred. Antiviral therapy consisted in oral oseltamivir (75 mg twice daily) and inhaled zanamivir (10 mg twice daily). Blood and urinary cultures, tracheal aspirate, and pharyngeal swab were obtained upon patient admission. Empiric antimicrobial regimen at ICU admission was initiated with levofloxacin and amoxicillin/clavulanate; eventually specific antimicrobial therapy was varied or ended on the basis of microbiological results. Steroids were administered at low dosage (20 mg metilprednisolon twice per day) to prevent lung fibrosis. Diuretics were administered at different dosages, depending on clinical judgment and the patient's renal function. Lung ultra sound (LUS) examinations were daily performed by the attending physician, with a multifrequency convex probe (3.5-5 MHz, Mylab TM 30CV, ESAOTE, Genova, IT). With the patient in semirecumbent position, lateral and anterior views were obtained from base to apex of the chest. Posterior axillary line was followed during lateral transversal examinations. Chest quadrants defined by the intercostal spaces and the parasternal, mid-clavicular, and anterior axillary lines were scanned on the anterior chest wall [8] . The occurrence and extension of parenchymal consolidations, alveolar interstitial syndrome (measured by the number of B-lines), and morphology of pleural line were evaluated [9] [10] [11] . Pleural effusions were estimated by using Balik's formula [12, 13] . In order to ensure a uniform record, and allow to follow the evolution of the findings over time, all exams were recorded in an electronic form, in which the description of the main LUS features was predetermined [14] . During the study period, 12 patients requiring invasive ventilation treatment and/or ECMO were admitted or transferred to our ICU. Baseline and clinical characteristics of patients admitted for H1N1-induced severe respiratory failure are summarized in Table 2 . The median time between initial, non specific, symptoms and respiratory failure was 7 days (IQR 6-8.25), and severe hypoxia, unresponsiveness to non-invasive ventilation, was the main clinical feature. Our patients were young, median age 44.5 years, none of them older than 58 years, and eight (80%) younger than 50. Two patients were severely obese (BMI > 40), one woman was pregnant (18 weeks), two patients had a history of chronic obstructive respiratory disease (COPD), and one had diabetes. Two patients had Legionella Pneumophila coinfection at admission, and one young patient (16 years old) with suspect viral myocarditis and heart failure. At admission the patients, with the exception of the two coinfected, presented low leukocyte and platelet count and low plasma procalcitonin levels, significant levels of lactate dehydrogenase (LDH), creatine kinase (CK), and C-reactive protein ( Table 2 ). Median duration of mechanical ventilation (days) was 11.5 (IQR 9.8-16.3) and median ICU length of stay (days) was 14 (IQR 12-16.5). The pregnant woman continued the pregnancy without significant complications. In ICU infection rate was low with two ventilator associated pneumonia and two asymptomatic positive blood cultures in two ECMO patients. One ECMO patient died due to a systemic secondary infection by Aspergillus: this patient was the only non-surviving patient (overall mortality rate 8.3%). RT-PCRs from bronchoalveolar lavage samples were positive in all patients included in this study. On the contrary, RT-PCR dosed on pharyngeal swab resulted positive in less than 70% of patients at ICU admission, and in 90% of patients in the second day ( Figure 1 ). Also efficacy of antiviral therapy was reliably followed through RT-PCR from bronchoalveolar samples, since analysis on pharyngeal swabs became negative quite early. Finally, no RT-PCR significant for H1N1 infection from subglottic aspirate sample was found. In one patient, intravenous administration of zanamivir was needed, since the patient remained positive to viral infection after two weeks of therapy. Intravenous formulation of zanamivir is still subjected to pre-phase 4 clinical trial investigation, even if some reports on its safety profile are already available in literature. Therefore, local Ethical Committee approval was requested and the manufacturer provided the drug for use. Zanamivir was administered intravenously for five days (600 mg twice daily), as indicated by the producer. The patient's respiratory function improved and RT-PCR became negative after the third day. No adverse reaction was noted. A total of 156 LUS have been performed. During every LUS, the following parameters were considered: pleural line aspect and motility, presence of consolidations, occurrence and severity of Alveolar Interstitial Syndrome (based on the number of B-lines), presence of pleural effusion and occurrence of pneumothorax. Pleural thickness was described in 100% of cases and mostly bilaterally. Lung base was always involved. Lung gliding was present in 70% of LUS, even if decreased (20%). Pathological Lung Pulse was found in 20% of LUS, often in proximity to large parenchyma consolidations. Pleural effusion occurred in 7 patients. Two spontaneous pneumothorax have been detected with LUS during ICU treatment. Alveolar Interstitial Syndrome was present in all ultrasound examinations, with the presence of normal lung pattern (spared areas). In 90% of cases, B-Lines were described as moderate/many. At lung recovery, residual B-Lines patterns were found mostly at both bases. White lung feature occurred in about 15% of LUS performed, mostly in the anterior and lateral scans. White lung was never uniformly distributed, but it was alternated to spared areas, or areas with a limited number of B-Lines. Consolidations were found in 100% of cases. Most of them were multiple (65%), and lung bases were always involved. Contiguous subpleural consolidations were also present, increasing the pleural thickness laterally, mostly at the base and the apical part. Aerial bronchograms were always found within the consolidation pattern. The routine use of LUS limited the number of conventional radiology examinations (Table 3 ). In ECMO patients group, the higher number of chest X-ray examinations was needed to verify the correct cannulae positioning. In both groups, bedside LUS limited the transportation to the CT-scan room, increasing patient safety and avoiding the transitory disconnection of the patient from the ventilator. ECMO was needed in 7 patients (Table 3 ). In 4 cases, the ECMO Team was alerted and extracorporeal oxygenation was implanted directly at peripheral ICUs. No major transportation related problems were faced, even in the case of a long distance journey (400 Km). Median duration of ECMO support was 8 days (IQR 6-16.5), with a median duration of mechanical ventilation (days) of 19 (IQR 12-36). Main clinical features and ventilatory and ECMO parameters of patients treated with ECMO are presented in Table 4 . Bleeding was the most important complication. In three cases, bleeding from vascular access sites due to heparin infusion required blood transfusions. Three patients presented prolonged oropharyngeal bleeding and transfusions were required. Among them, one needed electrical coagulation of a palatine injury, probably related to nursing manoeuvres. Two patients presented severe intra-bronchial bleeding, and several flexible bronchoscopy examinations and clot suctions were required. In one of these patients, bleeding from the lower airways during the weaning phase from ECMO, and ECMO removal has been hastened. Table 3 summarizes the main differences between patients who underwent to ECMO treatment and patients only ventilated. Despite the small sample, ECMO patients clearly showed a higher critical illness score (SAPS II), and worst pulmonary gas exchange compared to patients who did not required extracorporeal lung assistance. Coinfection and comorbidities at admission were present only in ECMO patients. Our study population is young, comprising mainly healthy subjects, as previously reported [1, 2, 15] . Risk factors are similar to other studies, such as obesity, diabetes and pregnancy. In the present case series, bacterial infection rate at presentation was low. Previous reports showed incidence of secondary superinfection by Streptococcus Pneumoniae, Staphylococcus Aureus, Pseudomonas Aeruginosa, Acinetobacter Baumannii, Escherichia coli [1, 3, 2] . In our experience, we found two cases (16.7%) of co-infection with Legionella Pneumophila, which is, to the best of our knowledge, a new epidemiological data, since no other case has been reported in literature. It is questionable whether Legionella Pneumophila infection occurred before or after H1N1 pneumonia. However, it could be that H1N1 pneumonia was associated with a lower reactivity of the immune system, as suggested by the low leucocytes count reported in our sample and by other Authors [3, 2, 1] . One young patient presented heart failure, and viral myocarditis was suspected. The association of influenza with myocarditis is debated [16] , and H1N1 related myocarditis, has rarely been reported [17] . Furthermore, in our patient prolonged pre-hospital hypoxia was present and myocardial hypoxemia damage might have been involved. The patient required inotrope/vasoactive support for several days and eventually recovered fully with normal heart function. Our observations confirm the responsiveness of this infection to antiviral therapy. We adopted a two-modality administration, both oral and inhaled. Our choice was made in consideration of the decrease in gut motility and adsorption usually observed in critically ill patients. The World Health Organization (WHO) has questioned the sensibility of RT-PCR analysis for H1N1 in pharyngeal swab sample, encouraging analysis on samples from the lower respiratory tract. We routinely monitor H1N1 infection on three compartments: pharyngeal swab, subglottic aspiration, and bronchoalveolar lavage. In our experience, bronchoalveolar lavage at admission was positive in all patients while pharyngeal swab resulted positive in only 75% of cases. As shown in Figure 1 , RT-PCR from pharyngeal swab at ICU admission failed to demonstrate the viral infection in 3 patients. Similarly, the time course showed that RT-PCR from pharyngeal swab resulted negative in an average time of 3 days after therapy start. Conversely RT-PCRs from bronchoalveolar lavage remained positive for a longer period and resulted more reliable for infection monitoring and assessment of the efficacy of administered therapy. Based on our experience, RT-PCR from bronchoalveolar lavage resulted to be the most reliable method to diagnose and monitor H1N1 infection, since pharyngeal swab does not offer enough sensibility, neither for antiviral therapy initiation nor for antiviral therapy management. As subglottic aspiration resulted persistently negative, we do not recommend this sampling for diagnosis and monitoring of H1N1 infection. Despite the severe clinical pictures, we experienced a very low mortality rate: only one patient out of 12 died (8,3%). One of the surviving patients presented a lung cavern for a past pulmonary infection, and deceased for a secondary superinfection by Aspergillus, probably already colonizing lung parenchyma before the onset of viral infection. Our mortality rate is surprisingly low in comparison to a larger series of H1N1 patients, even when extracorporeal support technique were employed [5, 18] . Our finding can be related to the small number of patients included the study and definitive comparison with larger studies could be misleading. However, despite the severity of symptoms and the rapid progression to ARDS, H1N1 respiratory failure presents a relatively benign course when adequately treated, if compared to non-H1N1 induced ARDS, reported to have a mortality rate from 37% to 43% [19] [20] [21] [22] . Several factors may account for the favourable outcome in our series. All patients received protective ventilation. In particular, ECMO support permitted the maintenance of patients under a protective tidal volume with a respiratory rate below 12 per min, and a FiO 2 below 60%, compared with non-ECMO patients who needed a higher respiratory rate and FiO 2 to maintain an acceptable pulmonary gas exchange. The availability of easily accessible tools for pulmonary mechanics evaluations on modern ventilators allowed an individualized and appropriate setting of ventilation pressure within the thresholds of so called "protective ventilation" [23] . Furthermore, early access to ECMO resource allowed the maintenance of protective ventilation even in more severe patients (Table 4 ). In this regard, lactate dehydrogenase is commonly considered a marker of lung damage, and in H1N1 pneumonia is reported as high [1] . In our ECMO patients, lactate dehydrogenase values presented lower levels than in non-ECMO patients (445 U/L vs 627 IU/L, respectively), suggesting that in ECMO patients the reduced need of pulmonary ventilation could reduce lung ventilatory stress and enhance healing, regardless of the more impaired lung condition. However, it is possible that, since the technique has gained popularity and experience gathered to demonstrate its feasibility, we used ECMO also in patients who might previously have been successfully treated conventionally, and this may have influenced mortality. Moreover, more than half of our ECMO patients needed to be land-transported from other hospitals in an advanced stage of respiratory failure. This may have further encouraged an early treatment with ECMO to ensure the safest transport. Bleeding is commonly reported during ECMO treatment [24] , and either anticoagulation or platelet and coagulation cascade activation through oxygenator and pump is involved [25] . In our population bleeding also occurred more frequently in ECMO patients, and they required more transfusions compared to non ECMO patients. Nevertheless, in our experience, bleeding from cannulas insertion site or from upper airways, despite requiring transfusion, were not life threatening, and could be managed. In only two cases did severe bleeding occur in the lower respiratory tract. Fortunately in one case it occurred during weaning from ECMO, and it ceased after extracorporeal support removal. The other patient died from pulmonary aspergillosis and the haemorrhage could be also related to parenchyma disruption caused by the fungus. Monitoring heparin regimen is extremely important during extracorporeal circulation, and activated clotted time is commonly measured bedside. Some debate exists regarding the optimal range and the accuracy of pointof-care measuring devices [26] [27] [28] . In our protocol, we usually measured aPTT every two hours with Hemochron Jr. in order to closely monitor heparin administration in the low range of dosage. In our clinical practice, lung recovery and response to treatment are daily assessed by LUS examination, following several recent reports which underline the reliability of LUS in the evaluation and management of chest disorders [10, 29] . Despite CT-scan is the reference technique for evaluating lung lesions, it requires a transitory disconnection of the patient from the ventilator to permit the transportation radiology suite with potential risk of alveolar de-recruitment and worsening of oxygenation. Moreover, severe complications have been reported in intra-hospital transportation of critically ill patients [30, 31] . As we recently reported [13] , the routine use of bedside LUS has significantly reduce of the number of CT-scan and chest X-ray examinations in critical patients. The potential clinical benefit of reducing in-hospital transport for diagnostic radiology, it can be particularly relevant in patients with ECMO. In these patients, in fact, transportation requires time and a significant commitment of resources, although it was proved feasible both for inhospital [32, 33] and for inter-hospital long distance transportations [34, 35] . Another advantage of LUS is the ability to evaluate the effectiveness of alveolar recruitment manoeuvres with the possibility to visualize real-time imagines of lung parenchyma re-aeration [8, 10, 29] . Finally, pleural effusions can be accurately diagnosed and monitored with LUS and in case of need for treatment an ultra-sound guided technique is recommended [36, 13] . This option seems to be particularly appropriate ECMO patients, where bleeding for conventional chest tube placement can occur in consideration of the need of heparin infusion. The present case series comprises a small number of patients, and naturally, it cannot be considered a high grade of evidence trial. However, our experience might be helpful for intensivists challenging H1N1-induced ARDS. For H1N1 infection monitoring (or diagnosis, if patient was intubated before) bronchoalveolar lavage can be more reliable than pharyngeal swab in order of the higher sensitivity. In our clinical practice, ECMO therapy resulted safe and feasible in the context of a life threatening condition, and it might be taken into consideration as a therapeutic choice rather than a rescue solution in experienced centers. • ECMO might be taken into consideration as a safe therapeutic choice rather than a rescue solution in ARDS. • RT-PCR from bronchial lavage is more accurate than from pharyngeal swab, in H1N1 diagnosis. • Lung ultrasonography is a safe and reliable method to follow the pathology evolution/recovery of lung. • Lung ultrasonography can limit the need of CT-scan and chest X-ray examinations. List of abbreviations ARDS: acute respiratory distress syndrome; BMI: body mass index; CVVH: continuous veno-venous hemofiltration; ECMO: extracorporeal membrane oxygenation; ICU: intensive care unit; LOS: length of stay; LUS: lung ultrasound; RT-PCR: real-time reverse transcriptase-polymerase-chain-reaction; SAPS: simplified acute physiology score.

Broadly cross-reactive antibodies dominate the human B cell response against 2009 pandemic H1N1 influenza virus infection

The 2009 pandemic H1N1 influenza pandemic demonstrated the global health threat of reassortant influenza strains. Herein, we report a detailed analysis of plasmablast and monoclonal antibody responses induced by pandemic H1N1 infection in humans. Unlike antibodies elicited by annual influenza vaccinations, most neutralizing antibodies induced by pandemic H1N1 infection were broadly cross-reactive against epitopes in the hemagglutinin (HA) stalk and head domain of multiple influenza strains. The antibodies were from cells that had undergone extensive affinity maturation. Based on these observations, we postulate that the plasmablasts producing these broadly neutralizing antibodies were predominantly derived from activated memory B cells specific for epitopes conserved in several influenza strains. Consequently, most neutralizing antibodies were broadly reactive against divergent H1N1 and H5N1 influenza strains. This suggests that a pan-influenza vaccine may be possible, given the right immunogen. Antibodies generated potently protected and rescued mice from lethal challenge with pandemic H1N1 or antigenically distinct influenza strains, making them excellent therapeutic candidates.

The 2009 pandemic H1N1 influenza pandemic demonstrated the global health threat of reassortant influenza strains. Herein, we report a detailed analysis of plasmablast and monoclonal antibody responses induced by pandemic H1N1 infection in humans. Unlike antibodies elicited by annual influenza vaccinations, most neutralizing antibodies induced by pandemic H1N1 infection were broadly cross-reactive against epitopes in the hemagglutinin (HA) stalk and head domain of multiple influenza strains. The antibodies were from cells that had undergone extensive affinity maturation. Based on these observations, we postulate that the plasmablasts producing these broadly neutralizing antibodies were predominantly derived from activated memory B cells specific for epitopes conserved in several influenza strains. Consequently, most neutralizing antibodies were broadly reactive against divergent H1N1 and H5N1 influenza strains. This suggests that a pan-influenza vaccine may be possible, given the right immunogen. Antibodies generated potently protected and rescued mice from lethal challenge with pandemic H1N1 or antigenically distinct influenza strains, making them excellent therapeutic candidates. most other H1N1 and H5N1 influenza strains, especially in high-risk populations such as immunosuppressed patients and the elderly. Influenza-specific plasmablasts are persistently induced throughout infection, providing a rich source of antiviral mAbs B cell responses were examined in nine patients infected with the pandemic 2009 H1N1 influenza virus. These patients had varying courses and severity of disease. The cases ranged from mild disease with rapid viral clearance within a few days after onset of symptoms to severe cases that shed virus for several weeks and required hospitalization with ventilator support. A majority of the patients were treated with antiviral drugs. The diagnoses were confirmed by pandemic H1N1-specific RT-PCR and serology. All patients had neutralizing titers of serum antibodies at the time of blood collection. A summary of the clinical patient data are shown in Table I . The majority of samples were obtained around 10 d after the onset of symptoms, with the exception of a particularly severe case where sampling was done 31 d after symptom onset. Antigen-specific plasmablasts appear transiently in peripheral blood after vaccination with influenza or other vaccines (Brokstad et al., 1995; Bernasconi et al., 2002; Sasaki et al., 2007; Wrammert et al., 2008) , but the kinetics of their appearance and persistence during an ongoing infection remain unclear. Here, we have analyzed the magnitude and specificity of the plasmablast response in blood samples taken within weeks after onset of clinical symptoms of pandemic H1N1 influenza virus infection. Using a virus-specific ELISPOT assay, it was possible to show a significant number of pandemic H1N1-reactive plasmablasts in the blood of the infected patients, whereas none were detectable in a cohort of healthy volunteers (Fig. 1, A and B ). These cells were also readily detectable several weeks after symptom onset in the more severe cases. Fig. 1 (A and C) illustrates that, of the total IgG-secreting cells, over half of the cells were producing antibodies that bound pandemic H1N1 influenza virus. Moreover, plasmablasts specific for HA occurred at 30-50% the frequency of virus-specific cells (Fig. 1, C and D) , the specificity most likely to be critical for protection. Most patients also had a relatively high frequency of plasmablasts, forming antibodies that bound to past, seasonal influenza strains ( Fig. 1 C) or recombinant HA from the previous annual H1N1 strain, A/Brisbane/59/2007. Based on the overall frequency of pandemic H1N1-specific cells, it is likely that the cells binding other strains were overlapping populations and cross-reactive. None of the induced plasmablast cells bound to recombinant HA from the H3N2 strain from the same vaccine (A/Brisbane/10/2007). These findings demonstrate that influenza-specific human plasmablasts are continuously generated throughout an ongoing infection and that a fairly high proportion of these cells make antibodies that also cross-react with previous annual H1N1 influenza strains. To analyze the specificity, breadth, and neutralizing capacity of these plasmablasts, we used single-cell PCR to amplify the heavy and light chain variable region genes from individually Hancock et al., 2009) . The Centers for Disease Control reports that there were an estimated 60 million cases of the 2009 H1N1 pandemic strain, which caused 256,000 hospitalizations. An unusually high frequency of severe disease occurred in younger and otherwise healthy patients (Hancock et al., 2009 ). In addition, rare infections with avian H5N1 influenza strains in humans had close to a 50% mortality rate (Subbarao and Joseph, 2007) . Emergence of a zoonotic or antigenically distinct strain that combined even a fraction of the morbidity and mortality of the pandemic H1N1 and H5N1 viruses would have dire consequences. Antibodies play a key role in protection against influenza infection in vivo (Puck et al., 1980; Gerhard et al., 1997; Luke et al., 2006; Simmons et al., 2007) . The fact that there were little or no preexisting antibody titers present before the emergence of this pandemic virus, and that the virus atypically caused such severe disease in young adult, illustrates the importance of comprehensively understanding the B cell responses and antibody specificities induced by infection with this influenza virus. Here, we have analyzed the plasmablast responses induced by pandemic H1N1 infection and generated a panel of monoclonal antibodies (mAbs) from these cells to analyze their characteristics in detail. In contrast to seasonal vaccination, we show that a majority of the neutralizing antibodies induced by infection were broadly cross-reactive with all recent annual H1N1 strains, as well as the highly pathogenic 1918 H1N1 and avian H5N1 strains. These neutralizing antibodies bound predominantly to conserved epitopes in the hemagglutinin (HA) stalk region (Ekiert et al., 2009; Sui et al., 2009) , with some binding to novel epitopes in the HA globular head. The high frequency of these HA-stalk binding antibodies is of particular interest, as this epitope is a promising target for a broadly protective influenza vaccine (Steel et al., 2010) . Furthermore, the cross-reactive antibodies carried highly mutated immunoglobulin genes, indicative of extensive affinity maturation. Together, these findings support a model in which infection predominantly activated broadly cross-reactive memory B cells that then underwent further affinity maturation. We propose that the expansion of these rare types of memory B cells may explain why most people did not become severely ill, even in the absence of preexisting protective antibody titers. Recent studies in mice strongly support the idea that consecutive immunizations with antigens from divergent influenza stains can indeed hone the antibody response to preferentially target broadly protective conserved epitopes Wei et al., 2010) . Our findings demonstrate that cross-reactive antibodies can be preferentially induced in humans given the right immunogen, providing further support for the feasibility of generating a pan-influenza vaccine. Finally, in vivo challenge experiments showed that the neutralizing antibodies isolated protected mice challenged with a lethal dose of pandemic H1N1 influenza virus, even when administered therapeutically 72 h after infection, and also provided protection against antigenically distinct H1N1 influenza strains. These antibodies are thus promising as therapeutics against pandemic H1N1, as well as activated by pandemic H1N1 infection. Consistent with the frequency of plasmablasts secreting antibodies binding annual influenza strains by ELISPOT analyses (Fig. 1 C) , a majority (29/46, or 63%) of the pandemic H1N1-specific antibodies also cross-reacted with seasonal influenza viruses (Fig. 2, A and B ). In fact, by ELISA, one third of these antibodies bind to the prepandemic strains at lower concentrations than they did to the pandemic H1N1 strain, suggesting higher avidity binding. By comparison, only 22% (11/50) of plasmablasts induced by annual H1N1 strains before the pandemic could bind the pandemic H1N1 influenza (Fig. S1 B) . We propose that the cross-reactivity of pandemic H1N1-induced cells derives from the activation of memory cells originally specific for past influenza immunizations in an original antigenic sin fashion. Based on the 10-15-fold induction of plasmablasts and expression of intracellular Ki67 during ongoing immune responses (Brokstad et al., 1995; Bernasconi et al., 2002; Sasaki et al., 2007; Smith et al., 2009; Wrammert et al., 2008) , we can assume that most plasmablasts result from the ongoing infection sorted cells (defined as CD19 + , CD20 lo/ , CD3  , CD38 high , CD27 high cells; Fig. 1 E; Wrammert et al., 2008; Smith et al., 2009) . These genes were cloned and expressed as mAbs in 293 cells, and the antibodies were screened for reactivity by ELISA. Thresholds for scoring antibodies as specific to the influenza antigens were empirically determined based on being two standard deviations greater than the background level of binding evident from 48 naive B cell antibodies (Fig. S1 A) . Of 86 antibodies generated in this fashion, 46 (53%) bound pandemic H1N1 (Fig. 1 F) and one third (15 antibodies) were reactive to HA ( Fig. 1 G and Fig. S2 A) , most of them at sub-nanomolar avidities (based on surface plasmon resonance analyses; Fig. S2 B) . On a per donor basis, 55% of the mAbs bound to purified pandemic H1N1 virions (range: 33 to 77%). Of the virus-specific antibodies, 31% bound to recombinant HA (range: 14 to 55%). We conclude that virus-specific plasmablasts are readily detected after pandemic H1N1 influenza virus infection and that virus-specific human mAbs can be efficiently generated from these cells. Plasmablasts from patients infected with pandemic H1N1 influenza were highly cross-reactive to prepandemic influenza strains As the plasmablasts are specifically induced by the ongoing immune response, we can learn about the origin of the B cells The mAb column indicates whether mAbs were made from the plasmablasts of these patients. (Table I) , mutations had accumulated at a significantly lower frequency than the IgG controls ( Fig. S3 A; P < 0.0001), suggesting a unique circumstance such as a low-level or lacking primary response. Detailed sequence characteristics for pandemic H1N1-induced plasmablasts are provided in Tables S1-S3. Though based on a limited or vaccine response. The ready detection of clonal expansions at a mean frequency of 16.5% of the cells for the six patients supports this view (based on CDR3 sequence similarity; Fig. 2 C). Since the discovery of somatic mutation, it has been appreciated that mutations progressively accumulate on variable genes after repeated immunizations (McKean et al., 1984) . Thus, we can gain insight into the origin of the pandemic H1N1 response by comparing the somatic mutation frequency of the plasmablasts present during H1N1 infection to that of other plasmablast responses. The PCR strategy allowed isolation of either IgG or IgA transcripts and identified 68% IgG and 32% IgA plasmablasts from the patients. Similar to plasmablasts induced by annual vaccination (Wrammert et al., 2008) , or after a fourth booster vaccine to anthrax, the variable genes of novel H1N1-induced cells from five of the six patients harbored high numbers of somatic mutations Antibodies were tested at 10 µg/ml and threefold serial dilutions until a nonbinding concentration was determined. Each antibody was tested in at least two (and typically more) replicates for specificity and affinity estimations. Note that only 14 of 15 HA-binding antibodies have curves in G because one of the HA-reactive antibodies only binds HA on whole virions, not on the recombinant protein. H1N1 reactive mAbs isolated from infected patients (1000, EM, 70, 1009) were assayed for binding to annual H1N1 influenza strain whole virus. The minimum detectable concentration is defined as two standard deviations above the mean binding of 48 randomly chosen naive B cell antibodies ( Fig. S1 A) . Bars are color coded to approximate levels of crossreactivity to the annual vaccine (circulating) strains of recent years. Panels A and B use the same color scheme. Each value is representative of at least two replicate ELISAs repeated until a single consistent minimum concentration was established. The center numeral equals total antibodies. (C) Analysis of the variable gene sequences from plasmablasts of the four pandemic H1N1-infected patients indicated that 16.5% of the pandemic H1N1-induced plasmablasts were clonally related (shared identical VH and JH genes and CDR3 junctions). (D) The average number of somatic hypermutations in the pandemic H1N1 patient plasmablast variable region genes compared with primary IgG plasmablast responses to vaccinia (small pox) or the anthrax vaccine, or after at least 4 boosters with the anthrax vaccine. To account for the obvious outlier in the pandemic H1N1 group (patient-EM), median values are indicated by the bar. Student's t tests excluding the outlier indicated a p-value of <0.04 for the remaining five pandemic H1N1 samples compared with the IgG memory and germinal center (GC) cells or the primary IgG plasmablast responses (0.2 with EM included) and a p-value of <0.0001 against the IgM populations. Notably, besides patient EM, each individual set of VH genes averaged significantly more mutations than the IgG memory and GC or the primary responses ( Fig. S3 A) . Each point represents one individual donor and is averaged from 25-75 sequences, except for the primary response to anthrax from which only 10 VH genes could be cloned from single cells because of the highly limited response. Mutations accumulated per individual sequence are depicted in Fig. S3 . Detailed sequence characteristics are provided in Tables S1-S3. The naive, IgG and IgM GC and memory populations are derived from historical data (Zheng et al., 2004 (Zheng et al., , 2005 Koelsch et al., 2007; Wrammert et al., 2008) . indicating that they bound to sites other than the HA active site. Interestingly, antibodies of the latter type were predominant in the response (Fig. 3 A) . This specificity is reminiscent of antibodies against the recently discovered broadly neutralizing epitopes found on the HA stalk, rather than those located on the HA globular head that is more typical for neutralizing antibodies (Ekiert et al., 2009; Sui et al., 2009) . Importantly, five of these antibodies are indeed of similar specificity (including antibodies 70-5B03, 70-1F02, 1000-3D04, and a clonal pair from donor 1009: 3B05 and 3E06). These five antibodies bind with high affinity to most H1 strains including all from the vaccines of the past 10 yr, the 1918 pandemic strain, and to the H5 of a highly pathogenic number of patients, the frequent cross-reactivity and high number of somatic mutations support a model in which many of the plasmablasts induced by pandemic H1N1 infection arose from cross-reacting memory B cells. A majority of the neutralizing antibodies bound to highly conserved epitopes in both the HA stalk and head regions A high frequency of the HA-specific antibodies was able to neutralize the virus in vitro (totaling 73% or 11/15; Fig. 3 A) . These neutralizing antibodies could be further categorized into two distinct groups: (a) neutralizing antibodies that displayed hemagglutination inhibition (HAI) activity (HAI + ) and (b) neutralizing antibodies that had no HAI activity, Fig. S2 A) . The antibodies are grouped based on whether they show HAI and/or neutralizing (neut) function. Antibody 1009-3B06 was only tested for binding to whole virus, as this antibody did not bind to rHA due to binding of a quaternary or conformationally sensitive epitope that is not present in the recombinant protein. HAI and neutralization assays were performed in duplicate and repeated at least three times. ELISA curves are provided in Fig. S2 A. (B) ELISA binding as shown by minimum positive concentration (defined for Fig. 2 ) of neutralizing mAbs to rHA or whole virions from pandemic H1N1 or other influenza strains (ELISA binding curves are provided in Fig. S2 A) . Three binding patterns (epitopes 1 and 2, and 3) were observed that coincided with specificity comparisons by competitive ELISA, as illustrated in Fig. 4 A. (C) Three representative neutralizing antibodies (EM-4C04, 70-1F02, and 1009-3B06) were used for HAI and microneutralization (MN) activity against pandemic H1N1 and several other annual or laboratory H1N1 influenza strains. Experiments were performed in duplicates and repeated at least three times. Minimum effective concentration is shown for both assays. FACS analysis showed that the five antibodies bound to all 13 H5 variants tested at levels quite similar to F10, for which a crystal structure had been generated to define this epitope. Thus, half of the neutralizing and a surprising 10% of all antibodies induced by pandemic H1N1 infection bound to a conserved, critical epitope on the HA stalk. By comparison, none of 50 H1N1 strain-specific antibodies that we had previously isolated after annual vaccination before the 2009 pandemic had this reactivity (unpublished data). The frequency of pandemic-induced, stem-reactive antibodies (5/46) versus those from annual vaccine (0/50) is significantly greater (Chi-square test, P = 0.02). Further, this specificity is only rarely seen in human memory B cells (Corti et al., 2010) or from phagedisplay libraries (Sui et al., 2009 ). These observations support avian influenza strain (Fig. 3 B and Fig. S2 A) . In addition, these five antibodies cross-compete for a similar epitope that was not over-lapping with the HAI + antibodies (Fig. 4 A, epitope-1). These antibodies are competitively inhibited by a commercial antibody referred to as C179 that binds this HA stalk region (Okuno et al., 1993) , and four of five of these antibodies are encoded by the hallmark VH1-69 gene (Ekiert et al., 2009; Sui et al., 2009) . To verify HA stalk reactivity, these five antibodies were tested for binding to H5 variants predicted to affect the stalk epitope by the crystal structure, and their binding patterns were compared with that of the prototypical stalk antibody (mAb F10; Sui et al., 2009; Fig. 4 B) . Each H5 variant has a single residue mutation in the stalk region and was transiently expressed on 293T cells. (A) Competition ELISA assays were used to determine the similarity in specificity between the various neutralizing antibodies. Shown is the percentage of competition of each antibody in an ELISA binding assay against all other neutralizing antibodies. A 10-fold molar excess of unlabeled antibody was used to inhibit a biotinylated antibody. Percent competition is calculated as the reduction in absorbance relative to the level of inhibition of any particular antibody against itself. Colors indicate degree of inhibition of antibody binding, as indicated. Antibody C179 is a commercial antibody that binds to the stalk region of the HA molecule identifying epitope-1. Epitope-2 and -3 are each on the HA-head active site. 1000-2G06 and the nonneutralizing, but HAbinding, antibodies had no competition with any of the other HA-reactive antibodies and are therefore not shown. VH gene usage of the individual antibodies is listed on the right. All assays were performed in duplicate. (B) Plasmids encoding full-length WT H5-TH04 (A/Thailand/2-SP-33/2004 [H5N1]) and its mutants were transiently transfected into 293T cells. 24 h after transfection, cells were harvested for FACS analysis, and binding of indicated antibodies was tested at 10 µg/ml. The cell surface HA expression of each of the mutants were verified with a ferret anti-H5N1 serum (not depicted). Antibody F10 was one of the antibodies used to characterize the HA stalk epitope by x-ray crystallography (Sui et al., 2009 ) and served as a positive control for the binding pattern expected of HA stalk-reactive antibodies to these HA mutants. to all recent H1 vaccine strains and reacted strongly to the 1918 pandemic strain (antibodies 1009-3E04 and 1000-3E01; Fig. 3 B and Fig. 4 A, epitope-3) . These mAbs bind to past vaccine strains with higher avidity than to the pandemic H1N1. Further studies are underway to precisely identify the epitopes of all neutralizing antibodies in this study. Only two of 11 neutralizing antibodies were highly specific for the pandemic H1N1 strain alone (Fig. 3 B and Fig. S2 A) , including a low avidity antibody, 1000-2G06, which only showed slight neutralization capacity in vitro, and EM-4C04, which was very effective at neutralizing the pandemic H1N1 influenza. We conclude from these experiments that a surprising 82% (9/11) of the neutralizing plasmablasts that we isolated during pandemic H1N1 influenza infections were broadly cross-reactive to multiple influenza strains. There is a distinct interest in developing monoclonal antibodies for use in a therapeutic setting. We selected three representative antibodies of the set we have identified for detailed functional analysis both in vitro (Fig. 3 C) and in vivo ( Fig. 5 and Fig. 6) , including: EM-4C04, 1009-3B06, and 70-1F02. The antibodies EM-4C04 and 1009-3B06 are specific for the active site of the HA molecule, whereas 70-1F02 binds to the stalk region. Furthermore, EM-4C04 is highly specific for pandemic H1N1, whereas 1009-3B06 and 70-F02 display the idea that a vaccine might be developed that preferentially targets the HA stalk, thus providing broad protection against many influenza strains. The remaining neutralizing antibodies were HAI + and therefore bound to the HA globular head. Based on crosscompetition analyses, these antibodies fell into two groups binding nonoverlapping regions of the HA head, including epitope-2 and epitope-3 (Fig. 3 B and Fig. 4 A) . Indeed, using spontaneous escape mutant selection, we found that the EM4C04 mAb binds to the Sa region of the HA globular head (unpublished data). Thus, by proximity based on the competition assay (Fig. 4 A) , we can predict that all of the epitope-2 antibodies bind near the Sa/Sb region (including EM-4C04, 1009-3B06, and 1009-3F01). Broadly reactive antibodies binding both pandemic H1N1 strains and common annual H1N1 strains have been identified both in humans Xu et al., 2010) and in mice (Manicassamy et al., 2010) . It is notable that three of five of the HA globular-head-binding antibodies induced by pandemic H1N1 infection were also broadly reactive to various H1N1 strains (Fig. 3 B) . One such novel antibody was the SF1009-3B06 antibody that reacts strongly with the pandemic H1N1 strain, as well as all recent H1N1 vaccine strains (Fig. 3 B and Fig. S2 A) . The precise epitope to which the 1009-3B06 antibody binds appears to be quite unique; it is only accessible on whole virions, not on recombinant HA, suggesting that the epitope is quaternary in nature. Finally, two antibodies cross-reacted and inhibited hemagglutination Figure 5 . In vivo prophylactic and therapeutic efficacy of human mAbs against pandemic H1N1 influenza virus. 6-8-wkold BALB/c mice were infected with a 3xLD50 dose of highly pathogenic, mouse-adapted 2009 pandemic H1N1 influenza (A/California/ 04/09). 24, 48, and 60 h after infection, 200 µg (10 mg/kg of body weight) of EM-4C04, 70-F02, or 1009-3B06 human mAb were injected intraperitoneally. All mice were monitored daily for body weight changes and any signs of morbidity and mortality. Percentage of initial body weight is plotted, and the number of surviving mice is shown in the lower right of each plot. Infected, untreated mice showed clear signs of sickness around day 4-5 after infection and perished by day 8-9. Prophylactic treatment is shown on the left for comparison. Antibody treatment conferred significant protection as determined by comparison of weights in untreated versus prophylaxis and at the time of treatment versus 12 d after infection (unpaired, twotailed Student's t test, P < 0.05). The log-rank test indicated significant survival as well (P < 0.001). Figure shows one representative experiments of at least three independent repeat experiments. for the pandemic H1N1, had no protective effect on infection with PR/8/34 or FM/1/47. In conclusion, the antibodies characterized herein show promise for development as broadly reactive therapeutic agents against the pandemic H1N1 influenza virus, as well as against the majority of H1N1 and H5N1 influenza strains. Our findings provide insight into the human B cell responses to a pandemic influenza virus strain. The unique genetic composition of the pandemic H1N1 influenza virus meant that our relatively young cohort probably had little or no preexisting specific antibody-mediated immunity to this virus before infection (Brockwell-Staats et al., 2009; Dawood et al., 2009; Garten et al., 2009; Hancock et al., 2009) . Thus, two sources of B cells could have contributed to this response: newly recruited naive B cells and preexisting memory B cells that bound to epitopes conserved between past seasonal strains and the pandemic H1N1 strain. We theorize that predominant activation of the latter, preexisting memory cells can account for the observed high frequency of neutralizing antibodies (11/15 HA-binding antibodies), the majority (9/11) of which are cross-reactive with seasonal H1N1 strains (Fig. 3 C) and other group 1 influenza strains, including H5 HA. Several observations support this conjecture. Most convincingly, there was a particularly high frequency of cross-reactive antibodies overall, with a high level of somatic mutations found particularly among the variable genes of cross-reacting cells (Fig. 2 and Fig. S3 ). In fact, by ELISA most antibodies were cross-reactive and one third of the broadly cross-reactive binding (Fig. 3 B) and have functional activity against multiple recent and older H1N1 strains (Fig. 3 C) . These antibodies were all highly effective at providing prophylactic protection against infection with a lethal dose of mouse-adapted pandemic H1N1 in 6-8-wk-old BALB/c mice (Fig. 5) . Moreover, all three antibodies were effective therapeutically, even when they were administered as late as 60 h after the lethal challenge infection, well after the mice were symptomatic. For EM-4C04, we have successfully treated mice as far out as 72 h post-infection (unpublished data). Infected mice were already showing measurable weight loss that was reversed by administration of the antibody, demonstrating therapeutic potential even after the onset of disease. Viral clearance was analyzed in mice treated at 48 h after infection with EM4C04 (Fig. S4) . As early as day 4, the antibody-treated mice exhibited more than a log reduction in viral titers; titers continued to decline, such that by day 6, virus was undetectable or present at very low levels. The untreated mice perished by day 7 or 8, whereas the treated mice cleared the infection with no detectable virus on day 12. Finally, 1009-3B06 and 70-1F02, which showed activity against several current and older H1N1 seasonal influenza strains in vitro (Fig. 3 C) , were also tested in vivo against antigenically distinct influenza strains. For these experiments, mice were treated with 200 µg of mAb intraperitoneally 12 h before infection with a lethal dose of either pandemic H1N1 influenza or either of the two common influenza laboratory strains PR/8/34 or FM/1/47. 1009-3B06 and 70-1F02 showed protection against these antigenically distinct H1N1 influenza strains, as illustrated in Fig. 5 . EM-4C04, which is highly specific Figure 6 . Breadth of in vivo prophylactic efficacy in mice. 6-8-wk-old BALB/c mice were treated with 200 µg (10 mg/kg of body weight) EM-4C04, 70-1F02, or 1009-3B06 human mAb intraperitoneally. Control mice were treated with PBS only, a control mAb or polyclonal human IgG. 12 h later, they were challenged with a 3xLD50 dose of mouse adapted pandemic H1N1, PR/8/34, or FM/1/47 influenza virus. All mice were monitored daily for body weight changes and any signs of morbidity and mortality. Percentage of initial body weight (left) and survival curves (right) are plotted. Infected, untreated mice showed clear signs of sickness 4-5 d after infection and perished by day 8-9. Figure shows one representative experiments of at least three independent repeat experiments. Antibody treatment conferred significant protection as determined by comparison of weights in untreated versus prophylaxis, and at the time of treatment versus 12 d after infection (unpaired, two-tailed Student's t test, P < 0.05). The log-rank test indicated significant survival as well (P < 0.003). seroconversion, or by the presence of highly potent antibodies, such as EM-4C04, whose activities were less likely to titer out. The highly specific nature of the response from this patient may have contributed to this advantage, ultimately better targeting the epitopes of the pandemic H1N1 strain. In contrast, patient 1009 had relatively low HAI and MN serum titers but the highest frequency of broadly neutralizing antibodies and a less severe disease course. One possibility is that our sampling from this patient was done before peak serological responses. Another possibility is that the high frequency of these potent antibodies in the memory B cell compartment may have resulted in rapid resolution of infection, precluding the development of a high serological response. A third possibility is that despite broader protection, the stalk-reactive antibodies are on the whole less potent and more rapidly titrated out than the highly specific antibodies to the HA globular head. These various possibilities will be of significant interest to study in the future. Finally, we report the development of a large panel of human mAbs induced by pandemic H1N1 infection. Prophylactic therapy with polyclonal or mAbs has successfully been used for RSV, rabies, Hepatitis A and B, and varicella. In the case of influenza, mAbs have been shown to provide prophylactic or therapeutic protection in mice and other animal models (Reuman et al., 1983; Sweet et al., 1987; Palladino et al., 1995; Renegar et al., 2004) . Passive transfer of maternal antibodies in humans has also been shown to confer protection (Puck et al., 1980) . Several of the antibodies we isolated have broad neutralization capacity in vitro against divergent influenza strains and show potent prophylactic and therapeutic activity when used to treat mice that were lethally infected with influenza. These antibodies could provide much needed pandemic therapeutics to treat severe cases of influenza and to protect high-risk populations. In conclusion, analyses of the 46 mAbs induced by pandemic H1N1 infection indicated frequent activation of broadly reactive B cells. We propose that these cells had a memory cell origin caused by cross-reactivity to conserved and functionally important epitopes. If true, it will be important to characterize the efficacy of the pandemic H1N1 vaccine to induce a similarly cross-protective response. All studies were approved by the Emory University, University of Chicago, and Columbia University institutional review boards (Emory IRB#22371 and 555-2000, U of C IRB# 16851E, CU IRB#AAAE1819). Patient clinical information is detailed in Table I. PBMC and plasma isolation. All work with samples from infected patients was performed in a designated BSL2 + facility at Emory University. Peripheral blood mononuclear cells (PBMCs) were isolated using Vacutainer tubes (BD), washed, and resuspended in PBS with 2% FCS for immediate use or frozen for subsequent analysis. Plasma samples were saved at 80°C or frozen in medium with 10% dimethyl sulfoxide for subsequent analysis. Viruses and antigens. The pandemic H1N1 influenza virus (A/California/ 04/2009) was provided by R.J. Webby (St. Jude Childrens Hospital, Memphis, TN). Influenza virus stocks used for the assays were freshly grown in eggs, antibodies bound to past annual viral antigens at lower concentrations, suggesting higher avidity to past influenza strains than to the current pandemic H1N1 virus. Further, crossreacting cells that bind with higher affinity to the pandemic H1N1 strain also have the highest frequency of variable-gene mutations (Fig. S3 B) . Antibodies that were broadly crossreactive were among the more highly mutated clones (Fig. S3 B) . We propose that many of these cells were specific for crossreactive epitopes present in annual influenza strains that then underwent further affinity maturation and adaptation to the infecting pandemic H1N1 virus. Supporting this conjecture, Corti et al. (2010) first demonstrated that naturally occurring HA stalk-reactive memory B cells could be isolated from the blood of people recently immunized with the annual vaccine, before the outbreak of pandemic H1N1. The nature of that study was to screen EBV-transformed memory cell lines, thus precluding the determination of precise frequencies of these stalk-reactive B cells. However, these antibodies were estimated to be quite rare; occurring at one in thousands to one in hundreds of influenza-binding B cells, varying by individual. In contrast, we show that plasmablasts activated by infection with the highly novel pandemic H1N1 influenza strain have substantially increased targeting to the HA stalk region epitopes, totaling 10% of all influenza-specific antibodies and half of the neutralizing antibodies (Fig. 4) . In fact, most specific antibodies isolated in this study were cross-reactive to past influenza strains. Collectively, the data described supports a model in which divergent viruses that are conserved only at the most critical regions for function will elicit a higher proportion of cross-reactive and neutralizing antibodies. Thus, although the activated plasmablasts of relatively few patients could be analyzed in detail at the monoclonal antibody level, we proffer that with the proper immunogen, the long-sought development of a pan-influenza vaccine might be possible. Interestingly, the highly specific antibody EM-4C04 was derived from a patient that had a very severe disease course, with persistent viral shedding over several weeks. In addition, the variable genes from the plasmablasts of this patient had the lowest average number of somatic mutations (Fig. 2 B , outlier, and Fig. S3 B) . Collectively, the unique specificity against pandemic H1N1, the low levels of somatic mutation, and the unusually severe disease in the absence of predisposing conditions suggest that this patient may have mounted a primary immune response to the pandemic H1N1 influenza infection. The complete lack of preexisting immunity may have contributed to the more severe disease observed in this patient. In contrast, the activation of broadly cross-neutralizing memory B cells in those with immune experience to annual strains might have contributed to the less severe disease of most infected patients during the pandemic. It is notable that there is a discrepancy between patients for serum MN titers, the severity of disease, and the frequency of plasmablasts expressing neutralizing antibodies (Table I and Fig. 3 ). For example, patient EM, despite having the worst disease course, had the greatest HAI and MN serum titers. This may be caused by the time from infection (day 31), allowing full conditions that were previously published (Wrammert et al., 2008; Smith et al., 2009) . Variable genes were determined using in-house analysis software compared with the Immunogentics V gene dataset and the IMGT search engine (Ehrenmann et al., 2010; Lefranc et al., 2009 ). Background mutation rates by this method is 1 base-exchange per 1,000 bases sequenced (based on sequences of constant region gene segments). Comparisons were made to historical data, some of which was previously published (Zheng et al., 2005; Wrammert et al., 2008; Duty et al., 2009) . Plaque assay and PRNT 50 assay. MDCK cells were grown in 6-well plates at a density of 8 × 10 5 /well. On the next day, cells were washed with PBS. 10-fold dilutions of virus were added in 500 µl DME and incubated at 37°C for 1 h, with mixing every 10 min. Cells were washed with PBS and overlayed with 199 media containing 0.5% agarose (Seakem), 1x antibiotics (100 U/ml penicillin and 100 mg/ml streptomycin), 0.2% BSA (Sigma-Aldrich), and 0.5 µg/ml TPCK-Trypsin (Sigma-Aldrich). Cells were incubated for 36-40 h and fixed with 2% PFA for 10 min. Agarose plugs were removed and cells were stained with 0.1% crystal violet in 25% EtOH for 1 min. After removal from the crystal violet solution, plates were dried and used to count plaques in each well. For PRNT 50 assay, MDCK cells were prepared as above. On the next day, mAbs were threefold-diluted (60-0.74 µg/ml). 100 PFU of virus in 250 µl DME were incubated with equal volume of diluted mAbs at 37°C for 1 h before the plaque assay. Plaques were counted and the final concentration of antibodies that reduced plaques to <50 PFU were scored as PRNT 50 . To determine the TCID 50 , MDCK cells were grown in 96-well plate at a density of 1.5 × 10 4 /well. On the next day, cells were washed with PBS and 10-fold diluted viruses in 100 µl DME were added into each well and incubated at 37°C for 1 h. After the incubation, cells were washed with PBS and 100 µl of DME containing 1x antibiotics (100 U/ml penicillin and 100 mg/ml streptomycin), 0.5% BSA (Sigma-Aldrich), and 0.5 µg/ml TPCK-Trypsin (Sigma-Aldrich) was added. Cells were further incubated for 60 h, and 50 µl of the supernatant was incubated with equal volume of 0.5% of PBS-washed Turkey red blood cells (Rockland Immunochemicals) for 30 min. Four replicates were performed for each dilution, and complete agglutination was scored as HA + . Virus titers were calculated by the Reed-Muench method. For MN assay, 100 TCID 50 of virus in 50 µl DME were incubated with 50 µl of threefold-diluted antibodies (60-0.082 µg/ml) at 37°C for 1 h. Cells were washed and incubated in the media as described for the HAI assay for 60 h. The MN titer was determined to be the final concentration of mAbs that completely inhibited infection. HAI and ELISA assays. Whole virus, recombinant HA, or vaccine-specific ELISA was performed on starting concentrations of 10 µg/ml of virus or recombinant HA and on 1:20 dilution of the vaccine, as previously described (Wrammert et al., 2008) . In brief, microtiter plates were coated with live virus strains totaling 8 HAU of total virus per well or with 1 µg/ml of recombinant HA protein. To standardize the various ELISA assays, common highaffinity antibodies with similar affinities and binding characteristics against each virus strain were included on each plate, and the plate developed when the absorbance of these controls reached 3.0 ± 0.1 OD units. Goat anti-human IgG (goat anti-human I-peroxidase-conjugate; Jackson ImmunoResearch Laboratories) was used to detect binding of the recombinant antibodies, followed by development with horseradish peroxidase substrate (Bio-Rad laboratories). Absorbencies were measured at OD415 on a microplate reader (Invitrogen). Affinity estimates were calculated by nonlinear regression analysis of curves from eight dilutions of antibody (10 to 0.125 ug/ml) using GraphPad Prism. The HAI titers were determined as previously described (Wrammert et al., 2008) . In brief, the samples were then serially diluted with PBS in 96-well v-bottom plates and 8 HAU (as determined by incubation with 0.5% turkey RBCs in the absence of serum) of live, egg-grown virus was added to the well. After 30 min at room temperature, 50 µl of 0.5% turkey RBCs prepared, and purified as previously described (Wrammert et al., 2008) . The hemagglutination inhibition activity was determined using turkey red blood cells (Lampire Biological Laboratories) as previously described (Wrammert et al., 2008) ELISPOT assay. Direct ELISPOT to enumerate the number of either total IgG-secreting, pandemic H1N1 influenza-specific, or vaccine-specific plasmablasts present in the PBMC samples were essentially done as previously described (Crotty et al., 2003) . In brief, 96-well ELISPOT filter plates (Millipore) were coated overnight with either the optimized amounts of purified pandemic H1N1 virions, recombinant HA from the pandemic H1N1 (as above), the 08/09 influenza vaccine at a dilution of 1/20 in PBS, or goat anti-human Ig (Invitrogen). Plates were washed and blocked by incubation with RPMI containing 10% FCS at 37°C for 2 h. Purified and extensively washed PBMCs or sorted plasmablasts were added to the plates in dilution series and incubated for 6 h. Plates were washed with PBS, followed by PBS containing 0.05% Tween, and then incubated with a biotinylated anti-huIgG () antibody (Invitrogen) and incubated for 1.5 h at room temperature. After washing, the plates were incubated with an avidin-D-HRP conjugate (Vector Laboratories) and, finally, developed using AEC substrate (3 amino-9 ethylcarbazole; Sigma-Aldrich). Developed plates were scanned and analyzed using an automated ELISPOT counter (Cellular Technologies, Ltd.). Flow cytometry analysis and cell sorting. Analytical flow cytometry analysis was performed on whole blood after lysis of erythrocytes and fixing in 2% PFA. All live cell sorting and single cell sorting was performed on purified PBMCs using either a FACSVantage or ARIAII cell sorter system. All of the following antibodies for both analytical and cell sorting cytometry were purchased from BD, except anti-CD27, which was purchased from eBioscience: anti-CD3-PECy7 or PerCP, anti-CD20-PECy7 or PerCP, anti-CD38-PE, anti-CD27-APC, and anti-CD19-FITC. ASCs were gated and isolated as CD19 + CD3  CD20 lo/ CD27 high CD38 high cells. Flow cytometry data were analyzed using FlowJo software. Generation of mAbs. Identification of antibody variable region genes were done essentially as previously described (Smith et al., 2009; Wardemann et al., 2003; Wrammert et al., 2008) . In brief, single ASCs were sorted into 96-well PCR plates containing RNase inhibitor (Promega). VH and V genes from each cell were amplified by RT-PCR and nested PCR reactions using cocktails of primers specific for both IgG and IgA using primer sets detailed in (Smith et al., 2009 ) and then sequenced. To generate recombinant antibodies, restriction sites were incorporated by PCR with primers to the particular variable and junctional genes. VH or V genes amplified from each single cell were cloned into IgG1 or Ig expression vectors, as previously described (Wardemann et al., 2003; Wrammert et al., 2008; Smith et al., 2009) . Antibody sequences are deposited on GenBank (accession nos. HQ689701-HQ689792 available from GenBank/EMBL/DDBJ). Heavy/light chain plasmids were cotransfected into the 293A cell line for expression and antibodies purified with protein a sepharose. Antibody proteins generated in this study can be provided in limited quantities upon request. Mutational analysis. Antibody anti-H1N1 induced plasmablast variable genes were amplified by single-cell RT-PCR using primer sets and PCR efficacy of the mAb, mice were treated intraperitoneally with 200 µg (10 mg/kg of body weight) of the specific mAbs. 12 h later, mice were challenged with 3xLD50 of one of the mouse adapted influenza viruses used in the study. All mice were monitored daily for any signs of morbidity and mortality. Body weight changes were registered daily for a period of 14 d. All mice that lost >25% of their initial body weight were sacrificed according to the institutional animal care and use committee guidelines. To determine the therapeutic efficacy of the mAbs, mice were challenged with 3xLD50 of the mouse-adapted pandemic H1N1 virus. At various times after infection (12, 24, 36, 48, 60 , and 72 h) mice were treated intraperitoneally with 200 µg (10 mg/kg of body weight) of the specific mAbs. All mice were monitored daily and the body weight changes were registered daily as described above. Statistical analysis. Data were collected and graphed using MS Excel and GraphPad Prism software. Efficacy of the therapeutic and challenge experiments was evaluated by analysis of variance using GraphPad Prism software. Online supplemental material. Fig. S1 shows the binding characteristics of control mAbs. Fig. S2 shows further binding characteristics of the neutralizing mAbs. Fig. S3 shows further analysis of pandemic H1N1-induced plasmablast somatic mutations. Fig. S4 shows experiments demonstrating the therapeutic control of pandemic H1N1 viral titers in lungs after mAb treatment. Tables S1-S3 provide detailed characteristics concerning the variable gene sequences cloned from pandemic H1N1 induced plasmablasts. Online supplemental material is available at http://www.jem.org/cgi/ content/full/jem.20101352/DC1.

Assessment of Virally Vectored Autoimmunity as a Biocontrol Strategy for Cane Toads

BACKGROUND: The cane toad, Bufo (Chaunus) marinus, is one of the most notorious vertebrate pests introduced into Australia over the last 200 years and, so far, efforts to identify a naturally occurring B. marinus-specific pathogen for use as a biological control agent have been unsuccessful. We explored an alternative approach that entailed genetically modifying a pathogen with broad host specificity so that it no longer caused disease, but carried a gene to disrupt the cane toad life cycle in a species specific manner. METHODOLOGY/PRINCIPAL FINDINGS: The adult beta globin gene was selected as the model gene for proof of concept of autoimmunity as a biocontrol method for cane toads. A previous report showed injection of bullfrog tadpoles with adult beta globin resulted in an alteration in the form of beta globin expressed in metamorphs as well as reduced survival. In B. marinus we established for the first time that the switch from tadpole to adult globin exists. The effect of injecting B. marinus tadpoles with purified recombinant adult globin protein was then assessed using behavioural (swim speed in tadpoles and jump length in metamorphs), developmental (time to metamorphosis, weight and length at various developmental stages, protein profile of adult globin) and genetic (adult globin mRNA levels) measures. However, we were unable to detect any differences between treated and control animals. Further, globin delivery using Bohle iridovirus, an Australian ranavirus isolate belonging to the Iridovirus family, did not reduce the survival of metamorphs or alter the form of beta globin expressed in metamorphs. CONCLUSIONS/SIGNIFICANCE: While we were able to show for the first time that the switch from tadpole to adult globin does occur in B. marinus, we were not able to induce autoimmunity and disrupt metamorphosis. The short development time of B. marinus tadpoles may preclude this approach.

The spread of the cane toad, Bufo (Chaunus) marinus, into the Australian environment following the initial introduction at Gordonvale near Cairns, Queensland in 1935 has been spectacularly successful. Cane toads are now present throughout most of tropical northern Australia and their range is continuing to expand into world heritage areas. Predicted warming due to climate change could extend the range of the cane toad south into what are currently temperate regions [1] . The cane toad is a highly toxic and hardy introduced species and presents wide ranging ecological and social impacts within the Australian landscape. Native species such as quolls, goannas and native frogs are particularly susceptible to the cane toad toxin and many populations have been severely impacted by the arrival of cane toads [2, 3, 4] . Attempts to halt the spread cane toads have so far been unsuccessful mainly due to the extensive, remote and inaccessible areas inhabited by the toads. An infectious biological agent appears to be the only viable option for controlling cane toads at such continental scales but as yet no known naturally occurring microbes have been confirmed as Bufo specific. An alternative option is to explore whether an infectious agent can be genetically modified to carry a gene that will specifically disrupt the cane toad life cycle, requiring selection of cane toad specific target genes as well as an infectious agent for delivery. The concept of using genetically modified infectious agents to deliver antigens to wildlife is not new. Recombinant vaccinia virus expressing rabies glycoprotein delivered in baits to wild foxes has proved to be a highly effective strategy to combat rabies [5] . Since then other vaccines developed against diseases of wildlife include a rabies virus based vector used to immunise wildlife against SARS [6] . Extension of this concept has seen recombinant viruses developed to control a host's biological processes. An example is recombinant viruses expressing zona pellucida antigen that successfully deliver immunocontraception to pest animal species in laboratory trials [7, 8] . Bohle Iridovirus (BIV) is a ranavirus in the family Iridoviridae. BIV was originally isolated from the Bohle River region in northern Queensland [9] and is the only documented isolation of a virus from amphibians in Australia. It is capable of infecting cane toad tadpoles [10] and is therefore a candidate for testing the viral delivery of genes in this species. Furthermore, in recent studies we have shown that BIV can carry and express foreign genes in vitro [11] . Selection of target genes for delivery to cane toads has focused on metamorphosis since it is a critical phase in the amphibian life cycle. Metamorphosis is characterised by rapid and extensive morphological changes [12, 13] , accompanied by strong shifts in the expression of genes and proteins at the molecular level [14, 15] . Cane toad genes that are expressed in metamorphs but not in tadpoles therefore represent ideal targets to block and thus manipulate aspects of development. One documented example of the transition from a larval to an adult form in amphibians is haemoglobin [16] and we have used microarray analysis to establish that adult haemoglobin is significantly upregulated during cane toad metamorphosis [17] . Injecting tadpoles with adult globin interfered with expression of this protein in Rana catesbeiana, and induced changes in gene expression profiles of metamorphs [18] . Thus we hypothesise that it may be possible to alter metamorphosis by immunologically sensitising larval stages (tadpoles) to proteins expressed only in later post-metamorphic stages. Adult globin is not a cane toad specific gene; we used it here to determine whether autoimmunity might affect the protein profile of metamorphs. If successful, the concept would be extended to cane toad specific genes that were upregulated at metamorphosis. This study outlines an investigation into the feasibility of an immunologically based biocontrol for cane toads. We demonstrate the presence of a clear larval to adult switch in haemoglobin mRNA and protein levels and hypothesise that this switch may be affected by early exposure, by either injection or viral delivery, to adult B. marinus haemoglobin. Our results indicate that the altered adult globin protein profile seen in Rana catesbeiana metamorphs after exposure of tadpoles to adult globin does not occur in B.marinus. The short larval stage in B. marinus compared with R. catesbeiana may preclude this approach to cane toad biocontrol. All animals used in these studies were sourced from a colony of B. marinus maintained at CSIRO according to the methods described in Hamilton et al. [19] . Briefly, when tadpoles were required, adults were injected subcutaneously with a 0.25 mg/mL solution of leuprorelin acetate to induce ovulation and stimulate amplexus. Eggs were hatched and tadpoles maintained in aged water without chlorine at a temperature of 23-27uC. Authority for the use of animals was provided by CSIRO animal ethics committees in accordance with the Australian National Health and Medical Research Council's code of practice [20] . These permits were (i) CSIRO Sustainable Ecosystems Animal Ethics Committee, Approval No. 08-05, exposure of pre-and post-metamorphic cane toads to proteins, DNA and RNA and produced RNA/cDNA and (ii) CSIRO Australian Animal Health Laboratory Animal Ethics Committee, Approval number 1132, biological control of cane toads. Production and purification of recombinant globin and antisera B. marinus adult and tadpole globins (GenBank Accession numbers EL342145 and EU877979, respectively) were amplified using the following full length primer sets: adult globin sense 59-ATGGTCCATTTGACAGATCAC -39, and antisense 59-TTA-GTGGTAACCCTTGCCAAG -39 (444 bp), or tadpole globin sense 59-ATGGTTCATTGGACCGCTGAAGA -39, and antisense 59-TTAGAAATAGCCATGGCTCAGG -39 (444 bp). The fragments were cloned into the bacterial expression vector pDEST17 and expressed as His 6 -tagged proteins in E. coli BL21-AI cells (Invitrogen). Cultures were grown overnight (37uC) in LB supplemented with antibiotics, then diluted 100-fold and grown to an OD of 0.6 (600 nm). L-arabinose (Sigma) was added (0.2% final conc.) to induce protein production and incubation continued for 3-5 h. Bacteria were harvested by centrifugation, rinsed and resuspended in Tris-buffered saline (TBS: 50 mM Tris, 500 mM NaCl; pH 7.5), disrupted by freeze/thaw cycles and centrifuged at 10,0006 g for 30 min. The pellet was solubilised in TBS containing 8 M Urea for 30 min and then centrifuged at 20,0006 g for 30 min to remove insoluble materials. His 6 -tagged proteins were purified in the denatured state using Ni 2+ NTA agarose (Qiagen), washed via imidazole-containing steps (TBS+20, 30 or 40 mM imidazol) and eluted in TBS+500 mM imidazole. Size-based secondary purification was then achieved by continuous-elution electrophoresis (Model 491 Prep Cell, Bio-Rad). Globin proteins were dialysed against amphibian Ringers solution [4.89 g NaCl, 0.298 g KCl, 0.265 g CaCl 2 .2H 2 0, 0.197 g MgSO 4 .7H 2 0, 1.495 g NaHCO 3 , 0.127 g NaH 2 PO 4 .H 2 O and 1.982 g glucose per litre dH 2 O] overnight at 4uC and concentrations determined using the Bio-Rad Protein Assay. Proteins were separated by polyacrylamide gel electrophoresis (SDS-PAGE) (15% gels) using the Bio-Rad Protean II System and visualised with Coomassie brilliant blue. Purified recombinant globin proteins were used as immunogens for the generation of rabbit antiserum. Briefly, two rabbits per immunogen were injected 10 days apart with vaccine containing 50 mg of either adult (rAdglob) or tadpole (rTadglob) recombinant globin. Two weeks later a third dose was administered if a boost to the antibody response was required. Each dose was prepared in CSIRO triple adjuvant (60% v/v Montanide; 40%, v/v rHb [combined with Quil A, 3 mg/mL, and DEAE-dextran, 30 mg/mL]), in water. The time course of adult and tadpole globin production in normal animals was determined. Animals were sampled at stages 27, 28, 34, 36, 40, 41, 42, 45 , 46 (metamorph) and 1 month post metamorph, $3 animals per stage. Developmental stages of B. marinus were determined according to the method of Limbaugh and Volpe [21] . Levels of mRNA and protein were measured individually except for tadpoles at stages 27 and 28 where .10 animals were pooled due to the small size of tadpoles in the early stages of B. marinus development. Recombinant BIV (rBIV) expressing the neomycin resistance gene and adult globin (rBIV/neo r /Adglob), and the control virus without adult globin (rBIV/neo r ) were constructed according to Pallister et al. [11] . To assess the effect of viral delivery of adult globin to B. marinus tadpoles, seven groups of tadpoles were infected at day 6 (stage 20) post hatching. Group A, uninfected cell culture supernatant; Groups B, C and D -10 2 , 10 3 and 10 4 TCID 50 /mL of the negative control virus, rBIV/neo r respectively; Groups E, F and G -10 2 , 10 3 and 10 4 TCID 50 /mL of the test virus, rBIV/neo r / Adglob respectively. Each group of 116 tadpoles was infected for approximately 6 h in 2 L of water containing the appropriate concentration of virus, rinsed in clean water for 5 min then divided into 4 tubs each containing 29 tadpoles. These tubs were randomly dispersed around 3 different rooms to allow for statistical variation due to the position of the tub. Sampling was carried out at 5 different stages, 20, 28, 33, 42 and 1-2 weeks post tail resorption, according to the schedule in Table 1 . Tadpoles were sampled at each of the 5 stages for the detection of virus by real time PCR, and at all stages except stage 20, where tadpoles are very small, for the analysis of native adult globin profiles. However, tadpoles were only sampled up to and including stage 33 (just before the onset of native adult globin production) for the detection of viral adult globin. Five tadpoles per group were sampled at all stages except post tail resorption, where, for statistical purposes, 30 animals per group were sampled. Tadpoles were euthanased by bathing for $5 minutes in 0.2% MS-222 (see Preparation of immunogen and inoculation trials). Tadpoles for real time PCR and for detection of viral adult globin were then frozen at 280uC until further processing. Tadpoles for blood sampling were euthanased as described then decapitated and blood was collected using a micropipette containing 0.2 M EDTA to prevent clotting. Red blood cells (RBCs) were pelleted at 13,0006g for 5 min, washed with PBS then lysed by resuspending in an equal volume of distilled water. Following centrifugation at 13,0006 g for 5 min the supernatant containing globin was removed and stored at 280uC. Total RNA (5 mg) was extracted from tadpoles and toadlets using Trizol reagent (Invitrogen) according to the manufacturer's instructions and reverse transcribed using the cDNA First Strand Synthesis Kit (Invitrogen). Real time PCR was used to detect larval and adult globin mRNA at selected stages of development. Oligonucleotide sets were larval globin; sense 59-GCTGAAGA-GAAAGCCGC -39, and antisense 59-ATGGCGGTGACATTG-GAC -39 (151 bp), or adult globin; sense 59-CAGATCAC-GAGCTCAAGAG -39, and antisense 59-ATGGCATCAG-CAGAGCCA -39 (151 bp). Results were standardised with cane toad actin (GenBank Accession no. EL595572) by amplifying a 117 bp fragment using sense 59-ATGACACAGATAATGTTT-GAGAC -39 and antisense 59-ATCACCAGAGTCCATCA-CAAT -39 primers. Reactions consisted of QuantiTect SYBR Green RT-PCR Master Mix (Qiagen), 0.5 mM oligonucleotides and 200 ng of first strand cDNA template, and run on a RotorGene 2000 (Corbett). The thermal profile was as follows: 50uC for 2 min, 95uC for 2 min, 40 cycles of 95uC for 30 sec; 58uC for 30 sec; and 72uC for 30 sec. PCR target products were used as reaction standards at concentrations ranging from 1610 2 to 1610 8 molecules per mL. Analysis of real time PCR reactions and melting point dissociation curve reactions were performed using the programme Rotorgene version 6.0 (Corbett). For DNA extraction the animal was thawed, weighed and Prepman Ultra (Applied Biosystems) was added directly to the tube at the following rates: #0.20 g, 450-500 mL; 0.21-0.30 g, 500-750 ml; .0.31 g, 900 mL. This was followed by approximately 100 mL sterile 1.0 mm zirconia silicone beads (BioSpec Products). The tadpole was homogenised in a mini-beadbeater (BioSpec Products) for 30 sec, microfuged at 13,0006 g for 30 sec then homogenised in a mini-beadbeater for a further 30 sec. To extract DNA the sample was heated to 100uC for 20 min, left to stand for 4 min at room temperature (RT), microfuged at 13,0006 g for 6 min and the DNA (350-650 ml) transferred to a clean tube. Each reaction in the TaqMan assay contained 12.5 mL of Platinum Quantitative PCR SuperMix-UDG (Invitrogen), 90 nM sense (59-CTCATCGTTCTGGCCATCAA -39), 90 nM antisense (59-TCCCATCGAGCCGTTCA -39) primers, 25 nM MGB TaqMan probe (59-CACAACATTATCCGCATC-39), 5 mL of a 1/10 dilution of template DNA in dH 2 0, 0.05 mL of Rox dye, and water to a final volume of 25 mL. The reaction was run on the ABI 7500 Fast Real-Time PCR machine with a thermal profile was as follows: 2 min at 50uC, 10 min at 95uC, 45 cycles of 15 sec at 95uC then 1 min at 60uC. Results were automatically plotted by the Sequence Detection System Software version 1.3.1 (Applied Biosystems). Proteins for western blot were extracted either from Trizol samples according to the manufacturer's instructions, or from virally infected tadpoles homogenised in 10 mM Tris with 100 mL sterile 1.0 mm zirconia silicone beads (BioSpec Products). The tadpole was homogenised in a mini-beadbeater (BioSpec Products) for 30 sec, centrifuged at 13,0006 g for 30 sec then homogenised in a minibeadbeater for a further 30 sec. SDS was added to a final concentration of 5% and the homogenate heated to 100uC for 1 min. After 30 min at RT to allow SDS to penetrate the sample, the homogenate was centrifuged, the supernatant decanted, heated to 100uC for 1 min and stored at 280uC until used. To extract globin from RBCs, blood was processed as previously described (see Infection and sampling) and the protein content of each sample was determined Samples were collected from stage 46 metamorphs anaesthetised by bathing for 2 min in 0.22% MS-222 followed by decapitation. Samples were collected by aspiration and pelleting of RBCs prior to the collection of sera-like fluids. Adult blood samples were collected as per Zupanovic et al. [23] . Microtitre plates (Beckton-Dickonson) were coated with 50 mL/well of rAdglob (10 mg/mL) and positive control ovalbumin (4 mg/mL; Sigma) in Ringer's solution overnight at 4uC and washed (x3) in TBS with Tween-20 (0.05%; TBST). The plates were blocked for 2 h with 5% skim milk powder (Bio-Rad) in TBST at 37uC. Metamorph sera (treated and control) and adult toad aovalbumin control sera were serially diluted in 1% skim milk/TBS and applied to plates (50 mL/well). After incubation for 1 h at 37uC the plates were washed with TBST three times. Rabbit antisera against toad IgG (1:1000; Zupanovic et al. [23] ) was added in 1% skimmed milk/TBS, incubated for 1 h at 37uC and washed (x3) in TBST. Next, goat a-rabbit IgG (1:2000; KPL) was added to each well, incubated for 1 h at 37uC and again washed (x3) in TBST. HRP-conjugated streptavidin (50 mL; KPL) was incubated for 30 min at RT, washed three times and developed with peroxidase substrate solution (50 mL of TBM; KPL) and the absorbance measured (405 nm). Fitness was assessed using burst swim speed for tadpoles and maximum jump distance for metamorphs. At 11 days post injection (,1 week prior to metamorphic climax/stage 46) average burst speed in tadpoles was measured according to the method described by Van Buskirk and McCollum [24] where specific speed is normalised to body length rather than measuring absolute speed. Temperature at the time of testing was constant at 23-24uC. Jump distances were calculated in stage 46 metamorphs approximately 2 days after removal from aquatic tanks according to Wilson and Franklin [25] . Jumps were recorded using a digital camera (Canon; 15 frames sec 21 ) and distances calculated with MouseZoom version 1.4 (Freeware 1998 1.4 (Freeware -2003 and Microsoft Windows Movie Maker version 5.1 (1998) (1999) (2000) (2001) . Absolute distances were determined from screen pixel units with conversion to mm via background graph paper. It was determined that optimal jumps were performed on plastic rather than paper. The time course for the appearance of adult globin mRNA and the disappearance of tadpole globin mRNA during normal cane toad metamorphosis were first assessed by real time PCR (Fig. 1a) . Tadpole globin mRNA was detected at all early stages until the level declined rapidly at the metamorphic climax (Stages 42-46) and was undetectable one month after metamorphosis. Conversely, adult globin mRNA was first detectable at stage 40 and at all following stages until the experiment was terminated. Analysis of total protein extracts at selected stages by Western blot showed that larval haemoglobin was expressed in stage 40 tadpoles but not in stage 46 toadlets (Fig. 1b) . Adult globin protein was first detected in stage 36 tadpoles (faint signal) and continued to increase in abundance after metamorphosis. The positive control used for the experiments was recombinant adult globin that migrates as a larger protein than the native globin on PAGE gels due to the addition of a 6XHis-tag and plasmid linker sequence. The detection system was specific; anti-larval globin antibody detected larval globin, but not recombinant and native adult haemoglobins, and vice versa for the anti-adult globin antibody. Thus we confirmed that the globin switch seen at the mRNA level was also seen at the protein level. rAdglob within inoculated tadpoles was clearly detected by rabbit antibody to adult globin as an 18 kDa band for several days after injection (Fig. 2) . rAdglob levels appeared reduced by half every 2-3 days, until 14 days post injection where no protein was detected. Gross morphology during metamorphosis determined by wet weight and length was similar for treated (rAdglob) and control groups (no injection, or Freund's adjuvant only) (Fig. 3a) . Treatment with adult haemoglobin did not significantly delay metamorphosis. There was no significant difference in tadpole fitness between treatment and control groups just before metamorphic climax measured by swimming performance (burst swim speed). A p-value of 0.086 was determined in Microsoft Excel using the student t-test, 2-tailed distribution, 2-samples of unequal variance. Likewise there was no significant difference between treatments in the fitness of metamorphs as measured by maximum jump length (Fig. 3b) . A p-value of 0.996 was determined using the same student t-test as for the swim speed. Metamorphs from each of stages 36, 40, 42 and 46 were pooled and analysed for differences in adult globin mRNA between treated and untreated groups. The results indicated no significant differences in adult globin mRNA levels between treated and untreated groups at any of these 4 stages (Fig. 4a) . A p-value of 0.914 was determined using the same student t-test as for the swim speed and jump length. As pooling animals from each stage could have masked effects in individual animals, 6 animals were taken from the treated and untreated groups at stage 46 and analysed individually. Again, no significant difference in globin mRNA levels was observed between individuals from treated and control groups. A p-value of 0.095 was determined using the student t-test outlined previously. As seen for the mRNA studies, we detected no change in the protein profiles of adult globin immunised metamorphs (n = 10) compared to control (n = 9) animals by Western blot (Fig. 4b) . Similar results were recorded in preliminary trials conducted using native globin purified from the blood of adult toads rather than recombinant globin. These inoculations had no effect at the morphological or mRNA and protein levels (data not shown). We were unable to detect reactive antibodies (IgY) against recombinant globin protein in any of the animals by ELISA. However, we were also unable to detect antibody to a normally highly immunogenic test antigen (ovalbumin) in metamorphs, although reactive antibodies generated by immunising one adult with ovalbumin were readily detectable (Fig. 5 ). The effect of recombinant virus infection on mRNA expression levels in tadpoles was assessed using real time PCR. Treated and control animal groups were sampled at stages 20, 28, 33, 42 and 1-2 weeks post tail resorption (Table 2) . At stage 20, after infection and rinsing, no virus was detected in any animals indicating that there was no background level of BIV detectable by real time PCR and that any virus detected at later stages was the result of virus replication. BIV was detected in all of the infected groups, with more infected animals detected as the inoculum increased. The control Analysis of adult globin profiles was carried out on RBCs from tadpoles taken at stage 42 and metamorphs at 1-2 weeks post tail resorption, by which time the switch has been made from tadpole to adult globin production. Blood from tadpoles given all three doses of the test virus (rBIV/neo r /Adglob that does express adult globin) and the control virus (rBIV/neo r that does not express adult globin) were analysed by polyacrylamide gel electrophoresis (PAGE) and silver staining (Fig. 6a) . The protein profiles of blood from animals infected with the rBIV/neo r control (n = 51) and rBIV/neo r /Adglob (n = 89) viruses were all similar. A western blot using rabbit anti toad adult globin confirmed that the main 14 kDa band detected by silver stain was adult globin (Fig. 6b) . We were thus unable to detect any change in the adult globin profile in RBCs taken from the animals that had been exposed to adult globin as tadpoles. In this report we outline the steps we have undertaken to determine whether interference with cane toad tadpole develop- Burst swim speed represents the absolute swim speed normalised to body length, and the full data range (vertical line), standard deviation (box) and mean (horizontal line) are indicated. c: Jumping performance of rAdglob treated and FCA control amimals at stage 46 is shown. Longest jump distance was normalised to body lengths. doi:10.1371/journal.pone.0014576.g003 ment can be achieved using an immunological approach. Our proof of concept approach was largely influenced by previous observations that in bullfrog tadpoles immunised with purified adult globin, the adult globin protein profile was altered in surviving metamorphs [18] . Here we extended the concept to test whether a similar effect could be induced in B. marinus by injecting tadpoles with purified native and recombinant globin as well as viral delivery of this antigen. We first established that the larval to adult globin switch reported in other amphibian species also occurred in B. marinus. It is well documented that a tadpole form of globin in anurans is replaced by adult globin during the course of metamorphosis [18, 26, 27] and here we demonstrate the existence of this switch for the first time in a Bufo species. We have previously demonstrated strong upregulation of adult globin genes during cane toad metamorphosis and that this was more pronounced than for any of the other genes induced at metamorphosis [17] . We have also previously shown that the recombinant BIV can be genetically modified to express adult globin in vitro [11] and confirmed here that this virus (rBIV/neo r /Adglob) is capable of infecting cane toad tadpoles. We therefore used rBIV/neo r /Adglob to assess the effect of viral delivery of an adult specific gene or protein to tadpoles on subsequent metamorphosis, and compared this to the effects of immunisation with purified protein and adjuvants. A number of parameters were used to assess the effect of adult globin delivery to tadpoles. These included behavioural (average burst speed in tadpoles and maximum jump length in metamorphs), developmental (time to metamorphosis, weight and length at various developmental stages, protein profile of adult globin by PAGE) and genetic (mRNA levels for adult globin) measures. However, we were unable to detect any differences between treated and control animals following immunisation with purified globin or exposure to recombinant virus. This contrasts markedly with the effects of globin immunisation in R. catesbeiana reported previously by Maniatis et al. [18] , who speculated that some form of immune response was instrumental in the altered adult globin profile observed in injected tadpoles. Possible explanations for the differences observed between these studies is that tadpoles of B. marinus are inherently less immunocompetent than those of R. catesbeiana and/or they had less time than R. catesbeiana to mount an effective immune response to the globin antigen. Firstly, the cane toad has a very short larval stage of approximately 50-60+days, depending on tadpole density and temperature [28] whereas the larval stage in the bullfrog lasts at least 120 days and up to 2 years depending on environmental conditions [29] . The long larval stage in the bullfrog enabled an initial immunisation using FCA, followed by a boost 1 month later [18] . By contrast, cane toad tadpoles in our study were only large enough to be first injected at stage 26 (approximately day 9), barely 3 weeks before the onset of adult globin synthesis at stages 36-40 (day 30-43) and so a similar boost was not given. Nevertheless we considered that there should be sufficient time for a primary immune response to develop before adult globin appeared, provided cane toad tadpoles recognise the adult globin as a foreign protein. Secondly, due to the large difference between the cane toad and bullfrog developmental time frames, and as age at inoculation was not specified, the bullfrog tadpoles may have been inoculated later in development and been more immunocompetent than in our study. In support of this, studies in Xenopus have demonstrated that immune responses improve with age. The affinity of specific IgY antibodies against dinitrophenol (DNP) in Xenopus larvae is reported to be less than in adults, and in turn much lower than the affinity of mammalian anti-DNP IgG antibodies [30, 31] . The range of antibodies produced to DNP were also less heterogeneous in larval than in adult Xenopus [32] . Cell mediated immunity may also be impaired in tadpoles. In mammals the antiviral response relies on cytotoxic T lymphocytes and these are Major Histocompatibility Complex Class I (MHC class I) restricted. Larval Xenopus reportedly lack MHC classical class I expression [33] suggesting the antiviral response in larvae may be compromised. Studies of the adaptive immune response in Xenopus adults and larvae to frog virus 3 (FV-3), the type virus of the ranavirus genus, indicate this may be so. While adult Xenopus cleared an initial infection and showed an accelerated response to a second injection, tadpoles were much more susceptible, suffering a high mortality rate and a reduced ability to clear the infection compared with infected adults [34] . In spite of lacking MHC class I, tadpoles do have CD8 T cells [35] and so the role played by the lack of MHC class I in the poor antiviral response is unknown. While most of these studies have been carried out in Xenopus, limited studies indicate bullfrogs are capable of mounting a detectable antibody response to an antigen, but apparently not to influenza virus despite repeated inoculations and a substantial and anamnestic response to bacteriophage T7 [36] . Our own studies indicate that B. marinus tadpoles did not mount a detectable antibody response to a widely used and well characterised immunogen, ovalbumin, to which adult B. marinus did respond. Precedents do exist for immune interference in development. Arif et al. [37] showed that when affinity purified antibody to a protein involved in insect metamorphosis was injected into late stage larvae, the development of the larvae into adult moths was defective. Other studies in rabbits [38] and mice [7] have shown Table 2 . Detection of viral DNA in treated and control animals. that the immune response to an antigen can be enhanced by viral delivery. However, all indications are that targeting autoimmune responses in larval amphibians may not be a useful strategy as the capacity of tadpoles to respond to immunological stimuli may be too weak to affect the chain of events at metamorphosis. In conclusion, we have shown that the globin switch occurs in B. marinus and, while we have not been able to perturb this switch immunologically, it remains a viable target for other approaches such as RNA interference. Given the short larval phase in the B. marinus life cycle, antigens produced later than globin may be more effective immunogens and we are currently investigating this possibility. It is also possible that this approach would be more successful in an amphibian with a longer larval phase. Finally, we have designed, developed and delivered a recombinant viral system that will enable the development of future strategies to prevent the spread of the toad.

Analysing the eosinophil cationic protein - a clue to the function of the eosinophil granulocyte

Eosinophil granulocytes reside in respiratory mucosa including lungs, in the gastro-intestinal tract, and in lymphocyte associated organs, the thymus, lymph nodes and the spleen. In parasitic infections, atopic diseases such as atopic dermatitis and asthma, the numbers of the circulating eosinophils are frequently elevated. In conditions such as Hypereosinophilic Syndrome (HES) circulating eosinophil levels are even further raised. Although, eosinophils were identified more than hundred years ago, their roles in homeostasis and in disease still remain unclear. The most prominent feature of the eosinophils are their large secondary granules, each containing four basic proteins, the best known being the eosinophil cationic protein (ECP). This protein has been developed as a marker for eosinophilic disease and quantified in biological fluids including serum, bronchoalveolar lavage and nasal secretions. Elevated ECP levels are found in T helper lymphocyte type 2 (atopic) diseases such as allergic asthma and allergic rhinitis but also occasionally in other diseases such as bacterial sinusitis. ECP is a ribonuclease which has been attributed with cytotoxic, neurotoxic, fibrosis promoting and immune-regulatory functions. ECP regulates mucosal and immune cells and may directly act against helminth, bacterial and viral infections. The levels of ECP measured in disease in combination with the catalogue of known functions of the protein and its polymorphisms presented here will build a foundation for further speculations of the role of ECP, and ultimately the role of the eosinophil.

Eosinophils were discovered in the blood of humans, frogs, dogs and rabbits in 1879 by Dr. Paul Ehrlich [1] . At that time, the German chemical industry was flourishing and Ehrlich took advantage of newly developed synthetic dyes to develop various histological staining techniques. The coal tar derived, acidic and bromide containing dye eosin identified blood cells containing bright red "alpha-granules" and the cells were named eosinophilic granulocytes. Due to the acidity of the staining solution Ehrlich could not at the time say with certainty that the eosinophilic granules contained protein, though he speculated that if present, protein might be denatured by the low pH of the dye [1] . Subsequently it was shown that eosin binds highly basic proteins which constitute the granules of these cells. These charged proteins are contained in on average twenty large granules dispersed throughout the cytoplasm of each cell, which the eosin stain awards the characteristic red spotted appearance that discriminates eosinophils from other leukocytes [2] . More than a century later the physiological roles of these granular proteins have yet to be fully identified. In eosinophil granules pH is maintained at 5.1 by an ATPase [3] where the basic proteins are packed forming crystals [2] . The main content of these granules are four proteins, the major basic protein (MBP) present in their cores, surrounded by a matrix built up of eosinophil peroxidise (EPO), the eosinophil protein X/eosinophil derived neurotoxin (EPX/EDN) and ECP. Vesicotubular structures within the granules direct a differential release of these proteins [4] . The granule proteins were all discovered and characterised about one hundred years after the discovery of the eosinophils [5] [6] [7] [8] . ECP is the best know of the proteins, assessed and used extensively as a marker in asthma and other inflammatory diseases. ECP has been scrutinized in a number of functional studies. The aim of this article is to review some of the findings of ECP quantifications in various diseases and set those in context of the experiments that have functionally analysed the protein. The findings will be used as guidance in a speculation of the biological role of eosinophil. ECP is mainly produced during the terminal expansion of the eosinophils in the bone marrow Eosinophil progenitors (EoP's) in the bone marrow are the first cell identified exclusively of the eosinophil lineages. These EoP's express the cell surface markers IL-5R + CD34 + CD38 + IL-3R + CD45RA -, haematopoietic lineage associated transcription factor GATA-1, ECP mRNA transcripts and have visual characteristics of early eosinophilic blast cell [9, 10] . Most of the granule protein production takes place as EoP's undergo the final stages of maturation [11, 12] . ECP is synthesised, transported and stored in the mature secondary granules at such a high rate as that when the eosinophils are ready to leave the bone marrow, they contain 13.5 μg ECP/10 6 cells [13] ( Figure 1B ). Eosinophils are the major ECP producing cell while monocytes and myelomonocytic cell lines produce minute amounts in comparison [14] . Activated [15] but not resting neutrophils also produce some ECP and have the ability to take up further ECP from the surrounding environment storing it in their azurophil granules [16, 17] . In the myelo-eosinophilic cell line HL-60 clone 15, ECP production is dependent on a nuclear factor of activated T-cells (NFAT)-1 binding site in the intron of the ECP gene (denoted RNASE3) [18] . The RNASE3 gene was formed by gene duplication of an ancestral gene about 50 million years ago, the other duplication gene product being the eosinophil granule protein EPX/EDN gene (RNASE2). ECP and EPX/EDN are two ribonucleases with such a high degree of homology that they are unique to humans and primates and not found in other species. After this gene duplication however, ECP lost part of its ribonuclease activity, but acquired cytotoxic activity, whereas EDN/EPX remained a potent ribonuclease [19] . ECP has homology to pancreatic ribonuclease and has the ability to degrade RNA [20] . The amino acid sequence of ECP has eight cysteine residues spaced all throughout the peptide establishing the tertiary structure of the protein by the formation of four cysteine double bonds. Two catalytic residues, a lysine and a histidine, responsible for the RNA degradation have been identified, K38 and H128 [20, 21] (Figure 2 ) and these residues together with the cysteines are present in all members of the pancreatic ribonuclease family [20] . Analysis of the crystal structure of ECP verified this relationship to these other members of RNase family; namely a β-sheet backbone and three α-helices [22] . In a grove between two of the alpha helices the catalytic site for RNA degradation is located, with ECP showing a preference for cleaving poly-U RNA but not doublestranded RNA [23] . ECP consists of a single-chain peptide of 133 a.a. containing three sites for N-linked glycosylation, a.a.'s 57-59, 65-67 and 92-94 [24] (Figure 2 ). The glycosylation is composed of sialic acid, galactose A. Blood, negative control B. Blood positive, ECP Figure 2 The RNASE3 (ECP) gene and ECP protein sequence with numbers referring to the amino acid sequence. Below the protein sequence is a schematic diagram of the peptide sequence where the beta sheet domains and the alpha helix domains are shown as red arrow and green barrel structures, respectively. Amino acids involved in RNase activity are represented by scissors. Amino acids involved in membrane interference, heparin binding and bactericidal activity are represented by red arrows. Glycosylated amino acids are represented with a glycomoiety while the letter N highlights the nitrated amino acid. A blue box shows the site of the amino acid altering polymorphism rs2073342. and acetylglucosamine [25] explaining the variation in its detected size by Western blot of between 16 and 22 kDa [26] . Nineteen arginine residues facing the outside of the protein giving rise to the proteins basicity (pI > 11) [27] and possibly also its extraordinary stability compared to other ribonucleases [28] . In the presence of H 2 O 2 ECP can be nitrated on tyrosine Y33 by EPO. This inflammation-independent nitration occurs during granule maturation and was suggested to enhance interactions after secretion between several of the otherwise repulsive, positively charged granule proteins ( Figure 2 ) [29] . ECP has been shown to interact with artificial lipid membranes [30] and two tryptophan residues, W10 and W35 facing the outside, similar to the present arginine's, have been associated with this lipid membrane interaction [31] . ECP also has RNase independent cytostatic activity on tumour cells and the tryptophan residues contribute to this activity [32] . W35 was additionally found necessary for killing gram negative and gram positive bacteria [31] . The tryptophan's also facilitate ECP binding to heparin [33, 34] . Another study found that the residues R34, W35, R36 and K38, all part of loop 3 (a.a.'s 32-41) contributed to heparin binding and cytotoxicity [35] (Figure 2) . Surprisingly, when purified from granules of circulating cells, large quantities of the protein were found to lack cytotoxic activity [36] . ECP has not, like EPX/EDN, been found have alarmin activity, stimulating dendritic cells during Th2 immune responses [37] , but ECP has the ability to bind lipopolysaccharide (LPS) and other bacteria cell wall components [38] which might have a priming influence on the immune system. The binding of LPS was mainly attributed to a.a.'s 1 to 45 [39] . The 1 to 45 a.a. region was found to retain bactericidal activity as well as membrane destabilization activity. One commonly occurring polymorphism in the gene is leading to the replacement of an arginine residue with a threonine, R97T [40] ( Figure 2) . The a.a. alteration reduced ECP cytotoxicity to the cell line NCI-H69 assessed by using both recombinant protein [36] and pools of naive protein variants [41] . RNase activity was however not influenced by the R97T alteration. Deglycosylation of the recombinant T97 restored the proteins cytotoxicity suggesting that glycosylation are responsible for this inhibitory role. The physiological function of the granule contained cytotoxic ribonuclease Eosinophils contain a large amount of ECP but the question is why? What is the function of this protein? There is a constitutive baseline level of the eosinophils in many tissues and certain stimuli cause elevated production and influx of eosinophils in different organs. Moreover levels of the ECP in tissue and peripheral blood robustly correlated with the number of eosinophils present, which might be indicative that the function of ECP is also key to the role of eosinophils (see table 1 ). Since the discovery of ECP in 1977 [8] it has been used and evaluated as a biomarker to assess activity in various inflammatory diseases. This analysis has given indirect information of the proteins role in disease. For a comprehensive review of advantages and pitfalls of the usage of ECP as a biomarker in allergic disease see ref [42] . Furthermore, a number of in vitro studies have addressed the direct functional activities of the protein. Detailed following is a comprehensive review of these studies with summaries in table 1 and 2. To simplify comparison the concentrations used have been recalculated to μg/mL using the mean M w of 19.000 for the native protein (average of 16-22 kDa). At homeostasis the eosinophil contributes 1 -4 percent of the circulating leukocyte pool. ECP is readily detectable in blood with plasma levels on the average 3 ug/L (serum 7 μg/L) in healthy individuals which correlates with circulating eosinophil numbers [43] . ECP in blood shows a turnover time (t 1/2 ) of 45 min [44] , and the plasma protein α 2 -macroglobulin (α 2 M) is found to be associated to the protein, in vitro at a molar ratio of 1.6 (ECP/α 2 M). This interaction is facilitated by proteolytic activity of cathepsin G or methylamine [45] , and conceivably takes place to facilitate the clearance of ECP [46] . When eosinophils encounter adhesion molecules expressed on the endothelial cells of post capillary venule wall, the cells adhere and emigrate through the cell layer [47] . Local signals do however drive a low level influx of eosinophils in specific tissues at homeostasis. Eosinophils are present in almost all mucosal associated tissues, nasal mucosa [48] (Figure 3B ), lungs [49] ( Figure 4B ), gastrointestinal mucosa [50] , the reproductive tract, the uterus [51] , breast mucosa of mice [52] and skin [53] . The chemokine eotaxin is responsible for homeostatic eosinophil influx in the gastrointestinal tract in mice [54] whereas the mechanism of influx in other organs remains unknown. In addition, lymphocyte-associated tissue: lymph nodes [50] , thymus [55] and spleen [50] will have some cells stained red by eosin (see Figure 5 ). The majority of ECP is released after the cell has left the circulation [56] . Several types of inflammatory stimulation have been shown to cause eosinophil degranulation. Interaction with adhesion molecules [57, 58] , stimulation by leukotriene B 4 (LTB 4 ), platelet activating factor (PAF) [59] , interleukin (IL)-5 [60] immunoglobulins and complement factors C5a and C3a [61] all cause ECP release. Upon stimulation of eosinophils small variants of ECP with sizes 16.1 and 16.3 kDa are released [62] . One line of studies have suggested that during Plasma cell line 0.5 ng/mL inhibition of Ig production anti ECP ab [87] B lymphocyte cell line 1 ng/mL inhibition of Ig production [88] Rat Peritoneal Mast Cells 17 45 min 50 percent increased histamine release [92] Human heart Mast cells 4.7 60 sec 10-80 percent increased histamine release PGD 2 synthesis Ca 2+ , temperature [94] Guinea-pig tracheal epithelium 103 6 hr exfoliation of mucosal cells [79] Feline tracheal epithelium 2.5 1 hr release of respiratory conjugates [99] Human trachea 2.5 [99] Human primary epithelial cells 10 6 hr rECP, necrosis [80] Bovine mucus 100 3 fold altered structure [97] Nasal epithelial cells 2.1 ng/mL upregulation of ICAM-1 [100] Human corneal epithelial cells 100 decreased cell viability [98] Epithelial cell line NCI-H292 20 ng/mL 16 hr upregulation of IGF-1 [102] Human fetal lung fibroblast (HFL1) 10 48 hr release of TGF beta, collagen contraction [81] Human fetal lung fibroblast (HFL1) 10 5 hr rECP and naive, migration anti ECPab [107] Human fetal lung fibroblast (HFL1) 10 6 hr 6 fold increased proteoglycan accumulation [108] Potential effects due to high ECP levels in circulation and skin inflammation whole eosinophil granules are released from disrupted cells ( Figure 4B ) and that internal proteins are subsequently released differentially through the process of piece meal degranulation [4] . Several diseases are associated with eosinophils and ECP. Most common are diseases associated with atopy and the T helper lymphocyte type 2 (TH2) phenotype. Cytokines such as IL-5 [63] , or chemokines such as eotaxin are produced in elevated levels and attract elevated numbers of eosinophils to the lumen and bronchi of the lungs in asthma [49] (Figure 4B ), the nasal mucosa in allergic rhinitis [48] ( Figure 3B ) and to the skin in atopic dermatitis [64] . In addition, the gastrointestinal tract and esophagus are infiltrated during conditions such as ulcerative colitis [65] and eosinophil esophagitis [66] . ECP has been measured in disease and the increase in number of activated eosinophils is associated with elevation of serum ECP (sECP) and plasma ECP levels [67] . Anticoagulants such as EDTA attenuate ECP release from eosinophils giving a snapshot of the in situ ECP level in plasma. sECP level on the other hand is often higher than plasma ECP as it's an artificial measure obtained by detection of the protein released during the blood clotting process in the test tube. sECP is thought to reflect the activation state of eosinophils [68] . ECP has also been detected in several other biological fluids such as bronchoalveolar lavage fluid (BALF), sputum, nasal lavage and in mucosa of the intestine [69] . ECP levels in various biological fluids in various diseases are presented in table 1. ECP measurements in allergic asthma have been found useful in monitoring the disease as sputum ECP correlates with forced expiratory flow (FEV) [70] and the need for glucocorticosteroid (GC) therapy while sECP correlate with eosinophil numbers in blood [71] . sECP is also elevated in some but not all cases of TH2 cytokine associated atopic dermatitis [72] eosinophil esophagitis [73] , parasite infection [74] and childhood respiratory syncytial virus (RSV) infection [75] . Raised levels of ECP have also been found in some cases that are not TH2 associated; a group of patients with bacterial infections had elevated sECP [76] , very high levels were found in nasal secretions from patients with bacterial sinusitis [77] and in sputum of a patient with tuberculosis and drug-induced acute respiratory distress syndrome (ARDS) [78] . Malignancies with primary eosinophilia are associated with the highest measurable sECP levels (see HES and malignancy section). Polymorphisms have been shown both to alter expression level and the function of the protein which might complicate the usage of the protein as a biomarker (see polymorphism section). The pathology attributed to eosinophils and ECP has been of both acute character such as defoliation of airway epithelium or activation of other cells [79] [80] [81] and of a chronic type, such as fibrosis in lungs [49] (Figure 5 ). Below we discuss the studies that indicate how ECP release influence other cell types locally ( Figure 6 ). Lymphocyte activation mutually with ECP level has been shown to correlate with acute exacerbations in asthma [82] . sECP is also reduced during immune therapy which is a regimen that suppresses lymphocyte activity [83] . Eosinophils have been shown to migrate to lymph nodes where they might interact with T-lymphocytes. Eosinophils up-regulate major histocompatibility complex class II [84] for antigen presentation, thereby possibly contributing to T-lymphocyte activation and the increased inflammatory response during allergic inflammation [85] . Eosinophils are also present in the lymphocyte rich organs, the thymus and spleen and lamina propria of the gastrointestinal (GI) tract [50] . Although no studies have shown any direct link between ECP release and lymphocyte function, ECP released during the inflammatory processes, co-localises with lymphocytes. In vitro ECP has been shown to influence the proliferation of T and B lymphocytes which indicate that the protein could regulate those cells in vivo ( Figure 6 ). This was shown when mononuclear cells (containing lymphocytes, 2 × 10 5 ) were incubated with or without phytohaemagglutinin (PHA) and low levels of ECP (1 nM -0.1 μM, 190 ng/mL-2 μg/mL) for 48 hr, resulting in 50-67 percent inhibition of proliferation of the lymphocyte fraction [86] . The cells were not killed by these low levels of ECP. B lymphocyte activity might also be influenced by ECP since low levels (0.5-1 ng/mL) inhibit immunoglobulin production by plasma cells [87] and by B lymphocyte cell lines [88] . This effect was inhibited by anti-ECP antibodies and ECP was not toxic to the cell lines as cell proliferation was not inhibited with these low concentrations. IL-6 could restore the immunoglobulin production by the plasma cells and IL-4 had the same influence on the B lymphocytes. Primary human A. Healthy Control B. Allergic rhinitis plasma cells and large activated B lymphocytes responded to ECP in a manner similar to that of the cell lines [87] . Thus, ECP might influence the immune system in that immature lymphocytes are inhibited in their proliferation by ECP while activated B lymphocytes respond by decreased immunoglobulin production (see Figure 6 ). Mast cells are found in the skin and in all mucosal tissues at homeostasis, and numbers are elevated in asthmatics lungs [49] . Mast cell and eosinophil numbers in mucosa are correlated to bronchial hyperactivity (BHR) [89] and mast cell products and eosinophil MBP but not ECP induces BHR [90] . Several lines of evidence suggest that there is a cross talk between eosinophils and mast cells [91] which to some extent are related to ECP release. Mast cells produce and secrete IL-5, PAF and LTB 4 known to augment ECP release from eosinophils. Rat peritoneal mast cells on the other hand incubated with moderate levels of ECP (0. 9 μM/17 μg/mL) for 45 min released 50 percent of their histamine. Histamine is not released from peripheral basophils by ECP treatment (as by MBP) [92] . However, the release of histamine may be location specific as no release was observed from human skin mast cells treated with up to 200 μg/ mL ECP [93] . Histamine and of some tryptase was though released from human heart mast cells, purified from traffic victims or from individuals undergoing heart transplantation, when stimulated with moderate levels of ECP (2.5 μM; 4.7 μg/mL). Between 10 and 80 percent of preformed mediators were released from these cells and MBP had a similar effect whereas EPX/ EDN did not induce any release [94] . This ECP induced histamine release occurred within 60 sec of stimulation and was found to be Ca 2+ -, temperature-and energy dependent, and ECP was not toxic to the cells. Another mast cell product, prostaglandin D 2 (PGD 2 ) was synthesised de novo by the same amount of ECP added. PGD 2 is a chemoattractant for eosinophils and TH2 lymphocytes, through binding the CRTH2 receptor [95] . Therefore these findings suggest that in some tissue the interactions between mast cells and eosinophils can be attributed to the positive feedback of ECP release. ECP is detected in nasal mucosa in association with damaged epithelium [48] , in damaged corneal epithelium [96] as well as in BALF (at 40 ng/mL, table 1) [97] . The function of ECP has been assessed using several assays in the view of the presence of the eosinophil in the airways. Both destructive and non-destructive consequences have been found when analyzing various concentrations of the protein in interaction with the epithelium. High levels of ECP (5.4 μM/103 μg/mL) caused exfoliation of guinea-pig mucosal cells after 6 hr incubation with tracheal epithelium [79] . Confluent primary human corneal epithelial cells incubated with 0-100 μg/mL ECP, displayed a concentration-dependent gradual increase in morphological change and with the highest concentration, 100 μg/mL, being cytotoxic [98] . Lower concentration of the ECP (2.5 μg/mL) caused release of respiratory glycoconjugates (marker of mucus secretion), with a peak after 1 hr, from feline tracheal B. Allergic asthma A. Healthy control Figure 4 Eosinophil granulocytes in the bronchial mucosa. Sections of bronchial biopsies from (A) a healthy control or (B) an individual with allergic asthma were stained with ECP antibody visualizing eosinophils in the mucosa. The figures show that only a few eosinophils are present in the tissue of the healthy control, but many eosinophils accumulate in areas of reduced epithelial integrity in a specimen from a patient with allergic asthma. Original magnification ×420; Mayer's haematoxylin. explants [99] . The short incubation time and possibility to repeat the stimulation suggested a non-toxic mechanism. MBP, which is almost as basic as ECP, in the same assay, showed the opposite effect; therefore these effects on mucus secretion are unlikely to be due to electrostatic charge. ECP at these moderate levels (2.5 μg/mL) displayed the same effect on human trachea [99] . However human primary epithelial cells underwent necrosis at higher levels (10 μg/mL) in another study [80] . ECP has also been shown to acting directly on airway mucus in vitro. At high levels (100 μg/mL) ECP altered bovine mucus three fold, as measured by a capillary surfactometer Helminth defence (F) Bacterial defence (F) Skin Ulceration (P) Figure 5 Known anatomical locations of eosinophil granulocytes and suggested activities of released ECP at these sites. On the left side are eosinophil granulocytes locations at homeostasis shown. On the right side are areas speculated to be affected by increased numbers of eosinophils and elevated levels of released ECP, in disease (pathology, P) and in physiological defense (function, F). This is a speculation by the authors of the review. [97]. At low levels ECP (0.1 nM; 2.1 ng/mL) was instead found to increase the expression of intracellular adhesion molecule (ICAM)-1 on nasal epithelial cells [100] . ECP has previously been shown to be released from eosinophils when the cells adhere with their β2 (CD18) integrins to ICAM-1. Therefore the ECP triggered up-regulation of ICAM-1 on epithelial cells might mediate a positive feedback mechanism [101] . ECP has also been proposed as a mediator of tissue remodelling, see the fibroblast section below. When low levels of the protein (20 ng/mL) were used to stimulate the bronchial epithelial cell line NCI-H292 for 16 hr, the insulin growth factor (IGF)-1 receptor was found to be up-regulated [102] . ECP was speculated therefore to be involved in IGF-1-dependent lung tissue repair processes perhaps present during homeostasis and abnormally amplified during inflammatory conditions. The persistent high number of eosinophils and ECP in the lungs of allergic asthmatics has led to the suggestion of their participation in the development of chronic lung tissue remodelling. Remodelling has also been found in the esophagus of patients with eosinophil esophagitis [103] and sECP has been found elevated in one case [104] . The remodelling in asthmatic lungs is in part caused by collagen and proteoglycan secretion from interstitial fibroblasts. Eosinophils have been suggested to participate in this by secretion of transforming growth factor (TGF) beta [105, 106] but here is additionally described how ECP could influences fibrosis development. Stimulation of a human fetal lung fibroblast cell line (HFL1) with moderate/high levels of ECP (5-10 μg/mL) for 24-48 hr resulted in increased release of TGF-beta [81] . ECP also augmented fibroblast mediated contraction of collagen gel and stimulated migration of HFL1 fibroblasts which could be blocked with antibodies to ECP [107] . In addition, ECP incubated with the fibroblast cell line for 6 hr resulted in a 6-fold increase of intracellular proteoglycan accumulation [108] . Bronchial smooth muscles cells are involved during the progression of asthma development by secretion of cytokines as well as remodelling due to proliferation. Eosinophils have been found located in close proximity with smooth muscle cells. ECP does not influence smooth muscle cells by causing BHR [90] but high levels of ECP, similar to used for epithelial cells, appears to be cytotoxic, inducing cell death by necrosis in 1 hr. TNF alpha in contrast causes apoptosis of the smooth muscle cells [109] . Conditions where eosinophils are overproduced lead to detrimental effects for the host. One such condition, HES is defined by the presence of more than 1.5 × 10 6 eosinophils/mL blood during a time period of at least 6 months, organ involvement and with no other etiology identified. One form of HES, the myeloproliferative form, is caused by an 800 bp deletion on chromosome 4 during the haematopoiesis in the bone marrow, resulting in a fusion between the gene FIP1L1 and the PDGFRA gene [110] . A fusion protein is produced which constitutively phosphorylates tyrosine residues leading to malignant expansion of eosinophils. Another form of HES is a clonal lymphocytic variant (L-HES) where aberrant cytokine production by malignant lymphocytes causes HES. For other cases the cause of the overproduction of the eosinophils is unknown but HES is associated with high levels of ECP in plasma and serum, of up to 0.2 μg/mL [111, 112] . It is not know however whether theses high levels of the protein are pathological. A few in vitro studies might relate to the etiologies of HES. Eosinophil infiltration of the skin of HES patients is the most common clinical manifestation [113] . Some of these patients present with erosive and ulcerative lesions and ECP was found both deposited and taken up by cells in those lesions [114] . ECP's ability to cause ulcerations in the skin has been analysed by injecting the protein intradermally into guinea pig skin, where it was found that the protein can persist there for two weeks [64] which is possibly attributed to its high stability [28] . Injections of high levels of ECP (48 and 190 μg/mL/2.5 and 10 μM) caused ulcerations which were most severe after seven days [114] . Inflammatory cells were found infiltrating the inflamed area and ECP was found taken up by cells within 48 hr. Injection of poly-lysine, other basic granule proteins MBP, EPO and the basic ribonuclease onconase showed that the severity of the lesions was not directly correlated with level of basicity. ECP and EDN were found to be more potent in lesion formations than MBP and EPO. Addition of RNase inhibitor or obliteration of the RNase activity by carboxymethylation of the RNase site of ECP reduced the ulcerations by 60 percent suggesting RNase activity is important, but not wholly responsible for the activity [114] . Some studies have shown that patients with HES have an slightly elevated risk for thrombosis formation systemically [115] and in the cardiac ventricle [116] . ECP has been shown to shortened the coagulation time for plasma which was dependent on an interaction with coagulation factor XII [117] . Eosinophils also infiltrate the endomyocardium of some patients and this has been suggested to be the cause of development of scaring in the ventricle [116] . High levels of ECP (16.25 μg/mL) degrade the muscle protein component, the myosin heavy chain in vitro [118] but it is not known whether ECP directly interacts with muscle fibres of the heart. The final stage is endomyocardial fibrosis in which eosinophils and ECP have been postulated to participate [119] by their influence on fibroblast function. Although a rare finding, a few patients with the myeloid form of HES have been reported to have central nervous system (CNS) manifestation [113, 120] . It is not known whether ECP can reach the brain but ECPs effect on the CNS has been assayed by direct intracerebral injection. Guinea-pigs injected with ECP, showed with doses of 0.1 μg and up, cerebral symptoms up until the end of the experiment at day 16 [121] . Purkinje cells in the brain were decimated in this model, suggesting that the circulating ECP could affect the CNS of some HES patients if the protein reached the brain. Eosinophils have occasionally been found to infiltrate developing tumours and have been suggested to have a role in fighting these malignancies [122] . The involvement of the eosinophils have been suggested by the finding of elevated sECP levels in patients with renal tumours (table 1) [123] . ECP assayed in urine from patients with urinary bladder tumours showed a twofold increase compared to normal's [124] . The elevated levels suggest presence of activated eosinophils in some patients with these malignancies. In the analysis of the possible involvement of ECP in tumour defence, ECP has been evaluated in respect of altering proliferation of various cell lines. The cell lines K562 and HL-60 were incubated with 1.1 uM (21 ug/mL) ECP and the cell line A431 with 4 μM (76 ug/mL) and this resulted in 50 percent inhibition of proliferation after four days. To analyse whether growth inhibition was related to positive charge or RNase activity, poly-lysine or RNase A was used with no effect [34] . ECP exists in two forms dependent on a polymorphism, R97 and T97. It was found that the T97 form had reduced capability to kill K562 and NCI-H69 cells [36] . These recombinant (r) ECPs were produced in a baculovirus system and deglycosylation restored the cytotoxic activity. Furthermore, high levels of bacteria expressed rECP had 50 percent cytostatic effect on HL-60 and HeLa cells [31] , compared to non-affected controls. ECP was found binding the surface of HeLa cells and caused cell death after 24 hr, accompanied by increases in intracellular radical oxygen species (ROS) generation and caspase 3-like activity [125] . A mix of ECP and EDN purified from urine and incubated with the Kaposi's sarcoma cell line KS Y-1 for 16 hr caused complete cell death at 0.625 μg/mL while 1 μg recombinant ECP produced in yeast and incubated with the same time span decreased the viability of the KS cell line by 29 percent. Proteins expressed in yeast lack glycosylation and the possible implications of this were speculated [126] . Levels of serum ECP are elevated in TH2 engaging parasitic and helminth infections and eosinophils have long been thought to be a major defence against these types of infection. Elevated ECP have also been reported in some cases of bacterial and viral respiratory infections. Given that ECP is a cytotoxic ribonuclease, the ability of the protein to exterminate parasites, bacteria and virus in vitro has been extensively investigated (see also Figure 6 ). Parasitic and helminthic infections drive the immune system towards TH2 cytokine production and concurrent eosinophilia. Since eosinophil infiltration in infected organs and skin is a common finding, eosinophils are thought to have a specific role in parasite killing [127] . Although, a challenged theory; the deposition of the cytotoxic protein ECP could be a mechanism by which the immune system kills off the intruders. Indeed, the eosinophilia in parasitic diseases is associated with elevated ECP in circulation (table 1) [72, 128] . ECP is also found released from eosinophils in proximity to parasites in skin and lymph nodes [129, 130] . The ability for ECP to kill or paralyse parasites and helminths have been analysed in vitro and high quantities were needed to influence the organisms. Three-hr-old larvae of Schistosoma mansoni were incubated with 10 μM (190 μg/mL) ECP and 60 percent were killed. S. mansoni, 3 days of age, were paralysed by the protein [131] while 50 μM (950 μg/mL) ECP killed 40 percent of Trypanosoma cruzi by 6 hr and 90 percent of Brugia malayi by 48 hr. This cytotoxicity of ECP to parasites was inhibited by heparin [132] and dextran sulphate, probably by interfering with the tryptophan and arginine residues as discussed earlier. In addition, heat obliterated the toxic effect of ECP to parasites, highlighting the importance of the conformation of the protein [133] . The RNase activity of ECP was clearly shown not to be important for parasite toxicity, similar to that observed for EPX/EDN. Eosinophils are found lining and degranulating in both the respiratory and gastrointestinal mucosa [50] . Eosinophils are generally not thought of as defendants during bacterial inflammation. However sECP has been found elevated in septic patients [76] and very high levels of ECP in nasal secretions from patients with normal cold (13 μg/mL) or severe community acquired rhinosinusitis has been described in one case (11.7 μg/mL, table 1) [77] . Moreover, a recent study has shown that eosinophils expel mitochondrial DNA coated with ECP and other granule proteins which are bactericidal in mice in vivo [134] . Additionally, a few studies have described neutrophils producing ECP [15] . In view of these findings the anti -bacterial properties of ECP has been evaluated. Bacterial strains chosen for analysis were Escherichia coli (E. coli) and Staphylococcus aureus (S. aureus). High levels of ECP (50 μg/ mL) decreased the number of colony-forming units (cfu) by 72 percent and close to 100 percent, respectively, for the two strains after a very short 2 hr of incubation. ECP only killed E. coli growing in logarithmic phase and acted on both the inner and outer membranes of E. coli [135] . Recombinant ECP was also cytotoxicity to S. aureus. Overnight incubation of rECP with the bacteria (16 kDa, 16 ug/ mL/1 uM) left 35 percent of the cfu. rECP in which a.a.'s involved in RNase activity had been substituted (K38R and H128D), terminating the RNase activity, had no effect on the bacterial killing activity [21] . In conclusion therefore, eosinophils and ECP might have a role in bacterial defence. Due to its stability, it might be feasible to speculate that ECP over time accumulate in mucus fluids such as nasal secretions and act as a first line of defence against bacterial intrusion. ECP has been found significantly elevated in sputum from atopic subjects subjected to experimental rhinovirus infection [136] and in nasal secretions from atopic infants with respiratory RSV infection (table 1) [75] . Eosinophils and ECP are associated with RSV infection in children's lungs [137] and RSV can infect, and replicate in eosinophils [138] . Recombinant ECP expressed in a baculovirus system was used to evaluate whether ECP can inactivate the B subtype of RSV. ECP (0.5 μM; 9.5 μg/mL) incubated with the virus showed a 6-fold reduction of the infectivity of the virus to a human pulmonary epithelial cell line [139] . This antiviral activity was lower than that found with EPX/EDN (54-fold reduction) [140] , but the infectivity was increased by addition of RNase inhibitor (RI) to both proteins during incubation. Mixing the two proteins did not mediate any synergistic effects on antiviral activity. RNase A, however [up to 4 mM (76 mg/mL)], did not exert antiviral activity, suggesting that the RNase site but not activity is important for inhibition of infectivity. Polymorphisms in the RNASE3 gene and association to production and disease Table 3 summarizes data from the NCBI entrez nucleotide site regarding polymorphisms detected in the ECP gene. Two polymorphisms are found in the protein coding region, two in intronic regions and two in the 3' untranslated region (UTR). ECP polymorphisms are differentially distributed according to ethnicity [141] . Two studies have evaluated polymorphisms in intronic and UTR regions of the ECP gene, and linked them with ECP production. One polymorphism rs11575981 (-393T > C) located in the promoter, in an C/EBP binding site was associated with decreased ECP level in serum, and decreased binding of C/EBP alpha [142] . Another polymorphism, in the 3'UTR, rs2233860 (499G > C or 562G > C) was associated with content of ECP in the eosinophils [143] . Three studies have analysed whether any polymorphisms are linked to allergic asthma and allergic rhinitis. The presence of the C allele in the nonsynonymous rs2073342G/C (371G > C/434G > C) polymorphism in the ECP gene, causing a.a. alteration Y97T, was found to be associated with absence of asthma in one Swedish study [40] . A study of Norwegian and Dutch subjects instead found that the haplotype C-G-G for the three polymorphisms rs2233859/rs17792481 (-38C/A), rs2073342 (371G/C/434G/C) and rs2233860 (499G/C/ 562G/C) being protective [144] . In a third, Korean study, which was the largest, the genotype rs2233860CC (499/562CC) was associated with allergic rhinitis [145] . Eosinophils occasionally infiltrate oral squamous cell carcinoma tumours. A study found a tendency for association of the rs2073342G/C C/C (371/434GC/CC) genotypes with a poor clinical outcome in patients with eosinophil rich such tumours [146] . As discussed earlier, eosinophils are present during helminth infections. The rs8019343 polymorphism T (1088TT) in the 3'UTR was exclusively present in the genome of a patient with tropical pulmonary eosinophilia [147] . Furthermore a study has found the rs2073342 with C (371/434C) polymorphism overrepresented in helminth infected Ugandans rs2233858 protein coding, Y > G is associated with allergic asthma [40] a , poor outcome in oral squamous cell carcinoma tumours [146] , C over represented in helminth infected Ugandans [148] rs12147890 A/G protein coding rs2233860 G/C 3' UTR,G is correlated to higher intracellular ECP [143] , G is associated with allergic rhinitis [145] , a rs8019343 A/T 499G > C, 562G > C 3' UTR T is only present in one patient with helminth infection [147] Polymorphisms found in the ECP gene and surrounding chromosomal sequence. Listed are Polymorphism i.d.'s, altered bases, alternative names, and types of associations a) C-G-G haplotype associated with allergic rhinitis [144] [148]. Interestingly, from the -550 polymorphism over a stretch of 272 bases to the mRNA transcription start site, thirteen polymorphism sites are located (NCBI Reference Sequence: NC_000014.7, J. Bystrom unpublished observation). Similar to the protein coding region and the 3'UTR, this region is highly homologous to the RNASE2 gene region, with the only differences being the sites of the polymorphisms. The replacement base's for twelve of the thirteen polymorphisms is to the same base as in the RNASE2 promoter sequence. This is also the case for two of the 3'UTR polymorphisms. This further highlights the extremely close relationship between RNASE3 (ECP) and RNASE2 (EPX/EDN). ECP was first discovered in 1977 and since then, evidence has been gathered to understand its roles in physiology and pathophysiology. ECP is a peptide of 133 a. a., with the first 40 a.a. necessary for membrane interfering, heparin binding and cytotoxic activity. The heparin binding ability of ECP might enable the protein to bind proteoglycans on other human cells for possible uptake [34] or heparan sulfate in extracellular matrix for later use such as is the case for CXCL10 [149] . In a similar manner ECP might bind microorganisms peptidoglycans for uptake and cytotoxicity [32] . The non-synonymous polymorphism rs2073342 reduces cytotoxicity suggesting an alteration of the three-dimensional structure influencing catalytic site elsewhere in the protein. ECP is glycosylated, and as recently discovered can be nitrated. The development of increasingly sophisticated assays will determine whether other modifications, perhaps function associated, are also important in ECP activity. Since the discovery of ECP, assays have been developed to determine its levels in biological fluids in various diseases (table 1) . ECP in serum can reach 0.1 -0.2 μg/mL for HES patients [111] and parasitic diseases infected individuals [72] and this is a 30 fold elevation compared to ECP in serum of healthy individuals. In BALF and nasal lavage from atopic patients the ECP levels are lower, 0.050 μg/mL but the sample are diluted during the collection process. In undiluted tears, sputum and nasal secretions the highest ECP levels have been found: 0.5, 0.7 and 10 μg/mL, respectively. The ECP measurements correlate with eosinophilic disease but have been found elevated also in some diseases without known eosinophil involvement [76] [77] [78] . The biological activity of ECP has been studied by incubation of the protein with several different cell types in vitro. Both human cells and pathogens have been assayed analysing different parameters (see table 2 ). In general, 10 -20 μg/mL and above, result in growth inhibitory and destructive consequences to mammalian cells, parasites and bacteria. ECP released in situ in diseases engaging high levels of eosinophils might reach these destructive concentrations (e.g. ECP accumulated in air way mucus of asthmatics, in nasal secretions of some sinusitis patients or released in skin of atopic dermatitis/HES patients, table 1 and Figure 5 ). Although it remains to be proven, there is a possibility that destructive activity to multiple cell types as well as induction of fibrosis is part of the etiology of disease where ECP levels are elevated during prolonged periods, e.g. in HES and helminth infection. There is also evidence that neutrophils are carriers of significant amounts of ECP. Using the murine system, granule proteins have been found associated with expelled eosinophil mitochondrial DNA and this DNA/protein complex trapping and killing bacteria in the gut [134] . It is intriguing to speculate whether the high levels of ECP present in various human mucosal secretions would equally be associated with eosinophil mitochondrial DNA and whether such complexes had the ability to capture and kill microorganisms. The role of eosinophils in asthma has been under scrutiny since clinical trials showing that anti-IL-5 therapy did not improve the disease symptoms for allergic asthmatics albeit eosinophil numbers were reduced [150] . However, two recent clinical trials have shown that anti-IL-5 antibodies actually could relieve symptoms in eosinophil rich, late onset asthma, suggesting that eosinophils can have a pathogenic role in this disease. In these trials inflammatory exacerbations were reduced when anti-IL-5 antibodies were administrated [151, 152] . Earlier studies using diagnostic ECP measurements seem to agree with these findings as ECP levels correlate with severity of asthma: FEV (sputum ECP) [70] , need for GC treatment (sputum ECP) [153] and blood eosinophilia (sECP) [71] . Results from in vitro studies presented in this review may well suggest several roles for ECP in this type of allergic asthma. The protein might act as an inflammatory amplifier by augmentation of release of, for eosinophils chemotactic, PGD 2 from mast cells in asthmatic patients. Moreover, protein released in the interstitium might influence fibrosis development ( Figure 3B , 4B and 5). One might speculate that blocking antibodies to ECP could be a symptom relieving addition to the already established GC and anti-IL5 therapies used in eosinophil rich asthma and other eosinophilic diseases [113, 151, 152] . Table 2 shows that the level of protein needed to influence proliferation of lymphocytes and their antibody production is 1000 times lower than the destructive levels described above, i.e. in the ng/mL range. In murine system eosinophils have been ascribed a novel role in inflammation; the cells enter and contribute to the well orchestrated process of inflammation resolution of by release of the pro-resolving lipid protectin D1 [154] (for a review see [155] ). Whether ECP is released during this resolution process for the dual role of sequestering subpopulations of inflammatory lymphocytes [86] and promoting tissue repair by TGF beta augmentation [81] is an intriguing speculation. Eosinophils are also present at homeostasis at low numbers in lymphocyte rich organs at various locations but degranulated only in the GI tract [50] . A single eosinophil contains 13 pg ECP. Do eosinophils have a role in maintaining homeostasis and do low levels of ECP also have a role here? EDN, the sister protein has been found to play an active role during inflammation development influencing the maturation of DC's [37] . If EDN is proinflammatory, perhaps the two proteins divergence could be because ECP might have acquired a novel role as yet unknown role. Finally, analysis of the DNA sequence of the ECP gene and surrounding regions have unravelled a number of polymorphisms. These studies have linked different polymorphisms and haplotypes to TH2 diseases, asthma, and allergic rhinitis. The studies have in some cases come to different conclusions but used different patients and different ethnic groups which might explain the variations. Diseases such as allergic asthma are multifactoral and to determine the role of certain polymorphisms one might need to look at larger defined groups to get a clear association. Altered expression levels might also influence both destructive functions and possible homeostatic roles. A careful analysis using all polymorphisms and corresponding haplotypes and large groups of defined populations would more clearly determine the role of ECPs genetic make-up, and its potential functions in physiology and disease. The eosinophil granulocyte was discovered 130 years ago but its roles are still being revealed. The most characteristic feature of the eosinophil is the large secondary granules filled with basic proteins. The purpose of these proteins is still not fully understood. One of the proteins, ECP is a highly basic, cytotoxic, heparin binding ribonuclease that seems to need its ribonuclease site but not activity for its activities. Sensitive assays have been developed for its measurement in biological fluids which have contributed to the understanding of the role of the eosinophils in disease. In vitro studies have shown that high levels of ECP are necessary for development its destructive actions. Diseases engaging high levels of eosinophils might reach these levels locally in the tissue. At those high levels polymorphisms altering expression level and protein sequence might play a role within certain populations. Whether ECP also has roles at lower concentrations, such as the growth inhibitory influences on lymphocytes found in vitro, remain to be shown with in vivo models or clinically. These additional roles for ECP when discovered, might provide critical answers to the functions of eosinophil granulocytes and is therefore well worth waiting for.

Excess healthcare burden during 1918-1920 influenza pandemic in Taiwan: implications for post-pandemic preparedness

BACKGROUND: It is speculated that the 2009 pandemic H1N1 influenza virus might fall into a seasonal pattern during the current post-pandemic period with more severe clinical presentation for high-risk groups identified during the 2009 pandemic. Hence the extent of likely excess healthcare needs during this period must be fully considered. We will make use of the historical healthcare record in Taiwan during and after the 1918 influenza pandemic to ascertain the scope of potential excess healthcare burden during the post-pandemic period. METHODS: To establish the healthcare needs after the initial wave in 1918, the yearly healthcare records (hospitalizations, outpatients, etc.) in Taiwan during 1918-1920 are compared with the corresponding data from the adjacent "baseline" years of 1916, 1917, 1921, and 1922 to estimate the excess healthcare burden during the initial outbreak in 1918 and in the years immediately after. RESULTS: In 1918 the number of public hospital outpatients exceeded the yearly average of the baseline years by 20.11% (95% CI: 16.43, 25.90), and the number of hospitalizations exceeded the corresponding yearly average of the baseline years by 12.20% (10.59, 14.38), while the excess number of patients treated by the public medics was statistically significant at 32.21% (28.48, 39.82) more than the yearly average of the baseline years. For 1920, only the excess number of hospitalizations was statistically significant at 19.83% (95% CI: 17.21, 23.38) more than the yearly average of the baseline years. CONCLUSIONS: Considerable extra burden with significant loss of lives was reported in 1918 by both the public medics system and the public hospitals. In comparison, only a substantial number of excess hospitalizations in the public hospitals was reported in 1920, indicating that the population was relatively unprepared for the first wave in 1918 and did not fully utilize the public hospitals. Moreover, comparatively low mortality was reported by the public hospitals and the public medics during the second wave in 1920 even though significantly more patients were hospitalized, suggesting that there had been substantially less fatal illnesses among the hospitalized patients during the second wave. Our results provide viable parameters for assessing healthcare needs for post-pandemic preparedness.

The 2009 pandemic H1N1 (pH1N1) virus spread swiftly to all parts of the world in a matter of a few months after it was first identified in Mexico in March. In August 2010, World Health Organization (WHO) declared the world to be in the post-pandemic period, when the pH1N1 virus is expected to continue to circulate as a seasonal virus for some years to come [1] . Moreover, active transmission of pandemic influenza virus still persists in some local areas, and it is still unclear whether the pandemic influenza activity has already transitioned into a seasonal pattern [2] . It is further speculated that groups identified during the recent pandemic as at a higher risk of severe or fatal illness will probably remain at a heightened risk during the post-pandemic period, although the number of such cases may decrease. In addition, a small proportion of people infected during the 2009 pandemic developed a severe form of primary viral pneumonia that is not commonly seen during seasonal epidemics and is especially difficult to treat. Therefore, quantitative ascertainment of the likely healthcare burden is an important aspect of post-pandemic preparedness planning. More than 90 years ago, the first pandemic of the last century was initially observed in the early spring of 1918. It was quickly followed by much more fatal second and third waves in the fall and winter of 1918-1920, causing an estimated 50 million deaths [3] . It still proves to be a major dilemma for the scientific community to understand what had happened precisely, how it had happened, and why it was in several ways unlike any other influenza pandemic in recorded human history [3] . Several studies have focused on quantifying the global impact of that pandemic, either by using records of cases and mortality of the affected countries (e.g., [4] [5] [6] ), or by using vital statistics data from the affected countries to estimate the excess mortality of these countries. For example, Murray et al. [6] estimated the excess mortality rate of each affected country and extrapolated to conclude that an estimated 62 million people would be killed in a similar pandemic in 2004. However, to the best of our knowledge historical healthcare records had not been used to directly assess the excess healthcare burden during a pandemic. The 1918-1920 pandemic swept through Taiwan in two distinct waves, both occurred during winter influenza seasons -the first at the end of 1918 and the second in early 1920, and with devastating loss of human lives. Increased influenza cases were initially reported in mid-October 1918 in Keelung, the main seaport in north [7] . A report published in the Taiwan Medical Association Journal in February 1920 [8] on the devastation brought by this first wave of influenza outbreak reported that 20.8% of the population had been infected with a case fatality rate of 3.26%. A second wave appeared at the end of 1919 and affected Taiwan through the early months of 1920, also with a severe death toll. A recent study [9] made use of the monthly mortality data in Taiwan during that time period to estimate that the total number of excess deaths during the pandemic months of November-December 1918 and January-February 1920 was 51,048 (95% CI 41,998-61,853). The 1918-1920 influenza epidemic in Taiwan was intriguing in several aspects. First, it was one of the few regions in the world that a wave had occurred as late as 1920 [5] . Moreover, the two waves of the epidemic were separated by almost a full year, in contrast to intervals of a few months in most countries in the world [3] , and both occurred during the months of yearly winter influenza season in Taiwan. The relatively late occurrence of the initial outbreak in November of 1918, as well as the second wave in early 1920, perhaps signifies the relative lack of international travel due to its status at that time as a fairly recent Japanese colony (since 1895), in contrast to nearby regions such as Singapore which is geographically similar but more globally connected [10] . Moreover, Taiwan is located in the tropical-subtropical zone with similar excess influenza deaths to those observed in temperate zone during periods of previously recognized influenza epidemics in Taiwan [11, 12] , and has been known to be one of the evolutionarily leading regions for global circulation of influenza [13] . Therefore, the island population in Taiwan could serve as a good model for studying spatial and temporal spread of influenza outbreak in a confined region during distinct waves of a pandemic. In this study, we will make use of the healthcare records from 1918-1920 in Taiwan to examine the level of excess healthcare burden under which healthcare system was extended in the years immediately following the initial wave of the pandemic in 1918, and to ascertain the possible post-pandemic demands on a modern healthcare system, such as we might face in the coming influenza seasons. Our main source of data is the 1895-1945 Statistical Abstract of Taiwan [14] which contains the complete and detailed vital statistics of Taiwan during all 50 years of the Japanese occupation including detailed yearly healthcare records. We will use this data to explore the public health events that had occurred during those years during and immediately after the initial epidemic in 1918. During 1918-1920, there were 12 large public hospitals, 18-19 smaller public hospitals, and 60-68 private hospitals in Taiwan. In addition, there was a large network of trained "public medics" which was responsible for, among other duties, providing basic and primary medical care in the local community for people with clinical symptoms and for reporting local incidence of illnesses (including epidemic intelligence) to the government [15] . However, only the numbers of outpatients, in-patient hospitalizations, and all-cause deaths for the large public hospital and the public medics system were given in the Statistics Abstract [14] . A measure of the severity of an epidemic and its burden on the healthcare system is the excess number of hospital visits and hospitalized patients during the epidemic and during the post-epidemic years. Yearly excess numbers of outpatients, in-patient hospitalizations (abbreviated to "hospitalization" hereafter), and all-cause deaths reported by the large public hospital and the public medics system during 1918-1920 were computed by the method of Serfling et al. [16] . We first computed the yearly mean numbers of outpatients, hospitalizations, and all-cause deaths over the two adjacent baseline years before the epidemic (1916, 1917) and the two adjacent baseline years after (1921, 1922) . We then subtracted these means from the corresponding yearly numbers of outpatients, hospitalizations, and all-cause deaths for each year during 1918-1920 to obtain the yearly excess numbers during 1918-1920. A yearly excess number is considered to be statistically significant if the number (of outpatients, hospitalizations, or allcause deaths) for that pandemic year exceeds the corresponding mean of the adjacent baseline years of 1916, 1917, 1921 , and 1922 by 2 SDs or more [9] . In order to compare the yearly excess healthcare burden of the pandemic years of 1918-1920, we computed the percentages of these yearly excess numbers over the means of the adjacent baseline years, to ascertain the impact of the pandemic on the healthcare system during each of the years in 1918-1920. The yearly excess number of patients and all-cause deaths reported by 12 public hospitals and public medics system during 1918-1920 compared with the yearly averages during the adjacent "baseline" years of 1916, 1917, 1921, and 1922 are shown in Figures 1, 2, 3 . The percentages of the excess number of medical treatments and hospitalization for each year during 1918-1920 over the averaged yearly numbers of the adjacent baseline years of 1916, 1917, 1921, and 1922 are given in Table 1 with the 95% confidence intervals (CI). In 1919, the numbers of hospitalizations and treatments by public medics are clearly excessive, exceeding even the corresponding numbers in the epidemic years in 1918 and 1920 in some instances (see Figures 1, 2, 3 Moreover, the percentages of excess yearly number of deaths reported by 12 large public hospitals and public medics for each year during 1918-1920 over the averaged yearly number of deaths of the adjacent years of 1916, 1917, 1921, and 1922 , are given in Table 2 . Only There is an underlying assumption of our method that there is no drastic change in the Taiwanese population during 1916-1922, when the population size increased steadily but only by less than 10%, from 3,596,109 to 3,904,692 [14] . Methods to detect significant changes over time can be found in, among others, [17] . We also note that the decline following 1919 in all the three sets of numbers reflecting healthcare burden might be partly attributable to a regression to the mean, given that the second wave was still in full force in the first two months of 1920. The limitation in the data, where only yearly numbers (and not monthly numbers) are given, makes it impossible to determine the months in which the drop had occurred, and whether the decline is attributable to the decrease in healthcare demand after the pandemic was over or to the low level of healthcare demand even during the pandemic months early in the year. The percentages of excess deaths reported by the public hospitals and the public medics in 1918 were both statistically significant, corroborating the results from another study [9] . Moreover, given that the excess hospitalizations in the public hospital were not statistically significant and yet the percentage of excess deaths reported by the public hospitals was statistically significant and exceeded even those reported by public medics, one could infer that a comparatively larger proportion of hospitalized patients had lost their lives in 1918. However, the corresponding percentages of excess deaths were not statistically significant in 1920, even though similar levels of excess numbers of deaths were found in both waves [9] . This gives indication that the second wave in early 1920, although with a significantly greater number of hospitalizations, had substantially fewer fatal illnesses among the hospitalized patients when compared with the initial wave in 1918. Our results indicate that there was a considerable extra burden on the public medic system during the initial wave of the epidemic in 1918, with a significant loss of lives reported by both the public medic system and the 12 large public hospitals. In comparison, only a substantial number of excess hospitalizations in the public hospitals was reported in 1920, indicating that the population was relatively unprepared for the first wave in 1918 and did not fully utilize the public hospital system. The most surprising part of our findings is the significant increases in the numbers of hospitalizations and treatments by the public medics for 1919, the year between the two waves when only the beginning of the second wave in December of 1919 had contributed 9 .24 (7.90, 11.12) *Excess deaths in 1918 are statistically significant (more than 2 SD). some initial influenza deaths over the monthly means of neighboring "baseline" years [9] . One possible reason for this is the contribution to hospitalization/treatment due to other diseases that were prevalent in 1919 (e.g., a cholera outbreak which led to 2,693 deaths in 1919 and 1,675 deaths in 1920 [14] ). However, limited by the retrospective nature of the study design, we are unable to identify or rule out other non-relevant diseases or conditions solely from our hospitalization/treatment data due to the lack of more detailed historical data. It is also possible that the severe first epidemic wave during the previous winter of 1918 had alarmed the population to being more readily willing to quickly seek medical assistance at the first sign of an ailment, even though many of these illnesses might be unrelated to influenza. That is, the populace was more readily alerted to seek treatment from local public medics with any initial symptom of illness (as compared to visiting large hospital), while patients with more severe illness (of any kind) are more likely to be hospitalized by physicians. This type of overreaction on the part of the healthcare system and the general public had also been observed during the 2003 SARS outbreak where many non-SARS patients were hospitalized unnecessarily as suspected SARS cases. Adding the fact that both the numbers of hospitalizations and treatments by the public medics dropped drastically next year in 1920, the last scenario seems plausible. Another possibility is that pathophysiological or social processes [18] may be at play where the end of World War I could have contributed to movement of people and affected the pandemic's spread, although the Taiwan data indicated no noticeable increase in migration,. Our results suggest that the excess burden on the healthcare system was high in the post-pandemic period, which would be a major challenge to any well-managed healthcare system. But it could contribute to fewer fatal illnesses. It has been noted that any present-day projection based on the 1918-1920 pandemic merely presents a worst-case scenario which we can avoid with diligence [19, 20] . However, one should note that the situation today, 90 years later, is very different in many aspects. While modern communication systems may facilitate more rapid spread of infections, implementation of interventions (school closures, masks, hand washing, bans on spitting in public, etc.) may reduce the overall transmission of influenza. Moreover, population demographics, health status and prior exposure to influenza are also different. In 1918-1920 life expectancy was shorter, so the population would have been on the average younger with less prior exposure to influenza, and therefore less compounded by past circulation of influenza as mentioned previously. Our results provide a basis to learn from the past to obtain projections of pandemic scenario and the viable hypothetical parameters for assessing healthcare needs specifically for the current post-pandemic preparedness in every country, including antivirals and vaccines needs for speedy, adequate, and equitable distribution. Finally, while this study is retrospective in design, the study methods can be easily modified for a prospective design and incorporated into a part of syndromic surveillance during a future influenza pandemic to monitor and adjust resources accordingly.

Multifunctional nanoparticles as simulants for a gravimetric immunoassay

Immunoassays are important tools for the rapid detection and identification of pathogens, both clinically and in the research laboratory. An immunoassay with the potential for the detection of influenza was developed and tested using hemagglutinin (HA), a commonly studied glycoprotein found on the surface of influenza virions. Gold nanoparticles were synthesized, which present multiple peptide epitopes, including the HA epitope, in order to increase the gravimetric response achieved with the use of a QCM immunosensor for influenza. Specifically, epitopes associated with HA and FLAG peptides were affixed to gold nanoparticles by a six-mer PEG spacer between the epitope and the terminal cysteine. The PEG spacer was shown to enhance the probability for interaction with antibodies by increasing the distance the epitope extends from the gold surface. These nanoparticles were characterized using thermogravimetric analysis, transmission electron microscopy, matrix-assisted laser desorption/ionization-time of flight, and (1)H nuclear magnetic resonance analysis. Anti-FLAG and anti-HA antibodies were adhered to the surface of a QCM, and the response of each antibody upon exposure to HA, FLAG, and dual functionalized nanoparticles was compared with binding of Au–tiopronin nanoparticles and H5 HA proteins from influenza virus (H5N1). Results demonstrate that the immunoassay was capable of differentiating between nanoparticles presenting orthogonal epitopes in real-time with minimal nonspecific binding. The detection of H5 HA protein demonstrates the logical extension of using these nanoparticle mimics as a safe positive control in the detection of influenza, making this a vital step in improving influenza detection methodology. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s00216-010-4419-8) contains supplementary material, which is available to authorized users.

Viruses are the smallest form of life on earth with the ability to replicate and spread within living cells [1] . As they pass from cell to cell, they adapt to evade host immunity and spread disease, creating some of the worst pandemics in history [2] . Improving diagnostics for viruses, such as influenza, would help slow the spread of infection in the event of an emerging virus. It is known that even with reassortment, a common viral defense mechanism, the majority of anti-hemagglutinin (anti-HA) antibodies recognize a specific nine amino acid sequence within the epitope, AYDPVDYPY, which has been the focus of many assays to improve the detection of influenza [3] . Using this immunodominant sequence for the HA epitope, the influenza epitope can be mimicked with a functionalized nanoparticle yielding a comparable affinity to the linear peptide [4] . Another virus, Ebola, has also been effectively mimicked with a monolayer-protected cluster (MPC) through functionalization of the MPC with the antigenic determinant of the Ebola glycoprotein [5] . Integrating biology and materials chemistry using biomimicry in this way has allowed materials chemistry the opportunity to improve current diagnostic treatments, techniques, and limits of detection, but further improvements are still yet to be made [6] . The utilization of immuno-molecular recognition in the assembly of nanoscale sensors has applications in medical diagnosis, treatment, and the understanding of diseases [7] . Enzyme-linked immunosorbent assay (ELISA) is widely utilized in clinics and hospitals as an initial screening for several infectious diseases. While ELISA can be effectively employed in laboratory settings for common infectious agents, several obstacles inhibit the adaptation of this standard clinical assay to portable or select agent detection schemes. Unfortunately, pathogenic agent detection requires calibration with irradiated or otherwise attenuated samples of the organism. This requirement limits the widespread use of immunosensor diagnostics because of the scarcity of these agents and the logistic difficulty in safe transportation to remote locations. Recent cases like the development of meningitis and tularemia infections in researchers who were working with the causative agents of these diseases alert scientists to the hazards accompanying work with live calibrants [8] . The possibility of exposure and high cost are enough to warrant investigation into a safer positive control for these disease detection assays and devices. Without a positive control, the operational status of the sensor cannot be determined. Thus, the development of a nanoparticle mimic would be a safer alternative to current methodology and could be extended to address the needs of other assays that incorporate well-defined epitopes. Many traditional clinical assays lack the ability for electronic adaptation and timely results. The need exists for rapid, sensitive, and inexpensive methods that could be utilized in clinical settings [9] . Therefore, the rapid realtime quantification of a QCM has been combined with the selectivity of monoclonal antibodies to create an immunosensor for evaluation of the multifunctional nanoparticles. The use of a QCM for immunological detection has been demonstrated previously, where it has been shown to detect Staphylococcus epidermidis in clinical samples using nanoparticle amplification [10] , SARS virus in sputum samples [11] , plant pathogens [12] , an antigenic mimic of Ebola [5] , and influenza A and B from nasal washes [13] . In QCM assays, nanoparticles have enabled simultaneous parallel detection and amplification for gravimetry, thus lowering the limit of detection [6, [14] [15] [16] [17] . A QCM-based sensor with a nanoparticle control would allow for the simultaneous rapid and accurate analysis of multiple viral or biological hazards, without posing a safety or health threat. The work reported here builds on this previous work and addresses the specificity of detection of antibody-antigen binding at the QCM using polyepitope-functionalized gold nanoparticles. Gold nanoparticles have favorable characteristics for their use as the basis for multifunctional microorganism simulants [18] . Nanoparticle size, shape, and capacity for surface modifications, based on chemical characteristics and environments, make the use of nanoparticles advantageous for detection, discovery, and diagnosis [7, 19] . Previous studies have shown that nanoparticles are capable of accepting a wide array of functional molecules via the Au-thiol bonds at the interface of the ligand and particle [20] . Since the physical and chemical properties of nanoparticles are dependent upon size [6] , the gold nanoparticles can be customized to simulate the variability in pathogen size. MPCs in this work had an average diameter of 2.6± 0.6 nm. Influenza virions vary in size, normally around 100 nm in diameter, but smaller nanoparticles were chosen in this study since they have a larger surface area to volume ratio and therefore increase the ratio of possible antigen presentation to gold core [21] . These nanoparticles were polyfunctionalized to increase the presentation of the nanoparticle epitope to the antibody. Previous studies have found multivalent ligand attachment to gold nanoparticles enhances the affinity measured in binding studies [22, 23] . The selectivity of antibody sensors for functionalized MPCs has been examined using two orthogonal epitopes: FLAG and HA. The FLAG epitope is a biological peptide sequence commonly used for identification of proteins in biological samples [24] . HA-and FLAG-functionalized MPCs can be used as a synthetic simulant and negative control for the HA epitope of the influenza virus [25, 26] . The binding of these nanoparticles was also compared with binding of negative-control Au-tiopronin nanoparticles and an authentic sample of H5 HA proteins from influenza virus (H5N1). The simulants work as safe controls whose application could be extended to address various pathological threats, in which using attenuated controls is problematic. Chemicals Gold shot was purchased from precious metal vendors (Canadian Maple Leaf, 99.99%) and was initially converted to HAuCl 4 ·3 H 2 O by boiling Au 0 in HCl/HNO 3 solution [27] . N-(2-Mercaptopropionyl)-glycine (tiopronin, reagent grade), bovine serum albumin (BSA, fraction V, 96%), and sodium phosphate (monobasic, reagent grade) were purchased from Sigma. Other chemicals were obtained as follows: protein G was purchased from Southern Biotech, Fmoc-dPEG-COOH from Quanta Biodesign, anti-FLAG mouse monoclonal antibody from Stratagene, and anti-HA mouse monoclonal antibody from the Vanderbilt Molecular Recognition Core facility. The following reagent was obtained through the NIH Biodefense and Emerging Infections Research Resources Repository, NIAID, NIH: H5 HA protein from influenza virus, A/Hong Kong/156/97 (H5N1), recombinant from baculovirus, and NR-652 (56 kDa) [28] . Analytical grade solvents for nuclear magnetic resonance (NMR) were obtained from Cambridge Isotope Laboratories, and water was purified using a Modulab Water Systems unit (∼18 MΩ/cm). Buffers were prepared according to standard laboratory procedure. Other chemicals were reagent grade and used as received. MPC synthesis and characterization Gold-tiopronin-protected MPCs were synthesized as previously described [29] [30] [31] . Briefly, tiopronin-protected gold nanoparticles were synthesized by dissolving tiopronin and HAuCl 4 ·3H 2 O (3:1) into a MeOH/acetic acid solution (6:1). The reaction was stirred for 30 min and then cooled in an ice bath. NaBH 4 was then added in 10× molar excess of gold. Reaction product was stirred for 1 h. The solvent was removed under vacuum, and the pH was lowered to 1 with concentrated HCl. Reaction product purification by dialysis with cellulose ester membranes (Spectra/Por CE, MWCO= 10,000) removed any excess tiopronin. Average particle diameter was determined by transmission electron microscopy (TEM). TEM images were taken on a Phillips CM20 instrument after applying aqueous MPC samples to Formvar-coated 200-mesh copper grids (Ted Pella). The microscope operated at 200 keV with magnification in the range ×150-750,000. Thermogravimetric analysis (TGA) was performed with a TGA 1000 (Instrument Specialists, Inc.) to calculate Au-tiopronin stoichiometry. From TEM and TGA data, the nanoparticle size and Au-tiopronin stoichiometry were obtained. MPCs used in the following experiments had an average diameter of 2.6±0.6 nm and a base composition of Au 544 Tiop 204 (140.4 kDa), calculated from TEM and TGA data (at 650°C), respectively (Electronic Supplementary Material Figs. S1, S2, and S4) [32] [33] [34] . Peptide synthesis and characterization The FLAG and HA epitopes were synthesized with standard 9fluorenylmethoxycarbonyl (Fmoc protocols on a solid resin support) [35, 36] . Epitope sequences were modified with a linker region comprised of a discrete polyethylene glycol (PEG, purchased as Fmoc-dPEG™ from Quanta Biodesign) and a C-terminal cysteine. PEG was added to attenuate nonspecific binding, while the C-terminal cysteine provided a thiol linkage to the gold surface. PEG formed the link between the cysteine and the rest of the peptide, HA-PEG-C for example. After initial MPC, studies performed without the PEG linker were found to have no binding; all future studies incorporated this linkage. The notations HA-Au, HA-FLAG-Au, and FLAG-Au are assumed to include the incorporation of this linkage. Epitope antigens HA (AYDPVDYPY-(PEG) 6 -C, 1562.6 Da) and FLAG (KDDDDKYD-(PEG) 6 -C, 1451.5 Da) were synthesized on an Apex 396 (Advanced Chemtech) equipped with a 96-well reaction block capable of vortex mixing. 2-Chlorotrityl resin was swollen in dichloromethane (Fisher) prior to synthesis [37] . Fmoc amino acids (Synpep) and Fmoc-dPEG 6 acid (Quanta Biodesign) were coupled using O-benzotriazole-N,N,N′,N′-tetramethyl-uronium-hexafluoro-phosphate (HBTU, 5 eq with respect to resin, Synpep), 1-hydroxybenzatriazole (Hobt, 5 eq, Synpep), and N,N-diisopropylethylamine (DIEA, 10 eq, Advanced Chemtech) in N,N,-dimethylformamide (DMF, Fisher). Peptides were cleaved with 90% (v/v) trifluoroacetic acid, 5% anisole (Sigma), 3% thioanisole (Sigma), and 2% ethanedithiol (Sigma). Final purification was performed on a Waters C18 semi-prep RP HPLC column using a water (0.05% trifluoroacetic acid and acetonitrile) gradient. Peptide identity was confirmed via matrix-assisted laser desorption/ionization-time of flight analysis. Place exchange Place exchange reactions between tiopronin-MPCs and thiolate-containing peptides, HA and FLAG, followed a method similar to Hostetler et al. [18, 38, 39] and assumes that epitope-tiopronin S N 2 place exchange is one for one. While it has been shown by 1 H NMR that the interchange of ligands is 1:1, the final ratio of the original ligand to exchanged ligand depends on ligand length, concentration of entering and exiting ligands, as well as the placement of the ligand (vertice, edge, or terrace) on the surface of the cluster [18, 38] . It should also be noted that the reported ratio will be an average exchanged ratio calculated from 1 H NMR, as the dispersity of MPC size will cause variations in the exact number of exchanged ligands on each cluster [18] . With the X-ray diffraction identification [40] of staple motifs, which are composed of one Au(I) and two thiolates, as part of the capping structure on gold-thiolate nanoparticles, place exchange must insert the new thiolate into the existing staples at the same time as the previous thiolate is lost in the S N 2 mechanism. Tiopronin MPCs were co-dissolved in DI water with free thiolated epitope (1:10). If necessary to aid the solvation of the epitopes into water, the epitopes were first dissolved into ethanol. The place exchange reaction took only one night for the exchange of a single epitope, while requiring 3 days for the exchange of two epitopes simultaneously. Solutions were dialyzed at room temperature for approximately 3 days as previously described and then dried under air [29] . The epitope composition of the MPCs was determined by TGA and 1 H NMR analysis (Electronic Supplementary Material Figs. S2 and S3). Nuclear magnetic resonance 1 H NMR experiments were run at 300 MHz on a Bruker DPX-300 instrument with 5 s relaxation times. Samples were dissolved in D 2 O. The extent of place substitution of tiopronin by epitope on gold nanoparticles was determined by 1 H NMR, using methods previously described [18, 31] . Enzyme-linked immunosorbent assay Initial tests to determine whether the functionalized nanoparticles were suitable as both a viral simulant and a synthetic calibrant were performed using ELISAs. Positive activity was established by adhering gold nanoparticles (250-500 ng/per well) with a variety of functionalities (Au-Tiop, HA-Au, FLAG-Au, and HA-FLAG-Au) to a 96-well Immulon 2HB plate. Buffer and free peptides (25 μg/well) were plated and tested in a similar fashion. Blocking was achieved with BSA (1 mg/mL), followed by exposure to either anti-FLAG or anti-HA primary antibodies (at recommended dilutions, monoclonal mouse). Horseradish peroxidase linked antimouse antibodies (1:5000 dilution) were then added, followed by exposure to tetramethylbenzidine (TMB, Sigma) substrate solution. The reaction was halted with 2 M H 2 SO 4 , and the plates were read at 450 nm (BioTek Synergy HT plate reader). Immunosensor assembly The immunosensor was assembled on a 5-MHz AT-cut quartz crystal. Before assembly, the gold electrode was triple cleaned in piranha, and then received a final ethanol or acetone rinse and dried in a stream of N 2 (grams). The quartz crystal was then mounted in a flow cell holder, rinsed with phosphate buffer (PB, 50 mM phosphate, pH 7.2), and brought to resonant frequency at room temperature. For the work described herein, a Stanford Research Systems quartz crystal microbalance model 200, which measures both frequency and resistance, was used. Solutions were passed through the cell at a flow rate of 28 μL/min controlled by a Masterflex peristaltic pump. The sequence for biosensor assembly was protein G (20 μg/mL, 10 min), BSA (1 mg/mL, 5 min), antibody (20 μg/mL, 16 min), and viral simulant (500 μg/mL, 11 min) all in PB. The nanoparticles were filtered prior to use with a 0.22-μm filter from Millipore. The crystal was washed for a minimum of 10 min with PB between each step. Deposition of protein G allowed for the immobilization of the Fc region of anti-HA and anti-FLAG antibodies. The binding of non-functionalized Autiop nanoparticles (500 μg/mL, 11 min) with anti-HA antibodies was used as a negative control and as a test for nonspecific binding. Positive control was established utilizing anti-HA's recognition of recombinant H5 HA protein (5 μg/mL, 10 min). The adsorbance of each of these molecules resulted in changes in both frequency and resistance. The bound analyte mass is proportional to changes in the oscillation frequency of the quartz crystal as described by the Sauerbrey equation (Eq. 1), where Δf is the change in frequency, C f is the known sensitivity factor of a 5-MHz crystal (56.6 Hz cm 2 μg −1 ), and Δm is the change in mass [41] : In addition to mass, the frequency of the crystal is also dependent on the density and viscosity of the contact medium, a consequence of the solution's resistance to crystal oscillation [42] . The crystal frequency and resistance were recorded during the QCM experiments to allow for corrections to be made for solution resistance, following the work of Kanazawa and Martin [43] [44] [45] , which modifies the standard Sauerbrey equation to: Frequency change for solution resistance is nominally −2.464 Hz/Ω given an active crystal surface area equal to 0.40 cm 2 per sucrose calibrations similar to work reported previously [32] . The resulting change in mass is therefore a combination of changes in resistance and frequency, while taking into account solution resistance. The QCM results corroborated results from the ELISAs. In the development of a synthetic positive control for this influenza immunoassay, functionalized gold MPCs were synthesized and characterized. TEM analysis of the tiopronin-gold nanoparticles (Au-tiop) resulted in an average cluster size of 2.6±0.6 nm, with a range from 1.5 to 4 nm (Electronic Supplementary Material Fig. S1 ). The TGA tiopronin MPC mass loss equaled 37.9% and yielded a 2.67 Au:Tiop stoichiometric ratio for the nanoparticles assuming the loss of the Au(I)-thiolate staples (Electronic Supplementary Material Figs. S2 and S4 ). The assumption of the loss of the Au(I)-thiolate staples is justified by the observation of gold-thiolate compounds upon thermal decomposition in a mass spectrometer [46] . The average number of gold atoms calculated per cluster equaled 544 atoms. Combining the TEM and TGA yields a calculated empirical formula Au 544 Tiop 204 for the 2.6-nm diameter particle. Following place exchange of epitope(s) for tiopronin, the epitope loading was determined through 1 H NMR (Electronic Supplementary Material Fig. S3 ). The epitopes HA and FLAG each contain an exclusive amino acid whose proton signal occurs at a unique location in the 1 H NMR spectrum. Specifically, valine (V) is unique to the FLAG epitope and lysine (K) to the HA epitope. Tyrosine (Y) is common to both epitopes, but occurs in different stoichiometric ratios. The integrated resonance for each functionalized nanoparticle can be compared with that for nanoparticles protected solely with tiopronin to determine the average epitope stoichiometry of the nanoparticles. Peaks used for quantification consisted of signals at 1.2 ppm (V, 2 CH 3 ), 1.45 ppm (tiopronin, CH 3 ), 2.90 ppm (K, 2 ε-CH 2 ), 6.75 ppm (Y, δ-CH), and 7.05 ppm (Y, ε-CH Material Fig. S4) . The ELISA results confirmed specificity of the functionalized gold nanoparticles to the anti-HA and anti-FLAG antibodies (Fig. 1 ). Evidence of cross reactivity between the gold nanoparticles presenting both peptides and both antibodies corroborates with QCM data. HA-FLAG-Au and FLAG-Au were detected with the anti-FLAG antibodies, while HA peptides and HA-Au were not. Correspondingly, HA peptides, HA-Au, and HA-FLAG-Au were recognized by anti-HA, while FLAG-Au was not. The LOD was calculated to be 0.146 OD using the average absorbance of wells incorporating buffer only in the initial step, which served as a negative control. The gold particles displaying only tiopronin (Au-tiopronin) appeared to nonspecifically bind with anti-HA. Nonspecific binding could be reduced in ELISAs with the use of higher concentrations of BSA or with the use of more stringent rinsing protocols, such as the use of Tween 20, a nonionic surfactant, between steps, which might also remove bound particles and lower the sensitivity of these assays [47] . MPCs functionalized without the PEG linker between the epitope and the cysteine linkage showed no binding with their respective antibodies in either ELISA or QCM experiments, but once the PEG link was incorporated into the mimic's a b designs, the antibodies were able to bind with the epitopefunctionalized nanoparticles. Calculations using Spartan computational software show the spatial projection of tiopronin molecules to be about 7 Å from the gold surface, and the peptides, due to the PEG addition, extend 20 Å, from the gold surface [48] . The addition of the six-mer PEG link between the epitope and the terminal cysteine enhanced the probability for cluster-epitope interaction with antibody by distancing the epitope from the gold surface. This enhanced binding was demonstrated with both ELISA and QCM studies. Several observations can be made by inspection of the QCM responses (Fig. 2 , recorded data and schematic of adsorption) during sensor formation. Both antibodies are found to bind well to protein G. Observed spikes during assembly occur when reagent flowing through the peristaltic pump is momentarily interrupted for reactant exchange. A time delay (1-2 min) between spike and sensor response was due to the requisite transport time for the analyte to move through the peristaltic pump to the QCM sensor. binding of the nanoparticles and the HA protein to the HAantibody (Fig. 3a) and to the FLAG-antibody (Fig. 3b) . Detection of the HA protein confirms the ability of this assay to detect HA specific to the H5 HA subtype, and thus the use of this HA-functionalized nanoparticle as a viral simulant. This change in mass is based on the relationship that both frequency and resistance changes have on mass load. Originally, the Sauerbrey equation was applied in air or in a vacuum, and the resulting equation was only valid for thin solid layers deposited on the resonator [49]. Since then, extensive work has been done to establish the use of the QCM to probe interactions in a liquid environment, involving suitable oscillator circuits, fluid modeling in viscous and lossy fluids, as well as determination of the relationship between motional resistance and mass load [12, 45, 50] . The sensing layers utilized in this study should yield at most a layer 30-nm thick, using liberal estimates, where the Sauerbrey equation can be applicable to thin films less than 250 nm [51] [52] [53] [54] . Previous work, which uses a sucrose calibration, modifies the Sauerbrey equation to account for the changes that do occur in part from motional resistance, and therefore allows the ideology behind the Sauerbrey equation to apply in environments where energy is dissipated in the non-rigid liquid environment [30] . The nanoparticles increase the resistance of the crystals, and thus the rigidity, as well as decreased the frequency, resulting in a detectable increase of mass adsorbed on the surface of the crystal. Binding to the HA-antibody (positive Δm) occurs if the gold nanoparticle is functionalized with either HA or both HA and FLAG epitopes, but does not occur if the cluster is only FLAG functionalized (Fig. 3a) . The nonspecific binding observed with ELISA of Autiopronin to anti-HA was not observed with the QCM. QCM naturally prohibits nonspecific binding through the acceleration of adhered particles. This acceleration is generated by the oscillation of the quartz and can help remove weakly bound or nonspecifically bound molecules [55, 56] . Similar binding of anti-FLAG to FLAG-Au and HA-Au-FLAG but not HA-Au was measured with the QCM (Fig. 3b) . Thus, orthogonal and normalized binding at the QCM is observed consistent with immunological results obtained from ELISA, but is observable in a much shorter time interval (10 min compared to the 4 h it took to complete the ELISAs) when using the developed immunoassay and the QCM. The demonstrated binding between bi-functionalized nanoparticles and their respective antibodies makes evident the practicality of their use as a simulant for microorganisms, while lacking the difficulties associated with use of an attenuated or killed pathogen. When a change in Fig. 4 Calibration of HA protein binding to anti-HA. a The average change in mass of H5N1 HA protein binding was determined at varying concentrations. Briefly, (black line) 40 μg/mL (n=2), (red line) 20 μg/mL (n=5), (blue line) 10 μg/mL (n=3), (teal line) 5 μg/mL (n=3), and (pink line) 1 μg/mL (n=4). Also shown for comparison is the binding of HA-Au (green line) and HA-FLAG-Au (orange line). b A linear representation of the Langmuir isotherm produced by this average binding is shown with a linear relationship of y ¼ 0:035 AE 0:004 ð Þ x þ 4:90 AE 15:5 ð Þ Â 10 À10 and R 2 =0.96. c Calibration curve for an assay time of 1.5 min where the black squares are HA proteins, and the red and blue dots are HA-FLAG-Au and HA-FLAG, respectively b mass is observed upon introduction of the functionalized MPC to the immunosensor, the immunological response is shown to be above the limit of detection (∼3 ng, calculated by three times the average noise). Results show the average change in mass for protein G to be 128±48 ng, for antibodies to be 276±75 ng, and for the nanoparticle simulants to be 37±8 ng (Table 1 ). While the time was held constant for each step, the binding varied slightly, which could be due to slight variations in surface roughness and the surface coverage of prior adsorption steps. Even with a standard deviation of 75 ng for antibody adsorption, a deviation of only 8 ng was measured for the final detection step. The binding of the nanoparticles demonstrates saturation behavior (Fig. 3) . Based on the shape of the QCM curves and the rapid increase in mass with time, it can be assumed that the kinetics would occur quickly and with presumably large equilibrium association constants. To test this theory, a calibration of anti-HA to HA binding was determined, and the binding of the functionalized nanoparticles was compared. The HA protein from H5N1 was exposed to anti-HA antibodies at concentrations ranging from 1 to 40 μg/mL (Fig. 4a, b) . This binding can be compared at any time point to generate a calibration curve (Fig. 4c , example shown is at 1.5 min). Based upon the desired separation of the lower data points, the binding can continue for several more minutes. Also, the amount of nanoparticle exposed to the surface can be lowered to prevent overloading of the sensor. Measuring the maximum binding that occurs, as opposed to lower time points, can be used to determine the equilibrium association constant (K a ) and increase our understanding of the affinity of our sensor. This constant was determined by fitting to a linear rearrangement of the Langmuir adsorption isotherm, where C was plotted versus C/Δm (Eq. 3, Fig. 4b ) [30] : This yielded an equilibrium association constant for the binding between the HA protein and anti-HA of 7.14± 0.26×10 7 . This K a is in the range expected for antibodyantigen interactions, from 10 6 to 10 10 M −1 [57] . In fitting the binding of the nanoparticles to this Langmuir isotherm calibration (at the experimentally used nanoparticle concentration of 500 μg/mL), the sensor response to HA-FLAG-Au would have the same binding as 0.92±0.01 μM HA protein, and HA-Au would generate the same sensor response as 1.50±0.02 μM HA protein. The large response seen is at the maximum of the Langmuir isotherm. This demonstrates that even at lower concentrations, these functionalized nanoparticles can be used as a positive control. The functionalized MPC-antibody binding is not inhibited by the presence of an additional non-interacting epitope (either FLAG or HA) on the polyfunctionalized nanoparticle; therefore, multiple binding interactions can be explored simultaneously. The ability to create multi-epitope-presenting nanoparticles that can orthogonally bind to specific monoclonal antibodies has been demonstrated using both ELISA and immunological QCM. Determination of the extent of antibody-functionalized nanoparticle binding is rapid using the QCM compared to ELISA. Also, like ELISA, the immunological response is specific, with QCM incurring less nonspecific binding. Interaction of the epitope with its antibody was improved through the use of a PEG linkage for epitope attachment to the MPC. Binding studies at the QCM show that polyfunctionalized gold nanoparticles exhibit the expected affinity to both antibodies that a normal immunological response is achieved from matched antibody-antigen couples and that an orthogonal response results otherwise. The results demonstrate that binding of polyfunctionalized gold nanoparticles could be used to determine sensor functionality, without resorting to the use of attenuated or killed microorganisms, or extracted and purified whole proteins.

Hepatitis G Virus associated aplastic anemia: A recent case from Pakistan

BACKGROUND: Aplastic anemia (AA) is a serious and rare disorder characterized by a hypocellular bone marrow. Hepatitis associated aplastic anemia (HAAA) is a variant of aplastic anemia in which aplastic anemia follows an acute attack of hepatitis. Several reports have noted an association between HGV and hepatitis-associated aplastic anemia besides other hepatitis causing viruses. CASE PRESENTATION: A female girl of age 11 year with a history of loose motion for one month, vomiting for last 15 days and poor oral intake for last few days is reported here. The physical examination presents fever, pallor whereas bleeding, hepatomegaly, Splenomegaly and bruising were absent, abdominal ultrasonography confirmed the absence of hepatomegaly, Splenomegaly and lymphodenopathy. The laboratory investigation parameters were: haemoglobin 6.2 g/L, total leucocytes count 1.51, neutrophils 0.47%, absolute reticulocyte count 0.5%, Monocytes 0.16%, red cell count 3.2 mil/uL, Picked cell volume (PCV) 30.13%, Mean Corpuscular Volume (MCV) 78 fL, Mean Corpuscular Hemoglobin (MCH) 26.3 pg. The liver enzymes were alanine aminotransferease (ALT) 98 IU/L, aspartate aminotransferase (AST) 114 IU/L. Serologic and molecular tests for hepatitis A, B, C, D, E, TTV, B19 were negative, whereas HGV RNA PCR test was found positive for hepatitis G virus. The bone marrow aspirate and trephine biopsy examination revealed hypo- cellularity, erythropoiesis, myelopoiesis and megakaryopoiesis. CONCLUSION: HAAA is an uncommon but severe condition, which may occur following idiopathic cases of acute hepatitis. Our finding suggests the involvement of HGV in the development of aplastic anemia. In patients presenting with pancytopenia after an episode of acute hepatitis, the definitive diagnosis should be considered and confirmed by RT-PCR and if possible by bone marrow biopsy.

Hepatitis G virus was reported first time has a non-A-E hepatitis and placed as flavivirus [1] .The induction of this new agent in the family of Hepatitis has attracted significant attention because of its etiology [2] . Hepatitis G virus has been marked as a cause of non-A through E acute viral hepatitis and sharp liver failure. Aplastic anemia complicating hepatitis is an uncommon but well recognized phenomenon. Hepatitis associated aplastic anemia is a severe disorder with a high mortality (85%) [3] .Hepatitis associated aplastic anemia (HAAA) is a deviation of aplastic anemia in which aplastic anemia follows an acute attack of hepatitis. The marrow failure can be severe and is usually lethal if untreated. Lorenz and Quazier has documented first time HAAA in two case back in 1955 [4] , by 1975 more than 193 cases had been reported [5] . Adil et al has reported, severe aplastic anaemia (SAA) 51.4%, very severe (VSAA) in 16.7% of 144 patients of aplastic anemia cases [6] . A number of reports have mentioned alliance between HGV and HAAA [7] [8] [9] [10] . Number of HAAA cases with a history of multiple blood transfusions has been reported [11, 12] . Crespo et al has documented a case of 24 year old man have community acquired HGV that later progress into severe aplastic anemia, point out HGV for both hepatitis and aplastic anaemia. However greater number of serum samples are needed to prove the association of hepatitis G virus and aplastic anaemia [13] . Moatter et al. has reported 5/43 patients of haemodialysis with raised liver enzyme, reduced platelet count and 21/100 patients of polytransfused b-thalassemia major children infected with HGV RNA form Pakistan [14, 15] . In this study well characterized samples of 93 aplastic anaemia patients before blood tranfusion were included. These characteristics include history, physical examination, haematological investigation, bone marrow aspirate and trephine biopsy examination, liver function test (LFTs), renal parameters, viral profile and abdominal ultrasonography. The diagnosis of HA-aplastic anemia was made on the basis of hepatitis (elevated serum aminotransferase enzymes, jaundice, absolute neutrophils counts, platelet counts and reticulocytes. All the 93 samples were checked for serological marker of HAV, HBV, HCV, HDV, HEV, HGV,TTV and B19. One of 93 samples from patients with HA-aplastic anemia has hepatitis G associated aplastic anaemia with positive HGV RNA. A female girl of age 11 year is reported here. The patient had a history of loose motion for one month, vomiting for last 15 days and poor oral intake for last few days. The physical examination presents fever, pallor whereas bleeding, hepatomegaly, Splenomegaly and bruising were absent, abdominal ultrasonography confirmed the absence of hepatomegaly, Splenomegaly and lymphodenopathy. The laboratory investigation parameters were: haemoglobin 6.2 g/L, total leucocytes count 1.51, neutrophils 0.47%, absolute reticulocyte count 0.5%, Monocytes 0.16%, red cell count 3.2 mil/uL, Picked cell volume (PCV) 30.13%, Mean Corpuscular Volume (MCV) 78 fL, Mean Corpuscular Hemoglobin (MCH) 26.3 pg. The liver enzymes were alanine aminotransferease (ALT) 98 IU/L, aspartate aminotransferase (AST) 114 IU/L. Serologic and molecular tests for hepatitis A, B, C, D, E, TTV, B19 were negative, whereas HGV RNA PCR test was found positive for hepatitis G virus. The bone marrow aspirate and trephine biopsy examination revealed hypo-cellularity, erythropoiesis, myelopoiesis and megakaryopoiesis. Flaviviruses belong to enveloped viruses with a single positive sence RNA about 10 kb. Hepatitis G virus medium of transmission mostly through blood. The possible roel of hepatitis G virus infection in the pathogenesis of rare non-liver disease has been suggested but need to be recognized. By some unknow reseason aplastic anemia some time proceded by hepatitis. Few reports have analysed the role of HGV in the development of HAAA [8, 9, 11] . Crespo et al in print a case through negative serological markes for HAV, HBV, HCV, HEV, hypoplastic marrow low platelet and white cell counts but detected HGV-RNA before any blood transfusion [13] . Byrnes et al has described hepatit G Virus positive case of 26 year old man before the use of medication, blood tranfusion or intravenous drug abuse [9] . Zaidi et al, reported a 19 year male before blood transfusion with positive HGV by RT-PCR and suggested that in the absence of any other clincial manifestions the possible infectious agent may be HGV for hepatitis G virus associated aplast anaemia [8] . In the list of studies of Hepatitis G Virus associated aplastic anaemia before blood transfusion we report a case of 11 years female girls. Whereas some reports confirms the presence of HGV viraemia in the patient aplastic anaemia after blood transufion [8, 9, 11, 12, [16] [17] [18] . But however concluded that HGV viraemia is frequent in patients with aplastic anaemia [19] . Kiem et al reported similar finding after logistic regression analysis, that HGV RNA in transfused patients was 5.9 times higher compare to untransfused patients (P = 0.001). This implicates transfusion as major source of HGV with aplastic anaemia [18] . The published literature point out that studies must be performed on many more aplastic anaemia patients prior to blood transfusion [20] . However, to find such patients in large number are not normally avaible to study, so far individual cases are reported. The ideal case regarding Hepatitis G associated aplastic anaemia are pre blood transfusion. In conclusion, HAAA is an uncommon but severe condition, which may occur following idiopathic cases of acute hepatitis. Our finding suggests the involvement of HGV in the development of aplastic anemia. In patients presenting with pancytopenia after an episode of acute hepatitis, the definitive diagnosis should be considered and confirmed by RT-PCR and if possible by bone marrow biopsy. Written informed consent was obtained from the patients a copy of which is available for review by Editor-in-Chief of this journal. Authors' contributions SARS aided in acquistion of data that was included in this case report and drafted the manuscript. MI aided in acwuision and interpretation of the data analysed. AH helped in statistical analysis of the data and in editing of this manuscript. All authors have read and approved the final version of this manuscript.

Travel Patterns in China

The spread of infectious disease epidemics is mediated by human travel. Yet human mobility patterns vary substantially between countries and regions. Quantifying the frequency of travel and length of journeys in well-defined population is therefore critical for predicting the likely speed and pattern of spread of emerging infectious diseases, such as a new influenza pandemic. Here we present the results of a large population survey undertaken in 2007 in two areas of China: Shenzhen city in Guangdong province, and Huangshan city in Anhui province. In each area, 10,000 randomly selected individuals were interviewed, and data on regular and occasional journeys collected. Travel behaviour was examined as a function of age, sex, economic status and home location. Women and children were generally found to travel shorter distances than men. Travel patterns in the economically developed Shenzhen region are shown to resemble those in developed and economically advanced middle income countries with a significant fraction of the population commuting over distances in excess of 50 km. Conversely, in the less developed rural region of Anhui, travel was much more local, with very few journeys over 30 km. Travel patterns in both populations were well-fitted by a gravity model with a lognormal kernel function. The results provide the first quantitative information on human travel patterns in modern China, and suggest that a pandemic emerging in a less developed area of rural China might spread geographically sufficiently slowly for containment to be feasible, while spatial spread in the more economically developed areas might be expected to be much more rapid, making containment more difficult.

Good morning/afternoon/evening ! With the spread of H5N1, potential next pandemic will be a terrible threat to human beings. It is necessary for us to develop feasible and effective pandemic preparedness and response strategy. The resident traveling survey is implemented by Chinese Center for Disease Control and Prevention. The purpose of the survey is to learn about the residents' traveling habits through collecting individual traveling data (including the distance from home to work/school and other traveling within past 7 days) to provide scientific evidence for establishing pandemic modeling and developing the strategy of pandemic influenza preparedness and response in China. We mainly collect the information about your everyday traveling such as going to work or school and other traveling within past 7 days. Simple demographic data such as gender and age will also be collected. There are not any sensitive or private questions in the questionnaire, so you will not feel any discomforts. Meanwhile we will not begin the face to face survey until obtain oral informed consent from you. We promise that all the information you provide will be kept secret strictly. You will be voluntary for the survey. If you are not willing to take part in our survey, it would not bring any bad impact for you. It will take you less than 10 minutes to finish the survey, we feel sorry for bothering you. If you have any question, please consult our investigators, we will try our best to explain for you. If you consent to participate, let's begin the survey. Please answer all the questions truthfully. We really appreciate you for your kindly help. 2. Fill the ID No. of the interviewees who has journeys caused by non-work/non-school in the last 7 days. and the sequence should be consistent with section one. 3. If the destinations are more than one place, please fill all the destinations in the blank. If the destinations are more than one place, please fill all the GPS distance in the blank. . For national travel, if the journey includes flows between rural area and urban area, please mark "√" in the blank. For international travel, if the destination is HK SAR, please fill①in "HK&Macao " column, if the destination is Macao, please fill②in "HK&Macao " column, if the destinations include other countries besides or except HK&Macao, please fill the actual address in Other column.(Other: all other countries except HK SAR and mainland China). 6. ①Business ②Tour ③Visit relatives and friends ④Other (If choose ④, write down travel reason directly in the blank). Don't need to fill. Good morning/afternoon/evening ! With the spread of H5N1, potential next pandemic will be a terrible threat to human beings. It is necessary for us to develop feasible and effective pandemic preparedness and response strategy. The resident traveling survey is implemented by Chinese Center for Disease Control and Prevention. The purpose of the survey is to learn about the residents' traveling habits through collecting individual traveling data (including the distance from home to work/school and other traveling within past 7 days) to provide scientific evidence for establishing pandemic modeling and developing the strategy of pandemic influenza preparedness and response in China. We mainly collect the information about your everyday traveling such as going to work or school and other traveling within past 7 days. Simple demographic data such as gender and age will also be collected. There are not any sensitive or private questions in the questionnaire, so you will not feel any discomforts. Meanwhile we will not begin the face to face survey until obtain oral informed consent from you. We promise that all the information you provide will be kept secret strictly. You will be voluntary for the survey. If you are not willing to take part in our survey, it would not bring any bad impact for you. It will take you less than 10 minutes to finish the survey, we feel sorry for bothering you. If you have any question, please consult our investigators, we will try our best to explain for you.

Spatial dynamics of the 1918 influenza pandemic in England, Wales and the United States

There is still limited understanding of key determinants of spatial spread of influenza. The 1918 pandemic provides an opportunity to elucidate spatial determinants of spread on a large scale. To better characterize the spread of the 1918 major wave, we fitted a range of city-to-city transmission models to mortality data collected for 246 population centres in England and Wales and 47 cities in the US. Using a gravity model for city-to-city contacts, we explored the effect of population size and distance on the spread of disease and tested assumptions regarding density dependence in connectivity between cities. We employed Bayesian Markov Chain Monte Carlo methods to estimate parameters of the model for population, infectivity, distance and density dependence. We inferred the most likely transmission trees for both countries. For England and Wales, a model that estimated the degree of density dependence in connectivity between cities was preferable by deviance information criterion comparison. Early in the major wave, long distance infective interactions predominated, with local infection events more likely as the epidemic became widespread. For the US, with fewer more widely dispersed cities, statistical power was lacking to estimate population size dependence or the degree of density dependence, with the preferred model depending on distance only. We find that parameters estimated from the England and Wales dataset can be applied to the US data with no likelihood penalty.

Spatially explicit models are critical to understanding the spread of infectious diseases through populations and to better inform policy aimed at controlling that spread. Indeed, recent outbreaks of communicable diseases in human populations have triggered a series of studies addressing the spread of directly transmissible infections at a country level [1 -5] . Identifying a possible backbone of high probability transmission paths through populations may underpin the development of effective interventions to curtail spread on the population network [6] . For example, in human diseases, spatial models and microsimulations can quantify the possible role of border control, quarantine or transport reductions in curtailing local and international spread [2 -4,6 -11] . Spatial microsimulation models like these are critical to making effective policy decisions. Spatial models also allow limited control resources to be used where they might be most effective. For instance, during the initial stages of an outbreak of a new virus, disease incidence tends to occur in spatial clusters with occasional long-range infection events [10] . As case numbers increase in the start location, the frequency of long-range infection events increases. This spatial pattern was seen during the early stages of the 2009 influenza pandemic in Mexico, leading to local foci seeded by long-range interactions to other countries [12, 13] . However, those epidemic models rely on a set of structural assumptions that need to be validated from data. A basic assumption of many spatially explicit transmission models is that flows between urban centres are a function of the distance between them and their attributes, most notably residential or worker population sizes [14, 15] , resulting in a so-called gravity model. However, for human diseases, little work has been done to validate the underlying assumption that human travel patterns are predictive of the spatial spread of diseases. Early models of spatial coupling in ecology assumed that connectivity between populations was inversely related to the distance between them [16] . For people (and most animals) distance-based coupling is too simplistic an assumption. Movement between large population centres is disproportionately more frequent than between smaller ones [17] . Xia et al. [18] found that a distance-only model for spread of measles in the UK was a poor fit to weekly measles data from England and Wales from 1944 to 1967. The gravity model used in that study measures connectivity between population centres as a function of distance and a function of the population sizes of the origin and destination cities. However, measles is a childhood disease, and so the spatial dependencies of the host are different than for infections that affect both adults and children, such as pandemic influenza. Gravity models and other spatial interaction models allow understanding of the movement of populations from one location to another in the absence of movement data. Models, once validated, can predict modifications in connectivity when populations grow or shrink, when workflows vary owing to economic changes or if restrictions are imposed on one city and not others. This is in contrast to movement surveys, which are context-specific and provide a snapshot of the movement habits of a population. The strength of connection between cities may be density independent, that is, the sum of connectivity of a city to all its neighbours does not depend on the number of neighbours that city has. In contrast, density-dependent connectivity links two cities at a strength solely determined by the sizes of those cities and their distance apart, so that the total connectivity of any one city scales with the number of close neighbours. Density-independent transmission gives a total force of infection, which is independent of the remoteness of the population, whereas density-dependent transmission will cause populations with many neighbours to experience a higher force of infection than those cities that have few neighbours. Thus, a densitydependent model will predict that isolated populations are less likely to become infected than populations with many neighbours, or few very large neighbours. The concepts of density dependence/independence have not only been used to model interactions between cities; they have also been used extensively in individual based models of disease spread. Most past studies tend to assume either density dependence (e.g. for animal epidemics; [19, 20] ), or density independence (e.g. for most human diseases) [3, 4] . Here, we explore the extent to which city-to-city (rather than individualto-individual) contacts are density independent by constructing a model that can capture intermediate levels of density dependence. In this paper, we analyse mortality datasets from England and Wales and the United States from 1918 to 1919 to examine the pattern of spatio-temporal spread and the extent to which gravity models can reproduce observed trends. The 1918 pandemic constitutes a rare example of a well-documented epidemic in a largely susceptible human population, where the high mortality gives a clear incidence signal, and is therefore a rare opportunity to validate models of epidemiologically relevant geographic coupling. We examine the effect of city-specific characteristics (e.g. location, distance from other cities, population size and the number of influenza-related deaths) on the pattern of spread seen. We also investigate the impact of the distribution of cities in each country. The analysis provides further insight on the spatial variation in the spread of the 1918 influenza pandemic at a country level, much of which remained unexplained in past studies [18, [21] [22] [23] . The 1918 pandemic H1N1 virus appears to have entered the general population of the UK and US in the spring of 1918 causing a reportedly mild disease [24] . This early wave was associated with increased mortality, but was probably only noticed because influenza is rare in summer. This epidemic waned later in the summer, but infection reappeared in the autumn with much increased mortality. It is unclear whether the viruses causing the spring and summer waves were closely related, but there is increasing evidence that the spring wave gave immunological protection against the autumn wave at a population level [25, 26] . By September 1918, the pandemic was a prominent global phenomenon. The autumn wave was virtually universal, albeit with some variation between countries in precise timing. In both the UK and the US, a third wave of influenza occurred in early 1919 although with greater heterogeneity in mortality rates between cities [24, 27, 28] . US cities had more variation in the severity of the major wave than the UK, probably in part because some enacted more stringent non-pharmaceutical interventions to mitigate the epidemic [27, 28] . The third wave was less pronounced in US cities than in England and Wales, again perhaps partly because of the effect of interventions. The England and Wales dataset shown in figure 1a was published in the Supplement to the 81st Registrar General Report [29] . It provides weekly death counts and annualized mortality rates per 1000 from 83 county boroughs, 84 municipal boroughs, 71 urban districts and three unclassified urban centres in England and Wales, for a 46 week interval, 29 June 1918 -10 May 1919. The first 10 weeks are designated as wave 1, the next 19 as wave 2 and the last 17 as wave 3. Our analysis focuses on the second wave because it occurred in all cities (unlike wave 3) and because recording of mortality had begun in all cities before its arrival (unlike wave 1). In addition, reporting of influenza and influenza-related mortality changed between the first and second waves. Point locations of all urban centres were determined from the current or historical records, with Euclidean distance used to quantify inter-centre separation. Further information is given in the electronic supplementary material. We compiled a US city dataset (figure 1b) from five publications reporting the Weekly Health Index as collated by the Bureau of the Census [24, [30] [31] [32] [33] . It covers the period 14 September 1918 to 15 March 1919 and contains weekly pneumonia and influenza death counts for 47 cities in the US. The US data therefore covers a period of two waves, with not all cities experiencing the later wave. There is very good agreement between different sources where they overlap. We used the Euclidean distance between cities (accounting for curvature of the Earth) to measure separation. Further information is given in the electronic supplementary material. The analysis requires an estimate of when each city became infected to allow potential sources of that infection to be identified. For each city infected in week t, the candidate infectors are those infected in any week before t. We define the infection week of city i, t i , to be the first that meets a set of conditions on mortality in weeks t i þ 1, t i þ 2 and t i þ 3. We use mortality values ahead of t i to include the time from infection to death. A week could be designated the infection week if either (or both) of two sets of criteria were met for mortality in the following weeks. The first set of criteria required the mortality rate in week t i þ 1 to be above a certain threshold, to have increased in t i þ 2 and to be above a higher threshold in t i þ 3. These criteria are intended to ensure that the epidemic in that city is patently increasing. The second set of criteria was designed to capture cities where there was a rapid onset of increased influenza-related mortality. They therefore used a higher threshold on mortality in week t i þ 1, but less strict conditions on rate of growth in the following two weeks. The week of infection determined was found to be relatively robust to the precise choice of thresholds used. For further details on the algorithm and a spatio-temporal display of the result, see the electronic supplementary material. In formulating our inter-city transmission model, we take the city as our unit of study. Each of N cities, i, has an infection time t i , an invariant population size P i and a time-varying mortality rate, r i,t at time t. Infected city i is separated from susceptible city j by distance d ij . Each week, each city can be in one of the three disease states: Susceptible, Latent or Infectious. We assume that all cities are susceptible at the start of a wave, they are latently infected for one week on infection, and that a city becomes infectious the week after it becomes infected. We assume that all transmission is endogenous to England and Wales or US after external seeding to the first infected city in each territory. If a city becomes infected in week t i , the candidate infectors are only those cities that are infectious in week t i . We assume that the transmission parameters are constant through time. The model formulation aims to capture the effect of distance and population size on the connectivity of cities. Three modes of spatial transmission are considered: density-independent connectivity, densitydependent connectivity and an intermediate form where the degree of density dependence is estimated. The model also examines the different assumptions regarding a city's infectivity over time. The force of infection, l is the hazard of infection from one city to another. From infected city i on susceptible city j at time t, it is: where source city population and distance are normalized together. n and m are estimated parameters on source and destination population sizes, respectively. d ij represents the distance between cities with power parameter g to be estimated. w is an estimated parameter relating the infectivity of a city to its mortality rate. When w ¼ 1, the infectiousness of a city at time t i is proportional to the death rate in that city at time t i þ 1. We use one week as a lag from infection to death [24, [34] [35] [36] . A value of w ¼ 0 gives a flat infectiousness profile, independent of the death rate in the source city. Intermediate values of w give variation in infectiousness, which scales sub-linearly with weekly mortality. Estimating w allows us to assess whether mortality rate is a good proxy for infectiousness in an infected city. b is a time-invariant estimated infectivity term. Parameter 1 describes the strength of connection of a susceptible city to all possible infectors. 1 ¼ 0 gives the density-dependent model and 1 ¼ 1 gives the density independent model. By allowing 1 to vary, we allow the model to estimate the degree of density dependence in connectedness between the cities. The total force of infection on city j at time t is given by: where Since the force of infection is a hazard, the probability that a susceptible city j is infected in a week t j is given by: t¼0 Àl j;t ! ð1 À expðÀl j;t j ÞÞ: We used a Bayesian framework for statistical inference. The log-likelihood is given by: ln P t j À Á : We explore the joint posterior distribution of parameters by Markov Chain Monte Carlo (MCMC) sampling [37, 38] . We sampled parameters on a logscale using a random walk update scheme. b and g were jointly updated, while 1, m and n were updated singly. Five MCMC chains were started from a variety of start points within a credible range to assess convergence. Convergence was achieved within 100 000 iterations for all models from all starting parameter values. For each model, the chain was run for 500 000 iterations including a burn in of 100 000. Parameter estimates and equal-tailed 95 per cent credible intervals were obtained from the posterior distribution of 80 000 values thinned from the last 400 000 samples of the MCMC chains. To investigate which components are most important for describing the spread of influenza, we consider a set of simplified variants of the model presented above. In those variants, each parameter can be either fixed at 0, at 1 or be estimated by MCMC. For a full comparison of each component of the model, see the electronic supplementary material. The Deviance Information Criterion (DIC) is used to compare models [39] . This is calculated using the median parameter values owing to non-normality in the likelihood [39, 40] . Lower values are preferable and a difference of around 5 units is considered important [41] . The model is used to generate epidemic trees [42, 43] . We sample 1000 parameter sets from the joint posterior distribution and calculate the probability of infection for each potential infector city. The most likely tree for each parameter set is generated by calculating which 'infector' city has the highest probability of infecting each 'infectee' city. The distance to this infector, the probability of the infector-infectee pair and the number of infectees each infector creates are calculated for each parameter set. Mean values are weighted by the frequency of infector -infectee pairs from 1000 trees. We examine the ability of the models to recreate the observed epidemic by simulation. We use 1000 parameter sets sampled from the joint posterior distribution and for each set, we simulate an epidemic using the first infected city as a source of infection. Once a city is infected, the observed mortality curve is used to model the infectiousness of that city through time. We also calculate the probability distribution of the week of infection for each city conditional upon the observed epidemic up to that time point. We use 1000 parameter sets sampled from the joint posterior distribution and for each set, we calculate the probability of infection each week for each city given the epidemic observed up to that time point. We use Welch's two-tailed t-test to differentiate outlying groups. Comparison of spatial and non-spatial population-independent models shows that inclusion of distance substantially improves model fit for both England and Wales and the US (D DIC ¼ 64.7 and D DIC ¼ 33.3, respectively). DIC and parameter estimates for the distance-only model are given in column 1 of tables 1 and 2. Previous formulations of the gravity kernel in the literature have considered either density-dependent (1 ¼ 0) or density-independent transmission (1 ¼ 1). Figure 2 compares the fit (expressed by the posterior deviance) of these two formulations and with that from the model where the degree of density dependence, 1, is estimated. This comparison is made for models assuming no linear or a fitted power-dependence of spatial coupling on both source and destination city population size. In figure 2a , w is estimated, whereas in b it is fixed at 1. See model components in the electronic supplementary material for further comparisons. For England and Wales (figure 2a), in each population context the variant that estimates the degree of density dependence (the lightest curve of each colour) gives a slightly better fit than models with no density dependence, with pure density dependence fitting substantially less well. The comparison also shows that the models which estimate the effect of origin and destination city population sizes on the connectivity of cities are much better than either the populationindependent or linear population size-dependent models. The same set of comparisons is made for the US in figure 2b . The situation is more complex with the posterior distribution of many model variants lying in the same area. Comparisons by DIC value cannot distinguish these models. Unlike in England and Wales, there is no density-dependence variant which has lower deviance for all three of the population sizedependence variants examined. Inclusion of nonlinear population size-dependence does not penalize the fit of the US model, and so cannot be definitively excluded as being consistent with the data. The models presented in columns 5 and 6 in table 2 have different population relationships, but the same DIC score. The credible intervals on the population parameters of the densitydependent population with infectivity model (column 5) are very wide suggesting that little information is added by the inclusion of these parameters. In England and Wales, the lowest DIC model is one where the degree of density dependence is estimated and the effect of population is also estimated. This is in contrast to the US where population-independent models either with density-dependent or estimated density dependence spatial interaction terms are indistinguishable. We tested models with three types of infectiousness profile through time: constant infectivity, a linear relationship between infectivity and mortality in the week ahead, and an estimated power-law relationship between mortality and infectivity. Mixing was poor when estimating w with the US data so we only compare the first two models in that setting. In England and Wales, the linear infectivity model has a DIC value of more than 25 above either the constant or estimated infectivity model. Parameter estimates for the constant-infectivity and estimated-infectivity model variants are shown in columns 4 and 5 of table 1 using the density-dependent population-dependent framework from the previous comparison in England and Wales. These two models are indistinguishable by DIC (D DIC ¼ 0.1). Estimates for all other parameters are very comparable between these two models. The estimated relationship includes two inputs from the infected city: the mortality rate and the population of the city. It can be more difficult to estimate parameters regarding infectivity, so we tested a model which takes only one piece of information from the infected city. The final column in table 1 shows a model which takes the mortality rate from the infected city into account but does not include the population size of that city. There is an improvement in the DIC score for this model of 4.4 over the constant infectivity model. Table 2 shows parameter estimates for models in the US. The difference in DIC score between a constant infectivity model and one with a linear relationship between mortality and infectivity is negligible in either a distance-only model framework (columns 1 and 6) or a population-dependent framework (columns 2 and 5). Adding infectivity information does not improve the fit of the model. In the England and Wales dataset, the lowest DIC model is the single infected city parameter model in column 6 of table 1. The model is dependent on the destination population size, has estimated dependence of infectivity on mortality and an estimated intermediate degree of density dependence. The distance power g was estimated as 1.18 (0.96, 1.39). A lower value was found for models in the US, where for the most parsimonious low DIC model (density dependent, population independent), g was estimated as 0.79 (0.54, 1.00). Figure 3 shows the distance kernels for the two datasets. The credible intervals for g overlap for the two datasets. The power parameter on the destination city population, m, was estimated at 0.40 (0.25, 0.54) in England and Wales. The credible intervals exclude 1, demonstrating that as population size increases, the susceptibility of the city increased more slowly. We used the posterior median parameter estimates fitted to the England and Wales dataset to calculate a likelihood value in the US dataset. By likelihood ratio test, this value was not different from the most parsimonious low DIC US model (255.23, 257.72, p . 0.97). We therefore cannot reject the assumption that spread had the same characteristics in the US and England and Wales, though clearly the smaller size of the US dataset reduces inferential power. the epidemic during which each city was infected. Inferred city-to-city infection events more frequent than 70 per cent (in 1000 trees) are shown in black, events of lower frequency are shown in grey. Interactions in weeks 0 -3 are longer range than those in weeks 4 -7 ( p , 0.01), which are in turn longer range than those in weeks 8 -10 ( p , 0.01) (figure 4d ). The probability that the most likely infector was responsible for each infection falls as the epidemic progresses because there are many more potential infectors available later (figure 4e). Cities infected early give rise to more infections than those infected late in the wave, as expected, but the range is large, with some early cities giving rise to no new infections (figure 4f ). Figure 5 shows results for two models using the US data. We compare the most likely infection trees for the distance-only constant infectivity model with parameters inferred from the US data (figure 5a) with a model where parameters used to generate the trees are taken from the England and Wales single-infected city parameter model (figure 5c). In the distance-only constant infectivity model, the nearest infected city is always the most likely infector. In contrast, with the England and Wales parameters, some links between cities are high frequency, while other cities have several potential infectors of intermediate frequency ( figure 5b) . As in England and Wales, infection events inferred early in the epidemic have a higher support than those later in the epidemic. There are some exceptions owing to the distribution of cities in the US dataset-Oakland and San Francisco are distant from all other cities but very close to each other. In the distance-only constant infectivity model, some cities may give rise to a large number of new infections (e.g. Pittsburgh gives rise to nearly a quarter of infections) (figure 5d ). The effect of a city acting as a hub of infection is reduced in the more complex model, as the risk of infection from one city to another is the combined effect of several factors including distance. For England and Wales, there is a relatively good agreement between observed and simulated epidemic curves (figure 6b). The observed epidemic curve rises more steeply than the simulation curves in the early stages of the epidemic, and peaks one week earlier than the simulation mean. This suggests that the model may underestimate the external infection pressure early in the wave. We calculated the probability that a city was infected in each week given the observed behaviour of all other cities up to that time. In England and Wales, 245 of 246 cities lie within the 95 per cent interval of their expected distribution. Figure 6a shows the cities which the observed infection week lies outside the stricter inter quartile interval. For further information see the electronic supplementary material. There are no population size ( p ¼ 0.36) or density trends ( p ¼ 0.11) in these cities, which are typically infected later in the epidemic ( p , 0.01 for difference in infection week). In the US, all cities lie within the 95 per cent probability interval and all but three lie within the inter quartile interval. Those three outlier cities are smaller than other cities ( p ¼ 0.01) but equally distributed in space ( p ¼ 0.88) and time ( p ¼ 0.06). We have tested the effect on parameter estimates in England and Wales of relaxing the single-introduction assumption inherent in the model. We re-estimated the parameters conditioning on infections that occurred from week 3 of the epidemic onward. There is a small increase in the kernel power parameter estimate, which causes the kernel to decay more rapidly with distance (electronic supplementary material, figure S9 ). This suggests that the very long-range interactions, which are forced to occur early in the epidemic impact the shape of the kernel. However, the credible intervals largely overlap which indicates this assumption does not affect the fit of the model to a large degree. In the US the simulated curves for the distance-only constant infectivity model are shown in figure 5e and for the England and Wales parameters in figure 5f. In both cases, the mean simulated and observed curve are very comparable, with the distance-only constant infectivity model giving peak incidence in the same week as observed. Figure 5g shows the observed week of infection against the simulated week of infection for all 1000 simulated epidemics. There is good correlation between the observed and simulated weeks of infection for both parametrizations. We have tested the effect of thinning the England and Wales dataset so that it more closely resembles the US dataset to determine if the differences in formulation between the best models for each dataset are owing to the smaller number of cities in the US dataset. We removed all cities with fewer than 90 000 inhabitants in England and Wales leaving 46 cities distributed quite evenly in England and Wales as shown in the electronic supplementary material, figure S10. There were identifiability problems in estimating the density-dependence parameter 1 using the thinned dataset. The best model by DIC comparison gave a distance-only interaction (no dependence on population size) with infectivity scaling linearly with mortality in a density-independent framework. As we found with the US data, it is difficult to disentangle the effects of population and infectivity parameters because these feature in different combinations in comparable DIC models. We have presented a statistical analysis of the spatiotemporal spread of the 1918 influenza pandemic between cities in England and Wales and the US. The results demonstrate that for England and Wales, a model with intermediate levels of density dependence in the connectivity between cities gives the best fit to the observed pattern. For the US dataset, where there are few, large and widely spaced population centres, estimating the degree of density dependence does not improve the fit. In both contexts, city population size affects inter-city coupling sub-linearly. Parameter estimates and model formulation inferred from the data of England and Wales explain the US dataset well. Gravity model parameter estimates generated in this study are comparable with values found in studies describing the spread of seasonal influenza [5, 44] . Our analysis demonstrates the degree of spatial locality in the large-scale geographical spread of influenza in both England and Wales and the US in 1918. However, it is difficult to directly compare the kernel power estimate from this study with those from other studies owing to differences in the functional forms used. For instance, Viboud et al. [5] estimate two power parameters above and below a given distance threshold when modelling the spread of seasonal influenza in the US. Gravity models used to describe the spread of measles in the UK by Xia et al. [18] assumed a kernel power of 1, rather than fitting this parameter. The distance power estimates we found for England and Wales and the US are quite different from each other. It is not surprising that there is a disparity in the distance kernel in England and Wales and the US, as the spatial scale in the US is much larger than in England and Wales. In comparing the US and UK, it should be noted that the mean distance between cities is of course much larger in the US (see electronic supplementary material, table S1). In theory, this gives better resolution for estimating the kernel shape, as the range of inter-city separations is an order of magnitude larger than for the UK. However, this is counterbalanced by the smaller size of the US dataset, which reduces inferential power. The low kernel power parameter estimates we have found in both England and Wales and the US suggests that long-distance interactions were important in spreading influenza between distant cities in both countries. At the start of the major autumn wave in 1918, the armistice was more than two months away and it is likely that travel relating to the war effort, including troop movements, might have enhanced the frequency of long-distance movements. Density-dependent gravity models are frequently used to explain the connectivity of urban centres for human diseases [5, 18] . There is good evidence from the estimates of 1 in this analysis for England and Wales that the density-dependent model underestimates the total force of infection on remote cities. The 95 per cent credible interval for 1 includes 1, which indicates that the density-independent model formulation cannot be definitively excluded as an explanation for the data. There was limited statistical power to estimate the degree of density dependence in the US context, but use of the England and Wales best-fit model to describe the US data gave a very similar DIC to the best-fit US model. Hence, it is not clear if the difference in the estimated density dependence found between the US and the England and Wales is because of the large differences in the degree of population coverage between the US and England and Wales datasets. Our results using a subset of the England and Wales dataset suggest that the degree of density dependence is a difficult parameter to estimate when coverage is low. The low power on destination city size found in fitting the gravity model to the England and Wales dataset shows that connectivity of a city increases sub-linearly with population increase. When modelling spread of influenza in the US, Viboud et al. [5] found very comparable low values of the population exponents with the infectious city lower than the susceptible. Differing results come from the analysis of measles data in Great Britain with a power coefficient on infectious populations estimated at approximately 1.5 [18] . We found the best-fit model in England and Wales does not include the population size of the infector (origin) city, a result which needs further examination in future work. Differences in our population parameter estimates and those from studies on contemporary populations are likely to differ owing to changes in human mobility patterns since 1918. Our estimate for England and Wales that the infectivity of a city is sub-linearly related to mortality, suggests that the rate of death in a city is not as important to infectivity as the presence or absence of disease. Other studies have used constant infectivity terms for the analysis of human seasonal influenza [21] . However, our estimates do support some level of mortality dependence, suggesting that cities with a very high influenza burden, usually later in the epidemic wave, are more infectious than newly infected cities. In the US dataset, the best-fit model gave constant infectiousness, but again this may be due to a lack of power to estimate such parameters from the US dataset. It may also be caused by non-uniform infection pressure from cities not in the dataset, which could mask an infectivity relationship for the cities that are given. Future possible extensions of this work include relaxing the assumption that all cities were equally susceptible at the start of the autumn wave of the pandemic. The variation in the onset of infection in cities may, in part, be due to the differing susceptibility of each city owing to differing attack rates experienced in the spring -summer wave, or population-level immunity from the 1890 pandemic or seasonal strains. However, the low case fatality of the first wave and age-specificity of infection between waves need to be understood before spatial heterogeneity in susceptibility can be discerned. There are varying reports on the magnitude and mechanism of the effect of infection during the first wave on attack rates in subsequent waves [24] [25] [26] [45] [46] [47] . Further analysis of the datasets considered here may provide an opportunity to disentangle these effects.

Charge-Surrounded Pockets and Electrostatic Interactions with Small Ions Modulate the Activity of Retroviral Fusion Proteins

Refolding of viral class-1 membrane fusion proteins from a native state to a trimer-of-hairpins structure promotes entry of viruses into cells. Here we present the structure of the bovine leukaemia virus transmembrane glycoprotein (TM) and identify a group of asparagine residues at the membrane-distal end of the trimer-of-hairpins that is strikingly conserved among divergent viruses. These asparagines are not essential for surface display of pre-fusogenic envelope. Instead, substitution of these residues dramatically disrupts membrane fusion. Our data indicate that, through electrostatic interactions with a chloride ion, the asparagine residues promote assembly and profoundly stabilize the fusion-active structures that are required for viral envelope-mediated membrane fusion. Moreover, the BLV TM structure also reveals a charge-surrounded hydrophobic pocket on the central coiled coil and interactions with basic residues that cluster around this pocket are critical to membrane fusion and form a target for peptide inhibitors of envelope function. Charge-surrounded pockets and electrostatic interactions with small ions are common among class-1 fusion proteins, suggesting that small molecules that specifically target such motifs should prevent assembly of the trimer-of-hairpins and be of value as therapeutic inhibitors of viral entry.

Bovine Leukemia Virus (BLV) and Human T-Cell Leukemia Virus Type-1 (HTLV-1) are related deltaretroviruses that cause aggressive lymphoproliferative disorders in a small percentage of infected hosts [1, 2, 3, 4, 5, 6] . Like other enveloped viruses, retroviruses must catalyse fusion of the viral and target cell membranes to promote entry of the viral capsid into the target cell. The retroviral class I fusion protein consists of the transmembrane glycoprotein (TM) component of the envelope glycoprotein complex [7] . Envelope is displayed on the surface of the virus or infected cell as a trimer, with three surface glycoprotein (SU) subunits linked by disulphide bonds to a spike of three TM subunits [8] . Experimentally validated models suggest that SU-mediated receptor engagement induces isomerisation of the inter-subunit disulphide bonds and initiates a cascade of conformational changes that activate the fusogenic properties of TM [9, 10] . Membrane fusion is achieved by re-folding of the TM from a native non-fusogenic structure through a rod-like pre-hairpin intermediate, in which the C-and N-terminal segments are embedded in the viral and target cell membranes respectively [7, 8] . The pre-hairpin intermediate then resolves to a trimer-of-hairpins structure, which pulls the membranes together and facilitates lipid mixing and membrane fusion [7, 8, 11, 12] . For several viruses membrane fusion is sensitive to inhibition by peptides that mimic a C-terminal region of the trimer-of-hairpins [13, 14, 15, 16, 17, 18, 19] . The C-terminal fragment of the HTLV-1 trimer-of-hairpins exhibits a short a-helical motif embedded in an extended non-helical peptide structure referred to as the leash and a-helical region (LHR) [20, 21] . The LHR-based mimetics are structurally distinct from the prototypic extensively a-helical peptide inhibitors of human immunodeficiency virus but are reminiscent of the leash regions observed in influenza haemagglutinin [20, 21, 22, 23] . Importantly, amino acid residues that are required for potent inhibitory activity of the HTLV-1 and BLV peptides are not fully resolved in the available HTLV-1 TM structure, yet this information is critical to the development of therapeutically relevant peptide or low-molecular-weight inhibitors of HTLV-1 entry [17, 22] . Moreover, other class I fusion proteins have extended non-helical elements in the C-terminal region of the trimer-of-hairpins and understanding how these elements contribute to the leash in a groove mechanism of fusion protein function will have broad relevance to anti-viral therapies [23, 24] . We therefore sought to examine the structure and function of the BLV TM and to compare this information with data derived from diverse class I fusion proteins and the related HTLV-1 TM structure. Here we show the structure of the post-fusion conformation of the BLV TM ectodomain and demonstrate that coordinated ions and a network of hydrogen-bonded water molecules make critical contributions to the assembly and stability of the trimer-of-hairpins form of the BLV TM and are essential for TM-mediated fusion. Additionally, we resolve a region of the LHR that is critical to the activity of peptide inhibitors of HTLV-1 and BLV entry. We provide evidence that basic residues in a membrane proximal helical element of the LHR interact with charged residues that surround an extended hydrophobic pocket on the coiled coil. This charge-surrounded pocket represents an attractive target for antiviral drugs. Our data indicate that coordinated ions and charge-surrounded hydrophobic pockets are functionally significant leitmotifs of class I fusion proteins. The structure of the BLV TM ectodomain To obtain crystals for structural studies, the N-and C-terminal limits of the coiled coil region of BLV TM were identified using LearnCoil VMF [25] . The TM ectodomain, including the predicted coiled coil and LHR from Ala326 to Trp418, were fused to the C-terminal end of maltose binding protein via a threealanine spacer following the methodology of Center et al [26] . The soluble purified protein was crystallised and the structure solved to a resolution of 2.0 Å (Fig. 1 ). As anticipated, the overall fold of the BLV TM ectodomain is that of a trimer-of-hairpins, with the Nterminal a-helices twisting around each other to produce the central triple-stranded coiled coil that is characteristic of class I fusion proteins. Buried leucine and isoleucine residues within the coiled coil predominantly mediate the interactions between monomers, but strikingly there are two polar layers within the coiled coil that establish interactions with ordered water molecules and a chloride ion respectively (see below). At the base of the coiled coil the peptide backbone undergoes a 180u loop forming the chain reversal region (Fig. 1A, B) , within which is a short helical segment containing the first cysteine of the conserved CX 6 CC motif [9, 10] . The predicted disulphide between Cys384 and Cys391 is reduced in the resolved structure but this does not appear to affect the overall protein fold. In support of this view, preliminary lower resolution structures obtained for crystals formed in the absence of TCEP-HCL reveal an intact disulphide (data not shown). These data indicate that the disulphide is not essential for constraining the chain reversal region in the folded protein, even though it might play a role in the folding process. Therefore, the known defects in membrane fusion associated with substitution of these cysteines are likely due to direct perturbation of the inter-subunit disulphide isomerisation step of the envelopemediated membrane fusion process [9, 10] . After Cys391, the LHR begins with an extended non-helical leash followed by an eightresidue a-helix. This helix is followed by a three-residue linker incorporating a single proline, allowing a sharp kink in the LHR prior to the start of a second a-helix that adopts an orientation almost 90u to the first helical segment of the LHR (Fig. 1A , B, C). A panel of mutants in BLV envelope are processed and expressed identically in vivo Guided by the crystal structure, a panel of mutants in BLV envelope were designed to perturb key features of the trimer-ofhairpins structure. A common difficulty with mutagenesis of viral envelopes, particularly the TM, is that at particular positions even conservative substitutions can dramatically impair the proteolytic processing and cell surface expression of envelope [27, 28] . Therefore, we compared each of the mutants with the parental ''wild-type'' envelope and confirmed appropriate expression, processing and cell surface presentation of the TM mutants. No significant differences in either the expression level of gp51 or the cleavage of the precursor protein, gp72, were observed ( Fig. 2A) . Moreover, flow cytometry analysis confirmed that in each case the mutant envelope was displayed on the cell surface at levels equivalent to wild type (Fig. 2B) . Subsequently, the envelopeexpressing cells were used as effector cells in syncytium formation assays, and the efficiency of each mutant envelope relative to wildtype in catalysing membrane fusion was calculated as the relative fusogenic index. The effect of each mutation on fusogenicity (Fig. 2C) , and the mutant phenotypes are interpreted with reference to the crystal structure below. Solvent molecules are critical for stable assembly of the core coiled-coil The trimerisation of the N-helices appears to be facilitated by presentation of aliphatic residues to the interacting faces of the TM monomers, thereby forming a hydrophobic core down the axis of the central coiled coil. Notably, in the BLV trimer-of-hairpins there are two positions along this interface that harbour polar residues, Thr353 and Asn367 (Fig. 3A, B ). Thr353 is located approximately half way up the N-helix, and it is oriented such that the methyl group of the side chain points toward the centre of the coiled coil and the hydroxyl group faces toward the neighbouring N-helix. In this position, Thr353 participates in a complex network of hydrogen bonding via several buried ordered water molecules. In particular, Thr353 makes a contact through one water molecule with the main-chain of Glu397 and through a separate water molecule to the side-chain of His354, one of the residues which forms the wall of the groove into which the LHR Human T-cell leukaemia virus types-1 (HTLV-1) and bovine leukaemia virus (BLV) are divergent blood borne viruses that cause hematological malignancies in humans and cattle respectively. In common with other enveloped viruses, infection of cells by HTLV-1 and BLV is dependent on the membrane fusion properties of the viral envelope glycoproteins. Here we have solved the crystal structure of the BLV transmembrane glycoprotein, and, through a functional and comparative analysis with HTLV-1, we have identified features that are critical to fusion protein function. In particular, we demonstrate that electrostatic interactions with small ions dramatically stabilize the assembly and fusion-associated forms of the BLV TM, but are not required for the cell surface display of native prefusogenic envelope. Moreover, we show that charged residues that border a deep hydrophobic pocket contribute directly to appropriate folding of fusion-active envelope and are critical to membrane fusion. Importantly, the charged residues that border the pocket are key features that determine the specificity and activity of peptide inhibitors of envelope function. Our study demonstrates that charge-surrounded pockets and electrostatic interactions with small ions are significant leitmotifs of diverse class-1 fusion proteins and that these elements represent ideal targets for novel small-molecule inhibitors of viral entry. binds (Fig. 3A ). In the context of envelope the T353V substitution, which replaces threonine with a non-polar residue of equivalent size, completely abrogates envelope-mediated membrane fusion (Fig. 2C) . Moreover, the introduction of the T353V substitution into the pMBP-BLVhairpin vector yields a recombinant protein that completely fails to trimerise (Fig. 3E) . Interestingly, the data demonstrate that substitutions at Thr353 do not affect expression, processing, or surface expression of native envelope but severely impair the fusogenic activity of envelope by preventing assembly of the trimer-of-hairpins. Our data therefore provides further evidence that the trimeric coiled coil likely forms during the fusion process. A spherical density feature was observed on the central axis of the molecule; comparison of refined temperature factors with those of the surrounding protein atoms suggests an entity more electron-dense than water. Based on the chemical environment, the shape of the density and B factor comparisons we have modeled this feature as a chloride ion. This chloride ion is situated towards the chain reversal end of the coiled coil between a group of three asparagines, one from each monomer of TM (Asn367). These asparagine residues establish electrostatic interactions with the chloride ion by creating a slightly positively charged microenvironment between the N-helices. An alignment of envelope sequences from diverse viral groups (Fig. 3C ) reveals that this asparagine is conserved. Moreover, comparison of the crystal structures for the fusion protein ectodomains of HTLV-1 [20] , MoMLV [29] , Ebola [24, 30] , and Syncytin-1 [31] , a protein derived from a human endogenous retrovirus that plays a critical role in the implantation of the trophoblast into the wall of the uterus [32] , show that the ability to coordinate chloride is retained (Fig. 3D) . To test the importance of the chloride interacting asparagines, we introduced amino acid substitutions of similar size and geometry. Introducing a leucine residue at position 367 of BLV TM renders envelope entirely non-fusogenic. The N367L substitution does not completely prevent trimerisation of the BLV coiled coil in vitro but it significantly destabilises the trimer and a large broad peak corresponding to higher order oligomers and aggregated material is observed by gel filtration chromatography (Fig. 3E ). Interestingly, a substitution whereby Asn367 is replaced by an aspartic acid residue produces a decrease in fusogenicity of almost 90% (Fig. 2C ). The N367D substitution replaces an amide with a carboxylate group in the centre of the coiled coil, which inverts the surface charge of the binding pocket, and therefore would not support interaction with a chloride. In contrast to N367L, the N367D mutation should maintain the hydrogen bonding in and around the chain-reversal region mediated by the carbonyl group that is common to both asparagine and aspartic acid, and therefore the disruption of fusion should be due primarily to the absence of the chloride ion, albeit we cannot exclude the possibility that the close proximity of negatively charged aspartate side chains impairs trimerisation. However, our biological assays demonstrate that the N367D substitution, unlike N367L, retains some membrane fusion activity. In vitro, rather than producing a range of higher order species, the introduction of the N367D mutation into pMBP-BLVhairpin produces a recombinant protein with a major peak at an elution volume consistent with the molecular weight of a tetramer rather than a trimer (Fig. 3E ). Tetramerisation is unlikely to be the reason for the compromised fusogenic activity of the N367D-envelope; nonetheless, the results demonstrate the importance of the chloride-interacting asparagines to the stability of the trimeric coiled coil. Along the length of the LHR multiple contacts are made with the coiled coil and such interactions are required for the activity of inhibitory LHR-mimetic peptides [14, 17, 21, 22] . Amino acid residues that are critical determinants of peptide potency map to the C-terminal region of the LHR [21, 22] , but key residues and the manner in which they interact with the coiled coil are not resolved in the published HTLV-1 TM structure [20] . A detailed view of the BLV LHR bound to the coiled coil and a sequence comparison of the HTLV-1 and BLV LHRs are shown in Fig. 1C and D. Towards the N-terminus of the BLV LHR, conserved residues Leu394 and Ile396 dock into a hydrophobic pocket on the coiled coil, and an array of polar side chains at one side of the pocket interact with the LHR peptide backbone (Fig. 1C) . We have previously demonstrated that interactions with the peptide backbone contribute to the activity of the BLV and HTLV-1 peptide inhibitors [17, 22] . By contrast, the side chain of Arg395 projects out into solvent suggesting that this residue is not essential for interaction of the LHR with the coiled coil or for trimer-ofhairpins formation (Fig. 1C ). In keeping with this view, the R395A substitution has no significant effect on envelope fusogenicity (Fig. 2C) and therefore serves as a useful control for our analysis of envelope structure and function. As the N-helices twist around one another the groove between them accommodates the first conserved a-helical segment of the LHR. Within this a-helix, BLV residues Ile401 and Leu404 extend down and pack into the groove (Fig. 1C) . Significantly, substitution of Leu404 with alanine results in an 80% loss of fusogenic activity relative to wild type (Fig. 2C) , indicating that this interaction contributes substantially to the stability of the LHR/coiled coil interaction. The first a-helix of the BLV LHR ends at Asp406, and between the end of this helix and the beginning of the next is a threeresidue linker, comprised of Leu407, Gln408 and Pro409. Although Leu407 is positioned such that it faces the groove between N-helices, a ridge across this groove prevents Leu407 from docking particularly deeply (Fig. 1C ). This Leucine is not conserved among leukaemia viruses. Instead, there is an Arginine residue at this position of the HTLV-1 LHR that makes contact with the coiled coil. Such differences may account, in part, for the specificity and significant differences in potency that are observed for the BLV and HTLV-1 peptide inhibitors [17] . The conserved proline (BLV Pro409) of the extended non-helical linker induces a sharp kink in the LHR thereby allowing a second a-helix to bind almost directly across the groove, at right angles to the rest of the LHR. Crucially, this second helix is not resolved in the published HTLV-1 TM structure [20] . At the point at which this change of direction occurs, a conserved leucine, Leu410, docks into the start of a deep hydrophobic pocket. Adjacent to Leu410 in this pocket and within the second LHR helix is Val414, which is also conserved in HTLV-1 (Fig. 1C, D) . Substitution of either Leu410 or Val414 with alanine markedly impairs the fusogenic function of envelope (Fig. 2C) . The substitution of L410A is somewhat more detrimental than substitution of Val414, with an activity loss of 75% and 65% respectively. Embedded within the second helical element of the LHR is an arginine (Arg413) that participates in electrostatic interactions with the coiled coil (Fig. 1C ). This residue is conserved in HTLV-1 and is critical to the inhibitory activity of the HTLV-1 LHR mimetic peptide P cr -400 [22] . Intriguingly, Arg413 projects back along the axis of the coiled coil and binds between Leu407 and Leu410, where it forms hydrogen bonds with Gln343 from the same Nhelix and Asp342 from an adjacent N-helix and also donates a hydrogen bond, through the e-nitrogen, to the main chain carbonyl of Glu408 of the LHR (Fig. 1C) . The substitution of Arg413 with alanine dramatically disrupts the fusogenic activity of envelope and reduces envelope function by more than 90% relative to wild type (Fig. 2C) . Moreover, Arg413 cannot be functionally replaced by lysine. The R413K mutant, while more fusogenic than the alanine substituted derivative, is still 85% less effective at catalysing membrane fusion than wild-type envelope (Fig. 2C) . These data indicate that the electrostatic and hydrogen bonding interactions made by Arg413 are essential to the envelope-mediated membrane fusion process. Contrasting effects of two arginine residues on the stability of the trimer-of-hairpins Based on homology modelling, we previously suggested that Arg403 of the LHR projects out from one side of the a-helix and is repelled electrostatically by N-helix residue Arg345 [17] . This prediction is substantiated by the crystal structure presented here (Fig. 1C) . Moreover, the R403A substitution yields an envelope that produces extensive syncytia and is significantly more fusogenic than wild-type envelope (Fig. 2C and Fig. 4C) . Thus, the substitutions R403A and R413A have very different effects on the fusogenicity of BLV envelope, increasing activity by over two-fold and almost completely abolishing activity respectively (Fig. 2C and 4C ). Using a modified thermal aggregation assay [33] , we assessed the relative effects of these two substitutions on the thermostability of the BLV trimer-of-hairpins. Control experiments revealed that after a five minute incubation at 40uC, on average 53% of the wild-type MBP-BLV-hairpin protein was present in high-order aggregates (Fig. 4A, B) . However, when the same heat treatment was applied to the protein bearing the R403A mutation, we found that only 37% of protein aggregated. The difference to wild-type hairpin was significant (P#0.005, t-test) (Fig. 4A, B) . By contrast, the R413A substitution resulted in aggregation of over 70% of total protein, again a significant difference to wild type (P#0.005, t-test) ( Fig. 4A and B) . Notably, even without heat treatment over 50% of the R413A protein was aggregated (Fig. 4A) . Furthermore, these relative differences in the propensity of the recombinant proteins to aggregate were maintained following heat treatment at 50 and 60uC (Fig. 4B) . Hence, the contrasting effects of the arginine substitutions on envelope fusogenicity are directly due to changes in stability of the post-fusion trimer-of-hairpins structure. A prediction based on the increase in thermal stability of the R403A substituted trimer-of-hairpins is that such substitutions should confer reduced sensitivity to LHR-mimetic peptide inhibitors. To test this view we compared the activity of the P BLV -391 antagonist of envelope fusion [17] against wild type or R403Asubstituted envelope. This peptide mimics residues Cys391 to Gln419 of BLV Env. In syncytium interference assays, P BLV -391 inhibited fusion catalysed by wild-type envelope with an IC 50 of 3.1760.09 mM (Fig. 4D ). However, P BLV -391 only inhibited fusion catalysed by R403A envelope with an IC 50 estimated to be .9 mM, and even at peptide concentrations up to 15 mM inhibition of syncytium formation was markedly damped indicating that both the potency and efficacy of the peptide was reduced against R403A substituted envelope (Fig. 4D) . Moreover, though more active against native envelope, a reciprocally substituted peptide antagonist, P BLV -R403A, failed to exhibit full potency against the R403 envelope derivative; P BLV -R403A inhibited wild-type envelope catalysed membrane fusion with an IC 50 of 1.360.1 mM, but inhibited the R403A envelope with an IC 50 of .6 mM (Fig. 4E) . In both experiments, C34 (a peptide mimetic of HIV-1 gp41 residues Gly627 to Leu661) was used as a negative control. Thus, improving the thermal stability of the trimer-of-hairpins form of a class-1 fusion protein significantly reduces the sensitivity of envelope to peptide inhibitors targeted to the coiled coil. The structure of the C-terminal segment of the BLV LHR and the accumulated data suggest a key role for the conserved arginine (Arg413) in the mechanism of envelope-mediated membrane fusion. Moreover, our recent data indicate that the equivalent arginine residue, Arg422, of the HTLV-1 LHR-mimetic peptide is critical to inhibitory activity [22] . Notably, for the HTLV-1 LHRbased inhibitor residues equivalent to Leu413, Arg416 and Leu419 are also of importance to the biological activity of the peptide [21, 22] . We therefore sought to establish whether or not the conserved arginine in the HTLV-1 LHR docks with the coiled coil in a similar manner to that of BLV and if this residue is important to membrane fusion mediated by the HTLV-1 TM. We generated a small panel of mutants in HTLV-1 envelope, substituting Arg416 of the LHR with alanine and lysine (R416A and R416K), and making similar substitutions for Arg422 (R422A and R422K). Using a monoclonal antibody recognising HTLV-1 SU, we confirmed by Western blotting and flow cytometry that all four mutant envelopes were expressed and post-translationally processed in a manner identical to wild type (Fig. 5A, B) . In syncytium formation assays, the individual alanine substitutions of Arg416 and Arg422 resulted in envelopes that are 91% and 97% less fusogenic than wild-type envelope respectively (Fig. 5C) . Significantly, the lysine substitution of Arg416 produced an envelope that was not significantly less fusogenic than wild type (P.0.05, t-test) (Fig. 5C ). However, Arg422 could not be functionally replaced with lysine, the R422K mutation resulted in an 88% reduction in fusogenic activity. Taken together with the data from the BLV R413K mutant, this suggests that HTLV-1 R422 adopts a near-identical conformation to BLV R413 at the Cterminal segment of the LHR. The likely conservation of position and orientation of HTLV-1 R422 in the trimer-of-hairpins allowed us to construct a model for the binding of the C-terminal segment of the HTLV-1 LHR to the coiled coil (Fig. 5D) . The model suggests that R422 could bind to a negatively charged ridge that is orientated across the groove between N-helices and that lies between the binding pockets for Leu413 and Leu419 and that R422 docks adjacent to R416. As such, four critical interactions between LHR side chains and the coiled coil are contained within a binding hotspot of ,16 Å x ,8 Å and disruption of these interactions profoundly impairs envelope-mediated membrane fusion. There is greater sequence divergence between BLV and HTLV-1 than, for example, between SIV and HIV-1 [34, 35] , nonetheless the crystal structures of the BLV and HTLV-1 trimerof-hairpins are similar. The incorporation of ordered water molecules and ions to facilitate trimerisation of the coiled coil and assembly of the trimer-of-hairpins is seen in several viral fusion proteins [20, 29, 30, 36] . The crystal structure of SIV gp41 implicates multiple water molecules in the folding of the trimerof-hairpins [37] . However, the number of water molecules concentrated in one region of the BLV coiled coil is unusual. The interaction of Thr353 with an array of water molecules appears to play a critical role in assembly of the coiled coil, as substitution of T353 eliminates both in vitro trimerisation and in vivo fusogenicity. Notably, for HTLV-1 the threonine residue is replaced by an asparagine, which by virtue of its larger side chain directly hydrogen bonds with the adjacent N-helix. By contrast, Thr353 of BLV makes this contact through water-mediated hydrogen bonds. However, the contribution of polar interactions to coiled coil assembly is maintained at this location in HTLV-1 by interaction of a chloride ion on the symmetry axis of the coiled coil. Moreover, for both viruses three conserved asparagine Fusogenicity of HTLV-1 C-terminal LHR arginine mutants. The fusogenic index was calculated as described (mean 6SD from triplicate assays). (D) A ''hotspot'' for binding of HTLV-1 fusion inhibitors based on the previously published HTLV-1 structure and modelled incorporating the data from this study. Four key residues (Leu413, Arg416, Leu419 and Arg422, shown as yellow sticks) interact with the HTLV-1 coiled coil (grey space-filling model, region interacting with LHR residues shown in red) within a compact binding site. The contact made by Arg422 is critical to the binding and functional activity of HTLV-1 peptide inhibitors. doi:10.1371/journal.ppat.1001268.g005 residues located towards the membrane distal end of the coiled coil interact with a chloride ion. The properties of the N367L substitution and the failure of N367D to rescue envelope function indicate that the observed chloride-mediated interactions are essential for stable assembly of the coiled coil and for TMmediated membrane fusion. Significantly, the asparagine (Asn367) and adjacent arginine (Arg368) residues of BLV are conserved between BLV, HTLV-1 and a number of disparate viral fusion proteins, and the ability of the asparagines to interact with a chloride ion is maintained. Critically, the chloride ion is present within the post-fusion conformation of Ebola GP2 (30), but is conspicuously absent in a pre-fusion structure resolved by Lee et al [38] wherein the conserved asparagines adopt a more open arrangement and the coiled coil is only partially trimerised. We suggest that Asn367, as a direct consequence of conformational changes in SU resultant from receptor binding, and in concert with the equivalent asparagines from neighbouring monomers, mediates electrostatic interactions with a chloride ion, which brings the N-helices together, thus driving the trimerisation of the coiled coil to yield the pre-hairpin intermediate in an event pivotal to successful fusion. In addition, our results indicate that trimerisation of the N-terminus of TM is not a requirement for Env maturation and surface display of trimeric pre-fusogenic envelope, offering a novel insight into the conformation of retroviral Env spikes prior to activation. Taken together, the data suggest that while the ''knobs-into-holes'' interactions of aliphatic residues at the centre of the coiled-coil regions of class 1 viral fusion proteins are important for trimerisation of the fusion-active coiled coil the polar layers among these hydrophobic interactions are critical. This information should be considered when designing trimeric immunogens based on the coiled coil of TM. Moreover, the structure in and around the chloride-interacting region has been conserved in diverse viral fusion proteins and may be susceptible to coiled coil destabilising drugs. It is interesting to note that removal of a steric clash between R403 and the BLV coiled coil yields an envelope that is more fusogenic than wild type. For HTLV-1 an isoleucine, at a position equivalent to residue Arg403 in the BLV LHR, also appears to make a steric clash with the coiled coil and an alanine substitution of this residue in the context of an LHR-based peptide inhibitor yields a peptide with improved coiled coil-binding properties and increased inhibitory activity [21] . It is therefore clear that viral envelope has not evolved to optimise LHR binding to the coiled coil or to maximise fusogenicity. One plausible explanation is that membrane fusion needs to be kept in check. An overly fusogenic envelope may induce rampant membrane fusion and syncytium formation leading to cell death by apoptosis and thereby provide additional stimuli for a robust anti-viral immune response. Clearly, sub-optimal LHR/coiled-coil interactions have been maintained during viral evolution and suggest that the development of compounds that bind to the coiled coil with higher affinity than the LHR may be an achievable anti-viral strategy. The most notable difference between the HTLV-1 and BLV TM structures is the resolution of a second helical motif within the LHR of the trimer-of-hairpins. Residues that form the start of the second a-helix are conserved in HTLV-1 and are critical to envelope-catalysed membrane fusion and to the inhibitory activity of LHR-mimetic peptides [21, 22] . Just three residues from the end of our structure is the first tryptophan of a conserved tryptophanrich membrane-proximal region (MPR). The MPR is believed to interact with the opposing membranes during the fusion process and substitution of the aromatic residues severely compromises, but does not abolish, membrane fusion [39] . The conformation of the MPR has received considerable attention because broadly neutralising anti-HIV-1 antibodies bind epitopes contained within the MPR [40, 41] . It is likely that the HIV MPR adopts a helical structure when interacting with membranes and the coiled coil during fusion [40, 42] . Examination of the BLV envelope sequence and TM structure suggests that the second helical segment of the LHR is likely to continue and that in the fusion-activated state the tryptophan-rich MPR is also helical and inserted into the membrane at an oblique angle. For the leukaemia viruses, the conserved arginine (Arg413 in BLV), assists in defining the orientation of the MPR, as the interactions of Arg413 with the coiled coil and the LHR backbone propagate a sharp kink in the LHR towards the membrane-proximal end of the coiled coil. A lysine substitution of Arg413 retains only ,10% of the fusogenic activity of wild-type envelope lending further support to this view. The BLV crystal structure and our analysis of HTLV-1 and BLV envelope-mediated membrane fusion reveal an important role for electrostatic interactions in binding of the LHR to the grooves of the coiled coil. For both BLV and HTLV-1 a conserved basic residue in the second helical segment of the LHR interacts with the charged rim of a deep pocket lined by non-polar residues. For HTLV-1 the contribution of charge is enhanced by the interaction of an additional basic residue (Arg416) with the opposite rim of the pocket. Thus, while the insertion of non-polar residues into hydrophobic pockets likely drives the docking of the LHR with the coiled coil, we suggest that it is the electrostatic attraction and hydrogen bonding of charged side chains around these hydrophobic interactions that serves to ''cement'' the LHR in place and thereby establish the precise geometry and affinity of this interaction. In keeping with this view, substitution of these basic residues ablates envelope-mediated membrane fusion and dramatically impairs the inhibitory and coiled-coil-binding properties of LHR-based peptides [22] . Moreover, only one of these important basic residues is conserved in BLV compared to HTLV-1 and it is notable that the synthetic BLV LHR-mimetic is approximately 10-fold less potent than the corresponding inhibitor of HTLV-1 [17] . Examination of the electrostatic surfaces of BLV, HTLV, HIV and MoMLV coiled coils reveals, in each case, the presence of a pocket surrounded by charged residues, though the polarity of the charge and positioning along the groove is variable (Fig. 6 ). The importance of the pocket and surrounding charge to the assembly of the trimer-of-hairpins and the potency of C-helixbased mimetic peptides has been demonstrated for HIV [43, 44] , and taken with our findings for BLV and HTLV-1 suggest that such charged regions are critical to retroviral envelope-mediated membrane fusion in general. We suggest that structure-assisted design of small molecules to target these charge-surrounded pockets is a viable objective for anti-retroviral therapy. We expect that the data produced in this study will be pivotal to the development of more drug-like inhibitors of HTLV-1 envelope catalysed membrane fusion, and thereby provide therapeutic antiviral agents for adult T-cell leukaemia and HTLV-associated myelopathy/tropical spastic paraparesis, diseases for which there are currently no effective treatments [45, 46] . The plasmids pCMV-BLVenv-RRE and pRSV-Rev have been described [17, 47] . To construct pCMV-HTLVenv-RRE the BLV env sequences in pCMV-BLVenv-RRE were replaced by a HTLV-1 env coding region amplified by PCR from pHTE-1 [48] . To construct the plasmid pMAL-gp30hairpin (MBP-BLV-hairpin), a modified fragment of the MBP open reading frame from pMAL-c2 (New England Biolabs) with a 59 BglII site and a 39 PvuII site replacing a SacI site was PCR amplified. A fragment of BLV env encoding amino acid residues 326 to 418 was PCR amplified with a 59 PvuII site and a 39 PstI site. The PCR fragments were ligated into a pMAL-c2 backbone digested with BglII and PstI. The Quikchange mutagenesis kit (Stratagene) was used to introduce point mutations into pCMV-BLVenv-RRE and pMAL-gp30hairpin following the manufacturer's instructions. Expression of the MBP fusion protein was carried out as previously described [14, 49] . Briefly, Escherichia coli BL21(DE3) cells transformed with the MBP-BLV-hairpin vector were grown at 37uC with vigorous shaking in LB supplemented with 10 mM glucose and 100 mgml 21 ampicillin until an optical density at 600 nm of ,0.6 was reached. Protein expression was induced with Isopropylthio-b-D-galactoside (IPTG) at a final concentration of 0.5 mM for 4 h at 37uC. Cells were harvested by centrifugation and the pellet was resuspended in column buffer (20 mM Tris-HCl [pH 7.5], 200 mM NaCl, 1 mM EDTA) supplemented with protease inhibitors (1 mM phenylmethylsulfonyl fluoride and 1 mgml 21 aprotinin) and frozen O/N at 220uC. The cell suspension was thawed and the cells lysed by sonication. Cell debris was pelleted by centrifugation at 9,0006g for 30 min. The crude lysate was diluted 1:5 in column buffer and loaded onto an amylose column pre-equilibrated with column buffer. The column was then washed with 15 column volumes of column buffer, and the bound protein was eluted with column buffer supplemented with 10 mM maltose. The trimeric recombinant MBP-BLV-hairpin protein was purified by Superdex 75 gel filtration, and the required fractions pooled and concentrated to 12 mg ml 21 . Tris [2-carboxyethyl]phosphine hydrochloride (TCEP-HCl) was then added to a final concentration of 5 mM. Protein concentration was estimated by absorbance measured at 280 nm. Crystallization conditions were identified using the sitting drop vapour diffusion technique, with crystal screens from Hampton Research and Emerald Biostructures. Optimal crystallization was achieved by streak seeding with a reservoir solution composed of 24.5% (v/v) isopropanol, 13.5% (v/v) PEG-4000 and 0.1 M sodium citrate pH 5.6, and optimal diffraction was achieved using a cryoprotectant composed of mother liquor supplemented with 20% (v/v) glycerol. Diffraction data was collected at beamline BM14 at the European Synchrotron Radiation Facility (ESRF) in Grenoble. The structure was solved by molecular replacement using a truncated model of MBP-HTLV Hairpin (PDB ID 1MG1). Refinement proceeded through cycles of model building using Coot and O [50, 51] and refinement using REFMAC5 [52] . Data and refinement statistics are given in Table 1 Peptides Peptides were synthesised using standard solid-phase Fmoc chemistry and unless stated otherwise have acetylated N-termini and amidated C-termini. The peptides were purified by reversephase high-pressure liquid chromatography and verified for purity by MALDI-TOF mass spectrometry. All peptides were dissolved in dimethyl sulfoxide (DMSO), the concentration of peptide stock solutions was confirmed by absorbance at 280 nm in 6M guanidine hydrochloride and peptides were used at the final concentrations indicated. HeLa cells were maintained in Dulbecco's modified Eagle medium supplemented with 10% foetal bovine serum (FBS). For syncytium formation assays, HeLa cells were transfected with equal quantities of pCMV-BLVenv-RRE (or empty pcDNA 3.1 for mock samples) and pRSV-Rev using the GeneJuice transfection reagent (Novagen). After 24 h, 3610 5 transfected cells were added to 7.0610 5 non-transfected HeLa target cells in 6-well dishes (Nunc), and cocultured for 16 hrs, in the presence of peptides where appropriate. The cells were washed with phosphatebuffered saline pH 7.4 (PBS), fixed using 3% paraformaldehyde in PBS, and stained with Giemsa. Assays were performed in triplicate. To calculate fusogenic indices, the fraction of nuclei contained within syncytia in a 20x light microscope field was expressed as a percentage of the total number of nuclei within the field. Western blotting was carried out using standard methods; bmercaptoethanol was used in sample preparation. BLV and HTLV-1 envelope was detected using supernatant from murine hybridoma cell lines expressing monoclonal antibodies raised against recombinant antigen derived from BLV or HTLV-1 envelope, and anti-mouse horseradish peroxidase conjugated secondary antibody at a dilution of 1:10,000 in PBS containing 5% (w/v) marvel and 0.025% Triton X-100. To detect surface-expressed protein, cells transfected with the appropriate envelope expression vector were detached from culture flasks using PBS +2 mM EDTA, and washed twice with PBS. 5.0610 5 cells were incubated with agitation for 1 hr at room temperature in 1 ml DMEM +10% FBS containing 15 mgml 21 immunoglobulin purified from the BLV anti-Env antibodyexpressing hybridoma supernatant, or 150 ml supernatant from the HTLV anti-Env antibody expressing hybridoma. Cells were washed twice with PBS and incubated in the dark at room temperature for 45 mins in DMEM containing anti-mouse FITC (Sigma) at 1/1,000 dilution. Cells were washed once with PBS and once with PBS +0.1% sodium azide before fixing with PBS +0.5% paraformaldehyde. Bound fluorescence was detected using a FACScan flow cytometer (Becton Dickinson). The oligomerisation states of MBP-BLV-hairpin and mutant derivatives were examined by gel filtration using a Superdex 200 column equilibrated with MBP elution buffer. Fractions containing the trimeric species were retained. To assess the thermostability of the trimers, samples from these fractions were re-run over the column either without heat treatment or with heat treatment at the temperature specified for 5 min, followed by cooling for 2 min on ice. The areas under the peaks observed were calculated, and the area of the peak corresponding to the aggregate was expressed as a percentage of the total area under both this peak and the peak corresponding to the trimer.

Tracheostomy and mechanical ventilation weaning in children affected by respiratory virus according to a weaning protocol in a pediatric intensive care unit in Argentina: an observational restrospective trial

We describe difficult weaning after prolonged mechanical ventilation in three tracheostomized children affected by respiratory virus infection. Although the spontaneous breathing trials were successful, the patients failed all extubations. Therefore a tracheostomy was performed and the weaning plan was begun. The strategy for weaning was the decrease of ventilation support combining pressure control ventilation (PCV) with increasing periods of continuous positive airway pressure + pressure support ventilation (CPAP + PSV) and then CPAP + PSV with increasing intervals of T-piece. They presented acute respiratory distress syndrome on admission with high requirements of mechanical ventilation (MV). Intervening factors in the capabilities and loads of the respiratory system were considered and optimized. The average MV time was 69 days and weaning time 31 days. We report satisfactory results within the context of a directed weaning protocol.

During winter in 2009, Argentina as well as several other countries in the world, was affected by Influenza A H1N1 pandemic [1] . Our Pediatric Intensive Care Unit (PICU) was the referral center for this pandemic in the province of Buenos Aires. Between June 1 and July 31 2009 113 patients were admitted in our PICU, 79 of them with respiratory pathology. Twenty presented positive PCR (polymerase chain reaction) by Influenza A H1N1and 22 presented positive IFI (indirect immunofluorescence) by RSV (respiratory syncitial virus). Out of the 69 patients who survived 3 presented difficult weaning and needed tracheostomy. This paper describes the clinical behavior and presentation of these 3 patients who were admitted into our PICU in that period and who had a prolonged stay due to respiratory complications after viral infection (Table 1) . We report the results obtained according to a directed weaning protocol. Below, we describe each patient's evolution before tracheostomy was performed and before the weaning plan was begun. A six-month-old male patient was admitted in our PICU after intubation owing to respiratory insufficiency due to bilateral pneumonia. Through nasopharyngeal secretions a positive PCR by Influenza A H1N1 virus was found. He was a 28-week preterm infant with a birth weight of 900 grams. He had bronchopulmonary dysplasia (BPD) because of mechanical ventilation during 2 months in the perinatal age. He also presented some malformations compatible with a genetic syndrome which was still being studied before admission (retromicrognathia, abdominal angioma in the medial line, low implantation of the ears, among others). On admission the patient presented severe acute respiratory distress syndrome (ARDS) for which he required high MV parameters ( Figure 1 ). He presented serious refractory hypoxemia and remained 36 days on MV. Using a low tidal volume ventilation protocol we were not able to achieve an adequate minimal oxygenation. So, as we do not count on high frequency oscillatory ventilation (HFOV) because of certain socioeconomic limitations within our hospital, we resorted to standard modalities, reaching mean airway pressure (Paw) of 15. 2 Extubation failed by respiratory insufficiency in 2 opportunities requiring reintubation in the first 24 hours. In both opportunities the patient had passed a spontaneous breathing trial (SBT). A respiratory endoscopy was performed which showed an inflammatory pathology of the upper airway (UA). These findings together with presenting prolonged MV (36 days) led to the decision of carrying out a tracheostomy and of beginning a weaning plan. A six-month old male patient born in due term with adequate weight, previously healthy. He was referred to our PICU for presenting bronchiolitis by co-infection with RSV and A H1N1 virus. He presented severe ARDS and hypoxemia and required high MV parameters remaining ventilated for 51 days (Figure 2 ). Three SBTs were performed during his evolution with 3 elective extubations, failing in all cases. He also presented failure during disconnection from positive pressure with non-invasive positive pressure ventilation (NPPV) due to respiratory insufficiency. A respiratory endoscopy was performed which showed an important edema of the larynx crown and posterior synechia due to which tracheostomy was performed on the 44 th day of hospitalization. Then, the weaning plan was started. A female patient of 1 month and 20 days of age, born in due term with adequate weight, with no perinatologic history. She was admitted with low acute respiratory failure following pneumonia in the right superior lobe. Co-infection with RSV and Influenza A H1N1 virus was found. During her prolonged stay in PICU (111 days) she required MV due to ARDS with severe hypoxemia ( Figure 3 ). In five opportunities she passed an SBT but in all cases re-intubation was required within 48 hs owing to low respiratory involvement. Also, repeated presence of condensation in the right superior lobe was observed. Due to everything mentioned above, a respiratory endoscopy was performed, reporting a right anomalous tracheal bronchus and mild subglottic edema. Given the fact that the patient had several respiratory infections which led to an obstruction of the segment corresponding to the anomalous bronchus, an exeresis of the right anomalous tracheal bronchus and a segmentectomy of the right superior lobe were performed. After 51 days on MV a tracheostomy was performed and respiratory weaning was started. Disconnection of positive pressure was achieved after 98 days. The 3 patients described coincided in positive PCR for Influenza A virus (2 of them with co-infection with RSV), prolonged MV, ARDS with severe hypoxemia and successful SBTs with failed extubations during the course of 48 hours. Those patients presented upper airway pathology by pathologic endoscopy and lower airway pathology (post viral sequelae). These two factors affected their weaning from MV. In our PICU upon meeting the following criteria we initiate SBT based on current consensus: resolution of the basal cause of respiratory failure, peak inspiratory pressure (PIP) < 25 cm H 2 O, positive end-expiratory pressure (PEEP) < 5 cm H 2 O, tidal volume 6-8 ml/kg, FIo 2 < 0.40, intermittent and decreasing sedation, patient alert, hemoglobin > 10 g/dl, and decreasing vasoactives [2] [3] [4] . In our PICU SBTs are performed with a T-piece and supervised by respiratory therapists. The variables to be considered are registered on charts that we have developed for that purpose and which are filled in at the start of the test and every 15 minutes during the two hours the test lasts. It is PICU's policy to perform respiratory endoscopies in those children who present two or more failed extubations. In all three cases, because of presenting a pathologic endoscopy the decision of performing a tracheostomy was discussed and accepted by the patients' parents. All of them were trained in the handling, hygiene and comfort of the tracheostomy. For the patients we presented, a strategy was designed with the objective of achieving disconnection from positive pressure by means of an orderly and progressive protocol. In the first place we measured the maximum tolerated time on CPAP in each patient. We have defined maximum time as the longest period the patient is able to remain on CPAP without altering his vital signs in +/-20%. This determination was carried out comparing the The mean value of the maximum tolerated time on CPAP was 4 hours (3-6 hours). Out of the maximum time reached, between 50 and 75% was used for the respiratory training of the patient. The objective was to train the patient without exposing him to fatigue and muscular tiredness due to the decrease in his respiratory capacitance. This allowed to program rest periods which became less frequent as training progressed. This plan was carried out during day time while during the night, sleeping time was respected supplying MV in PCV [5] . A gradual diminution of the ventilatory support, combining PCV with increasing periods on CPAP + PSV, was performed. Once the patient managed to remain 12 hours on CPAP + PSV, this modality was alternated with T-piece whose duration was progressively incremented [6] (Figure 4) . The objectives of the programed tasks for the day were described in a chart in which the physiological parameters found during the training period were constantly recorded: respiratory rate, heart rate, blood pressure, oxygen saturation, maximal inspiratory pressure (PIMAX), maximal expiratory pressure (PEMAX), ventilatory mechanics, use of accessory muscles, comfort index and rapid shallow breathing index (respiratory rate/tidal volume per kg of weight) [2] . For the measurement of PIMAX and PEMAX a Marshalltown ™ manual vacuum meter was connected to the endotracheal tube. We selected the most negative and positive value of 3 efforts during spontaneous breathing over a period of 20 seconds occluding the airway with a one-way valve [2, 3] . In case the patient presented signs of muscular fatigue which kept us from accomplishing the programmed task we returned to the work plan of the previous day (when he tolerated the exercise adequately) and he was reassessed the following morning with the same parameters. With this muscular training, nutritional support and the supply of vitamins and minerals, an improvement of the respiratory mechanics was observed, focusing on the measured parameters managing to effectively disconnect patients from positive pressure [7] . This objective was considered as accomplished when the patient remained on a T-piece over a period of 48 hs. The average MV time was 69 days. The average weaning time was 31. Finally, the 3 patients were referred to a general pediatric ward, two of them with supplementary oxygen. Children with severe respiratory disease due respiratory virus presented a high percentage of ARDS on admittance to PICU with severe hypoxemia, requiring high MV parameters. Among the surviving patients, a torpid evolution with long periods on MV and difficulties for its disconnection were observed [8] . Unfortunately, up to date there aren't available protocols based on evidence to guide a difficult weaning process in tracheostomized children. The bibliography on adults is ample, but not always applicable to our infant population due to anatomic, functional and cognitiveaffective reasons [9] . Due to the functional anatomic differences between children's and adults' air way, the ideal time to indicate tracheostomy after a prolonged period on MV is still being discussed [10] . This is why, in the bibliography, more prolonged periods on MV through an endotracheal tube (ETT) are found in children than in adults. Tracheostomy facilitates weaning with respect to ETT because it produces: dead space reduction, less airway resistance, decreased breathing work, better secretion removal with suctioning, less likelihood of tube obstruction, improved patient comfort, less need for sedation and better glottic function with less risk of aspiration [5, 6, 11] . The term weaning refers to the transition from ventilatory support to completely spontaneous breathing, during which time the patient assumes the responsibility for effective gas exchange while positive pressure support is withdrawn [12] . Weaning consumes between 40 and 60% of mechanical ventilation time. About 25% of ventilated patients present difficulties in the disconnection from the ventilator, being unable to be disconnected by means of SBT, requiring a progressive procedure which can take several days, weeks or even months [13] . These patients would benefit from a directed weaning protocol. Ventilatory failure may be generated by a deterioration of the neuromuscular capacity or by increasing the loads of the respiratory system. The decrease of the capacity may be owed to a depression of the respiratory center (whether by sedation, alkalosis or encephalic lesion), neuropathies (polyneuropathy of the critical patient), muscular disorders (malnutrition, hyperinsuflation, electrolyte disorders and the use of muscle relaxants and corticoids) and chest wall abnormality (unstable chest, post-operative pain). The increase in the load may be due to the high ventilation requirements (sepsis, anxiety and pain), resistive loads (bronchospasm, secretions), elastic loads (auto PEEP, decrease in compliance) and related to the ventilator and the endotracheal tube (valves, small tubes) [6, 12] . All these factors were considered and optimized in our patients before proceeding to their extubation. Among the elements we count on, we opted for the use of the described combination of ventilatory modalities as part of an orderly and progressive protocol. This permitted the favoring of the muscular training, the improvement of the patient-ventilator interaction and therefore the enhancement of the respiratory work. The use of PCV enabled us to ensure total ventilation during night rest, eliminating energetic waste used in training which allowed the recovery. PSV allowed the patient to control his own frequency and therefore the circulating volume and minute volume. In this manner the PSV generated an adequate respiratory work and oxygen consumption of the respiratory muscles, diminishing fatigue and favoring re-training [12] . The combination of this modality with CPAP allowed to increase the functional residual capacity thus achieving a better oxygenation. This is owing to the increase in the number of effective alveolar units in gas exchange (alveolar recruitment). CPAP is able to offset the abnormal closing of the airway and the effect of the auto PEEP, noticing the beneficial effects when the level of inspiratory efforts to be developed during spontaneous breathing diminishes. The use of a T-piece allowed for greater independence from the ventilator with less resistance in the airway since the opening of an inspiratory valve is not necessary. In addition it permitted a clinical evaluation of the diaphragmatic function which was not possible if the patient received any kind of inspiratory support [14] . The three described patients had pulmonary sequelae for infection caused by respiratory virus with a pathology of the airway evidenced by respiratory endoscopy, and in spite of all this, their effective disconnection from positive pressure was achieved. This was possible combining tracheostomy with an orderly, monitored and progressive muscular training plan (directed weaning protocol). Even though more studies are necessary to develop weaning protocols based on evidence, we believe that this type of approach may be useful in pediatric patients with difficult weaning due to other etiologies. Written informed consent was obtained from the parents of the patient for publication of this report and accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

The activity of the HIV-1 IRES is stimulated by oxidative stress and controlled by a negative regulatory element

Initiation of translation of the full-length messenger RNA of HIV-1, which generates the viral structural proteins and enzymes, is cap-dependent but can also use an internal ribosome entry site (IRES) located in the 5′ untranslated region. Our aim was to define, through a mutational analysis, regions of HIV-1 IRES that are important for its activity. A dual-luciferase reporter construct where the Renilla luciferase (Rluc) translation is cap-dependent while the firefly luciferase (Fluc) translation depends on HIV-1 IRES was used. The Fluc/Rluc ratio was measured in lysates of Jurkat T cells transfected with the dual-luciferase plasmid bearing either the wild-type or a mutated IRES. Deletions or mutations in three regions decreased the IRES activity but deletion or mutations of a stem-loop preceding the primer binding site increased the IRES activity. The wild-type IRES activity, but not that of an IRES with a mutated stem-loop, was increased when cells were treated with agents that induce oxidative stress. Such stress is known to be caused by HIV-1 infection and we propose that this stem-loop is involved in a switch that stimulates the IRES activity in cells infected with HIV-1, supporting the suggestion that the IRES activity is up-regulated in the course of HIV-1 replication cycle.

coding sequences. This PCR introduced three stop codons to terminate the Rluc coding sequence. The rest of the 5'UTR was amplified directly from a proviral molecular clone of HIV-1 group M subtype B (pLAI) (4) . The initiation codon for Fluc expression is located within the 5'UTR IRES and the context of the AUG encompassing 30 nt from the Gag sequence was included. A peptide linker (GGGGSGGGGS) was inserted by PCR immediately upstream of the Fluc coding sequence. The first half of the linker was inserted using PCR amplification on pLAI with the 5'UTR-AflII(+) and 5'UTR-BamHI(-) primers and the second half of the linker with the 5'UTR(+) and Linker-BamHI(-) primers. The linker was cloned in the AflII and BamHI restriction sites of pDual-HIV(-1)ΔAflII-TAR to generate pDual-IRES-HIV. This last step removed the frameshift region originally present in pDual-HIV(-1). Finally, we cut the fragment containing the Rluc coding sequence, the 5'UTR region of HIV-1 RNA and the Fluc coding sequence, using PmeI and ApaI restriction sites. This fragment was inserted into pcDNA5FRT (Invitrogen) previously linearized with SciI and ApaI to produce pFRT-IRES-HIV. Prior to this last cloning step, the KpnI and BamHI restriction sites from pcDNA5FRT were eliminated to facilitate subsequent cloning of mutant IRESes. To this end, an oligonucleotide cassette formed by the K7ΔKpnIΔBamHI(+) and K7ΔKpnIΔBamHI(-) primers was inserted in the HindIII and EcoRV restriction sites of pcDNA5FRT. The different mutants of HIV-1 IRES were made by PCR amplification with four primers. The external primers were Ext-KpnI(+) et Ext-BamHI(-). The details for all the primers used can be found in Supplementary Table 1 . The resulting PCR products were cloned in the KpnI and BamHI restriction sites of pFRT-dual-IRES-HIV. ΔAflII-SpeI(+) 5'GACCCGGGGTACCAAGCTTGAGTTTAAACGCTAGCCAGC'3

Further Characterisation of the Translational Termination-Reinitiation Signal of the Influenza B Virus Segment 7 RNA

Termination-dependent reinitiation is used to co-ordinately regulate expression of the M1 and BM2 open-reading frames (ORFs) of the dicistronic influenza B segment 7 RNA. The start codon of the BM2 ORF overlaps the stop codon of the M1 ORF in the pentanucleotide UAA UG and ∼10% of ribosomes terminating at the M1 stop codon reinitiate translation at the overlapping AUG. BM2 synthesis requires the presence of, and translation through, 45 nt of RNA immediately upstream of the UAA UG, known as the ‘termination upstream ribosome binding site’ (TURBS). This region may tether ribosomal 40S subunits to the mRNA following termination and a short region of the TURBS, motif 1, with complementarity to helix 26 of 18S rRNA has been implicated in this process. Here, we provide further evidence for a direct interaction between mRNA and rRNA using antisense oligonucleotide targeting and functional analysis in yeast cells. The TURBS also binds initiation factor eIF3 and we show here that this protein stimulates reinitiation from both wild-type and defective TURBS when added exogenously, perhaps by stabilising ribosome-mRNA interactions. Further, we show that the position of the TURBS with respect to the UAA UG overlap is crucial, and that termination too far downstream of the 18S complementary sequence inhibits the process, probably due to reduced 40S tethering. However, in reporter mRNAs where the restart codon alone is moved downstream, termination-reinitiation is inhibited but not abolished, thus the site of reinitiation is somewhat flexible. Reinitiation on distant AUGs is not inhibited in eIF4G-depleted RRL, suggesting that the tethered 40S subunit can move some distance without a requirement for linear scanning.

Eukaryotic viruses have evolved a variety of translational control strategies to facilitate expression of downstream open reading frames (ORFs) on polycistronic mRNAs and examples have been described at all three steps of protein synthesisinitiation, elongation and termination [1] . These include leaky scanning of 40S ribosomal subunits past the most 59 AUG [2] , the de novo recruitment of ribosomes to intercistronic internal ribosomal entry sites (IRESs; [3] ); programmed ribosomal frameshifting [4, 5] and the circumvention of normal termination by programmed stop codon readthrough [6, 7] . Generally, these processes allow the expression of two (or more) proteins from a single mRNA and may also permit a level of control over their relative quantities. Another way of accessing a downstream ORF in viral mRNAs is by termination-dependent reinitiation of translation (termination-reinitiation), a phenomenon first described in the expression of the influenza B virus BM2 protein [8] . The dicistronic mRNA that is derived from genomic segment 7 of this virus has two ORFs encoding matrix protein 1 (M1) and BM2, with the termination codon of M1 in close proximity to the start codon of the BM2 ORF (UAAUG; stop codon of M1 in bold, start codon of BM2 underlined) [8] [9] [10] . Following translation of M1, some 10-20% of ribosomes terminating at the M1 stop codon go on to reinitiate translation at the immediately adjacent BM2 start codon [8, 11] . This capacity to reinitiate protein synthesis following translation of a long upstream ORF was unexpected. During the elongation phase, initiation factors are likely to be rapidly lost, thus reinitiation of translation following termination was believed to be restricted to cases where the upstream ORF (uORF) is very short [12] [13] [14] . Our knowledge of the mRNA signals that allow efficient reinitiation following translation of a long upstream ORF has largely been obtained from studies of caliciviruses, namely in the expression of the VP2 protein of feline calicivirus (FCV; [15] [16] [17] ) and the VP10 protein of rabbit haemorrhagic disease virus (RHDV; [18, 19] ). Here, expression of the downstream ORF by termination-reinitiation requires a stretch of mRNA (between 69 and 87 nt in length) upstream of the stop codon of the first ORF (termed the termination upstream ribosome binding site or TURBS) and the close proximity of the stop and start codons of the two ORFs [15, 17, 19] . Within the TURBS, two essential sequence motifs (motifs 1 and 2) have been described. Motif 1 is highly conserved amongst caliciviruses and is composed of a short stretch of 6-9 nt (with a conserved core of UGGGA) with complementarity to the apical loop of helix 26 of 18S rRNA. This motif, which lies towards the 59 boundary of the TURBS, likely acts to tether the 40S subunit to the mRNA posttermination, facilitating the recruitment of the initiation factors necessary for reinitiation [15, 19] . Motif 2, which lies some 15-20 nt upstream of the termination-reinitiation window (AUGA in FCV), is a second essential stretch of 4-8 nt but shows little sequence conservation amongst caliciviruses. Recent work has revealed that this motif in fact forms the 39 arm of a stem-loop structure whose formation is necessary for efficient terminationreinitiation [16] . The features identified in caliciviral TURBS suggest a model for termination-reinitiation in which post-termination 40S subunits are tethered to the mRNA through interactions between the mRNA (through motif 1) and 18S rRNA, initiation factors are recruited and the AUG restart codon located, processes which may require precise RNA folding (involving motif 2) within the TURBS. The TURBS is not highly active as an IRES and 40S subunit recruitment probably requires high local 40S subunit concentrations [17, 18] . An important question, therefore, is how the interaction between TURBS motif 1 and 18S rRNA helix 26 is facilitated. Analysis of primary and secondary structural features of the BM2 signal has revealed that it contains a short TURBS (of 45 nt) which is largely single-stranded, with motif 1 likely to be located in the apical loop of a metastable stem-loop structure when the ribosome is positioned at the termination codon of the upstream ORF [11] . We have suggested that motif 1 may thus be ''presented'' to the solvent-accessible apical loop of helix 26, promoting the interaction and subsequent 40S tethering. However, this hypothesis remains contentious. The paucity of stable RNA secondary structure within the BM2 TURBS [11] , the TURBS of FCV [I. Brierley, unpublished observations] and the calicivirus murine norovirus (MNV; [20] ), limits the predictive power of RNA folding programs in the assessment of the likely RNA folds present before and after ribosomal transit through the TURBS. It is also known that eukaryotic initiation factor 3 (eIF3) plays a role in the termination-reinitiation process, perhaps interacting with the TURBS to provide another link between mRNA and the 40S subunit [17] . In this paper, we describe further analysis of the BM2 TURBS. In contrast to the caliciviral TURBS, which contain stretches of non-essential bases between motifs 1 and 2, all of the BM2 TURBS appears to be required for function perhaps due to its shorter length, with sequence-specific and sequence-independent elements. Evidence is provided, from oligonucleotide targeting and from expression studies in yeast cells, to support the hypothesis that BM2 motif 1 interacts directly with helix 26 of 18S rRNA, consistent with a tethering role. We also provide evidence that a close spacing between the M1 stop and BM2 start codons is not critical; rather, there is dependence upon the distance between the terminating ribosome and the TURBS. Indeed, the ribosome is able to locate start codons placed some distance downstream of the wild-type position if termination occurs at the normal distance relative to the TURBS. The role of RNA secondary structure in TURBS function was also investigated by mutagenesis, but the experiments gave only limited support for the predicted secondary structures. However, the capacity of eIF3 to stimulate terminationreinitiation, particularly from a defective TURBS, was confirmed, including a TURBS rendered defective through a point mutation in motif 1. Together, these data support the view that efficient termination-reinitiation requires both mRNA-rRNA interactions and the participation of eIF3. The p2luc-BM2 plasmid series was prepared by subcloning sequences encompassing the influenza B virus terminationreinitiation signal (prepared by RT-PCR) into the dual-luciferase reporter plasmid p2luc [21] . Most of these plasmids have been described previously [11] . The p2luc-BM218S-AG, 18S-AT and 18S-AC plasmids were generated by site directed mutagenesis of p2luc-BM2wt or p2luc-BM2-204 (see below). The CrPV-p2luc-BM2 plasmid series was generated by digestion of p2luc-BM2wt with PstI and NheI, and insertion of a PCR fragment comprising the cricket paralysis virus (CrPV) IRES downstream of a bacteriophage T7 promoter. The T7-CrPV fragment was generated by PCR from plasmid CrPV/+8norm [14] . The resulting plasmid contains a T7 promoter located 25 nt upstream of the CrPV IRES. Translation starts on an alanine codon within the inserted CrPV fragment, resulting in an Nterminal extension to rluc of 10 amino acids relative to that of the p2luc-BM2wt. Derivative plasmids were generated by site-directed mutagenesis using primers described previously [11] . Termination-reinitiation in yeast cells was studied using the yeast dual-reporter plasmid pAC99 [22, 23] . The wild-type termination-reinitiation signal of BM2 was prepared by PCR as an EcoRV fragment that was subsequently cloned into MscIdigested pAC99. pAC99-BM2 derivative plasmids were generated by site directed mutagenesis as described below. Sequences were confirmed by commercial dideoxy sequencing (using the facility at the Department of Biochemistry, University of Cambridge). Site-directed mutagenesis was performed using the Quikchange II site-directed mutagenesis kit (Stratagene) according to manufacturer's instructions. Mutagenesis to introduce insertions longer than 6 bp was performed in two steps [24] , by first subjecting the mutagenesis reactions (containing either the sense or antisense primer) to three cycles of PCR, then mixing the reactions and performing a further 18 cycles as previously described [11] . In vitro transcription and translation p2luc-BM2 reporter plasmids were linearised with HpaI and capped run-off transcripts generated using T7 RNA polymerase as described previously [25] . Messenger RNAs were recovered by a single extraction with phenol/chloroform (1:1 v/v) followed by ethanol precipitation. Remaining unincorporated nucleotides were removed by gel filtration through a NucAway spin column (Ambion). The eluate was concentrated by ethanol precipitation, the mRNA resuspended in water, checked for integrity by agarose gel electrophoresis and quantified by spectrophotometry. Unless otherwise stated, mRNAs were translated in FlexiH rabbit reticulocyte lysate (FlexiHRRL, Promega) programmed with template mRNA at 50 mg/ml. Typical reactions were of 10 ml and composed of 60% (v/v) FlexiHRRL, 20 mM amino acids (lacking methionine), 500 mM MgOAc, 2 mM DTT, 5U RNase inhibitor (RNAguard, GE Healthcare Life Sciences), 130 mM-160 mM KCl (optimised for each batch of FlexiHRRL) and 0.2 MBq [ 35 S]methionine. Reactions were incubated for 1 h at 30uC and stopped by the addition of an equal volume of 10 mM EDTA, 100 mg/ml RNase A followed by incubation at room temperature for 20 minutes. Samples were prepared for SDS-PAGE by the addition of 4 volumes of 4X Laemmli's sample buffer, boiled for 3 minutes and resolved on 12% SDS-PAGE gels. The relative abundance of products on the gels was determined by direct measurement of [ 35 S]-methionine incorporation using a Packard Instant Imager 2024. eIF4G-depleted RRL was prepared and used as described previously [26] . Purified eIF3 was a kind gift of Dr. Chris Fraser (Department of Molecular and Cell Biology and Howard Hughes Medical Institute, University of California). Dominant-negative eIF4A-R362Q was prepared as described [14] . Reporter gene assay in yeast cells pAC99 reporter plasmids were transformed into yeast strain Y349 using the LiOAc/ssDNA/PEG method [27] . For each experiment three transformants cultivated under the same conditions were assayed. Cells were disrupted by vortexing with acid washed glass beads (Sigma) at 4uC for 30 minutes. Cell debris was removed by centrifugation and reporter enzyme assays carried out as described [23] . Termination-reinitiation efficiency was calculated as the ratio of firefly luciferase activity to b-galactosidase activity relative to that of constructs in which the open reading frames were fused in frame. The BM2 TURBS is streamlined into 45nt of RNA essential for termination-reinitiation, containing sequence-specific and sequence-independent elements In RHDV and FCV, TURBS motifs 1 and 2 are separated by some 30 nt, much of which appears to be non-essential [15] [16] [17] 19] . However, given that the minimal sequence requirement for termination-reinitiation of BM2 (,45 nt; [11] ) is shorter than that documented in the caliciviruses (69 and 87 nt; [15, 17, 19] ), we wished to test whether any of the residues were non-essential. To do this, we introduced 6 nt deletions throughout the length of the sequence upstream of the M1 termination codon in the context of the BM2 termination-reinitiation reporter plasmid, p2luc-BM2-204 ( Figure 1A ; [11] ). As expected, translation of the BM2-204 reporter mRNA generated products corresponding to the capdependent upstream product (rlucM1-204 ,36 kDa) and the termination-reinitiation product (BM2fluc ,62 kDa). The latter protein was not observed in translations of the negative control reporter mRNA, p2luc-BM2ps (rluc'M1 ,33 kDa, [11] ), which contains an in-frame stop codon immediately upstream of the TURBS. Significantly, deletion of any part of the minimal required region led to a strong inhibition of BM2fluc synthesis, suggesting that the entire TURBS is required for efficient termination-reinitiation on the segment 7 RNA ( Figure 1A ). Alternatively, it was possible that the defect in BM2 synthesis in this deletion series may be a consequence of the altered spacing between the M1 stop codon and upstream features required for termination-reinitiation. To test this, we selected three of the 6 nt deletion mutants surrounding motif 1 (mutants 6.1, 6.4 and 6.6) and replaced the deleted regions with two different random 6 nt sequences ( Figure 1B ). No restoration of BM2 synthesis was observed with either of the 6.1 nucleotide replacement mutants ( Figure 1B) . Replacement of nucleotides into the 6.4 deletion mutant had no effect with the first of the mutants (6.4r1), but some recovery was observed in the second of these mutants (6.4r2), albeit the frequency of termination-reinitiation was diminished compared to that of the wild-type mRNA ( Figure 1B ). In contrast, insertion of 6 nt back into the 6.6 deletion mutant fully restored BM2 synthesis in both cases ( Figure 1B) . These results suggest that the minimal 45 nt TURBS region is split into both sequencespecific and sequence-independent elements, where the sequence towards the 59 end of the minimal region is important in basespecific interactions, whereas the 39 end of the TURBS may act solely as a spacer, important in placing motif 1 relative to the terminating ribosome. Targeting the 18S rRNA:mRNA interaction using antisense oligonucleotides Termination-reinitiation on the BM2 ORF is likely to be dependent on interactions between the influenza B segment 7 RNA and helix 26 of the 18S rRNA [11] . However, other work has revealed that motif 1 may also be important in eIF3 binding [17] . To investigate the mRNA-18S rRNA interaction further, antisense 2-O-methyl oligonucleotides (AONs) were synthesised that would target either the loop of helix 26 of 18S rRNA ( Figure 2A ) or motif 1 of the BM2 TURBS ( Figure 2B ). Control oligonucleotides were also prepared to target sequences both upstream and downstream of motif 1 to control for potential nonspecific effects on translation ( Figure 2B ). An AON designed to target helix 26 of the 18S rRNA had no effect on BM2 synthesis ( Figure 2A ), although high concentrations (320-fold to 2960-fold molar excess of the AON to the mRNA) inhibited global translation ( Figure 2A and data not shown). Conversely, an AON targeting motif 1 specifically inhibited BM2 synthesis, with little effect on overall translation ( Figure 2B ). Importantly, the upstream and downstream control AONs had little effect on BM2 synthesis or global translation within the range tested for the BM2 motif 1 complementary AON ( Figure 2B ). It should be noted that the 80S ribosome will strip annealed AONs from the mRNA as it translates through the TURBS. However, both the upstream and motif 1 complementary AONs will be able to re-anneal once the ribosome reaches the UAAUG. However, it is possible that any effect of the downstream oligo is masked due to it being unable to reanneal in the presence of the terminating ribosome. Nevertheless, these data indicate that the effect of the BM2 motif 1 complementary AON is specific and not an effect on ribosome processivity, and provide further evidence in support of an interaction between motif 1 and 18S rRNA. However, we cannot rule out that these effects may also be due to perturbations of RNA secondary structure within the TURBS or due to interactions of motif 1 with an unspecified molecule (see later). The dependence of BM2 synthesis on interactions between the mRNA and 18S rRNA raises the possibility that increasing 18S rRNA complementarity within motif 1 would lead to increased synthesis of the BM2fluc polypeptide. To test this, the bases adjacent to motif 1 were substituted such that the complementarity to 18S rRNA was increased to the 59, to the 39 or in both directions ( Figure 3A ). In all cases, however, the substitution mutations reduced the frequency of termination-reinitiation events, albeit to a lesser extent with the 39 complementary extension mutant ( Figure 3A ). We went on to examine the possibility that multiple copies of motif 1 would stimulate termination-reinitiation, through the provision of an increased number of tethering sites. As a precedent for this, the murine Gtx mRNA (also known as Nkx 6-2) contains a short 9 nt IRES element that is known to act more efficiently when present in multiple copies. Similarly to motif 1, ribosomes are recruited through interactions between the Gtx IRES and helix 26 of the 18S rRNA [28] [29] [30] , although with the arm of the helix [31] rather than the apical loop ( Figure 3C , left panel). Two sets of constructs were prepared. In the first, one or two additional copies of the 6 nt (AUGGGA) core of motif 1 were introduced alongside the original ( Figure 3B ). Control constructs were also generated in which one or two copies of the hexanucleotide AUGCGA were looped in as a negative control, since this motif 1 mutation was shown previously to be unable to support termination-reinitiation, presumably due to the abolition of mRNA:rRNA base pairing [11] . In the second set of constructs, the bases downstream of the wild-type AUGGGA sequence were substituted to generate one or two extra copies of the AUGGGA motif ( Figure 3B ). In vitro translations revealed, however, that none of the mutations had a stimulatory effect on reinitiation on the BM2 ORF. Looping in a single extra copy of the AUGGGA (or AUGCGA) hexanucleotide had very little effect on the process, whilst insertion of two copies led to inhibition in both contexts ( Figure 3B ). In constructs where motif 1 was duplicated by virtue of nucleotide substitutions, termination-reinitiation was inhibited to a similar extent in constructs containing either one or two copies of the 18S complementary motif ( Figure 3B ). These data indicate that there is no additive effect on termination-reinitiation of adding extra copies of the 18S complementary region. In fact, the inhibitory effects seen most likely reflect the importance of the precise spacing of the ribosome with respect to motif 1 (in the loop-in mutants) or, alternatively, some effect on TURBS RNA secondary structure. As discussed above, the Gtx mRNA is thought to be able to recruit ribosomes by virtue of interactions between helix 26 of the 18S rRNA and the mRNA [28] [29] [30] ). It was therefore of interest to determine whether the minimal 7 nt Gtx motif could functionally replace motif 1. Two constructs were generated, one in which the inserted Gtx element was complementary to murine 18S rRNA (which would result in a mismatch with helix 26 of rabbit 18S rRNA) and one with full complementarity to the rabbit 18S rRNA, such that efficient tethering would be expected in rabbit reticulocyte lysates (RRL; Figure 3C , top-right panel). Termination-reinitiation was found to be inhibited in either mutant and to a similar extent as mutants in which a single nucleotide substitution was present to abolish 18S rRNA:mRNA base pairing ( Figure 3C ; the single nucleotide mutants have been described previously [11] ). These results suggest that motif 1 and the Gtx IRES element cannot functionally replace each other. Termination-reinitiation in yeast cells: further evidence for mRNA:18S rRNA interactions in ribosome tethering It is known that nucleotide substitutions within motif 1 that are predicted to destabilise the motif 1:18S rRNA interaction are inhibitory to termination-reinitiation, but it is possible that these mutations may affect TURBS structure or interaction with other translational components, like eIF3 [17] . In an attempt to resolve this issue, we assayed termination-reinitiation in yeast cells, whose 18S rRNA helix 26 equivalent has a somewhat different primary sequence. This experiment was carried out in the context of the yeast dual reporter vector pAC99, into which we had cloned the relevant BM2 information appropriately framed between bgalactosidase and firefly luciferase reporter genes. In pAC99-BM2wt, b-galactosidase is synthesised by cap-dependent translation, whereas firefly luciferase should only be synthesised following reinitiation on the BM2fluc ORF ( Figure 4A ). To confirm that firefly luciferase was being synthesised as a result of a genuine termination-reinitiation event (as opposed to internal ribosome entry, for example), we generated a premature stop (ps) mutant (pAC99-BM2-ps) in which a stop codon was inserted into the M1 sequence 233 nt upstream of the start codon of BM2fluc. In pAC99-BM2-yeast, motif 1 was changed to give full complementarity to the yeast 18S rRNA partner ( Figure 4B ). The plasmids were transformed into yeast strain Y349, and b-galactosidase and luciferase assays were carried out. Termination-reinitiation frequency was determined in comparison to the reporter gene activities observed in a vector in which the reporter ORFs were fused in-frame (pAC99-BM2IFC). As expected, termination reinitiation was significantly (p,0.01) reduced in pAC99-yeastps, although there was still some background synthesis of fluc ( Figure 4C ). Importantly, mutation of the motif 1 homologue such that it was fully complementary to yeast 18S rRNA led to a significant (p,0.01) increase in BM2fluc synthesis relative to the wild-type BM2 reporter ( Figure 4C ), supporting the view that mRNA:rRNA base pairing is a key determinant in BM2 ORF expression. It has been suggested previously that termination-reinitiation is dependent on the stop codon of the uORF and the start codon of the downstream ORF lying in close proximity [8, 11, 15, [17] [18] [19] 32, 33] . Indeed, movement of the M1 stop codon more than 24 nt downstream of its original context in the terminationreinitiation 'window' of the BM2 signal results in inhibition of reinitiation on the BM2 ORF [11] . However, in all previous investigations into proximity effects of the start and stop codons, the termination site was moved downstream of the start codon of the upstream ORF [8, 11, 15, [17] [18] [19] 32] ). Given that terminationreinitiation almost certainly depends on interactions between the terminating ribosome and the TURBS, it is conceivable that these experiments have overestimated the importance of the relative proximity of the stop and start codons and that the distance between the terminating ribosome and the TURBS is the crucial issue. To investigate this, a series of mRNAs were prepared in which the UAAUG overlap was moved downstream en masse such that the distance between the stop and start codons was conserved but the ribosome would terminate at varying distances relative to the TURBS ( Figure 5B) . A series of control mRNAs were also examined in which the stop codon alone was shifted downstream of the BM2 start codon ( Figure 5A ; details in [11] ). In all constructs, nucleotides at -3 and +4 relative to the BM2fluc AUG were maintained to control for context effects of the start codon. As observed previously in the control series [11] , reinitiation occurred efficiently when the stop codon was moved up to 24 nt downstream of its original placement but was abolished when the stop was moved further ( Figure 5A ). Importantly, BM2 translation The 44 kDa product produced from this mRNA corresponds to the rlucM1 product, the 62 kDa product is the BM2fluc reinitiation product. The molar excess of AON (oligo) in relation to the mRNA is shown above the autoradiograph. (B) The sequence upstream of the termination-reinitiation site (UAAUG) is shown, with the A residue at the start of the TURBS underlined. Also shown is the core sequence of motif 1 in bold. The regions targeted followed the same pattern in mRNAs where the UAAUG motif was moved as a unit, with ablation of synthesis observed when this sequence was moved any further than 24 nt downstream of its original site ( Figure 5B) . These experiments reveal that the placement of the 40S subunit relative to the TURBS, posttermination, is quite critical. As reinitiation is not absolutely dependent on the close proximity of stop and start codons ( Figure 5A) , it seemed conceivable that termination-reinitiation could still occur efficiently if the start codon of the BM2fluc ORF was moved downstream of the stop codon of rlucM1. A series of mRNAs were prepared in which the original BM2fluc start codon was mutated to AGC and new AUG restart codons inserted independently at +12, +24, +36, +48 and +63 nt, again preserving the wild-type Kozak consensus. As can be seen in Figure 6A , movement of the start codon to +12 resulted in 50% inhibition of fluc synthesis, but there was no further reduction in product yield when the restart codon was displaced further downstream. However, the electrophoretic mobility of this reinitiation product did not show the clear increase that would be expected given that the restart product would become progressively smaller with increased displacement of the restart codon. Rather, the reinitiation product band seemed less sharp and more diffuse with the +48 and +63 mRNAs. This suggested the possibility that reinitiation might be occurring at two sites: one in the region of the wild-type reinitiation site (and so necessarily at a non-AUG codon), and the other at the displaced AUG restart codon. The most obvious potential non-AUG codon for the first type of reinitiation event is the AUA codon located immediately upstream (-1 position) of the wild-type reinitiation site at +1 ( Figure 6A ). We therefore tested whether reinitiation can indeed occur at this -1 position, by generating an AUG in this position and removing the wild-type +1 AUG. The results showed that the efficiency of reinitiation at an AUG in the -1 position was similar to when it was in the normal (+1) position ( Figure 6B , compare lanes 8 and 10), in agreement with the finding that reinitiation efficiency in the RHDV system was only slightly compromised if the restart codon was moved one codon upstream [19] . Given that quite efficient reinitiation still occurs in FCV and BM2 RNAs when the wild-type AUG is mutated to non-AUG codons [11, 15, 17] , this observation suggests that when the wild-type +1 AUG has been mutated, there is a strong possibility of some reinitiation occurring at the -1 AUA codon. As for reinitiation at the displaced AUG codon, the data of Figure 6B (lanes 1-6) demonstrate that reinitiation certainly occurs at these downstream sites, because on this higher resolution gel the product size clearly decreases as the displacement is increased, although the efficiency is only 20-25% of the reinitiation frequency in the wild-type mRNA (lane 1). In fact, in this particular experiment there is much more downstream cistron product from the displaced AUG than product from the putative -1 AUA reinitiation site, which was only just visible on the autoradiogram (highlighted by the asterisk). In reviewing the results of many experiments of this type we conclude that the relative use of the two potential reinitiation sites shows quite a large degree of variability, which seems related to the use of different batches of reticulocyte lysate and so may possibly be due to small batch to batch differences in the endogenous ionic conditions. The variability seems to mainly affect the yield of putative -1 AUA reinitiation product, whereas the yield of product from the displaced AUG codon is relatively invariant at ,20% of the wild-type mRNA yield. The two extremes of the variability range are well illustrated by Figure 6B (very minimal use of the -1 AUA site) and Figure 6C , where there is actually more putative reinitiation at the -1 AUA than at the displaced AUG codon. The fact that reinitiation, albeit at reduced efficiency, can occur when the first AUG is moved 63 nt downstream raises the question of whether the reinitiating 40S subunits access this site by linear scanning from the TURBS. We examined this by exploiting the dependence of ribosomal scanning on eIF4G [34] . The BM2wt and BM2start +63 mRNAs were modified such that translation of the upstream M1 ORF was driven by the cricket paralysis virus (CrPV) IRES, which does not require eIF4G for function. These mRNAs (CrPV-BM2wt, CrPV-BM2start+63) were then translated alongside capped brome mosaic virus RNAs (BMV, Promega) and BM2wt RNAs (as a control for the efficacy of eIF4G depletion) in mock-and eIF4G-depleted rabbit reticulocyte lysates ( Figure 6C ). Translations were also performed in the absence or presence of dominant-negative eIF4A (eIF4A R362Q [35] ) to inhibit the activity of any residual eIF4G that had escaped depletion. As expected, the translation of the capped wildtype RNA (BM2wt) and BMV RNAs were inhibited in the eIF4Gdepleted RRL, but the CrPV IRES-containing mRNAs were translated efficiently in both control and depleted RRL ( Figure 6C ). Importantly, depletion of eIF4G had no effect on the reinitiation efficiency in either the CrPV-BM2wt or CrPV-BM2start +63 mRNAs, suggesting that the reinitiation site can be located in a scanning-independent manner. When two AUG codons were present, one at +63 and the other at either the +1 (wild-type) or -1 position, the upstream site took complete precedence over the +63 site ( Figure 6B, lanes 8 and 10) , and this precedence was maintained even when the stop codon was moved 12 codons downstream ( Figure 6B, lanes 11-13) . Thus the strongly preferred site for reinitiation is an AUG codon just downstream of the TURBS. If there is no AUG in this region, the reinitiation mechanism apparently seeks alternatives: either an acceptable non-AUG codon in this same region or an AUG codon further downstream. Reinitiation in the later case does not involve eIF4G/4A-dependent linear scanning, and so it is more likely to involve a direct transfer of the 40S subunit from the TURBS to the AUG, a transfer which might be facilitated by looping out the mRNA between the tethered 40S ribosomal subunit and the AUG. Studies of the FCV signal have indicated a role for eIF3 in termination-reinitiation as, for example, addition of supplementary eIF3 to RRL is able to specifically stimulate synthesis of the ORF3 reinitiation product [17] . Furthermore, greater stimulation is observed with mRNAs that show a partial defect in reinitiation in the absence of eIF3, suggesting that the increase in eIF3 concentration may rescue the negative phenotype, perhaps by allowing increased binding of eIF3 to the mRNA [17] . To analyse whether the same is true of reinitiation on the BM2 ORF, we examined the effect of exogenous eIF3 on reinitiation on mRNAs containing a fully functional signal (BM2wt, BM2-204), or mutants by AONs are shown above the sequence, colour coded to match the relevant AON sequence. The BM2wt RNA was translated as above in the presence of increasing AON concentrations as indicated above each autoradiograph. doi:10.1371/journal.pone.0016822.g002 defective by virtue of a deletion (BM2-207, BM2-210, described in [11] ) or a substitution in motif 1 (BM2-AG, BM2-AC). To focus specifically on the effect of eIF3 on termination-reinitiation rather than 59-dependent initiation, the BM2 sequences were sub-cloned such that the rlucM1 ORF would initiate on an alanine codon provided as part of the CrPV IRES. As can be seen in Figure 7 , exogenous eIF3 specifically stimulated synthesis of the BM2fluc reinitiation product with all of the mRNAs tested, including those mutants that were essentially inactive (BM2-210 and BM2-AC; most evident in long exposures, see right panel, Figure 7) . Importantly, the addition of an unrelated, similarly purified protein had no effect on reinitiation in any of the RNAs described above (Figure S1 ), further highlighting the specificity of eIF39s effect on reinitiation. Given the likely importance of the interaction between motif 1 and 18S rRNA, we previously carried out RNA structure mapping and minimal free energy predictions to assess whether motif 1 was in basepaired or single stranded conformation in the native mRNA. This work implicated two potential structures, mfold 1 and mfold 2 ( Figure 8A ; [11] ). In mfold 1 the 18S complementary region is sequestered in a base-paired region of stem 2. We previously hypothesised that translation through the segment 7 mRNA and termination of ribosomes at the M1 stop codon would prevent these interactions, creating a structure similar to mfold 2 ( Figure 8B ). In this structure, the 18S complementary region would then be presented to helix 26 of the 18S rRNA on the apical loop of a metastable stem-loop structure. Given that translation through the TURBS up to the termination-reinitiation window is a prerequisite for efficient reinitiation [11] , one can speculate that mfold 2 is the biologically relevant fold, and that destabilisation of the central stem of mfold 1 (stem 2) would have little effect on the reinitiation process. To test this hypothesis we created destabilising mutations within either arm of stem 2 (chosen to avoid disruption of the 18S rRNA complementary region) in the context of the p2luc-BM2-204 parental plasmid (which encodes the minimal required region for termination-reinitiation [11] ). The mutations created were of either 2 or 3 nt to control for context effects on the BM2 AUG, creating the arm 1 mutants p2luc-BM2A1-xnt, and the arm 2 mutants p2luc-BM2A2-xnt (where x = the number of nucleotides mutated). We also prepared double mutants (A1/2-xnt) to create a pseudowild-type structure where base-pairing between the two arms is restored. In the context of both the 2 nt and 3 nt mutations, substitution of bases in arm 1 abolished terminationreinitiation, however substitution of the bases at the bottom of arm 2 had little effect on BM2fluc synthesis relative to the BM2-204 control. The double pseudowild-type mutation also demonstrated abolition of BM2fluc expression ( Figure 8A ). Taken together these results suggest that the structure presented in mfold 1 is unlikely to play a role in termination-reinitiation as disruption of arm 2 of stem 2 has little effect on BM2 translation, and no restoration of BM2 synthesis was observed in the pseudowild-type construct. As opposed to mfold 1, the structure presented for mfold 2 would still be able to form when the ribosome has translated through the upstream ORF and is in the process of termination. We tested this structure in the same way, by introducing destabilising mutations independently in both arms at the base of stem 29 (creating mutants p2luc-BM2A1' and p2luc-BM2A29) and preparing a double mutant that would yield a pseudowild-type structure, which would be expected to exhibit BM2fluc synthesis (p2luc-BM2A19/29). It should be noted that the A19 mutation is very similar to that of the A1-2/ 3 nt mutant, given that this arm base pairs to form both stem 2 and stem 29 in the two putative folds. As expected, this mutation dramatically inhibited synthesis of BM2fluc similarly to that observed with the A1 mutation ( Figure 8B ). However, as before, no inhibition of BM2 synthesis was observed when stem 2 was destabilised in the 39 arm ( Figure 8B ). Nevertheless, the partial restoration of BM2 expression (to around 50% of wild-type) seen with the A19/29 double mutant, indicates that base-pairing in this region may play some role in termination-reinitiation. The simplest conclusion from these experiments, however, is that neither mfold 1 nor mfold 2 represent the sole active confirmation, although it is possible that these mutants may form another structure that can present the TURBS motif 1 to helix 26 of 18S rRNA. Previous work has revealed that termination-dependent reinitiation on the BM2 ORF of segment 7 of influenza B virus is dependent on a relatively short (45 nt), largely unstructured, TURBS containing a typical motif 1 element towards the 59 end [11] . In the present study, we investigated features of the TURBS required for efficient reinitiation, focusing primarily on the proposed interaction between TURBS motif 1 and helix 26 of 18S rRNA. Whilst the effects of point mutations within motif 1 are consistent with such an interaction, direct evidence is lacking in the BM2 system. We began by attempting to block the interaction by targeting the binding partners with antisense oligonucleotides. Such an approach was successfully employed in studies of the IRES-like properties of the Gtx leader (which recruits ribosomes de novo by virtue of interactions between the 18S rRNA and the mRNA) and revealed that the IRES activity could be blocked by AONs that bind either the mRNA or the rRNA [36] . Whilst we were able to observe specific inhibitory effects with an oligonucleotide that targeted motif 1, no effect was observed with an oligonucleotide targeting the ribosomal RNA ( Figure 2 ). Our failure to observe an effect of the AON complementary to the apical loop of helix 26 may be due to a failure of the AON to access the target. The Gtx mRNA:rRNA interaction occurs at a In the p2luc-BM2Gtx mutants, the 7 nt core of motif 1 (bold italics in p2luc-BM2wt) was substituted for nucleotides complementary to the region of 18S rRNA where the Gtx IRES is believed to bind. Note the p2luc-BM2Gtx Rabbit sequence is altered from that of the p2luc-BM2Gtx Mouse reporter such that 100% complementarity with the different rabbit 18S rRNA could be achieved. Nucleotides differing between Gtx Mouse and Gtx Rabbit are underlined. RNAs were translated and analysed as in the legend to Figure 1 . A series of previously described motif 1 mutants [11] were also translated as controls. doi:10.1371/journal.pone.0016822.g003 different region of helix 26 than the putative segment 7 mRNA:rRNA interaction, and this may be more accessible. Indeed, termination-reinitiation could not be reconstituted in reporter mRNAs in which motif 1 was replaced by the Gtx 18S rRNA binding motif ( Figure 3C) , and thus different responses to oligonucleotides targeting the rRNA may be expected. It may also be that the 18S rRNA-targeting oligonucleotide may need to adopt a similar structure to the mRNA if it is to effectively bind to the 18S rRNA and block reinitiation. The inhibition of termination-reinitiation by the AON that targeted motif 1, in contrast, is highly consistent with its proposed role in binding to helix 26, although effects via RNA conformation or binding to an unknown molecule cannot be ruled out. However, the weight of evidence from mutational analysis, the yeast data described below and the AON titrations strongly suggest that the inhibitory effect of AONs directed against motif 1 are likely due to their blocking of the interaction between mRNA and 18S rRNA. Subsequently, we sought to confirm the interaction by investigating termination-reinitiation in yeast cells, exploiting the fact that the helix 26 equivalent in yeast 18S rRNA has a primary sequence distinct from that of the rabbit. Using a dicistronic reporter mRNA with variant motif 1 sequences, we found that increasing the complementarity of the motif to that of yeast 18S rRNA stimulated termination-reinitiation, supporting strongly the view that intermolecular interactions are important in reinitiation on the BM2 ORF. However, two aspects of this experiment require comment. First, a high background activity was observed in control assays (Figure 4) , the origin of which is uncertain Secondly, in yeast cells, the BM2wt reporter mRNA, whose motif 1 is not fully complementary to the helix 26 target, showed an efficiency of termination-reinitiation only two-fold lower than the BM2yeast mRNA. In RRL, a single base mismatch between mRNA and rRNA is sufficient to inhibit BM2 synthesis 10-20 fold [11] . One possible explanation for this difference is that the higher concentrations of mRNA and rRNA in an intact cell can act to stabilise the (presumably) weakened mRNA:rRNA interaction. During preparation of this manuscript, Luttermann and Meyers (2009) published a similar, more sophisticated, investigation of FCV motif 1:18S rRNA interactions using a yeast expression system in which both mRNA and rRNA partners could be The rlucM1/BM2fluc overlap window was changed from UAAUG to CAAUG such that termination occurs at the next in-frame stop codon, some 63 nt downstream of the original stop site (+63). New stop codons were then inserted, preserving the stop codon context, at +12, +24, +36 and +48 nt downstream of the natural termination site. Messenger RNAs derived from these plasmids and a control mRNA (ps) were translated and analysed as detailed in the legend to Figure 1 . This experiment is a repeat of that found in [11] and is shown here as a control. (B) The rlucM1/BM2fluc overlap window was changed from UAAUG to CAACU, and the UAAUG sequence reintroduced at +12, +24, +36, +48 or +63 nt downstream preserving both the initiation codon and stop codon context. Messenger RNAs derived from these plasmids were translated and analysed as above. doi:10.1371/journal.pone.0016822.g005 Figure 6 . Effect on termination-reinitiation of altering the BM2 start codon position. (A) The termination-reinitiation overlap was mutated from AUAAUG to AUAAGC and the BM2fluc start codon reintroduced at +12,+24, +36, +48 or +63 nt, preserving the start codon context at the -3 and +4 positions. The codon immediately 59 of the native reinitiation site is denoted -1, the reinitiation site is shown as +1. Messenger RNAs derived from these plasmids were translated and analysed as above. (B) Left panel: Translations were performed as in Figure 6A and the samples were separated by SDS-PAGE and autoradiographed. The putative AUA reinitiation product is marked with an asterisk. In addition, the three right hand modified. This work provided strong evidence in support of a requirement for mRNA:rRNA interactions in termination-reinitiation in the expression of FCV VP2 [16] . Since the reinitiation mechanism shows a fairly strong preference for an AUG codon, it seems a certainty that initiator Met-tRNA is involved. This, in turn, implies that the ribosome lanes of the left panel show translations of mutant mRNAs in which the codon -1 of the AUG start codon and the start codon (+1) were mutated to the indicated codons in the context of the BM2start +63 RNA. Lane numbers are shown below the figure. Right panel: Translations of RNAs containing stop and start codons at the indicated positions. C) Capped BMV (BMV) RNAs, BM2 wt reporters (Cap-wt), CrPV-IRES driven BM2wt (CrPV-wt) and BM2start +63 (CrPV+63) RNAs were translated in control (C) and eIF4G-depleted (D) RRLs in the absence or presence of ,1 mM dominant-negative eIFA (4A-R362Q). RNAs were translated as described in [26] and analysed as above. doi:10.1371/journal.pone.0016822.g006 which has just terminated at the stop codon must dissociate into subunits prior to reinitiation, in order to allow eIF2/GTP/Met-tRNA i ternary complex binding to the 40S subunit. So it is a 40S subunit rather than an 80S ribosome that interacts with TURBS motif 1. The fact that reinitiation efficiency decreases quite abruptly with increasing distance of displacement of the stop [11] ) is shown with motif 1 shown in red, the termination-reinitiation overlap shown in cyan and the sequences likely to be occupied by a ribosome terminating at the M1 ORF are 39 of the purple line. Mutations were introduced at the bottom of stem 2, with both the 2 nt and 3 nt substitution mutations shown next to arm 1 (A1) or arm 2 (A2). Messenger RNAs derived from these plasmids and control mRNA (204, ps) were translated and analysed as detailed in the legend to Figure 1 . RRF denotes the relative reinitiation frequency adjusted for relative methionine content as compared to the wt (set as 100%). (B) The mfold 2 structure (from [11] ) is shown with motif 1, the termination-reinitiation overlap, and the ribosome protected region marked as above. Mutations were introduced at the bottom of stem 29 and are shown next to arm 1 (A19) or arm 2 (A29). Messenger RNAs derived from these plasmids and a control mRNA (204) were translated and analysed as above. RRF denotes the relative reinitiation frequency adjusted for relative methionine content as compared to the wt (set as 100%). doi:10.1371/journal.pone.0016822.g008 codon further downstream ( Figure 5 ) favours a model in which this 40S subunit is transferred directly from the termination site to the TURBS, rather than dissociating from the mRNA after termination followed by reassociation with the TURBS. Recent work has implicated eIF3 in the termination-reinitiation process [17] , which is of interest because eIF3 plays an important role in disassembly of ribosomes following the termination event which leaves the 80S ribosome with bound deacylated tRNA and still associated with the mRNA [37, 38] . At low (sub-optimal) Mg 2+ , this disassembly requires eIF1 and eIF3, but no other protein factor. First, eIF3 binds to the solvent face of the 40S subunit, promoting dissociation of the 60S subunit, and leaving the 40S subunit (with bound deacylated tRNA) still associated with the mRNA. Then eIF1 ejects the deacylated tRNA, while the eIF3j subunit promotes dissociation of the 40S/eIF1/eIF3 complex from the mRNA [37] . At higher, more physiologically relevant Mg 2+ , as would pertain in our translation assays, there is also a requirement for ABCE1 and ATP [38] , which catalyses the first step of dissociation of the ribosomal subunits, ejecting the 60S subunit and again leaving the 40S subunit (plus bound deacylated tRNA) still associated with the mRNA. Then eIF3 binds to the solvent face of the 40S subunit, thereby preventing any ribosomal subunit reassociation, and events thereafter are exactly as at low Mg 2+ [37, 38] . Initiation factor eIF3 has been shown to crosslink to the FCV TURBS [17] . More important, supplementary eIF3 has been shown to stimulate reinitiation at the wild-type FCV TURBS, but more especially at TURBS derivatives which are partially defective due to deletions or point mutations [17] , just as has been observed here (Figure 7) . These results, particularly the stimulation of the partially defective TURBS, have suggested a model in which it is specifically the eIF3 associated with the TURBS that binds those 40S subunits which are destined to reinitiate translation [17] . There is no conflict between this hypothesis and the model in which tethering the 40S subunit to the TURBS is due to base-pairing between TURBS motif 1 and 18S rRNA. eIF3 binds predominantly to the back or solvent face of the 40S ribosomal subunit, and is known to make direct contacts with the 18S rRNA component of the small ribosomal subunit [39, 40] . A bridging interaction in which eIF3 binds simultaneously to the 40S subunit and to the TURBS could help stabilise the binding of the 40S subunit to the TURBS for sufficient time as is required to acquire an eIF2/GTP/Met-tRNA i ternary complex, and other necessary initiation factors. After all, the analogous Shine-Dalgarno interaction of prokaryotic 30S subunits is generally considered to be very transitory unless it is accompanied and stabilised by a P-site fMet-tRNA base-pairing with an AUG (or GUG) at an appropriate distance further downstream. An important question that remains to be resolved is whether there is a role for TURBS RNA secondary structure. We previously proposed that the TURBS may fold into two main configurations, in the first of which (mfold 1) motif 1 is sequestered in a base-paired region [11] , which may explain why it does not have significant IRES activity (as is seen with Gtx [28] [29] [30] ). Translation through and termination at the M1 ORF would remodel mfold 1 to a structure similar to that shown in mfold 2 ( Figure 8B ) such that motif 1 would then be presented to the 18S rRNA on the apical loop of a stem-loop structure. The experiments described in Figure 1 agree well with dependence on a structure similar to that presented in mfold 2, in that the substituted nucleotides in the 6.1 and 6.4 replacement mutants would act to disrupt the stem, whereas substitution of nucleotides in the 6.6 replacement mutants would be expected to have no effect as they would lie within the mRNA channel of the terminating ribosome. However, we also carried out a mutational analysis of the main stem regions present in mfold 1 and mfold 2 but the data did not corroborate our structural predictions. Whilst disruption of either of the putative stems in one arm inhibited reinitiation, little effect was observed when stem formation was disrupted in arm 2 of either stem-loop structure, although some restoration of BM2 synthesis was observed in an mRNA containing a pseudowild-type mfold 2 structure. It should be noted that RNA structure mapping of the TURBS of BM2 [11] , FCV (Brierley et al. unpublished observations) and MNV [20] TURBS reveals that they are largely single-stranded, and probably metastable. It may be that TURBS are able to adopt a variety of conformations, some of which are able to facilitate terminationreinitiation, for example the 6 nt replacement mutant 6.4r2 ( Figure 1B) . Alternatively, the requirement for translation through the TURBS may not be due to a dependence on translational remodelling and 'unzipping' of motif 1 from paired regions, but rather just to place the ribosome in proximity to motif 1 (similar to the case for reinitiation in bacteriophage where the ribosome is placed close to the vestigial Shine-Dalgarno [SD] motif [41] ). It is clear that further work is required to understand the putative role of RNA secondary structure in termination-reinitiation. Previous studies on the termination-reinitiation process have suggested the importance of the close proximity of the stop codon of the upstream ORF and the start codon of the downstream ORF [8, 11, 15, 17, 18, 32] . This is believed to reflect a restricted mobility of the tethered ribosome; that is, it may be able to undergo only limited movement following termination. However, we show here that the distance between the terminating ribosome and the TURBS is more critical to reinitiation efficiency ( Figure 5 ) and that when the start codon of the BM2 ORF is placed downstream of the stop codon, reinitiation is detectable even when the start codon is moved up to 63nt downstream of its original position ( Figure 6 ). This suggests that it is not solely the distance between the start and stop codons per se that affects reinitiation efficiency but rather how well the ribosome can be tethered (by virtue of the distance between where the ribosome terminates and the TURBS). However, whilst efficient reinitiation requires an AUG at, or very near, the wild-type site ( Figure 6A and 6B) , other reinitiation sites can be used when the native reinitiation codon is absent, albeit at lower efficiency. Under the latter circumstance, ribosomes can reinitiate at a downstream AUG codon ( Figure 6B ), or at a near-cognate initiation codon (such as the -1 AUA codon, Figure 6 ) close to the native reinitiation site. However, if given a choice of a wild-type AUG and one located downstream, the ribosome always selects the wild-type AUG, even if the ribosome artificially terminates downstream of the reinitiation window ( Figure 6B) , suggesting that the TURBS may cause the tethered 40S subunit to 'snap back' to the proper site of reinitiation. Importantly, we show that there is no requirement for eIF4G in location of either wild-type or downstream reinitiation sites. As such, location of downstream AUGs is likely due to the 40S subunit being transferred directly to the AUG whilst tethered to the TURBS (perhaps in complex with eIF3 and other factors) by an RNA-looping mechanism. Figure S1 Recombinant 4EBP1 has no significant effect on reinitiation on the BM2 ORF. Motif 1 and TURBS deletion mutants were translated in the absence or presence of 250nM 4EBP1 (details as in Figure 7) . The fold-stimulation over the reinitiation levels observed in the absence of 4EBP1 are shown below the autoradiographs. (EPS)

RNA and DNA Bacteriophages as Molecular Diagnosis Controls in Clinical Virology: A Comprehensive Study of More than 45,000 Routine PCR Tests

Real-time PCR techniques are now commonly used for the detection of viral genomes in various human specimens and require for validation both external and internal controls (ECs and ICs). In particular, ICs added to clinical samples are necessary to monitor the extraction, reverse transcription, and amplification steps in order to detect false-negative results resulting from PCR-inhibition or errors in the technical procedure. Here, we performed a large scale evaluation of the use of bacteriophages as ICs in routine molecular diagnosis. This allowed to propose simple standardized procedures (i) to design specific ECs for both DNA and RNA viruses and (ii) to use T4 (DNA) or MS2 (RNA) phages as ICs in routine diagnosis. Various technical formats for using phages as ICs were optimised and validated. Subsequently, T4 and MS2 ICs were evaluated in routine real-time PCR or RT-PCR virological diagnostic tests, using a series of 8,950 clinical samples (representing 36 distinct specimen types) sent to our laboratory for the detection of a variety of DNA and RNA viruses. The frequency of inefficient detection of ICs was analyzed according to the nature of the sample. Inhibitors of enzymatic reactions were detected at high frequency in specific sample types such as heparinized blood and bone marrow (>70%), broncho-alveolar liquid (41%) and stools (36%). The use of T4 and MS2 phages as ICs proved to be cost-effective, flexible and adaptable to various technical procedures of real-time PCR detection in virology. It represents a valuable strategy for enhancing the quality of routine molecular diagnosis in laboratories that use in-house designed diagnostic systems, which can conveniently be associated to the use of specific synthetic ECs. The high rate of inhibitors observed in a variety of specimen types should stimulate the elaboration of improved technical protocols for the extraction and amplification of nucleic acids.

Real-time (rt) PCR and reverse transcription (RT) PCR techniques are rapid and versatile diagnostic procedures broadly used in clinical virology where there are mostly considered as diagnostic ''gold standards'' [1] . Monitoring rt-PCR and rt-RT-PCR assays and validation of the results rely on the use of relevant external or internal controls (ECs or ICs) [1, 2] and commercial kits including such control systems are being increasingly improved for the molecular diagnosis of a number of pathogens such as HIV, hepatitis viruses, influenza viruses etc.. However, one of the main strengths of rt-PCR is versatility, which provides the opportunity to set-up ''in-house'' protocols for specific pathogens. The scientific literature now includes an impressive number of 'home made'' assays for various viral agents. Whilst most commercial kits include both ICs and ECs allowing accurate validation of the results [3] , ''home made tests'' are frequently performed in the absence of ICs and therefore without any possible individual monitoring of each diagnostic reaction. For example, the detection of technical errors or PCR amplification inhibitors is intrinsically impossible if only ECs are used. In addition, ECs are usually undistinguishable from the native genome. Here, our objective was to develop and test on a large number of clinical samples a bacteriophage-based IC system suitable for a standard laboratory of medical virology. We present results obtained by using T4 and MS2 bacteriophages as ICs in a routine-based evaluation including Important criteria for the design of ECs include (i) the possibility to use quantified or semi quantified controls (in order to manage the detection level of the diagnostic test used), (ii) the possibility to distinguish amplicons obtained from ECs from those obtained from the detected pathogen (in order to detect false positives due to accidental amplification of EC) and (iii) the use of relevant nucleic acids (ie, RNA and DNA molecules for RNA and DNA viruses, respectively). Simple protocols for preparing plasmid DNA or synthetic RNA quantified controls are described in Supporting Information S1 and schematised in figure 1. Such controls include specific sequences for the hybridisation of detection primers and probe, but also an exogenic ''Not I'' sequence (detectable by a specific probe or cleavable by the Not I restriction enzyme). 2a. rt-PCR assays for the detection of T4 and MS2 bacteriophages. The criteria to be addresses regarding ICs were: (i) to monitor all steps of the diagnostic procedure (extraction, RT, PCR); (ii) to be amenable for DNA and RNA viruses ; (iii) to rely on a detection system specific of the phage(s) to avoid complex molecular constructs; (iv) to be usable in either simplex or multiplex format, and in one-step or two-step RT-PCR reactions. We used freeze-dried E. coli Enterobacteria phage T4 (T4) and Enterobacteria phage MS2 (MS2) obtained from the American Type Culture Collection (ATCC ref. 11303-B4 & 15597-B1, respectively). Protocols for real time PCR detection of phages TA and MS2 were elaborated in various formats and are described in Supporting Information S2. Briefly, primers and probes targeting T4 phage (T4F CCATCCATAGAGAAAATATCAGAACGA, T4R TAAATAATTCCTCTTTTCCCAGCG, T4probe VIC-AACCAGTAATTTCATCTGCTTCTGATGTGAGGC-TAM-RA) and MS2 phage (MS2F CTCTGAGAGCGGCTCTATTGGT, MS2R GTTCCCTACAACGAGCCTAAATTC, MS2probe VIC-TCAGACACGCGGTCCGCTATAACGA-TAMRA) were designed from genomic sequences. Optimised oligonucleotide concentrations were 10pmol for primers and 4pmol for probes for both T4 and MS2 detection assays. 2b. Spiking and validation procedures. A suspension of T4 and MS2 phages was used for spiking clinical samples. Both phages were diluted to provide a similar level of detection by rt-PCR. The following format was used: Similarly, the influence of spiking on the sensitivity of rt-PCR detection was evaluated by comparing detection of enterovirus (EV) and CMV in serial dilutions of cell culture supernatant media and a series of positive clinical samples spiked with T4-MS2 mix in either Two-step or One-step rt-PCR format (see Supporting Information S4). Table 1 were included in the study. The approval of the relevant Ethics Committee (IFR48, Marseilles, France) was obtained for investigating the benefit of using phage spiking, but individual consent from patients was not required since French national regulations under the term of Biomedical Research (Loi Huriet-Sérusclat (loi 881138)) indicate that the signature at the hospital entrance office warrants that all samples taken during hospitalization for diagnostic purposes are accessible for research (excluding human genetic research) without the specific consent of the patient. 2b. Spiking and rt-PCR assays. A 200ml volume of each 8,950 clinical specimens was spiked (according to the procedure described above) before extraction which was performed onto the MagNA Pure LC instrument (Roche) using the High Pure Viral Nucleic Acid Kit (DNA extraction) or the MagNA Pure LC RNA isolation High Performance kit (RNA extraction) according to manufacturer's recommendations. CSF, amniotic fluids, biopsy tissues, effusions and aqueous humor were processed for DNA+RNA extraction using the BioRobot EZ1 with the Virus Mini Kit v2.0 (both from Qiagen). Reverse transcription was performed for RNA viruses using the TaqMan Reverse Transcription Reagents kit (Roche) and random hexanucleotides as per manufacturer's instructions. For each specimen: (i) rt-PCR reactions were carried out according to medical prescription (Table 1) , and (ii) distinct rt-PCR reactions for detection of T4 or MS2 were performed under a 15 mL reaction format (7,5 mL of mastermix, 3 pmol of each primer and 1,2 pmol of probe) and a standard cycling protocol (50uC for 2 min, 95uC for 10 min and 45 cycles 95uC for 15 sec, 60uC for 1 min). 2c. Interpretation of results. For each series of T4 and MS2 rt-PCR, the mean Ct value and the standard deviation within the series were calculated. Each individual reaction was subsequently analysed as follows: (i) If the Ct value was equal to or lower than the mean Ct value of the series +1SD, it was recorded as ''correct detection of the phage'' (CDP), and associated with the absence of detectable inhibitor or technical problem while processing the corresponding sample. (ii) If the Ct value was higher than the mean Ct value of the series +1SD (or undetectable), it was recorded as ''inefficient detection of the phage'' (IDP), and associated with the presence of amplification inhibitor(s) or technical problem while processing the corresponding sample. N When IDP was associated with a positive PCR (detection of a pathogen), this result was validated despite the presence of inhibitors (this would not apply to the case in which quantification of viral load is necessary). N When IDP was associated with negative PCR detection results, a new assay was performed using a tenfold dilution of the nucleic acid extract. All negative results were considered unresolved (UNR). Positive results were validated. The detection of EV and CMV in serial dilutions of supernatant cell culture media or in series of positive clinical samples spiked with T4-MS2 mix was performed and the comparison of the Ct values showed that neither the addition of the phage mix itself (Wilcoxon test, p = 0.18 for CMV and 0.45 for EV), nor the presence of phage-specific primers and probe in the reaction mix (Wilcoxon test, p = 0.77 for CMV and 0.18 for EV) did interfere with the Ct value associated with viral detection in Two-step (Table 2 and 3) and One-step rt-PCR. During the period of study, 8,950 clinical samples were spiked with phages and tested: 7,397 for the presence of DNA viruses only, 337 for the presence of RNA viruses only and 1,216 for both, corresponding to a total of 45,530 results transmitted to the prescribers (see details in Figure 2 ). Inhibitors could be detected for all types of samples, but with highly variable rates (analysis performed for series including a minimum of 10 samples): less than 10% in the case of pleural effusions and sputum samples, genital and pharynx swabs, lymph nodes and placentas; 10% to 30% for EDTA whole blood, white blood cells, sera, biopsies, pericardial and peritoneal effusions, bronchial aspiration, CSF, amniotic fluids, others swabs, urine, aqueous humor and culture cells; 30% to 50% for plasma, BAL and stools; more than 50% for heparinized plasma and bone marrow. IDPs were significantly more frequently detected for MS2 (RNA) than T4 (DNA) in plasma, BAL and CSF samples (p,0.05, Khi2 test, Yates corrected). Conversely, IDPs were significantly more frequently detected for T4 than MS2 in stools (p,0.05, Khi2 test, Yates corrected). Finally, it was noted that clinical samples stored at 280uC prior extraction exhibited an IDP rate lower than 30%. Rt-PCR and rt-RT-PCR techniques are now widely used for the molecular diagnosis of viral infections. Our study was focused on the importance of quality controls for such diagnostic assays. We proposed a simple protocol for synthesizing specific external controls which combines standard techniques for obtaining quantified DNA or RNA positive controls. Importantly, these controls were designed to include an extrinsic sequence that contains a (very rare) cutting site for the NotI restriction enzyme. This provides a simple tool for using adapted and reproducible amounts of positive controls, and also for identifying PCR contamination due to carry over of the positive control. This detection can be performed by real time amplification using the specific Not I probe (Figure 1 [4, 5, 6, 7] . Our study confirms that the performance of 'home made' tests can be significantly improved by the used of phage-based internal controls, but, most importantly, shows that such controls can be used for routine virological diagnosis and usable for a variety of clinical samples. Here, clinical samples were spiked with both T4 and MS2 phages, allowing the detection of inhibitors for both DNA and RNA viruses. Thirty-six different types of clinical samples were tested (including various blood samples, cerebrospi- nal fluids, stools, respiratory samples, swabs, biopsies or effusion fluids) and a large number of samples were tested in the context of an hospital routine molecular virology laboratory. The use of phages as internal controls proved to be extremely versatile and could be adapted to a broad range of methods and pathogens. It was validated for both PCR and RT-PCR real time techniques, in simplex or multiplex format and, in the case of RT-PCR assays, one-step or two-step amplification formats. In addition, whilst the current report relies on probe-based real time amplification techniques, the method could also be conveniently adapted to a real time SYBR Green detection assay [8] . This is important and suggests that a strategy including phage-based internal controls can be implemented in diagnostic laboratories irrespective of the technical characteristics of the amplification methods used for routine tests. In our experience, the simplex format strategy (i.e. based on testing phages and viral pathogens in distinct amplification reactions) proved to be the most simple and costeffective for routine molecular diagnosis since it does not require the specific development of multiplex reactions and relies on a unique control reaction for DNA viruses and another for RNA viruses. Our study identified different frequencies of inhibitors according to the nature of the clinical samples. In samples such as CSF, sera or pharynx swabs, inhibitors were identified in ,10% and ,15-20% for DNA and RNA virus detection, respectively. These values are unexpectedly high, and imply that a significant number of samples with results believed to be ''negative'' in the absence of an internal control should be considered ''unresolved''. Specific samples such as heparinized blood (72,9% of inhibitors for DNA virus detection), bone marrow (73.8% of inhibitors for DNA virus detection) and stools (36.6% and 20.8% of inhibitors for DNA and RNA viruses detection, respectively) may also benefit from the detection of inhibitors and the identification of ''unresolved'' tests. It should be noted that the frequency of amplification inhibition not only relates to the nature of the sample tested, but is also intrinsically linked with the technical protocols used for the extraction and amplification of nucleic acids. In our experience, no major differences were observed when different silica column-or magnetic beads-based extraction techniques, or different commercialized PCR or RT-PCR kits, were tested (data not shown). However, a detailed analysis may reveal that specific techniques give better results when used for testing specific samples. We suggest that, in the future, phage-based internal controls may constitute cost-effective tools with which to measure the frequency of amplification inhibition in specific samples. Estimation of this parameter may become a major criterion for the evaluation of extraction (and to a lesser extent amplification) techniques. Finally, in the context of diagnostic virology laboratories, our study shows that standard extraction and amplification techniques used for the molecular diagnosis of human pathogens led to a significant proportion of 'unresolved' results, which cannot be identified if an internal control is not used. In the absence of an internal control, such samples are commonly identified as 'negative', which is with hindsight incorrect: in our hands, detecting amplification inhibition using phages as internal controls, and testing tenfold dilutions of the nucleic acid extracts demonstrated that some of these samples were actually positives. This cost-effective and convenient strategy can therefore be used for enhancing the quality of routine molecular diagnosis, but it may also be adapted in other contexts such as testing of large numbers of animal or environmental samples. Supporting Information S1 Preparation of DNA and RNA synthetic ECs.

Pandemic H1N1 2009 influenza virus with the H275Y oseltamivir resistance neuraminidase mutation shows a small compromise in enzyme activity and viral fitness

BACKGROUND: Resistance to the neuraminidase inhibitor oseltamivir can be conferred by a well-characterized mutation in the neuraminidase gene, H275Y. In human H1N1 viruses that circulated in the first years of the 21st century, this mutation carried a fitness cost and resistant viruses were rare. During the 2007–08 influenza season, oseltamivir-resistant viruses of H1N1 phenotype emerged and predominated. March 2009 saw the emergence of a novel H1N1 influenza pandemic. We examined whether the H275Y mutation affected neuraminidase enzyme activity or replication of the pandemic influenza virus. METHODS: Using reverse genetics we engineered the H275Y mutation into the neuraminidase of a 2009 pandemic H1N1 virus and assessed the ability of this enzyme to desialylate mono- and multivalent substrates. The growth kinetics of wild-type and mutant viruses were assessed in Madin–Darby canine kidney (MDCK) and fully differentiated human airway epithelial (HAE) cells. RESULTS: The presence of H275Y was associated with a 1.3-fold decrease in the affinity of the neuraminidase for a monovalent substrate and a 4-fold compromise in desialylation of multivalent substrate. This was associated with a fitness cost to viral replication in vitro, which only became apparent during competitive replication in the mucus-rich HAE culture system. CONCLUSIONS: The neuraminidase protein of pandemic influenza isolates tolerates the H275Y mutation and this mutation confers resistance to oseltamivir. However, unlike seasonal H1N1 viruses isolated since 2007, the mutation is not associated with any fitness advantage and thus is unlikely to predominate without further antigenic drift, compensating mutations or intense selection pressure.

March 2009 saw the emergence of a novel strain of influenza with the ability to transmit readily between humans. The rapid global spread of the new influenza A/H1N1 virus resulted in the first influenza pandemic in 40 years. 1 Two therapies are currently licensed to treat influenza infections: the M2 ion channel blocking adamantane drugs (amantadine and rimantadine) and the neuraminidase (NA) inhibitors (oseltamivir and zanamivir). Pandemic influenza A/H1N1 2009 virus (pH1N1) crossed into humans carrying a well-characterized amantadine-resistance mutation, S31N, within the M2 ion channel protein, rendering the adamantane class of antiviral drug ineffective against the virus. Initial isolates were susceptible to NA inhibitors, including oseltamivir and consequently this drug was used extensively in the treatment and prophylaxis of pandemic influenza. 2 Oseltamivir resistance emerged infrequently in pH1N1 2009 influenza viruses. 3, 4 In contrast, seasonal H1N1 influenza viruses from the 2007-08 season onwards were predominantly resistant to oseltamivir with resistance conferred by the H275Y mutation in the viral NA. 5 Traditionally, the H275Y mutation was associated with compromised viral fitness amongst H1N1 isolates. 6 However, isolates from the 2007 -08 season with this mutation suffered no attenuation. 7 It is likely that certain other sequence variations, such as the D344N (N1 numbering) change found within the NA gene of oseltamivir-resistant viruses from the 2007-08 season, counteract the decrease in enzyme function that H275Y confers. 8, 9 Residue 344 is tyrosine in most avian influenza NA genes but mutated to asparagine in two N1 viruses, which crossed from birds into mammals shortly thereafter (the 1918 pandemic virus and the Eurasian lineage swine H1N1 viruses). 10 During circulation in humans, an aspartic acid at residue 344 was eventually selected in seasonal H1N1 viruses. Mutation back to asparagine occurred in circulating human H1N1 viruses prior to the 2007 season and was associated with increased NA activity. 9 The NA enzyme of pH1N1 virus is derived from a Eurasian lineage H1N1 swine virus, and harbours asparagine at residue 344. This suggested that it may tolerate mutations such as H275Y that would concomitantly decrease NA activity and confer oseltamivir resistance. This study aimed to assess in vitro the effect of the H275Y mutation in the NA of a prototypic pH1N1 2009 isolate. Seven RNA segments of laboratory-adapted strain [A/Puerto Rico/8/34 (H1N1)] were combined with the NA of a pH1N1 strain [A/England/195/09 (pH1N1)] to rescue a pair of isogenic viruses that differed only at position 275 (wild-type PR8+E195 and mutant PR8+E195 H275Y ). Similarly a pair of isogenic viruses based on the whole genome of A/England/195/09 were rescued according to published methods. 11 The sialidase enzyme properties of both wild-type NA and the mutant H275Y NA were assessed using a fluorescent substrate [2 ′ -(4-methylumbelliferyl)-a-D-N-acetylneuraminic acid (MuNANA)] as previously described. 12 Enzyme kinetic experiments were performed on a standardized amount of each virus for which the NA metabolized 10 nmol substrate/h. A chicken erythrocyte elution assay, where NA is required to cleave sialic acid from multivalent cell surface moieties of the red blood cell, was established. Equal haemagglutination titres of each virus were diluted in 2-fold steps and mixed with PBS containing 1% chicken red blood cells in a V-bottomed microtitre plate and left at 48C for 1 h to determine the haemagglutination endpoint titre. The plate was then incubated for 2 h at 378C. The last dilution of virus at which haemagglutination was lost was defined as the desialylation endpoint titre. To assay virus growth kinetics, two isogenic viruses containing eight segments of pH1N1 (A/England/195/09) were grown on either Madin-Darby canine kidney (MDCK) cells (originally sourced from the European cell culture collection) or MucilAir TM human airway epithelial (HAE) cells (Epithelix-Sà rl) at a multiplicity of infection (moi) of 0.01. At 12, 24, 48 and 72 h post-infection (MDCK) or 24, 48, 72, 96 and 120 h post-infection (HAE), viruses were collected from the apical surface and subsequently plaqued on MDCK cells to assay for viral titre. In competition assays, the two viruses were mixed at a defined ratio of either 50 :50 or 80 : 20 (mutant:wild-type) and used to infect MDCK cells or HAE cells at a total moi of 0.01. After 72 h, the released virus was pyrosequenced to quantify the proportion of each genotype. Mutation H275Y in A/England/195/09 NA confers resistance to oseltamivir The N1 NA of pH1N1 that carried the H275Y mutation showed an 300-fold increase in oseltamivir IC 50 values (Table 1 ), but the mutation did not confer resistance to zanamivir. There was a 1.3-fold decrease in the binding affinity of the enzyme for the small MuNANA substrate resulting from the resistance mutation as indicated by the difference in K m values (Table 1) . Furthermore, the EC 50 of oseltamivir for virus containing wild-type NA was between 10 and 100 nM, whereas for the H275Y mutant this value was .10 mM since the virus still formed plaques at this concentration of drug (data not shown). Both recombinant viruses grew to equivalent high titres in MDCK cells and haemagglutinated chicken red blood cells with a geometric mean titre (GMT) of 128. Following incubation of the microtitre plates at 378C, wells containing high titres of virus showed haemagglutination reversal. This is accounted for by the digestion of the sialic acid receptor from the erythrocyte surface by active NA enzyme. The ability of the wild-type pH1N1 to elute virus from red blood cells was 4-fold greater than that of the H275Y isogenic mutant ( Table 1) . The ability of isogenic A/England/195/09 recombinant viruses that differed only at residue 275 in the NA to infect and spread in MDCK or in differentiated HAE cells was assessed by Recombinant influenza viruses containing the NA from representative pandemic influenza A/England/195/09 were assessed for their ability to catabolize mono-and multivalent substrates. The MuNANA substrate was used to determine IC 50 values for two NA inhibitors and assess enzyme K m values. IC 50 and K m values are given as the mean of two determinations, and the SEM is indicated for K m values. For red cell elution, chicken erythrocytes were first mixed in equal volumes with 2-fold serially diluted virus and incubated on ice to determine haemagglutination titres. The microtitre plate was then moved to 378C to allow NA to cleave sialic acid residues and reverse haemagglutination. The geometric mean titre derived from duplicate wells shown is the last dilution at which an effect was seen. A lower desialylation titre indicates a reduced ability of the NA to cleave sialic acid and reverse haemagglutination. Oseltamivir-resistant pandemic influenza A H1N1 virus 467 JAC performing infections at low multiplicity. Both viruses replicated efficiently in either cell culture system. Although there was an 0.75 log 10 decrease in the amount of mutant virus released from HAE cells at 72 h post-infection, this was not statistically significant (Figure 1a and b) . The two viruses were used to co-infect triplicate wells of either MDCK or HAE cells at defined ratios and after 72 h released virus was analysed for the presence of H275Y mutation. Pyrosequence analysis showed that in MDCK cells there was no growth advantage for virus with either H or Y at residue 275 ( Figure 2 ). The output virus from all three wells of MDCK cells contained the same mixture of wild-type and mutant genomes as the input. Thus, the mean percentage of wild-type genotype (H275) following inoculation of the 50: 50 mixture was 50.3% and after inoculation of the 80 : 20 mutant:wild-type mixture this value was 21.4%. In contrast, after 72 h of propagation in HAE cells where mucus could accumulate, the wild-type oseltamivirsusceptible H275 genotype accumulated to a higher degree in two out of three wells than did the drug-resistant variant (Figure 2c) . The mean percentage of wild-type genome in the mixture after 72 h was 59%. The standard deviation of the pyrosequencing assay using this primer set was calculated by performing triplicate runs on six different extracted mixtures and was measured at ,1% (data not shown). Thus, the observed enrichment of wild-type genome after propagation in HAE cells was not a result of assay variation but rather indicative of a selective advantage of H275 in this system. The biological consequences of the H275Y mutation in the NA gene of pH1N1 influenza virus, which confers resistance to oseltamivir, are important because the drug is a first-line treatment for patients who present with pandemic influenza infection. Drug resistance was already observed in infected individuals in the community and in the clinic during the first and second waves of the swine flu pandemic 4,13 although it did not spread widely. Whether oseltamivir-resistant pH1N1 viruses might disseminate in subsequent waves through the community is key to future public health planning. The NA gene of the new pandemic H1N1 virus was acquired from Eurasian swine influenza H1N1 virus, a lineage of virus that crossed from bird to pigs in the late 1970s. In this background, the small compromise in enzyme affinity for sialic acid substrate (observed by a 1.3-fold increase in the NA K m value) and the decrease in cell surface expression that results from the H275Y mutation, 14 had no effect on virus growth in MDCK cell culture. On the other hand, in vivo NA must cleave complex substrates to mediate virus release from an infected airway cell and gain access through a complex layer of mucins to the new target cell. Subtle decreases in NA activity or cell surface expression may have more profound consequences in the airway than in monoculture. To probe this, we tested the NA activity of the mutated virus in assays that presented large sialylated substrates: in a red cell elution assay we detected a 4-fold compromise in the ability of the virus with H275Y mutation to mediate desialylation of chicken erythrocytes, although the biological significance of this assay is not entirely clear. In differentiated cultures of human airway cells, the difference detected in replication of wild-type and mutant virus was not statistically significant on either of two separate occasions (Figure 1 and data not shown) . However, in competition assays in HAE cultures, the wild-type virus out-competed growth of the drug-resistant strain suggesting that, in the absence of drug, the 275Y motif carries a fitness cost in the environment of the human airway. In both recent seasonal H1N1 strains and in H5N1 highly pathogenic viruses, mutations that increase the NA activity, protein stability or cell surface transport likely compensate for the effects of the mutation at 275 that would otherwise decrease the function of the enzyme. 8, 9, 14, 15 Examples of such mutations are D344N, 9 and V235M and/or R223Q. 14 Neither of Brookes et al. the latter two mutations currently exist in the NA protein of pH1N1 isolates, but further circulation of the pH1N1 virus in humans may select for these or other NA or haemagglutinin (HA) mutations that better prime the virus to accommodate or even select for the H275Y mutation. For contemporary pH1N1 viruses, the cost or advantage of drug resistance is so subtle that different groups have come to different conclusions about its relevance. In a hospital setting there have been reports that suggest that patient-to-patient transmission of drug-resistant virus has occurred amongst immunocompromised individuals. 16 Hamelin et al. 17 showed that oseltamivir-resistant pH1N1 virus was equally virulent as its wild-type counterpart in mice and ferrets and did transmit to co-housed animals, though they did not assess droplet transmission. Seibert et al. 18 used guinea pigs and ferrets in both contact and droplet transmission studies and concluded that the drug-resistant mutant could potentially circulate in the community. However, a detailed analysis of their data reveals that for one of the viruses they tested, there was a 2 day delay in transmission to half the contact-exposed guinea pigs and for the other strain of pH1N1 there was a reduction in droplet transmission to 88% rather than 100% of exposed sentinels. The ferret experiments were conducted with n¼ 1 so it is difficult to be sure of their significance. Similarly, in the manuscript from Kiso et al., 19 the conclusion is again that the H275Y mutant transmits through the air between ferrets, but there is a 2 day delay in transmission of one of the mutant strains of virus studied. Conversely, Duan et al. 20 found that the drug-resistant virus did not transmit between ferrets by the respiratory droplet route and that in co-infected animals, the wild-type virus outgrew the resistant mutant and was uniquely transmitted to contact animals. Thus, the current picture from animal experiments is confused and discrepant. This might be partly due to the use of different strains of pH1N1 virus as well as different experimental protocols used by the various investigators. Anecdotal evidence from the clinic shows that, in most instances, contemporary drug-resistant variants of pH1N1 were replaced by drug-susceptible variants when the selective pressure of oseltamivir was removed, suggesting that wild-type isolates are fitter in vivo in humans. 21 Whether the subtle fitness deficit reported here for one particular strain of drug-resistant mutant pH1N1 virus explains the epidemiological observation that mutant virus has not circulated through the community is not clear, since many other factors including heterogeneous mixing of populations and stochastic effects may influence whether a particular virus mutant predominates. Nonetheless the virus competition assay conducted in HAE cultured cells described here offers an alternative biologically relevant model as a useful adjunct to animal studies and this system may more accurately reflect the environment in which virus replicates in otherwise healthy humans. Information from a variety of model systems should be combined to guide the appropriate use of oseltamivir. Such knowledge clearly needs to be revised specifically for each novel influenza virus that emerges either as a seasonal strain by drift or as a pandemic virus by antigenic shift.

Ketamine inhibits tumor necrosis factor secretion by RAW264.7 murine macrophages stimulated with antibiotic-exposed strains of community-associated, methicillin-resistant Staphylococcus aureus

BACKGROUND: Infections caused by community-associated strains of methicillin-resistant Staphylococcus aureus (CA-MRSA) are associated with a marked and prolonged host inflammatory response. In a sepsis simulation model, we tested whether the anesthetic ketamine inhibits the macrophage TNF response to antibiotic-exposed CA-MRSA bacteria via its antagonism of N-methyl-D-aspartate (NMDA) receptors. RAW264.7 cells were stimulated for 18 hrs with 10(5 )to 10(7 )CFU/mL inocula of either of two prototypical CA-MRSA isolates, USA300 strain LAC and USA400 strain MW2, in the presence of either vancomycin or daptomycin. One hour before bacterial stimulation, ketamine was added with or without MK-801 (dizocilpine, a chemically unrelated non-competitive NMDA receptor antagonist), APV (D-2-amino-5-phosphono-valerate, a competitive NMDA receptor antagonist), NMDA, or combinations of these agents. Supernatants were collected and assayed for TNF concentration by ELISA. RESULTS: RAW264.7 cells exposed to either LAC or MW2 in the presence of daptomycin secreted less TNF than in the presence of vancomycin. The addition of ketamine inhibited macrophage TNF secretion after stimulation with either of the CA-MRSA isolates (LAC, MW2) in the presence of either antibiotic. The NMDA inhibitors, MK-801 and APV, also suppressed macrophage TNF secretion after stimulation with either of the antibiotic-exposed CA-MRSA isolates, and the effect was not additive or synergistic with ketamine. The addition of NMDA substrate augmented TNF secretion in response to the CA-MRSA bacteria, and the addition of APV suppressed the effect of NMDA in a dose-dependent fashion. CONCLUSIONS: Ketamine inhibits TNF secretion by MRSA-stimulated RAW264.7 macrophages and the mechanism likely involves NMDA receptor antagonism. These findings may have therapeutic significance in MRSA sepsis.

Infections caused by community-associated strains of methicillin-resistant Staphylococcus aureus (CA-MRSA) present a major public health problem because of recent increases in the incidence of these infections [1, 2] . In a 2007 report, the Centers for Disease Control concluded that Staphylococcus aureus is now the most important cause of serious and fatal infection in the United States [3] . The prototypical USA400 strain, MW2, (CDC nomenclature for this strain of MRSA) was first isolated in 1999 from a Midwest child with fatal CA-MRSA pneumonia [4] . In 2003, the prototypical USA300 CA-MRSA strain, LAC, was isolated from Los Angeles County patients with skin and soft tissue infections, severe pneumonia and sepsis. Recently, concerns about CA-MRSA infections were heightened after reports of severe invasive staphylococcal infections in some patients infected with the novel 2009 H1N1 influenza A virus [5, 6] . CA-MRSA isolates express many virulence factors [7, 8] , including several cytolysins: α-toxin, γ-toxin, Panton-Valentine leukocidin (PVL), phenol-soluble modulins (PSMs), δ-toxin and, unlike traditional hospital-associated (HA-MRSA) isolates, may express superantigens such as TSST-1 [9] . These bacterial components can stimulate massive cytokine release and lead to septic shock, acute respiratory distress syndrome (ARDS) and death. It is likely that strategies designed to modulate the excessive and prolonged host inflammatory response could improve the outcome of fulminant MRSA infections. Monocytes and macrophages play important roles in host defense against staphylococci and other pyogenic bacteria [10] , but excessive systemic or local production of inflammatory mediators by macrophages could be deleterious in patients with severe staphylococcal infections. We previously reported that RAW264.7 murine macrophages exposed to any of a series of six pediatric clinical isolates of S. aureus (two CA-MRSA, two HA-MRSA, and two methicillin-susceptible strains) in the presence of daptomycin (vs. vancomycin) secreted less TNF and accumulated less inducible nitric oxide synthase (iNOS) protein [11] . Vancomycin is a cell-wall active antibiotic that triggers bacterial lysis; it is the antibiotic most commonly used to treat severe MRSA infections in children [12] . Daptomycin is a novel antibiotic that is rapidly bactericidal against staphylococci but does not appear to cause rapid bacterial lysis; the mechanism of its action is not certain but it is reported to trigger depolarization of the bacterial membranes and inhibition of both DNA and RNA synthesis [13, 14] . The rapid lysis of staphylococci, streptococci and other pyogenic bacteria exposed to cell-wall active antibiotics such as beta-lactams and vancomycin results in exaggerated release of bacterial products and an augmented and potentially harmful host inflammatory response [15, 16] . Therefore, optimal treatment of sepsis and other severe bacterial infections might include the use of antibiotics and/or other medications that blunt the host inflammatory response and dampen the cytokine cascade [16] . Ketamine is one of the recommended anesthetics in pediatric septic shock [17] [18] [19] , which is frequently caused by staphylococci [12, 20] . The reasoning for ketamine's use in staphylococcal septic shock is its blood pressure supporting effect. It increases cardiac output and blood pressure, possibly via a catecholamine release mechanism [17, 21] . Some data suggest that ketamine has anti-inflammatory effects [22] [23] [24] [25] . For example, it has been reported that ketamine suppresses macrophage TNF secretion in response to Gram-negative bacterial LPS in vivo and in vitro [22, 23, 25] . There is also one report that ketamine suppresses TNF production by human whole blood in vitro after exposure to staphylococcal enterotoxin B [24] . The mechanisms responsible for the anti-inflammatory effects of ketamine are not known [22] [23] [24] [25] .The present study examined the hypothesis that ketamine could suppress macrophage TNF production in response to whole bacteria, in this case clinical isolates of methicillin-resistant Staphylococcus aureus (MRSA). Given the important role of TNF in sepsis [26] [27] [28] [29] , and the importance of staphylococcal sepsis in children, such suppression could have a therapeutic impact. Although membrane-bound Toll-like receptors (TLR2 and TLR4) are essential for lipopolysaccharide (LPS)induced TNF production [30] , this is not the case for Staphylococcus aureus. Because S. aureus is able to "attack" or form pores in macrophages, TNF secretion occurs even in the absence of TLR 2 and TLR4 sensors (possibly via Nod1 and Nod2, intracytoplasmic sensors of peptidoglycan-derived muropeptides) [31] . Therefore, another mechanism independent of Toll-like receptors must exist for ketamine's anti-inflammatory action, at least in staphylococcal infections. We also tested the effects of two chemically unrelated NMDA receptor antagonists, the anti-convulsant MK-801 (dizocilpine) [32, 33] , a non-competitive inhibitor of NMDA receptors, and APV (D-2-amino-5-phosphonovalerate), a competitive NMDA receptor antagonist [34, 35] , as well as the NMDA substrate itself, on macrophage TNF secretion in response to antibiotic-treated CA-MRSA bacteria. For these studies, we utilized two well-characterized clinical isolates: LAC (Los Angeles County), representative of the USA300 group of organisms and closely related to the dominant CA-MRSA clone associated with soft tissue infections and serious invasive disease in the Memphis area [1] , and MW2, a clinical isolate from a midwestern child with fatal CA-MRSA sepsis [4] , representative of the USA400 group of organisms that constitute the other main lineage of CA-MRSA isolates in the United States. Bacteria were grown to late logarithmic phase at 37°C in tryptic soy broth (Becton Dickinson and Co., Sparks, MD) and washed three times in endotoxin-free phosphate-buffered saline. Concentrations were determined by colony counts. A range of concentrations of bacteria (10 5 -10 7 CFU/mL) was studied, based upon our previously published data with other CA-MRSA strains [11] and our preliminary experiments using LAC and MW2 (data not shown). Minimum inhibitory concentrations (MICs) for these strains were determined by the microbiology laboratory at Le Bonheur Children's Hospital using the E-test method: both strains were fully susceptible to vancomycin and daptomycin (LAC: MIC vancomycin 1.0 μg/mL; daptomycin 0.75 μg/mL; MW2: MIC vancomycin < 0.5 μg/mL; daptomycin 0.75 μg/mL). Cell culture RAW264.7 murine macrophage-like cells were purchased from the ATCC and cultured in Dulbecco's modified Eagle's medium (Mediatech Inc., Herndon, VA) supplemented with 10% fetal bovine serum (HyClone, Logan, UT) and 2 mM glutamine (GIBCO, Carlsbad, CA). Experiments were done in 24-well tissue culture plates (Becton Dickinson, Lincoln Park, NJ) with 1 × 10 6 cells per well. Either vancomycin or daptomycin was added to the cell cultures immediately before the addition of live staphylococci (10 5 -10 7 CFU/mL). Cells were then incubated for 18 hours. Daptomycin was obtained from Cubist Pharmaceuticals (Lexington, MA). Vancomycin was purchased via the Department of Pharmacy at Le Bonheur Children's Hospital (LBCH) from Hospira (Lake Forest, IL). Clinically achievable concentrations of each of the antibiotics, as previously tested in our laboratory [11] , were used (20 μg/mL). These experiments were repeated in parallel in the presence of ketamine (100 μM) and/or MK-801 (dizocilpine, 150 μM), APV (D-2-amino-5-phosphonovalerate, 300 μM ("low") or 3 mM ("high"), or NMDA (30 μM). The modulation of MRSA-stimulated macrophage TNF production by ketamine was subsequently examined also at a range of concentrations of 10 μΜ, 50 μΜ, 100 μΜ and 150 μΜ. The selected concentration (100 μM) is based on the achievable anesthetic concentrations [36] [37] [38] [39] and on the pre-existing literature related to ketamine's TNF suppressive effect on murine macrophage models when stimulated by LPS [23] [24] [25] 27] . The concentrations for the other factors were selected from the available literature, MK-801 [40] [41] [42] , APV and NMDA [32] have previously been studied in cell culture models and have been shown to not cause cytotoxicity at the tested concentrations. Ketamine and/or MK-801 or APV or NMDA were added to the macrophage cultures one hour prior to bacterial challenge. The source of ketamine was Ketalar ® , a racemic mixture (1:1) of optically active isomers (R and L) of this drug, purchased from the LBCH pharmacy. Emphasis in the experiment was placed on correlation with the clinical situation; thus racemic ketamine, the most commonly clinically used product, was selected. Dizocilpine (MK-801), APV and NMDA were purchased from Sigma Chemical Co. (St. Louis, MO). After incubation, cell-free supernatants were collected and assayed for TNF concentrations by using a solid-phase sandwich enzyme-linked immunosorbent assay as specified by the manufacturer (eBioscience, San Diego, CA). TNF is a key cytokine produced by macrophages during MRSA stimulation. In our preliminary studies, we also measured secretion of other cytokines and found that IL-1, IL-6, and NO secretion were strongly correlated with TNF secretion in response to these bacteria (r 2 = 0.84, 0.87 and 0.93, respectively). We focused on TNF secretion for these studies. The tested concentrations of vancomycin, daptomycin, ketamine, MK-801, APV, and NMDA had no effect on the viability of the RAW264.7 cells, as determined by visual inspection of the monolayer, low power microscopic inspection of the monolayer and exclusion of 0.2% trypan blue dye. For the single comparison experiments (ketamine or MK 801 or APV), TNF secretion measurements were validated with an average of at least three well replicates and each of the experiments was repeated at least three times (a total of at least nine samples). The four preliminary runs and all the exposures (total of 16) where the inocula were different from 10 5 to10 7 CFUs/mL at the verifying colony count were excluded from the final analysis. Experiments with different exposure times (6, 10, 14, 24 hours) were conducted to determine whether the inhibition increased over time. In the multiple comparison experiments (ketamine and MK 801 synergistic action), TNF was measured from at least four well replicates. All experiments were performed separately for LAC and for MW2 MRSA strains. There is an intrinsic experimental variation of absolute values of TNF production (up to 25%) because of cell culture and macrophage growth characteristics. The design was composed of factorial multiple measurements and the results were analyzed according to a mixed linear model, (GLIMMIX) SAS 9.2 (SAS Institute, Cary, NC) and R 2.9.1 and ggplot2 software. We set pre-planned (a priori) contrasts, i.e., we set all our comparisons in advance of multiple setting experiments. Significant differences were presumed at a probability value of p < 0.05. The results were graphed using error bars with 95% confidence intervals. Differences in the means were estimated either with asymptotic techniques for normally distributed data or bootstrapping techniques for non-normally distributed data. CA-MRSA strains MW2 and LAC stimulated less TNF secretion by RAW264.7 murine macrophages in the presence of daptomycin than in the presence of vancomycin As previously observed with two USA300 CA-MRSA strains isolated from Memphis children with invasive staphylococcal infections [11] , macrophages exposed to either of the two prototypical CA-MRSA strains studied (the USA300 strain, LAC, or the USA400 strain, MW2) secreted significantly less TNF in the presence of daptomycin as compared with vancomycin (more than 50% reduction in each strain; Figure 1 ). Macrophage TNF secretion in response to MW2 was 34,535 ± 1,536 pg/ mL in the presence of vancomycin and 15,377 ± 1,267 pg/mL in the presence of daptomycin, a reduction of 55%, significant at p < 0.05. Similarly, macrophage TNF secretion in response to LAC in the presence of vancomycin was 33,345 ± 1,535 pg/mL, and 14,432 ± 1,536 pg/mL in the presence of daptomycin, a reduction of 57%, significant at p < 0.05. We previously reported similar findings in six S. aureus clinical isolates (including two pediatric CA-MRSA isolates of the USA300 group), suggesting that this effect of daptomycin is conserved in many different S. aureus isolates. The addition of ketamine (100 μΜ) to macrophage cell cultures inhibited TNF secretion in response to vancomycin-or daptomycin-exposed CA-MRSA isolates ( Figure 2) . The effect was similar on both strains, LAC and MW2, in the presence of vancomycin (upper panel) or daptomycin (lower panel). In the initial experiments we analyzed the effect of one hour pre-incubation with ketamine on the macrophage response to vancomycin-exposed CA-MRSA bacteria (MW2 and LAC). In response to vancomycinexposed MW2, pre-incubation with ketamine reduced macrophage TNF secretion by approximately 29% (p < 0.05), i.e., from 33,085 ± 867 pg/mL to 23,347 ± 862 pg/mL. Pre-incubation with ketamine led to a similar reduction (25%; p < 0.05) in macrophage TNF secretion response after stimulation with vancomycinexposed LAC (from 28,365 ± 735 pg/mL to 21,432 ± 736 pg/mL). We next studied the effect of ketamine pre-incubation on macrophage TNF secretion after stimulation with daptomycin-exposed MW2 or LAC. Once again, the addition of ketamine resulted in significant inhibition of macrophage TNF secretion in response to MW2 (23,185 ± 1,267 pg/mL to 17,354 ± 853 pg/mL, a reduction of approximately 25%; p < 0.05) or LAC (approximately 18% reduction, p < 0.05; Figure 2 ). Adding ketamine after the MRSA inocula did not alter the response. Figure 1 The CA-MRSA isolates LAC (USA300) and MW2 (USA400) stimulated less TNF secretion by RAW264.7 murine macrophages when exposed to daptomycin (DAP) than when exposed to vancomycin (VAN). LAC or MW2 were added to RAW264.7 cells at a final concentration of 10 5 to 10 7 CFU/mL (retrospective confirmation) in the presence of either vancomycin or daptomycin at 20 μg/mL. Cells were incubated for 18 hours; supernatants were collected and analyzed for TNF content by an enzyme-linked immunosorbent assay (ELISA). Results are depicted as means with 95% confidence intervals shown as "error bars" (See Methods). The "*" indicates significance at p < 0.05. Control represents the mean TNF macrophage production by macrophages not stimulated with bacteria. MRSA MW2 and LAC exposed to Ketamine One hour prior to stimulation, ketamine (100 μM) was added to the indicated wells. Cells were then incubated for 18 hours; supernatants were collected and analyzed for TNF content by ELISA. Results are depicted as means with 95% confidence intervals shown as "error bars" (See Methods). The "*" indicates significance at p < 0.05. Control represents the mean TNF macrophage production by macrophages not stimulated with bacteria. The NMDA inhibitor MK-801 (dizocilpine) inhibited macrophage TNF secretion after stimulation with antibiotic-exposed CA-MRSA strains Pre-incubation of RAW264.7 cells for one hour with the NMDA receptor antagonist, MK-801 (150 μΜ), also inhibited TNF secretion by these cells after stimulation with antibiotic-exposed CA-MRSA strains (MW2 or LAC, Figure 3 ). In response to stimulation with MW2 in the presence of vancomycin, pre-incubation with MK-801 significantly inhibited TNF secretion by these cells, i.e., from 32,407 ± 1,188 pg/mL to 23,337 ± 1,272 pg/mL (approximately 28% reduction; p < 0.05, Figure 3 , upper panel). MK-801 also inhibited macrophage TNF secretion in response to vancomycin-exposed LAC, causing a 34% reduction (Figure 3, upper panel) . Pre-incubation with MK-801 also significantly inhibited macrophage TNF secretion in response to daptomycin-treated MW2 or LAC (Figure 3 , lower panel). In response to stimulation with MW2 in the presence of daptomycin, pre-incubation with MK-801 inhibited TNF secretion by these cells by approximately 26% (from 22,305 ± 648 pg/mL to 16,437 ± 642 pg/mL, p < 0.05). MK-801 inhibited macrophage TNF secretion in response to daptomycin-exposed LAC by approximately 33% (from 22,164 ± 864 pg/mL to 14,647 ± 832 pg/mL, p < 0.05). Pre-incubation of RAW264.7 cells with combinations of MK-801 and ketamine did not affect the magnitude of inhibition of macrophage TNF secretion observed in the presence of ketamine (or MK-801) alone. Figure 4 depicts results for macrophages stimulated with vancomycin-or daptomycin-exposed MW2; responses to antibiotic-exposed LAC were similar (data not shown). One hour prior to stimulation, either ketamine at 100 μM, MK-801 at 150 μΜ, or both were added to the indicated wells. Cells were then incubated for 18 hours; supernatants were collected and analyzed for TNF content by ELISA. Lane 1 (control) represents the mean TNF production by macrophages not stimulated with bacteria. The mean includes wells exposed to ketamine, MK-801, both ketamine and MK-801, and neither. In the absence of bacteria, TNF secretion was minimal and was not affected by ketamine and/or MK-801. NMDA augments macrophage TNF secretion in response to antibiotic-treated CA-MRSA bacteria: both ketamine and a competitive NMDA receptor antagonist, APV, block this effect We further examined the role of NMDA receptors in modulating the macrophage TNF response to the CA-MRSA bacteria by studying the effects of a competitive NMDA receptor antagonist, APV, and the effects of the NMDA substrate itself ( Figure 5 ). We found that APV (at either 300 μM or 3 mM) also inhibited macrophage TNF secretion in response to vancomycin-exposed MW2 (p < 0.05, Figure 5 ). The magnitude of the inhibition was comparable to that observed with either ketamine or MK-801 (and, as in the case of MK-801, was not additive or synergistic with ketamine). Furthermore, the addition of the NMDA substrate (30 μM) resulted in a marked augmentation of the macrophage TNF response to the antibiotic-treated CA-MRSA bacteria (p < 0.05), and this effect was blocked by ketamine and by the competitive NMDA receptor antagonist, APV ( Figure 5 ). We next studied the effects of a range of concentrations of ketamine and found that inhibition of macrophage TNF secretion in response to vancomycinexposed LAC or MW2 was consistently observed at concentrations of ketamine at the lowest concentration tested (10 μM) and was greater at concentrations of 50 -150 μM ( Figure 6 ). We also examined the kinetics of inhibition of macrophage TNF secretion by incubating RAW264.7 cells for 6, 10, 14, 18 and 24 hours after exposure to ketamine at a concentration of 100 μM 1 hour prior to stimulation with vancomycin-exposed LAC or MW2. We found that the magnitude of suppression of TNF secretion was similar at all times studied (Figure 7 ). We found that exposure of murine macrophages to ketamine inhibited TNF secretion by 18-34% after stimulation with CA-MRSA bacteria in the presence of antibiotics. The magnitude of the effect was comparable in response to both MW2 (USA400) and LAC (USA300) bacteria and was similar in the presence of either vancomycin (a lytic antibiotic associated with a greater TNF response to the bacteria) or daptomycin (a non-lytic antibiotic associated with a blunted TNF response to the bacteria). Our data suggest that ketamine administration to macrophages stimulated by CA-MRSA is associated with blunting of the TNF response to these virulent pathogens, and suggest that these findings may have therapeutic significance in MRSA sepsis. Furthermore, these data confirm and extend our previous observations that CA-MRSA bacteria exposed to daptomycin (versus vancomycin) trigger less TNF secretion by macrophages. The potentially beneficial antiinflammatory effects of daptomycin and ketamine were additive (Figures 2, 3) . Bacteria were added at a final concentration of 10 5 to10 7 CFU/mL (retrospective confirmation) in the presence of vancomycin at 20 μg/mL. One hour prior to stimulation, APV ("low" concentration of 300 μM or "high" concentration of 3 mM), ketamine (100 μM), or NMDA (30 μM) were added, alone or in combination, as indicated. Cells were then incubated for 18 hours; supernatants were collected and analyzed for TNF content by ELISA. The control lane represents the mean TNF macrophage production by macrophages not stimulated with bacteria. The mean includes wells exposed to APV, ketamine, or NMDA alone or in combination. In the absence of bacteria, TNF secretion was minimal and was not affected by APV, ketamine, or NMDA. Lanes 0-9 depict mean TNF secretion by macrophages exposed to vancomycin-treated MW2 alone (lane 0) or in the presence of the indicated concentrations of APV, ketamine, and/or NMDA (lanes 1-9). TNF secretion was reduced by approximately 30-40% when macrophages were pre-incubated with APV, ketamine, or APV + ketamine (lanes [1] [2] [3] [4] [5] . The magnitude of inhibition by ketamine and high-dose APV was similar and there were no additive or synergistic effect observed with combinations of ketamine and APV. Addition of NMDA (30 μΜ) led to a substantial increase in the amount of TNF secreted in response to the MW2 strain (lane 9), and this augmented response was blocked by both APV and ketamine. The "*" on "0.Control" and "8.NMDA+lo_APV" bars indicates significance at p < 0.05. The "**" on "0.Control_MW2" and "9.NMDA" bars indicates differences between the pretreated wells, and that TNF production after MRSA stimulation with NMDA substrate (9.NMDA) is significantly higher than that at the baseline MRSA stimulation (0.Control_MW2) at p < 0.05. An improved understanding of the pathogenesis of sepsis and other life-threatening infections caused by CA-MRSA bacteria could expedite the development of novel strategies for the diagnosis, treatment, and/or prevention of these serious infections. CA-MRSA infections often are associated with severe and prolonged host inflammatory responses [43] [44] [45] [46] . Prompt antibiotic treatment of these and other serious bacterial infections is indicated, but paradoxically has the potential to trigger excessive release of bacterial products and the subsequent augmentation of the host inflammatory response [15, 16] . Macrophages are important sources of many of the proinflammatory cytokines (including IL-1β, IL-6, IL-8, IL-12, and TNF) secreted in response to staphylococci and other Gram-positive bacteria [15, 16, 41] . Although the cytokine cascade is essential for normal host defense, excessive or inappropriate inflammation can be harmful. Therefore we need an improved understanding of these interactions in order to develop better adjunctive therapies for patients with severe bacterial infections. In a previous study, we found that exposure of either of two CA-MRSA strains isolated from Memphis children (or any of four other S. aureus isolates from children with invasive staphylococcal infections) to daptomycin (compared with vancomycin) led to a less pronounced macrophage inflammatory response, characterized by diminished secretion of TNF and reduced accumulation of the inducible nitric oxide synthase (iNOS) [11] . In this study, we found that this differential effect of daptomycin (versus vancomycin) was also observed when macrophages were stimulated with either of the two prototypical CA-MRSA strains most widely studied today: the USA400 isolate, MW2, and the USA300 isolate, LAC. Importantly, ketamine pre-incubation inhibited macrophage TNF secretion in response to both CA-MRSA strains in the presence of daptomycin as well as in the presence of vancomycin, and the greatest suppression of TNF secretion was noted in the presence of both daptomycin and ketamine. The mechanism(s) responsible for the anti-inflammatory properties of ketamine are not known, but its neurological and psychotropic actions are believed primarily to be mediated by antagonism of NMDA receptors [21, 47] . Glutamate is the brain's primary excitatory neurotransmitter. NMDA receptors are found in many cell Bacteria were added at a final concentration of 10 5 to10 7 CFU/mL (retrospective confirmation) in the presence of vancomycin at 20 μg/mL. The ketamine concentration was 100 μM. Results are depicted as percentile reduction with 95% confidence intervals, i.e., the percent of TNF reduction at the specific exposure time that occurs in comparison to inoculation without ketamine at the same time. The "*" indicates statistically significant difference at p < 0.05. types, including blood lymphocytes, lung macrophages, and multiple hematopoietic precursors in bone marrow cells [40, 42, 47, 48] . Both ketamine and the chemically unrelated anticonvulsant dizocilpine (MK-801) are noncompetitive antagonists of the NMDA receptor, one of the three known glutamate receptors [32, 33, 47] . APV is a competitive inhibitor of the classical NMDA receptor and acts on the NR2 component of the receptor (30, 33) . We found that MK-801 and APV also inhibited macrophage TNF secretion in response to antibiotictreated MW2 or LAC cells. The magnitude of the inhibition by MK-801 (approximately 30%) and APV (25-35%) was comparable to that observed with ketamine (18-34%), and combinations of MK-801 and ketamine or of APV and ketamine did not exhibit additive or synergistic inhibition of TNF secretion. Furthermore, adding NMDA led to augmented macrophage TNF secretion in response to antibiotic-treated CA-MRSA bacteria, and the NMDA receptor antagonist, APV, blocked this effect. The suppression of TNF induced by ketamine was observed across a range of concentrations and throughout the incubation period. Our study has its limitations. To translate the present findings, we are currently working on a clinical model to assess the clinical significance of ketamine's anti-inflammatory effects in patients with bacterial sepsis. Although studies of the effect of ketamine on macrophage responses to purified bacterial components such as Gram-negative lipopolysaccharide (LPS) or Gram-positive lipoteichoic acid (LTA) are instructive [23, 24, 49] , we argue that characterization of the macrophage responses to whole organisms is more likely to provide clinical insights. Indeed, the pioneering experiments of Carswell and Old that identified TNF used whole bacteria as stimuli in macrophage sepsis simulation settings [49] , and we have previously demonstrated that macrophage responses to live, antibiotic-treated staphylococci serve as a powerful model system. Furthermore, the model examines the effect of ketamine only in the presence of antibiotics (either vancomycin or daptomycin). In practice, this is a common clinical scenario. Our data suggest that clinically achievable concentrations of both ketamine and daptomycin could potentially inhibit the excessive macrophage inflammatory response that is observed in patients with severe staphylococcal infections. In the battle of sepsis everything counts. Adjunctive therapies of sepsis are greatly needed. Studies in animal models and clinical trials will be required to determine whether the anti-inflammatory effects of ketamine and/or other agents that block NMDA receptors could be beneficial in the treatment of severe staphylococcal infections.

Avipoxviruses: infection biology and their use as vaccine vectors

Avipoxviruses (APVs) belong to the Chordopoxvirinae subfamily of the Poxviridae family. APVs are distributed worldwide and cause disease in domestic, pet and wild birds of many species. APVs are transmitted by aerosols and biting insects, particularly mosquitoes and arthropods and are usually named after the bird species from which they were originally isolated. The virus species Fowlpox virus (FWPV) causes disease in poultry and associated mortality is usually low, but in flocks under stress (other diseases, high production) mortality can reach up to 50%. APVs are also major players in viral vaccine vector development for diseases in human and veterinary medicine. Abortive infection in mammalian cells (no production of progeny viruses) and their ability to accommodate multiple gene inserts are some of the characteristics that make APVs promising vaccine vectors. Although abortive infection in mammalian cells conceivably represents a major vaccine bio-safety advantage, molecular mechanisms restricting APVs to certain hosts are not yet fully understood. This review summarizes the current knowledge relating to APVs, including classification, morphogenesis, host-virus interactions, diagnostics and disease, and also highlights the use of APVs as recombinant vaccine vectors.

Avipoxviruses (APVs) are among the largest and most complex viruses known. APVs belong to the Chordopoxvirinae subfamily of the Poxviridae family [1] . They infect and cause diseases in poultry, pet and wild birds of many species which result in economic losses to the poultry industry. Infections have also been reported in a number of endangered species or species in captivebreeding recovery programs [2] [3] [4] . APVs are transmitted via biting insects and aerosols and are usually named on the basis of the bird species from which the virus was first isolated and characterized [4] . The disease, which is characterized by proliferative lesions of the skin and diphtheric membranes of the respiratory tract, mouth and oesophagus has been described in avian species [4, 5] . Although APV infections have been reported to affect over 232 species in 23 orders of birds [6] , our knowledge of the molecular and biological characteristics of APV is largely restricted to fowlpox virus (FWPV) and canarypox virus (CNPV) for which fullgenome sequences are available [7, 8] . Currently, only ten avipoxvirus species are listed under the genus by the International Committee on Taxonomy of Viruses (ICTV) [1] ; Table 1 . Thus, it is safe to assume that many APVs have yet to be characterized. Recombinant APVs have been evaluated for use as vaccine vector candidates against infectious diseases [7, 9] . APV-vectored vaccines are already in use in veterinary medicine [10] [11] [12] [13] [14] , and it is likely that such vaccines will also be used against human diseases in the future. This fact emphasizes the need to learn more about the molecular characteristics of APVs, which underpins the development of safe APV-vectored recombinant vaccines. This review summarizes current knowledge of APVs as avian pathogens, including classification, morphogenesis, hostvirus interactions, diagnosis, as well as issues relevant to their use as recombinant vaccine vectors. Avipoxviruses are large, oval-shaped enveloped viruses whose genome consists of double stranded DNA ranging in size from 260 to 365 kb [8] . Unlike most other DNA viruses, APVs replicate easily in the cytoplasm of infected avian cells which results in a characteristic cytopathic effect (CPE) 4 to 6 days post infection depending on the virus isolate [4] . APVs also multiply on the chorioallantoic membrane (CAM) of embryonated eggs, resulting in the formation of compact, proliferative pock lesions that are sometimes focal or diffuse [15] . However, some isolates, especially from the host species great tit (Parus major), have failed to multiply on CAM of chicken embryos [16] . APVs are the etiologic agent of disease characterized by skin lesions in both wild and domestic birds [4, 5] . Histologically and ultrastructurally, APVs undergo morphologic stages that are similar to other chordopoxviruses, including the formation of intracytoplasmic inclusions bodies, a characteristic which has been observed in some epithelial and mononuclear cells of permissive hosts. APV particles can be detected and further characterized by use of transmission electron microscopy (TEM) [17, 18] . Great discoveries made in the mid-nineteenth century facilitated major advances in pox virology. Based on the report by Bollinger [5] on poxvirus infected cells in chickens, and subsequent work by Fenner and Burnet [19] , APVs and other poxviruses were classified on the basis of original host, growth and morphological characteristics in the CAM of embryonated eggs or cell cultures and on clinical manifestations in different diseases of humans, birds and animals [20] rather than on genetic identity, which may provide both rapid and reliable virus identification [21] [22] [23] . These criteria have remained the basis for subsequent classification of APVs despite development of new molecular tools that have the capability of resolving the issue of species specificity of APV. Members of the genus Avipoxvirus belong to the subfamily Chordopoxvirinae which shares several biological features with other poxviruses [7, 8] . Currently, little is known of the number of species within the genus. While only ten strains have so far been identified and classified Worldwide as APV [1] , avian poxvirus infections have been reported to affect a wide range of bird species [6] . These strains vary in virulence and host specificity, demonstrating an urgent need for further analyses and characterization of new isolates. Avipoxviruses share several morphological, biochemical and physiochemical features with other poxviruses. Virus particles measure 270 × 350 nm and are composed of an electron dense, centrally located core and two lateral bodies that are visible in fixed and stained ultra-thin sections. In negative stained preparations, such as phosphotungstic acid (PTA) the membrane displays an outer-coat composed of a random arrangement of tubules [24] ; Figure 1 . APV particles have been shown to be resistant to ether, but sensitive to chloroform treatment [25] , although resistance to both chloroform and ether have been reported for pigeonpox virus and two pigeonpox virus mutants [26] . Avipoxviruses have low G+C content (30 to 40%) and consist of a single linear molecule of double-stranded DNA of between 260-365 kb. The central region of the genome is flanked by two identical inverted terminal repeats (ITRs) which are covalently linked by hairpin loops and contains several hundred closely spaced open reading frames [7] . The central region contains about 90-106 homologous genes that are involve in basic replication mechanisms, including viral transcription and RNA modification, viral DNA replication, and proteins involved in structure and assembly of intracellular mature virions and extracellular enveloped virions [8] . In general, genes located in this region have common molecular functions and are relatively conserved among poxviruses [8] . This is in contrast to the more variable, terminally located genes that have been shown to encode a diverse array of proteins involved in host range restriction [8] . Complete genomes of the two most studied APVs, FWPV US (FP-challenge virus; Animal Health Inspection Service Centre for Veterinary Biologicals, Ames Iowa, USA), FWPV-FP9 (a plaque-purified tissue cultureadapted attenuated European virus) and a CNPV virulent strain (Wheatley C93, American Type Culture Collection; ATCC VR-111) have been sequenced [7, 8, 27] . Although nucleotide and amino acids sequences for these two viruses are known, the functions of some putative genes and proteins remain to be fully assigned. Comparison of the FP9 strain with FWPV US revealed 118 differences; of which 71 genes were affected by deletion (26 of 1-9334 bp), insertion (15 of 1-108 bp), substitution, termination or frame-shift [27] . FP9 strain is a derivative of European FWPV HP1 which was obtained through over 400 passages in chicken embryo fibroblast (CEF). Analysis of FWPV HP1 sequences at the loci in which differences exist between FP9 and FWPV US show that 68 of 118 loci differ from the FWPV US, but were identical to FP9. This thus indicate that more than half of the differences between the two geographic FWPV lineages represented differences between the parent virulent viruses FWPV HP1 and FWPV US [27] . Further comparison of molecular data; show that FWPV and CNPV share high aminoacid identity, significant gene-sequence rearrangements, deletions and insertions [8] . The CNPV genome is about 80-100 kbp larger than the FWPV genomes. Both FWPV and CNPV express cellular gene homologues with immunomodulatory functions, which might be responsible for their different virulence and host-range [8] , but CNPV shows a broader tissue tropism in permissive avian hosts [17] than FWPV. CNPV has additional sequence of over 75 kbp, 39 genes lacking in FWPV homologues and approximately 47% amino-acid divergence [8] . These divergences are primarily found in the terminal nonconserved regions [7, 8] . Genes located in the non conserved regions are more prone to mutation and recombination and are implicated in host range, immunomodulation and pathogenesis [28] , and may be responsible in some aspects of cell and/or tissue tropism or perform other cellular functions [8] . Virulence genes are generally of non-conserved nature and influence the pathological profile of viruses in an infected host. These genes are important in viral evolution and have been used in studies to provide insight into how some poxviruses evolve strategies to ensure their replication [29] . Many of these strategies can be traced back to discoveries made with knockout (KO) viruses, in which a targeted disruption of a specific viral gene produced phenotypic changes reflective of the normal biological function of its protein product. Deletions of some non-conserved genes have also resulted in conditional replication defects in specific cell types [30] , such as the demonstration of spontaneous deletion of host range genes of vaccinia virus resulting in the compromised growth in the mammalian cell [31] . The K1L and C7L genes of vaccinia virus have been shown to be essential for completion of the replication cycle of vaccinia virus in human cells [32, 33] . In a knockout experiment, vaccina virus was unable to complete its replication cycle in Chinese hamster ovary (CHO) cells, since replication was aborted shortly after virus binding and entry, at the stage of intermediate gene expression [34] . But the insertion of another host range gene, CHOhr from cowpox virus into vaccinia virus allowed vaccinia virus to grow in CHO cells in which they are normally restricted [35] [36] [37] . Through use of these techniques, we now have a better understanding of the biology of vaccinia and other poxviruses, including their host range restriction. Although, some major advances have been made in genome sequencing and in vitro characterization of APVs [7, 8, 38, 39] , studies on APV host range genes are scarce. A wide array of gene homologues with likely host range functions such as NK-cell receptors, chemokines, serine protease inhibitors and homologues of genes involved in apoptosis, cell growth, tissue tropism and avian host range, have been identified in APVs, which suggests significant viral adaptation in the avian host [7] . Molecular knockout studies that target identification and further characterization of viral genes involved in regulation of cell proliferation, chromatin remodelling, virulence and apoptosis, in different APV-infected mammalian and avian cells are needed to better understand the tissue tropism and host range characteristics of APVs, including the abortive infection in mammalian cells. Compared to other poxviruses, such as vaccinia virus, mechanisms that account for APV pathogenesis is poorly understood. APVs have evolved a variety of elegant mechanisms to deliver their genes and accessory proteins into host cells. Like many other DNA viruses, APV probably devotes much of its genes to allow it to evade host immune responses. Such viral genes commonly encode proteins that are critical for the virus to undergo molecular transformation that leads to successful membrane fusion, penetration and intracellular transport. These includes genes that encodes proteins which act on early innate pathways such as pathways involving interferon [40] , pattern recognition receptors as Toll-like receptor (TLR) [41] , chemokines [42] and cytokines [43] , as well as pathways that act on subsequent adaptive responses [44, 45] . The infection of a cell by a virus is a complex process, during which the virus must overcome several host factors restriction points and the host immune response. Host protein interaction networks and biochemical pathways are in most cases altered by the viral proteins that free the virus from normal cellular controls and allow nucleotide metabolism in cells that have shut down DNA synthesis [46] . Hence, understanding viral protein functions and their interactions with host proteins is a prerequisite, not only to understand the infection biology of the virus-host system in question, but also for the rational development of target vaccines, based on specific antigens and possibly immunomodulatory factors, as well as antiviral compounds. Since the first isolation of APVs in cell culture, these viruses have been recognized as highly host specific. They are believed to replicate only in avian cells, notably chicken embryo fibroblasts (CEF; American Type Culture Collection; ATCC, Rockville, Maryland, USA; CRL-1590) [47] . CEF cells have a good split ratio compared to other cell lines, and are thus useful for large-scale propagation of virus, such as antigen production for vaccines or as a diagnostic tool. APVs have also been shown to replicate in chicken embryo kidney, chicken embryo dermis [48, 49] and quail cell lines, such as QT-35, although the presence of viable endogenous herpesvirus and Marek's disease virus (MDV) in QT-35 cells, limits their use for preparation of vaccines [4, 50] . APV have been isolated once from a mammal. In 1969, viable FWPV was isolated from a terminally ill rhinoceros [51] . The isolate was identified as atypical FWPV, based on pathological, virological and serological characteristics [51] . Nelson (1941) [52] reported mild pathology in mice following intranasal inoculation with FWPV, with no virus replication. Recent studies have also shown replication of APV in mammalian cell cultures, such as embryonic bovine tracheal cells [53] and baby hamster kidney cells [54] that are defined by the presence of infectious viral particles and CPE. These studies raise questions about the species specificity and mechanisms that restrict these viruses to certain hosts, and challenge the hypothesis that APV cannot undergo a full replication cycle in mammalian cells. The cellular entry and exit of APV is complicated by the existence of at least two distinct forms of virus that can productively infect cells, namely the intracellular mature virus (IMV) and the extracellular enveloped virus (EEV). These two forms are surrounded by different lipid membranes and surface proteins that are yet to be fully characterized. After virus binding to cellular membranes, a fusion step, which is generally poorly understood, results in the release of the virion core into the cytoplasm of the cell [39] . The released core, which contains the endogenous RNA polymerase and transcription factors, initiates the first wave of early viral gene transcription by synthesizing viral mRNA under the control of early viral promoters. This is followed by the uncoating stage, the release of viral DNA into the cytoplasm where it serves as a precursor for viral DNA replication as well as the source of intermediate and late viral gene transcription. As a late viral gene product accumulates, the virus undergoes assembly and morphogenesis of infectious virus particles. During morphogenesis, APVs induce the formation of inclusion bodies in the cytoplasm of infected cells (Figure 2A and 2B). The inclusions, which may also be termed viral factories, viroplasms, or viral replication complexes, are generally believed to be the sites of active viral replication and particle assembly within infected cells [17] . One model for the function of viral inclusion bodies is that they act to concentrate and sequester proteins, nucleic acids, and other small molecules essential for viral processes. In permissive cells, the first viral structures detectable by electron microscopy are the crescent-shaped forms ( Figure 3A ), consisting of a membrane with spicules on the convex surface [39] . These structures develop into non-infectious spherical immature viruses (IV) ( Figure 3B ) from which the intracellular mature virus (IMV) is formed by a series of maturation steps ( Figure 3C ). The IMV represents the majority of infectious progeny from each infected cell [17, 39, 55] . There are three possible mechanisms by which poxviruses are released from host cells depending on the strain of virus, cell type and the post-infection time [54] [55] [56] . They can be released by cytolysis, in which case IMV are released when the cell undergoes lysis as a result of CPE at the advanced stage of infection. It can also be released via virus-induced exocytosis. The third way of release is by budding, in which case IMV migrates out of the virus factory through the plasma membrane. Budding is shown to be the main exit route for APV [39] in contrast to the orthopoxviruses which exit by exocytosis of intracellular enveloped virus (IEV) [57] . These exit processes all result in the acquisition of an additional double membrane [39, 55] . In non-permissive African green monkey cells (CV-1, Vero) and in a human cell line (MRC-5), there is a blockade of the APV morphogenesis cycle. This occurs in steps following the formation of immature virus and is shown to be devoid of an alteration in early gene expression [47, 58, 59] , indicating that this blockade may not be associated with cell receptors. Poxvirus tropism may not be dependent upon specific cell surface receptors, but rather upon the ability of a given cell to provide intracellular complementing factors needed for productive virus replication, and on the ability of the specific virus to successfully manipulate intracellular signaling networks that regulate cellular antiviral processes following virus entry [28] . APVs have large genomes that would enable them to express unique collections of viral proteins that function as host range factors, which specifically target and manipulate host signaling pathways to establish optimal cellular conditions for viral replication. However, in some cells, especially mammalian cells, APV replication is blocked. This may be due to the ability of APV to specifically activate signalling pathways or mediators, for example interferon pathways, in those cells. The role of mediators and immunopathology of APV is complex and not well understood. However, considering the numerous steps involved in APV morphogenesis, it is relevant to note that these viruses induce several mediators that allow them to survive and interact with the host cells. Some potential mediators have been identified [7] and are awaiting proper functional characterization. These molecules alone may not fully explain the events that have been documented in mammalian cells that have supported the replication of APV [54] . Hence, identification of new mediators that are up or down-regulated in response to APV infected mammalian and avian cells could help advance our knowledge of immune responses against APV and the related immune-mediated pathology and cell tropism. It would be of vital importance to investigate these characteristics further, especially for the cell types that were recently shown to support APV replication [53, 54] . APV infections are associated with significant levels of morbidity and mortality in domestic and wild bird populations [6, 60] . Most of the investigations and reported cases are based on single APV isolates, which makes it difficult to address the pathogenicity of different APVs in different bird species. Chickens are commonly used to determine the pathogenicity of new isolates, but chickens may not be the ideal host, since APVs from wild birds may not multiply in chickens. In an attempt to identify and characterize the pathogenicity of APVs, Tripathy and others [4] found that wild isolates of Hawaiian crowpox virus had a generally mild pathogenicity in domestic chickens, characterized by relatively minor lesions of short duration at the sites of inoculation, which were in contrast to the general ability of FWPV strains to produce extensive proliferative lesions [4] . In another experimental study, two APV isolates obtained from endangered Hawaiian wild birds, the Hawaiian Goose (Branta sandvicensis) and the Palila (Loxioides bailleui), were compared with FWPV in specific-pathogen-free chickens. Immune responses were measured by ELISA before and after immunization with Hawaiian APVs and after challenge with FWPV. Both isolates from Hawaiian birds developed only a localized lesion of short duration at the site of inoculation in chickens and did not provide protection against subsequent challenge with virulent FWPV, in which severe lesions were observed. In contrast to high antibody response in chickens immunized with FWPV, birds immunized with either of the two Hawaiian isolates developed low to moderate antibody responses against viral antigens [61] . Pathogenicity studies of APVs in parrots [62] , turkeys, pigeons and canaries have also been reported. Canaries were highly susceptible to CNPV, but showed resistance to turkeypox virus, FWPV and pigeonpox virus [4, 63, 64] . A poxvirus from a Canada goose (Branta canadensis) was transmissible to domestic goose, but not to chickens or domestic ducks [15] . Pigeonpox virus produced mild infection in chickens and turkeys, but was more pathogenic for pigeons [62] . Poxvirus isolates from magpies (Pica pica) and great tits (Parus major) did not infect young chickens [16] , however, poxvirus isolated from blackbacked magpie (Gymnorhina tibicen) produced lesions in chickens. These studies were based on clinical manifestations in the chickens and suggest host specificity and pathogenicity. Despite the worldwide prevalence of APV infections, experimental infection studies in birds using APVs have centred on relatively few viral isolates. Analyses of variation have essentially focused on a FWPV strain termed the prototype, while a minority of experimental studies have been reported on CNPV, quailpox, juncopox, and pigeonpox virus isolates [4, 48, 49] . In fact, in the last twenty years, approximately 50% of published studies on APVs have been on the FWPV isolate directly (based on a PubMed search on APVs). The important nature of APVs which has been used successfully for vaccine development mandates that a larger pool of viral strains should be analyzed both for consideration of pathogenesis and determination of immune correlates of protection. Our present understanding of the antigenic variation of APVs has been based on a limited number of virus isolates in assays that includes complement-fixation, passive hemagglutination, agar-gel precipitation, immunoperoxidase, virus neutralisation and immunofluorescence [48, 49, 65] . In addition to the immunological assays, variation of APVs has also been addressed through genetic assays, such as restriction enzyme analysis. Genomes of FWPV and quailpox virus isolates were compared by using BamHI, EcoRI, and HindIII endonucleases and distinct fragment patterns were observed between the isolates. The patterns of three quailpox virus isolates were similar to each other with a high proportion of comigrating fragments. However, when immunogenic proteins of three FWPVs, two quailpox viruses, a juncopox virus, and a pigeonpox virus isolates were examined by immunoblotting, shared as well as unique antigens were detected. The greatest disparity was observed between quailpox virus and FWPV [48, 49] , indicating extensive variation between the quailpox virus and FWPV, which would predict differences in immunogenicity and antigenicity, including neutralization sensitivity. Nucleotide sequence based studies for rapid identification of poxvirus species by PCR with specific primers and hybridisation are well established [21] . These approaches have concentrated on single genes or portions of genes that exhibits variations in their sequence and are important for quick analysis of genetic variability [22] . Understanding the phylogenetics of APVs is essential to the understanding of host specificity and virulence, but also to provide insights into the variation of different viruses. Although the complete genome sequences of FWPV and CNPV are available [7, 8] , little is known about APV phylogeny. This is probably because of the difficulty in identifying pan-genus or species-specific PCR primers that can be used to amplify different genes. The most common PCR locus used until now has been the P4b locus [21] . Recent phylogenetic studies of APV isolates based on this locus [22, 23] indicated that most isolates clustered around either CNPV or FWPV, while another study based on the same locus demonstrated a third cluster, from psittacine birds [66] . Amano and coworkers [38] showed that the CNPV thymidine kinase locus was highly diverged from that of FWPV. The extent of this divergence was further illustrated by the fact that the amino acid similarity between CNPV and FWPV orthologue P4b was only 64.2% [23, 38] . A recent study, based on three different genes including the P4b, revealed that penguinpox virus, isolated from lesions around the eyes of African penguins (Spheniscus demersus), was most closely related to turkeypox virus, ostrichpox virus and pigeonpox virus [67] . During avipox outbreaks, mortality can reach 80 to 100% in canaries and other finches. This is in contrast to a generally lower mortality seen in chicken and turkey [60] . Transmission of virus can occur through a break in the skin or, more commonly, when vectored by biting insect such as mosquitoes and mites [68] . Aerosols generated from infected birds, or the ingestion of contaminated food or water have also been implicated as a source of transmission [69] . The disease is most commonly characterized by cutaneous proliferative lesions consisting of epithelial hyperplasia of the epidermis that resulting in proliferative, wart-like projections. They are primarily confined to unfeathered parts of the body, such as legs, feet, eyelids and the base of the beak (Figure 4 ). Scars are usually visible after recovery and healing of skin lesions. The mortality in wild birds is usually low, depending on the number and size of the proliferative lesions. However, if infection occurs in feather-free areas of the skin, with secondary bacterial infection, mortality may be high. The other and less common form of APV infections is the diphtheritic or wet form [70] which occurs as fibrino-necrotic and proliferative lesions in the mucosa of the digestive and upper-respiratory tracts, and generally has a higher mortality than the cutaneous form [60] . In some instances, birds display both cutaneous and diphtheritic forms and in those cases, mortality rates are often higher compared to the cutaneous form alone. Despite the variety of hosts and virus strains, associated pathology remains the same in infected domestic birds, although clinical signs vary depending on the virulence of the virus, susceptibility of the host, distribution and type of lesions [60] . There exist a relationship between FWPV and the avian retrovirus, reticuloendotheliosis virus (REV) (see section on APVs and REV). However, the possible roles that simultaneous REV infection arising from the provirus integration into the FWPV genome might play in the expression FWPV during disease outbreak remain unresolved. It is well known that REV infection leads to immunosuppresion [71] in affected birds. Thus, it is plausible to suggest that the presence of REV in FWPV infection may exacerbate disease progression. In spite of the fact that some mammalian cell lines seems to be able to support the replication of APVs, there is no evidence that APVs have caused clinical disease in humans, in contrast to what is known for other poxviruses, such as several parapox and orthopoxviruses. Clinical features of infected birds show multiple skin lesions varying from papules to nodules. Gross lesions in both the cutaneous and the diphtheritic forms, seen on birds and during necropsy, are usually sufficient to suspect APV infection [60] . However, these signs are sometimes not sufficient for definitive diagnoses of APV infection as other agents, such as papilloma virus, scaly leg mites [72] and mycotoxins may produce similar lesions in the skin [60] , and conditions like candidiasis, capillariasis and trichomoniasis may give lesions in the oral cavity similar to the diphtheritic form of APV infection [73] . It is therefore crucial to secure samples and confirm the viral etiology of the condition. Suspicion of clinical signs of APV infection can if possible be supported by necropsy, especially if the oral cavities to reveal the diphtheritic form. Further, histopathology on tissue sections using the classic Wright's Giemsa stain may reveal typical large, solid or ring-like, eosinophilic intracytoplasmic inclusions known as Bollinger bodies [5] ; Figure 2A and 2B. Transmission electron microscopy (TEM) may also reveal definite proof of APV infection, demonstrating the typical APV particles within inclusion bodies. APV identification may also be carried out by negative staining electron microscopy with 2% phosphotungstic acid (PTA) on infected cells (Figure 1 ). This method has typically been used by national reference or research laboratories to identify APV [18] . Demonstration of infectious virus by inoculation of homogenates of clinical samples of typical APV skin lesions onto the CAM of embryonated hen's eggs is the gold standard method for diagnosis of APV, although some strains of APV do not grow readily on chicken embryos [16] . Eggs are first swabbed with 70% alcohol and a pore is made in an area over the air-cell and another one on the other side of the egg to make a false air sac and lower the CAM by negative pressure using a rubber bulb. Inoculation of infectious samples by the CAM route is performed with sterile disposable 1 mL syringe with approximately 0.1-0.2 mL of inoculum. Eggs are incubated at 37°C for 5 days with daily candling to check for embryo death. Pock lesions measuring in size 0.5-1.5 mm are observed on the membrane 3-5 days after inoculation, depending on the virulence of the virus [15, 16] . Another method of isolation of APV requires the excision and homogenization of clinical skin lesions and inoculation of a homogenate supernatant onto a permissive cell culture, such as CEF cells. This results in the formation of CPE within 4-6 days post inoculation, depending on the virus isolate and on the multiplicity of infection (MOI) [4] . APV are increasingly being detected and characterized by PCR, Restriction fragment length polymorphism (RFLP), Southern blot hybridization, and cycle sequencing, directed at specific genes such as the 4b core protein gene [22, 23] . PCR allows for sensitive and specific detection of viral nucleic acids and has been shown to increase the diagnostic sensitivity for many viral pathogens when compared to culture. A PCR amplicon sequence allows a rapid search for homologous sequences in gene databases, to verify and identify the virus in question and to address phylogenetic relationships. Detection by realtime PCR has been used to identify recombinant APV from individual plaques [74] . This method eliminates the need for amplification and hybridization from the transient dominant protocol and results in significant savings of time at each round of plaque purification [74] . The conventional serological techniques of passive neutralization and agar-gel immunodiffusion are in continued global use for surveillance and disease control efforts in domestic poultry species [75, 76] , despite the availability of modern molecular and immunoassay techniques. The tests are time consuming, especially when carried out with large numbers of sera, and sensitivity appears to be low when compared with other detection method, such as enzyme linked immunosorbent assay (ELISA) [77] . ELISA has been described as a non-species specific test approach for birds [78] . It is a faster and easier method to detect antibodies against APV, particularly when large numbers of sera are to be tested. The technique is also more sensitive than the neutralization test [18, 78] . ELISA protocols have also been developed and used to test the efficacy of FWPV vaccines in commercial and wild bird species where agar-gel immunodiffusion is ineffective due to lack of precipitating antibodies [61, 79] . The challenges of controlling APV disease in poultry are driven by economics, and require strategies that keep cost low while maintaining treatment efficacy. Prophylaxis can be achieved by vaccination [39] . Doyle [80] reported the use of live FWPV or Pigeonpox virus for vaccination against APV infection. Since then, recombinant and live modified vaccines have been developed and used to prevent APV infections in chickens, pigeons, turkeys and quails [79, 81, 82] ; Table 2 . These vaccines are very effective and have undoubtedly contributed immensely to the prevention of the disease in commercial poultry farming [47, 81] . Since different APVs are isolated from a wide range of bird species and since only a few isolates have been characterized, development of a taxon-specific vaccine, directed to all species, has been difficult. Thus, available vaccines are often applied on the basis of experimentation, and more knowledge of molecular biology, pathology and epidemiology of these viruses is necessary to develop vaccines that effectively can protect a range of bird species. As in most viral infections, there is no specific treatment for avian poxvirus infections in birds [39, 83] . Available treatments include the use of iodine-glycerin application on proliferating skin lesions to aid healing [84] , antibiotics to control secondary bacterial infections and vitamin A to aid healing [85] . In the poultry industry, prophylactic measures against FWPV are achieved primarily by vaccination with live FWPV or antigenically similar pigeonpox virus strains produced in CEF cells [60] . In the past two decades, numerous outbreaks have been reported in vaccinated flocks, suggesting that vaccines used against the disease were not effective. In the United States a commercial FWPV vaccine was shown to be contaminated with REV and caused lymphoma among broiler chickens [86] . It has been shown that sequences of REV have been integrated into the DNA of FWPV vaccines as well as in field FWPV isolates [81, [87] [88] [89] [90] . The integration site is constant, while the size of the integrated fragments differs between various isolates and strains. Two different types of integrated sequences are reported; long terminal repeats (LTRs) with size of approximately 200 to 600 bp and the near-full-length REV provirus of about 800 bp [87, 90, 91] . Most vaccine strains carry only an LTR remnant while most FWPV field isolates carry the nearfull-length provirus. Singh and others [81] , however, detected REV LTRs of various lengths in the genome of two commercial FWPV vaccine strains and four field isolates, while several studies have shown that the source of REV infection was REV-contaminated FWPV [86, [92] [93] [94] and herpesvirus of turkeys vaccines [92, [94] [95] [96] [97] . Reticuloendotheliosis is a tumorigenic and immunosupressive disease. REV strains have been reported to cause diseases characterized by chronic lymphoma, non-neoplastic lesions and a runting-stunting syndrome in chickens, turkeys, and quails [98, 99] . REV are group of avian retroviruses and representatives include the defective REV-T and the non-defective REV-A, spleen necrosis virus (SNV), duck infectious anemia virus, and chick syncytial virus (CSV) [98] . The presence of REV in FWPV vaccines and the failure of currently used FWPV vaccines to evoke high level immunological protection against field challenge of FWPVs are of major concern to the poultry industry [100] , which emphasizes the need for research into alternative vaccines. In 1796 Edward Jenner [101] published his landmark findings that vaccination of humans with cowpox virus could prevent infection with variola virus, the causative agent of smallpox [101, 102] . This traditional vaccine technology, based on live viruses and immunological cross protection, has given rise to a wide range of effective vaccines against a wide variety of infectious agents, both in veterinary and human medicine. However, the emergence of new deadly human pathogens and cancers, have proven less amenable to the application of traditional vaccine platforms, indicating the need for new approaches. The use of a live virus vector represent an attractive way to deliver and present vaccine antigens that may offer advantages over traditional platforms, by improving the quality and strength of the immune response, such as in the case of HIV-1 where two different strains of vaccinia virus have been used as vectors. The NYVAC vector has been shown to induce the CD4 + T cell-dominant response, whereas modified vaccinia virus Ankara (MVA) induces a stronger CD8 + T cell response with accompanying CD4 + T cell responses that are required for protection [103] . Although this assertion remains unproven (there are to date no virally vectored vaccines licensed for human use), virally vectored vaccines offer an avenue of possibilities, either as homologous regimens, or as heterologous (prime-boost) regimens in which different serotypes of a given vector, different vectors or vectors and traditional technologies such as recombinant protein in adjuvant are administered sequentially. Currently, representatives of a wide range of virus families are under intensive development as vaccine vectors for human or veterinary use. Of these, FWPV and CNPV appear to be of great interest as vectors, and some veterinary APV-vectored vaccines are already licensed and in commercial use in North America, South America and Europe ( Table 2 ). The most important characteristics of APVs as vaccine vectors are that unlike most other DNA viruses, APV replicate in the cytoplasm of the infected cell and enzymatic functions used for transcription and replication are provided by the virus itself. This has several consequences regarding the use of these viruses as vaccine vectors. For example, APV promoters must be used for efficient transcription of recombinant genes and as APV transcripts are not spliced, genes cloned into APV vectors cannot contain introns [70, 104] . Other reasons include (1) their ability to accommodate and effectively express large amounts of foreign DNA or multiple genes that encode antigens [47] , (2) their inability to conduct a full replication cycle in non-avian species [105] [106] [107] , (3) antisera against orthopoxviruses do not neutralize APV and thus, prior Table 2 . Notably among them is the Trovac AI H5, a recombinant FWPV that express the H5 antigen of avian influenza virus. This product has had a conditional license for emergency use for chickens in the United States since 1998 and has been widely used in Central America, with over 2 billion doses administered [110] . ALVAC vectored vaccines have recently been registered for veterinary use in the European Union (Proteq-Flu) [111] and the United States (Recombitek). The equine influenza virus vaccine with CNPV vector expresses the hemagglutinin genes of the H3N8 Newmarket and Kentucky strains and contains a polymer adjuvant (Carbopol; Merial Ltd.). With the induction of both cell-mediated and humoral immunity, it is claimed that the vaccine produced sterile immunity 2 weeks after the second of two doses. The new vaccine is also designed to protect horses against the highly virulent N/5/03 American strain of equine influenza virus and to prevent the virus from spreading through the elimination of viral shedding. Despite these notable advances in APV-vectored vaccine development, the list of licensed viral vectored vaccines for human medicine is short, with only a few vaccines that have entered clinical trials [112] ; Table 3 . This may be in part owing to stringent safety requirements that must be met for viruses, that in their natural state have the potential to be human pathogens, to be used as viral vaccine vectors that may replicate in vivo in a manner similar to their wild-type parental viruses. Another reason may be fear of risk of spontaneous recombination between virus vectors and naturally occurring viral relatives in the ecosystems in which the vaccine is used. Even if APVs are not generally expected to replicate in mammals, the vaccine vectors may reach bird populations via animal populations. It is also possible that the vector, through spontaneous recombination and mutation events, may restore its replication competence. To cater for this, the aim during design and development of a virus vector is always to introduce at least two gene deletions crucial for viral to undergo a full replication cycle to assure a very low probability that replication competence could be restored. To our knowledge, such reversions have not been identified in clinical trials of APV-vectored vaccines. In addition to concerns regarding reversion or recombination, another safety signal was recently identified in an in vitro experiment that showed APV replication in cell clones derived from embryonic bovine trachea [53] and Syrian baby hamster kidney (BHK) cells. In this experiment, infectious IMV was observed; indicating complete virus replication had taken place [54] . These findings are in contrast to the general dogma that APVs are restricted to infection of cells of avian origin, and are an indication that there is still more to learn about the replication mechanisms and virus-host interactions of these viruses, including evasion of immune responses, cell tropism and host range mechanisms. APVs cause disease of economical importance for the poultry industry, and also in pet and wild birds. Thus, prophylactic measures, such as vaccination, will always be required, and there is a need for more efficient and safe vaccines. One promising approach is the use of APVs as vectors for recombinant vaccines, increasing the efficacy and avoiding the potential contamination with REV and other agents. Many recombinant APV constructs are already licensed for use in veterinary medicine, and a range of vaccine candidates are currently being tested for use in vaccines against numerous infectious diseases in animals and man. Thus, it is likely that recombinant APV-vectored vaccines in the near future will also be used against human diseases. APVs have many advantages as vaccine vectors, including a large genome which allows for the inclusion of many heterologous genes, such as genes coding for antigens, cytokines and other immuno-modulating factors. The major safety argument for using APVs rather than vaccinia virus or other mammalian viruses as vectors, is that APVs are not zoonotic and are not able to conduct a full replication cycle in mammals. However, it was recently shown that FWPV was able to replicate and produce progeny virions in some established mammalian cell lines. This illustrates the fact that general knowledge of APVs is scarce. Indeed, only a few isolates have been characterized and classified. New molecular tools have led to a greater resolution of factors and mechanisms that restrict viruses to certain hosts, for example HIV and SARS. Mechanisms of host restriction, pathogenicity, host immunity and viral immune evasion strategies are of crucial importance regarding use of APVs as vectors in multispecies-targeted vaccines. A good understanding of the molecular properties of APVs underpins the development of safe APVvectored vaccines.

Apolipoprotein M Gene (APOM) Polymorphism Modifies Metabolic and Disease Traits in Type 2 Diabetes

This study aimed at substantiating the associations of the apolipoproein M gene (APOM) with type 2 diabetes (T2D) as well as with metabolic traits in Hong Kong Chinese. In addition, APOM gene function was further characterized to elucidate its activity in cholesterol metabolism. Seventeen APOM SNPs documented in the NCBI database were genotyped. Five SNPs were confirmed in our study cohort of 1234 T2D and 606 control participants. Three of the five SNPs rs707921(C+1871A), rs707922(G+1837T) and rs805264(G+203A) were in linkage disequilibrium (LD). We chose rs707922 to tag this LD region for down stream association analyses and characterized the function of this SNP at molecular level. No association between APOM and T2D susceptibility was detected in our Hong Kong Chinese cohort. Interestingly, the C allele of rs805297 was significantly associated with T2D duration of longer than 10 years (OR = 1.245, p = 0.015). The rs707922 TT genotype was significantly associated with elevated plasma total- and LDL- cholesterol levels (p = 0.006 and p = 0.009, respectively) in T2D patients. Molecular analyses of rs707922 lead to the discoveries of a novel transcript APOM5 as well as the cryptic nature of exon 5 of the gene. Ectopic expression of APOM5 transcript confirmed rs707922 allele-dependent activity of the transcript in modifying cholesterol homeostasis in vitro. In conclusion, the results here did not support APOM as a T2D susceptibility gene in Hong Kong Chinese. However, in T2D patients, a subset of APOM SNPs was associated with disease duration and metabolic traits. Further molecular analysis proved the functional activity of rs707922 in APOM expression and in regulation of cellular cholesterol content.

The human apolipoprotein M gene (APOM, Gene ID: 55937) is located on chromosome 6p21.33 and contains six exons spanning a region of 2.3 kb in length with gene structure conserved across species [1, 2] . In human and mice, APOM mRNA is highly expressed in liver and kidney [2] . The human apoM protein (MIM 606907) of 188 amino acids is mainly associated with HDL and to a minor degree with LDL, very low density lipoprotein, and chylomicrons [2] . Plasma apoM has been positively associated with plasma total cholesterol (TC), LDL cholesterol (LDL-C), and HDL cholesterol (HDL-C) [3] . APOM knockdown in mice by siRNA revealed its anti-atherosclerotic effect by participating in pre-b HDL formation and reverse cholesterol transport [4] . Kruit et al., recently reported the effect of cellular cholesterol accumulation on beta cell dysfunction in type 2 diabetes [5] . Such finding implies that factors (i.e., apoM) affecting the balance of cellular cholesterol content are likely to modify beta cell function and thus the susceptibility to or progression of type 2 diabetes. Several additional lines of evidence also indicated the possible involvement of APOM in the development of diabetes and metabolic disturbances: 1) the human APOM gene is located within a high susceptibility region (6q21-q23) to type 2 diabetes (T2D) in genome-wide linkage analyses [6] . 2) SNP rs805296 (T-778C) in APOM promoter has been associated with the levels of plasma total cholesterol (TC) and fasting plasma glucose (FPG) in non-diabetic participants, 3) SNP rs805296 has also been associated with the susceptibility to T2D and coronary artery disease among the Northern Chinese [7, 8] . In 2010, China became the country with the largest diabetic population in the world. The Northern and Southern Chinese populations are distinct in genetic marker analyses [9] , meaning disease markers identified in northen populations may not be shared by the Southern populations. The primary aim of the current study is to establish the association between APOM and T2D susceptibility in a Southern Chinese cohort in Hong Kong. By assuming the same effect size (OR = 1.934) and disease allele frequency as observed in the studies of Northern Chinese [8] , the power of the current case-control study is over 95% with 1234 cases and 606 controls. The secondary aims are to examine for association between APOM and component metabolic traits as well as to further assess the function of the gene. The pilot cohort consisted of 103 male and 95 female controls (average age = 43 yrs). They were Hong Kong Chinese adults recruited from a community health screening program of cardiovascular risk factors with normal response at a 75 g oral glucose tolerance test [10] . The study cohort had 1234 unrelated T2D patients and 606 controls. All participants gave written informed consent at the time of blood sampling. Ethics approval was obtained from the Clinical Research Ethics Committee of Chinese University of Hong Kong, Shatin, NT, Hong Kong. All T2D participants were selected from the Hong Kong Diabetes Registry. Control participants were recruited in a community health screening program for cardiovascular risk factors and some were hospital staff (3.1%, n = 19). No subdemographic differences were detected in control participants. All control participants had no known history of diabetes and had fasting plasma glucose (FPG) , 6.1 mmol/l. Clinical assessments of participants had been described elsewhere [11] . Body mass index (BMI), blood pressure (BP) as well as fasting blood biochemical and metabolic profiles were measured. Among the 1234 T2D patients, 9.8% (n = 121) were on diet treatment only, 41.3% (n = 510) were on oral anti-diabetic drugs only, 12.5% (n = 154) were on insulin only, 9.5% (n = 117) on both oral anti-diabetic drugs and insulin, and 7.3% (n = 90) were treated for dyslipidemia. 2.1. SNP selection and genotyping analyses. Genomic DNA was prepared from whole blood as previously described [12] . Seventeen APOM SNPs including rs6921907, rs1266078, rs9267528, rs805297, rs4947251, rs9404941, rs805296, rs805264, rs3117581, rs34490746, rs11462733, rs2273612, rs707922, rs707921, rs28432254, rs3132449, rs3178094 enlisted in the NCBI database [13] were selected for genotyoping in the pilot cohort of 198 controls by multiplex reactions using the Mass ARRAY system (Sequenom, San Diego, CA, USA) at the Genome Quebec Innovation Centre, McGill University (Montréal, Quebec, Canada). Six of the seventeen SNPs were confirmed in the pilot cohort: rs1266078(T-1628G), rs805297(C-1065A), rs9404941(T-855C), rs805264(G+203A), rs707922(G+1837T) and rs707921(C+1871A). These six SNPs were further genotyped in the study cohort of 1840 participants (1234 cases and 606 controls). Case and control DNA samples were genotyped in parallel on the same plates. Two hundred ninety one duplicate samples (15.8%) were used to assess intra-plate and inter-plate genotype quality. No genotyping discrepancies were detected. The overall call rate was 98.0%. Five out of the six SNPs (except for rs1266078) were successfully genotyped in the study cohort. 2.2. Plasma lipids and apoM levels. Plasma apoM concentration was estimated by dot-blot analysis using monoclonal mouse anti-human apoM antibody (ABNOVA, Taipei, Taiwan) following previously established protocols [14, 15] . Recombinant human apoM (ABNOVA, Taipei, Taiwan) was used as protein standard after serial dilution. The mean signal densities of each specimen and protein standards in triplicate measures were determined by ImageJ 1.42q software (http://rsbweb.nih.gov/ij/). ApoM concentration was derived from the standard curves developed using the recombinant apoM protein. ectopic APOM1 and APOM5 expression. WRL-68 and HepG2 hepatic cell lines were purchased from American Type Culture Collection (Rockville, MD, USA) and maintained in RPMI-1640 medium supplemented with 10% FBS and 100 units/ ml penicillin and 100 mg/ml streptomycin in a humidified atmosphere containing 5% CO 2 at 37uC. Prior to the expreiments measuring cellular and medium cholesterol content, cells were switched to serum free and phenol red free RPMI medium (Invitrogen, Carlsbad, CA, USA). Ectopic expression of APOM was achieved by transient transfection of APOM1 or APOM5 cDNA (cloned into the pCMV-Myc vectors) into cultured cells at 70% confluence using LipofectamineTM 2000 reagent (Invitrogen, Carlsbad, CA, USA) following the manufactorer's protocols [16] . Cellular lipids were extracted as previously described [17] . Total cellular cholesterol was measured using the InfinityTM Cholesterol Liquid Stable Reagent (Thermo Fisher Scientific Inc., Middletown, VA, USA) following the manufacturer's instructions. The measured amount of total cholesterol was normalized by cell protein concentration. 3.1. Comparative genomic and protein sequence analyses. APOM transcript and gene sequences were obtained from the NCBI human Genome Browser [18] , the Ensembl Genome Browser [19] and the human Expressed Sequence Tags (EST) databases. The Evolutionary Conserved Regions Browser [20] and the Ensembl Genome Browser were used to identify sequence conservation. 3.2. Rapid amplification of cDNA ends (RACE). The Human Liver FirstChoice RACE-ready cDNA kit (Ambion, Austin, Texas, USA) was used to amplify the 59 and 39 ends of novel APOM transcripts (Supplementary Figure S1 , and Supplementary Table S1). 3.3. Semi-quantitative RT-PCR analysis. Human normal adult tissue RNA samples were purchased commercially (Stratagene, La Jolla, CA, USA or Millipore Chemicon, Billerica, MA, USA). cDNA was synthesized using the GeneAmp RNA PCR kit (Applied Biosystems, Foster City, CA, USA) in combination with RNase inhibitor (Roche Applied Science, Indianapolis, IN, USA) and M-MuLV reverse transcriptase. The subsequent PCR amplification of cDNA was performed using the AmpliTaq Gold DNA Polymerase (Applied Biosystems, Foster City, CA, USA) following standardized protocols [21, 22] . The vulgate transcript APOM1 (Ensembl: ENST00000375916) and the novel transcript APOM5 were amplified using primer RT-PCR-YY12 paired with RT-PCR-YY13 and RT-PCR-YY12 paired with RT-PCR-YY14, respectively (Supplementary Table S4 ). GAPDH was amplified as a house-keeping gene control for RNA integrity and equal loading using the GAPDH primers, RT-PCR-Tao1 and RT-PCR-Tao2 [23] (Supplementary Table S1 ). Continuous variables were compared using Student's t test or one-way analysis of variance (ANOVA) for traits with normal distribution. Plasma triglycerides (TG) were skewed and logarithmically transformed. Association tests between genotypes and quantitative traits were performed in T2D patients and nondiabetic controls separately. Categorical variables, including genotype distributions were compared by x 2 tests. Genotype distributions were tested for Hardy-Weinberg equilibrium using goodness-of-fit test (1 df). Informative missingness was checked by coding successful genotypes into one group and failed genotypes into another group followed by 262 Chi-square test for T2D and t-test for the quantitative traits of interest. One SNP (rs805264) with significant result (p,0.001) indicative of informative missingness (IM) was excluded for further T2D association analyses (Supplelmentary Table S2 ). Pairwise LD of D' and r 2 analyses were performed using Haploview (Broad Institute of MIT and Harvard, USA, version 4.0). 262 contingency tables were used for comparing the differences of allele frequencies and 263 contingency tables were used for detecting the differences of genotype frequencies between cases and controls. Allelic, dominant, recessive and additive genetic models were used to test the association between each SNPs and T2D. Multivariate logistic regression analysis was used to assess the significance of covariates and adjusted for confounders in the association of genetic factors with T2D. The independent contributions of all traits, covariates, SNPs, and haplotypes were determined by multiple regression analysis. Association between haplotypes and T2D or metabolic traits were tested by Haploview (version 4.0, Broad Institute of MIT and Harvard, USA) and PHASE software (version 2.1, UW Tech-Transfer Digital Ventures, University of Washington, Seattle, WA, USA) [24] . When using PHASE software, the probability thresholds were set at 90% for haplotype inference to deal with ambiguous haplotypes. It was only used to infer the haplotype of each individual and thus the case-control permutation test was not conducted. The PHASE-imputed haplotypes were counted 20 times using different seed numbers. No difference between runs was detected. To account for multiple testing, we used the Bonferroni correction and a statistical significance was considered only when an SNP association with T2D/metabolic traits was p,0.017 (equivelent to 0.05/3), and a haplotype association with T2D/ metabolic traits was p,0.0125 (equivelent to 0.05/4). The human plasma apoM concentrations determined by dot-blot assays were compared by a nonparametric Kruskal-Wallis H test. A value of p,0.05 was considered significant. Results from functional analyses were analyzed by Student's ttest for two-group comparison, and one way ANOVA for multiple group comparisons. A statistical significance is considered at p,0.05 level. All statistical analyses were performed using the SPSS program (SPSS version 15.0, Chicago, IL, USA) unless otherwise specified. Seventeen APOM SNPs enlisted in the public databases were selected for genotyping in a pilot cohort of 198 control participants. Six SNPs were confirmed polymorphic in this pilot cohort of Hong Kong Chinese. These six SNPs were further genotyped in the full study cohort of 1840 participants. Five of the variants were successfully genotyped: rs805297(C-1065A), rs9404941(T-855C), rs805264(G+203A), rs707922(G+1837T), and rs707921(C+1871A) with genotype distributions fitting Hardy-Weinberg equilibrium. Table 1 summarized the allele and genotype frequencies of SNPs in non-diabetic controls and T2D patients. SNP rs805264 was removed from further association analysis for T2D due to informative missingness (Supplementary Table S2 ). 2.1. APOM SNPs and T2D susceptibility. No significant association was detected between individual SNPs and T2D (Table 1) . Further multiple logistic regression analysis adjusting for age, BMI, SBP (systolic blood pressure), DBP (diastolic blood pressure), TC and TG again detected no significant association between individual SNPs and T2D (data not shown). 2.2. APOM SNPs and T2D duration. We next examine the association between APOM SNPs and T2D disease duration. T2D patients were subgrouped into disease duration of # 10 years (n = 583) and disease duration of .10 years (n = 586). As shown in Table 2 , the C allele of rs805297 was associated with T2D duration of longer than 10 years (odds ratio = 1.245, p = 0.015). 2.3. APOM SNPs and metabolic traits. We next analyzed the association between SNPs and metabolic variables in patients and controls separately. SNP rs805297(C-1065A) was not associated with metabolic traits in either patients or controls. Since rs707922(G+1837T) was in near perfect LD with rs805264(G+203A) and rs707921(C+1871A) (Supplementary Figure S2B) , similar association results were expected and observed. Table 3 showed the representative results using rs707922 as the marker SNP. Under recessive model, homozygous minor allele TT of rs707922 was associated with significantly higher TC (p = 0.006), LDL-C (p = 0.009) in T2D patients. In controls, no association between SNPs and metabolic traits was detected. When plasma apoM concentration was measured in T2D patients and controls subgrouped by their rs707922 genotype, the TT genotype was found associated with significantly higher apoM level as compared to the GT (and GG) genotype(s) (p = 0.002). As mentioned above, rs805264, rs707922, and rs707921 are located within the same LD block. Therefore, in the subsequent haplotype analysis, rs707922 was used to 'tag' the three SNPs. Haplotype construction was conducted for rs707922(G+1837T) with other independent SNPs rs805297(C-1065A) and rs9404941(T-855C). Four haplotypes (A-T-G, C-C-G, C-T-G, and C-T-T) accounting for 99.9% of all possible haplotypes were detected in our Hong Kong Chinese population (Supplementary Table S3 ). No significant association was found between these haplotypes with T2D. Homozygous C-T-T was significantly associatiated with elevated TC, LDL-C and HbA1c in T2D patients (p,0.0125, Supplementary Table S4 ). Figure S3) . SNPs rs707922 (G+1837T) and rs707921 (C+1871A) which associated with plasma levels of TC, LDL-C and apoM in T2D patients fell within the evolutionary conserved region. Figure 1 (top panel) illustrated the cross-species sequence conservation of the rs707922-and rs707921-flanking region. BLAST search using this conserved sequence as the template returned unique human EST clones BI757556 (human brain) and AA975560.1 (human kidney) which are likely other APOM transcripts. The Ensembl Browser also displayed three APOM transcripts (APOM1: Ensembl-ENST00000375916, APOM2: Ensembl-ENST00000375920 and APOM3: Ensembl-ENST00000375918). 4.2. Molecular cloning of APOM transcripts. Since the sequences of EST clones BI757556 and AA975560.1 were different than the known APOM transcripts, we proceeded with cloning alternative transcripts of APOM. A novel transcript, designated APOM5, was identified by 59 RACE and 39 RACE. As shown in Figure 1 , the 39 end of APOM5 was identical to the 39 end of APOM3. The 59 end, however, was similar to that of the vulgate APOM transcript (designated APOM1) except that the transcription start site of APOM5 was 21 nucleotides downstream that of the APOM1. The full-length sequence of APOM5 is provided in Supplementary Figure S4 . It is important to note that the two SNPs rs707922(G+1837T) and rs707921(C+1871A) associated with metabolic traits in T2D are located to the exon 5 of APOM5. On the contrary, when reference to the vulgate APOM1 transcript, rs707921 and rs707922 are located to intron 5. These observations support the cryptic nature of exon 5 of the gene. The results of this study did not support an assoiation between APOM and T2D suseptibility in Hong Kong Chinese. For a subset of SNPs, we presented evidence of association between APOM and disease duration as well as metabolic traits in T2D patients. Further characterization of rs707922, one of the metabolic traitassociated SNP at molecular level lead to the discoveries of a novel transcript APOM5 and its SNP-dependent effect on cellular cholesterol content. The LD block formed among rs805264, rs707922, and rs707921 in our cohort agreed with the LD structure reported in the Northern Chinese [25] . It is currently unknown whether this subset of SNPs is also associated with metabolic traits in Northern Chinese with T2D. Among the four common haplotypes constructed from rs805297, rs9404941, and rs707922, only homozygous haplotype C-T-T was significantly associated with higher TC, LDL-C and HbA1c levels in T2D patients. Given the established association between rs7070922 and plasma TC and LDL-C levels, these association results did not support additional effects of the haplotypes on serum cholesterol levels. Interestingly, the association between the homozygous haplotype C-T-T with HbA1c indicated the interaction among the three alleles to control systemic glucose level in T2D patients. It would have been ideal if the previously reported association between SNP rs805296(T-778C) and T2D in Northern Chinese were reproduced in this Hong Kong Chinese population. Unfortunately, genotyping of this SNP failed to produce results in this study, precluding it being used for discussions attempting to reconcile the current findings with prior results. Although rs805296 is physically close to rs9404941, the existing information/data does not allow the relationship between SNP rs805296 and T2D/T2D metabolic traits to be predicted in our cohort. It is noteworthy that the case-control study design adopted by the current and other studies tend to be limited by the heterogeneity of the prevalent cases with regards to T2D ascertainment, i.e., both those have developed T2D and those have survived in the setting of T2D were included as cases. Therefore, those 'susceptible to' the disease were not distinguished from those 'survived' the disease'. One possibility to circumvent such issue is to examine for similar duration of diabetes across studies being compared and test for difference in duration of T2D by SNP. Interestingly, while our results did not support an association between APOM and T2D susceptibility, stratification of our cases by disease duration allowed us to detect an association between rs805297(C-1065A) and T2D duration. This result implied the possibility that relative to the rs805297-A carriers, the rs805297-C carriers better survived the diabetic condition over the long term and such possibility can be further tested. Interestingly, Zhao et al., recently reported a positive association between rs805297-A and the risk of stroke in Norhtern Chinese (OR = 1.38, p = 0.002) after adjusting for other risk factors including history of diabetes [26] . Whether such association is present among the Southern Chinese requires further investigation. Nevertheless, losing rs805297-A carriers with T2D to stroke over time provides a plausible explanation for the observed higher frequency of rs805297-C allele in the T2D duration .10 years subgroup (relative to T2D duration #10 years subgroup). Previous studies attempting to correlate plasma apoM and cholesterol levels have generated inconsistent results [3, 27] . In this study, rs707922 homozygous minor allele (TT) was associated with elevated TC and LDL-C as well as plasma apoM levels in diabetic cases (average BMI of 25.26). These observations are consistent with previously reported positive association between plasma apoM and plasma TC and LDL-C in overweight-obese individuals [3] . Results presented by Han et al. from the study of a Northern Chinese cohort showed significant association between rs707922 T allele and increased risk of cerebral infraction (OR = 1.78, p = 0.000). In parallel they also confirmed hypercholesterolemia as an independent risk factor for cerebral infraction [25] . These results implied the possibility that rs707922 is also a modifier of serum cholesterol in Northern Chinese. The association between rs707922 TT genotype and elevated serum total-/LDL-cholesterol levels in type 2 diabetes found in the current report deserves to be further substantiated in strict replicate studies. The mechanism underlying the effects of rs707922 on plasma TC and LDL-C levels in diabetes remains elusive. Richter et al., reported HNF-1 alpha being a potent transcription activator of APOM [28] . The decreased serum apoM level in maturity-onset diabetes of the young subjects as compared to the controls could be explained by the HNF-1 alpha mutations in these patients [28] . Given the association between APOM and metabolic traits found in this study, one may speculate that SNP rs707922 (G+1837T), in the capacity of an intronic SNP (reference to the APOM1 transcript), may modify APOM expression through SNP-specific recruitment of transcription factors (i.e., PAX 6 showed an allelespecific interaction with rs707922 T by computer prediction as presented in Supplementary Figure S5 ) and subsequently affect cellular cholesterol homeostasis in liver and possibly other tissues. More interestingly, we found that rs707922 can also assume the capacity as an exonic SNP (i.e., reference to the APOM5 transcript). While the function of APOM5 requires further elucidation, the high renal and hepatic expression levels of APOM1 and APOM5 indicated the possibility of these transcripts coordinate to regulate cholesterol homeostasis in these tissues. Such possibility is further supported by the results showing the activities of ectopically expressed APOM5 in modifying hepatic cell cholesterol content. With regards to systematic cholesterol homeostasis, we observed that homozygous rs707922-T allele associated with elevated total-and LDL-cholesterol levels. One possible mechanism of such elevation is through reducd hepatic and/or pheripheral clearance of circulating cholesterol. Consistent with this notion, our in vitro data showed that hepatic cells overexpressing APOM5-T transcript had lower cholesterol content relative to cells expressing the APOM5-G counterpart. In conclusion, the APOM SNP frequencies and the LD structure reported in this study of Hong Kong Chinese population will facilitate future population genetics studies. While our results did not support an association between APOM and T2D susceptibility in Hong Kong Chinese, subgroup analyses found SNP as well as haplotype associations between APOM and metabolic traits in T2D. Bioinformatics/molecular analyses revealed the cryptic nature of exon 5 responsible for the expression of a novel transcript APOM5, predominantly in liver and kidney. The activity of APOM5 on modifying cellular cholesterol content revealed another layer of regulation underlying the expression and function of APOM. (Mus musculus; chr17) , rat (Rattus norvegicus; chr20), cow (Bos Taurus; chr23) and dog (Canis familiaris; chr12) genes are shown. Conserved sequences were defined as coding exons (blue), The Evolutionary Conserved Regions (ECRs) were indicated by pink lines (on top of the panel for each species) with a default value of 70%. The human APOM was depicted as a horizontal blue line above the graph, with strand/transcriptional orientation indicated by arrows. APOM coding exons were shown as blue boxes along the line, while untranslated regions (UTR) were indicated as yellow boxes. Peaks within the conservation profile which corresponded to these five exons of APOM were similarly coloured within the plot. Peaks within the conservation profile that did not correspond to transcribed sequences were highlighted in red colour. Regions of transposable elements and simple repeats were highlighted in green color. Figure S5 Computer-predicted transcription factor interaction sites in nucleotide sequences spanning SNPs rs707922(G+1837T) and rs707921(C+1871A). This figure is generated by the MATCH program. Top panel: The transcription factors predicted to interact with the nucleotide sequences spanning the major allele of SNPs rs707922 (G allele) and rs707921 (the C allele). Bottom panel: The transcription factors predicted to interact with the nucleotide sequences spanning the minor allele of SNPs rs707922 (the T allele) and rs707921 (the A allele). The predicted transcription factors are marked by blue text with scores of matrix match indicated in parentheses. The locations and orientations of the binding sites for these predicted transcription factors are marked by black horizontal dashed lines with arrows. Highlighted in pink boxes are allele-specific transcription factors (PAX6 and AREB6 for rs707922-T; HNF4 and OCT1 for rs707921-A) and their corresponding binding sites. The vertical dashed lines indicate the locations of SNPs rs707922 and rs707921. The precise nucleotide positions of SNP rs707922 and rs707921 are also highlighted with grey boxes in the DNA sequences represented by red colored text. (TIFF) Table S1 Sequences of primers used in this study. (PDF) Table S2 Summary of data quality of the five successfully genotyped APOM SNPs in the full cohort (n = 1840). The clinical traits tested for IM include T2D, TG, HbA 1c , FPG, TC, HDL-C, and LDL-C. (PDF) Table S3 Frequencies of common haplotypes constructed by APOM SNPs rs805297, rs904941, and rs707922. (PDF) Table S4 Haplotype C-T-T formed from SNPs rs805297(C-1065A), rs9404941(T-855C), and rs707922 (G+1837T) and clinical characteristics of T2D patients. Values are either number of subjects, mean 6 SD, or geometric mean (95% confidence interval). p values here represent the comparisons between subgroup of homozygotes of C-T-T haplotype vs. subgroup with one C-T-T haplotype and without C-T-T haplotype (a recessive model). p values are adjusted for age, sex, BMI and disease duration in T2D. In non-diabetic controls, the p values without parentheses are adjusted for age, sex and BMI. ''+/+'' represent the homozygote of haplotype C-T-T, ''+/2'' represent the heterozygote of haplotype C-T-T, ''2/2'' represent the subgroup who do not have the haplotype of C-T-T. Individuals on lipid lowering medications (n = 90) were excluded for association analysis with lipid traits. * statistical significance (p,0.0125). (PDF)

Non-Invasive Microstructure and Morphology Investigation of the Mouse Lung: Qualitative Description and Quantitative Measurement

BACKGROUND: Early detection of lung cancer is known to improve the chances of successful treatment. However, lungs are soft tissues with complex three-dimensional configuration. Conventional X-ray imaging is based purely on absorption resulting in very low contrast when imaging soft tissues without contrast agents. It is difficult to obtain adequate information of lung lesions from conventional X-ray imaging. METHODS: In this study, a recently emerged imaging technique, in-line X-ray phase contrast imaging (IL-XPCI) was used. This powerful technique enabled high-resolution investigations of soft tissues without contrast agents. We applied IL-XPCI to observe the lungs in an intact mouse for the purpose of defining quantitatively the micro-structures in lung. FINDINGS: The three-dimensional model of the lung was successfully established, which provided an excellent view of lung airways. We highlighted the use of IL-XPCI in the visualization and assessment of alveoli which had rarely been studied in three dimensions (3D). The precise view of individual alveolus was achieved. The morphological parameters, such as diameter and alveolar surface area were measured. These parameters were of great importance in the diagnosis of diseases related to alveolus and alveolar scar. CONCLUSION: Our results indicated that IL-XPCI had the ability to represent complex anatomical structures in lung. This offered a new perspective on the diagnosis of respiratory disease and may guide future work in the study of respiratory mechanism on the alveoli level.

Many lung diseases alter the morphology of the lung tissue [1] [2] [3] [4] [5] [6] . Unfortunately, due to the limitation of resolution and contrast, at the early stage of the disease, minor pathological changes can not be discerned by conventional absorption-based imaging techniques. For example, lung cancer is one of the most common diseases with low survival rates worldwide; early detection of the lung cancer is known to improve the chances of successful treatment. But in most cases, cancer has been detected in the terminal stage and is impossible to be cured [1, 5, 7] . As a result, it is of great importance to image micro-structures in lung to enable early detection. Histological biopsy is the most commonly used method for micro structures observation. But this is invasive and usually nonrepeatable [8] [9] [10] . As one of the non-invasive methods, conventional radiography has been based purely on absorption contrast. However, the differences in X-ray absorption of biological soft tissues are quite small; this technique may lead to very low contrast and poor spatial resolution. Phase contrast imaging (PCI) is a relatively new imaging technique, which can provide high contrast images by using phase shift of the X-ray. It is well known that X-ray is a form of electromagnetic wave. When it propagates through an object, both the amplitude and the phase of the wave are modified. It can be described by a complex reflective index [11] , given by n = 12d2ib. The imaginary part b is related to the attenuation based on absorption of the object, which we use in the conventional X-ray imaging. Unfortunately, the variations in Xray absorption between different biological soft tissues are quite small; this technique may lead to very low contrast and poor spatial resolution. The real part d is responsible for the X-ray phase shift. For biological soft tissues, the phase shift is almost one thousand times greater than the absorption term. Therefore, the aim of phase contrast imaging is to use the information of X-ray phase shift and convert it into image contrast [12] . This technique can greatly improve the image quality of soft tissues, particularly at the interface of tissues where the refractive index changes significantly. During recent years, it has been widely used by researchers for imaging of small animals. Among all the main PCI methods, in-line X-ray phase contrast imaging (IL-XPCI) is the simplest and the most straightforward, making it more suitable for clinical applications than other methods [13] . IL-XPCI has previously been used to study soft tissues of both human and small animals such as mice, rats, and rabbits. The results are satisfactory: high resolution images were obtained [14] [15] [16] . It may be an alternative method for lung observation and it is totally noninvasive. To date, only a few studies have involved in-line X-ray phase contrast imaging to investigate lung of small animals. Due to the significant difference in density between air and tissues, the inner edges of the airway can result in marked edge enhancement in the phase contrast images. This has previously been proven by Suzuki et al [16] . Nevertheless, for the large number of alveoli, the fine structures were averaged out. Other researchers like Parsons and Sera et al. have utilized IL-XPCI to study lungs, and successfully generated the three-dimensional models of mouse airways [7, 17] . Despite advances in lung imaging using IL-XPCI, threedimensional visualization of the alveoli in the intact animals has remained elusive. Nevertheless, the alveoli are important tiny structures through which gas exchange occurs. The morphology and distribution of alveoli can reveal information regarding lung health. For the diagnosis of drowning or other types of respiratory disorder like pneumonia, asthma, chronic bronchitis and emphysema, the shape of alveoli can be of considerable value [3, 18] . Furthermore, the morphology and growth of alveoli in embryos can reveal the secret of lung development, which has long fascinated biologists and mathematicians [19] . Therefore, the aim of this study is to explore these small alveoli. In this study, IL-XPCI technique was used to image a 1-day-old mouse in situ. For the demonstration of complex microstructures, we combined IL-XPCI with computed tomography (CT) technique to provide three-dimensional images of an intact mouse lung. Moreover, we described our investigations on the anatomical structures of alveolar duct, alveolar sacs, and alveoli. Quantitative assessment was carried out on the morphology of the alveoli. The lung is an essential respiration organ with complex threedimensional configuration. About 90% of the lung is filled with air. IL-XPCI Computed tomography study was performed on a 1-dayold mouse. The results showed that the major features of the lung, including bronchi, bronchioles, and alveoli can be clearly displayed. Moreover, it is remarkable that the alveoli can be observed in three dimensions. A CT projection image of the mouse lung is shown in Figure 1A . Although in this image, all the tissues overlapped with each other, some bronchi, lobes and many alveoli can be discerned. Figure 1B and C are two slices of the CT reconstruction images. Compared to conventional X-ray CT, in-line X-ray phase contrast CT had a much higher resolution (about 13 mm in our experiment). Bone, bronchi and alveoli were easily discernible with clear edges, which made it possible to separate different anatomical structures using the image segmentation technique. In order to extract the lung tissue and remove background noise, we applied a threshold-based image segmentation method to the CT slices. A comparison between original CT slice and threshold-based segmentation result is presented in Figure 2 . The threshold was chosen at the valley of the histogram ( Figure 2B ). Other pixels with intensity lower than the threshold were set to zero. After gray scale transformation, the resulted image was shown in Figure 2D . From the profiles at the white line drawn in the same place of the two CT slices, the background noise in the image was considerably reduced. Surface rendering method was used to visualize the lung, which allowed a clear depiction of complex spatial relationships and provided a strong space sense. By manually selecting the value of the reconstruction iso-surface, 3D models of tissues in the mouse chest cavity were obtained. All 3D models can be varied in size and rotated in real time to facilitate a detailed assessment of each structure (Video S1). As shown in Figure 3 , a surface rendering view of lung is displayed. The three-dimensional relationships among bronchiole, alveolar duct and alveoli were all visible. Through the segmentation of the CT slices, three models were acquired. Figure 3A and B give an overview of chest which combines ribs, bronchi, bronchioles, and alveoli. Figure 3C focuses on the display of bronchi and alveoli. Figure 3D is the result of image segmentation; it highlights the bronchial tree. The virtual endoscope video is also provided in the supporting information (Video S2). Moving down the respiratory tract from the trachea, the air ducts divided more into smaller branches. The alveoli are the final branching of the respiratory tree, where the gas/blood exchange occurs. The distribution of the alveoli reveals information regarding lung health. Immature 1-day-old mouse alveoli range in diameter from 100 to 150 mm [15] . The diameter of most alveoli in our reconstruction image was in this range. Figure 4 is a detailed demonstration of one small section of the lung. These images provided a magnified view of the alveolar sacs and the alveoli. The model could be cut at any angle, varied in size, and rotated in real time. As shown in Figure 4 , it was easy to measure the morphological characteristics, which was of great importance in the diagnosis of diseases related to alveolus and alveolar scar. In order to illustrate the alveoli more precisely, we only focused on several of them. Twenty-one single alveoli were chosen from different part of the lung. The three-dimensional images of the alveoli were shown in Figure 5A . The maximum diameter and the minimum diameter were measured in Figure 5B . The ratio of maximum and minimum diameters reflects the morphology of the alveoli to some extent. If this ratio approximates one, the shape of the alveolus is closer to a spheroid. The alveoli provide an enormous surface area for respiratory exchange; the change of alveolar surface area is closely related to lung health. For this reason, we calculated the surface area of each alveolus, as shown in Figure 5C . The lungs are composed of sponge-like soft tissues. Detection of soft tissues using conventional absorption-based radiography is limited by the variation in tissue density. Since PCI is a phasesensitive technique, it facilitates the visualization of fine structures, particularly in soft tissues. This technique has opened a new window for the visualization of lung. As a current imaging modality, clinical thin-section CT can be used to visualize airways with an internal diameter larger than 2 mm. It is better than conventional radiography and can provide more detailed information [6] . However, the terminal airspaces cannot be detected via this technique. Micro-CT for laboratory studies has a much higher resolution on the order of microns. Nevertheless, it is still based on the absorption information of the X-ray. The alveolar wall and the air space can not provide sufficient contrast to be observed [20] . A few studies have applied magnetic resonance imaging (MRI) to the lung imaging. Although MRI can provide excellent soft tissue imaging and functional imaging, the limitations lie in the low spatial resolution and long acquisition times [21] . The measurement of bronchial diameter will be hampered by these limitations, because normal spatial resolution of MRI is in the range of 4-6 mm and only a few can reach up to 1 mm [22] . The lung is an air-filled organ. At the air-tissue boundary, the refractive index changes significantly, so the lung becomes highly visible in PCI images. The lung images in our experiment have shown dramatic improvement in quality, and the alveoli can be clearly discerned. The theoretical spatial resolution of our experimental system is about 1 mm, while the diameter of mature mouse alveoli range from 38 to 80 mm and human alveoli are larger at 200-250 mm [15] . Thus, the resolution of PCI is adequate for micro-structures observation in lung. Though the radiation dose in our experiment was relatively larger than that used in clinical imaging, it can be reduced by the use of high energy X-ray, since contrast is not dependent on absorption of the beam [12, 15] . In addition, compared to other existing PCI methods, IL-XPCI requires relatively simple instrument. Most importantly, conventional laboratory X-ray source can be used as the light source, although longer exposure times are required [13] . The equipments are similar to conventional radiography. In this regard, it has the potential to be used in clinical diagnosis. When it comes to studying the architecture of micro-structures in lung, histological biopsy is the most commonly used method [9] . But it is invasive and can not be repeated [10, 23] . During the process of invasive sampling, anesthesia is often required and tissue deformation frequently happen which is a major disadvantage when studying the morphology of the organs. There has long been a desire to explain breathing mechanism at the alveolar level. Unfortunately, a full understanding of these fine structures' original morphology is not available, because the lung will collapse when opening the thoracic cage at autopsy, due to the loss of negative pleural pressure [18] . Some researchers used confocal microscope to evaluate alveolar dynamics in the mouse lungs. However, this technique is limited by the depth of imaging (about 50 mm) [24] . How lungs develop has long fascinated biologists and mathematicians. Nevertheless, this process cannot be visualized in living embryos with current techniques [19] . Therefore, we adopted IL-XPCI to observe the lung non-invasively. An intact mouse was used to the maximum extent to preserve the organ's original morphology. In addition, previous studies have shown that deceased lungs can maintain a sufficient aeration level to be visualized [15] . In conventional radiography, it is impossible to conduct alveoli imaging without contrast agent. Although contrast agents are generally safe, adverse effects do sometimes occur [25] . In PCI experiment, some researchers instilled saline into the nasal airway of the mouse in order to see the alveoli [17] . Actually, the alveoli in human lungs had been visualized by conventional radiography. Some researchers used micro-CT to image human lungs [20] . In this case, an autopsy lung from a dead person was used and silver nitrate was needed as the contrast agent. However, in practice, when small pathological changes happen, it is impossible for doctors to open a live patient's chest or stain his/her alveoli to see where the lesion is. Therefore, non-invasive and non-staining technique is preferable. In our study, the imaging of alveoli has shown promising results without any contrast agent. The 3D reconstruction technique provided an excellent view of lung and revealed structural details that were invisible to conventional radiography. The surface rendering results gave a perfect description of complex spatial relationships among bronchi, bronchioles and alveoli. The 3D volume data allowed virtual endoscope of the lung airways which helped the doctor visualize the 3D model of lung and perform a diagnosis without having to operate on the patient. Moreover, the measurement of alveoli showed the ability of PCI to observe the morphology of lung. Therefore, this technique has the potential use in disease diagnosis, such as asthma, chronic bronchitis and emphysema, which are associated with the size or morphology of lung airways. There are also potential uses in pharmacology when determining the optimum diameter of aerosolized drugs in lungs. One limitation in our research was the study sample. Live animals were not imaged. Despite the advances in spatial resolution and contrast of phase contrast imaging, real-time Xray phase tomography of the live animals has remained difficult. Live animals can only be imaged in two dimensions. The major obstacle in realizing this is the motion artifacts caused by breathing and cardiac motion of the animals, which will lead to serious blur in the reconstructed CT images. It is well known that the alveoli in lung are very small; any noise can influence the accuracy of the imaging result. Some attempts have been made to live animals imaging by taking the projection images at a specific breathing phase in synchrony with ECG signals [26] . But the animals have to be anaesthetized and with their breathing controlled by a ventilator. High-speed X-ray phase tomography may be another possible solution to live animal imaging by substantially reducing the imaging time. In fact, finishing a CT in one respiration phase is possible but the rotation speed of the sample stage and the time resolution of the detector need to be improved [27] . Another limitation was the lack of pathological samples. In this study, a normal 1-day-old mouse was used. The results proved that IL-XPCI is possible for lung micro-structures observation. Lung disorder models were missing, such as asthma, emphysema and lung cancer. We plan to evaluate these disorder models in our IL-XPCI imaging experiment as next steps, which are in progress now. In this study, in-line X-ray phase contrast imaging was used to visualize a lung of an intact mouse. Our findings showed that IL-XPCI had the ability to represent complex anatomical structures in lungs. The three-dimensional model of lungs has been successfully established which provided an excellent view of lung airways. Thanks to the high resolution of the imaging, deep study was carried out on the tiny alveoli. A full understanding of the alveoli's architecture can facilitate the study of respiratory mechanism on the alveoli level. Single alveolus was displayed and measured. This offers a new perspective on the diagnosis of respiratory disease. Pathological changes can be detected by measuring the size of the lung airways, although further research and experimentation on lung is required to test this hypothesis. All experiments and procedures carried out on the animals were approved by the animal welfare committee of Capital Medical University and the approval ID is SCXK-(Army) 2007-004. The study sample was a 1-day-old mouse, provided by Laboratory Animal Science, Capital Medical University. Before imaging, the mouse was humanely sacrificed by intraperitoneal (ip) sodium pentobarbital injection overdose. The mouse was imaged after four hours of its death in order to avoid the small shifts of the body caused by the development of rigor mortis [7] . And then a piece of Kapton (Dupont, DE, USA) was used and rolled up into a tube. This kind of film is an electrical insulation material with outstanding thermal, mechanical and chemical properties. The mouse was constrained in the tube, and placed on the sample stage in a vertically up-side-down position. The principle of IL-XPCI is based on Fresnel diffraction theory which can provide an edge-enhancement effect. Pogany, Gao and Wilkins first established the theoretical formalism for phase contrast image formation of weakly absorbing thin objects [28] . When X-ray beams travel through the object, the downstream beams carry the information of absorption and phase shift. The interaction between the object and the beam can be described by this transmission function where (x, y) is the spatial coordinates in the plane perpendicular to the propagation direction z, m and Q are the attenuation and phase shift induced by the object. They are given by where l is the wavelength of the X-ray, b and d are the imaginary and real parts of the refractive index, respectively. After propagating a sufficient distance, the phase shifts in the downstream beams are transformed into measurable intensity variations by means of Fresnel diffraction. An image detector records them as the phase contrast image. According to Fresnel diffraction theory and the wave front function, the Fourier transform of the recorded intensity can be approximated bỹ where d(u, v) is unit impulse function, (u, v) is the coordinate in Fourier domain, M and W are the Fourier transform of m and Q respectively at the distance z. From Eq. (4) one can find that the recorded intensity of the image is determined by the phase shift and the absorption of the wave. Apparently, the optimal contrast depends on the spatial frequency, wavelength, and the objectdetector distance [14] . The in-line X-ray phase contrast imaging experiment was performed at X-ray imaging and biomedical application beamline (BL13W1) of Shanghai Synchrotron Radiation Facility (SSRF). SSRF is the third-generation synchrotron radiation source in China. Figure 6 shows the schematic image of the experiment setup. It consisted of two monochromator crystals, one automatic rotation sample stage and one X-ray sensitive CCD detector. The incident white synchrotron X-ray beam was first monochromatized by two Si (111) prefect crystals. The tunable energy range was from 8 to 72.5 keV, with the energy resolution of about 0.5%. In our experiment, it was adjusted to 18 keV. The theoretical spatial resolution of the system was about 1 mm. Subsequently, the highly parallel and monochromatic beam projected on the object was imaged. The CCD detector then recorded the transmitted beam at a distance of 1.2 m from the sample. During this distance, the downstream image was enhanced by Fresnel diffraction and the phase modulation was transformed via amplitude modulation. An X-ray sensitive CCD camera, which had maximum resolution of 4008 pixels62672 pixels with 13 mm613 mm each, was used as a two-dimensional detector to transform the beam into an image. During the CT data acquisition, the specimen was rotated around its cylinder axis for 180u. The number of projections was 1296, with exposure time of 80 milliseconds for each projection, and the total scanning time was 208 seconds. The surface dose was about 8 mGy for each projection. All the parameters were selected in order to obtain high quality images of the mouse lung. Although IL-XPCI can provide high-resolution images of soft tissues, image processing is required during the analysis of the image. Two-dimensional projection images always suffer from spatial superimposition of the lesions; small lesions inside the tissue can not be detected. Therefore, 3D visualization method is preferable. First of all, the background image was used for normalization of the projection images. For the reconstruction of the CT slices, conventional filtered back projection (FBP) algorithm was used [29] . The ribs are bone structures, which had a distinct gray scale in PCI reconstructed CT slices. Threshold-based method was used for the segmentation of ribs. Then an image segmentation method was applied to separate the lung tissues from the background. Since the histograms of our lung images had two peaks and a valley between them, the threshold for image segmentation can be chosen automatically at the bottom of this valley [30] . The other pixels with intensity lower than the threshold were set to zero. After this step, the ribs associated with many background signals considered as noise were removed. The whole lung airways can be segmented from this volume dataset by manually set a gray-level threshold. Due to the complexity of the lung structures and their irregular shapes and similar gray levels on the images, threshold-based image segmentation method was inadequate to separate the bronchial tree from the whole lung airways. In this study, we developed an image segmentation method based on 3D region growing to separate bronchial tree which was a semi-automatic procedure. It started in manually placing the seed point in the section of bronchus as well as setting the intensity threshold. Under the control of the intensity threshold, the growing would stop when the bronchial tree were separated. After the above steps, three volume datasets of the ribs, the whole airways of the lung and the bronchial tree were obtained. Finally, the 3D models were generated by the use of surface rendering method. The surfaces are reconstructed by an isosurface detection algorithm which allowed a clear visualization of the complex spatial relationships of anatomical features. Video S1 Animated view of rendered bronchi and alveoli. This is the same 3D model shown in Figure 3C . The rotation of the model permits the viewers to observe the lung from different angles. The model can be cut from any angle in order to give a detailed view of the inner part of the lung. The bronchi, bronchioles, and alveoli are all visible. (AVI) Video S2 The virtual endoscope from the bronchi to one alveolar duct. This is the same model shown in Figure 3D . The volume data of the lung airways is separated from CT slices using 3D image segmentation method. 360u rotation of the model displays the spatial relationships of different lung airways. And the virtual endoscope of the model reveals the internal lung airways. (AVI)

Evolution of size and pattern in the social amoebas

A fundamental goal of biology is to understand how novel phenotypes evolved through changes in existing genes. The Dictyostelia or social amoebas represent a simple form of multicellularity, where starving cells aggregate to build fruiting structures. This review summarizes efforts to provide a framework for investigating the genetic changes that generated novel morphologies in the Dictyostelia. The foundation is a recently constructed molecular phylogeny of the Dictyostelia, which was used to examine trends in the evolution of novel forms and in the divergence of genes that shape these forms. There is a major trend towards the formation of large unbranched fruiting bodies, which is correlated with the use of cyclic AMP (cAMP) as a secreted signal to coordinate cell aggregation. The role of cAMP in aggregation arose through co‐option of a pathway that originally acted to coordinate fruiting body formation. The genotypic changes that caused this innovation and the role of dynamic cAMP signaling in defining fruiting body size and pattern throughout social amoeba evolution are discussed. BioEssays 29:635–644, 2007. © 2007 Wiley Periodicals, Inc.

The social amoebas or Dictyostelia represent one of nature's several independent inventions of multicellularity. The dictyostelia are members of the amoebazoans, a genetically highly diverse group that is the closest sister group to the clade containing the animals and fungi. (1) Except for the myxomycetes, all other known amoebazoans are microscopic unicellular organisms. The myxomycetes alternate a trophic amoeboid stage with a syncytial form. Here, a single cell with millions of nuclei, can grow up to several meters across. (2) Social amoebas also have a trophic amoeboid stage, but they achieve macroscopic dimensions by aggregation. (3) This occurs in response to starvation, which triggers regulated secretion of chemoattractant by the amoebas (Fig. 1) . Cellular agglomerates are formed, which can consist of up to a million amoebas. Sophisticated cell-cell signalling mechanisms between the amoebas orchestrate the differentiation of up to five different cell types and coordinate an intricate progression of cell movements. In combination with the synthesis of a flexible skin-like matrix, cell differentiation and cellular movement first generate the formation of a motile structure, called the ''slug''. The slug responds to chemical gradients and to light and warmth, which cause it to move to the soil's top layer. Here, the slug projects upwards and forms the fruiting body. This again involves highly ordered movement and differentiation and yields a slender column of stalk cells that bears aloft a global mass of spores. Depending on the species, the stalk can show different patterns of side branches and/or be decorated with disc, root or cup-shaped support structures (Fig. 2) . Unlike the ontogeny of sessile organisms like plants and fungi, which depends largely on series of directional cell divisions, the formation of fruiting bodies in social amoebas is more similar to the ontogeny of animal form. Both depend strongly on an intertwined program of cell movement and cell differentiation. Seminal work of Raper showed similarity of principle in the establishment of the body plan in Dictyostelium and vertebrate development. In vertebrate development, a small group of cells known as ''the organizer'' releases signals that coordinate cell movement during gastrulation and neurulation and thereby generate the animal's head-to-tail body axis. (4) Raper demonstrated that, in Dictyostelium aggregates, small groups of cells, recognizable as tips, secretes signals that generate anteroposterior polarity of all or a subpopulation of cells in the aggregate, yielding one or several slugs with a distinct head and body region. (5) More recent work shows that animals and Dictyostelia share many conserved pathways for processing external signals. Particularly the elucidation of the processes that control chemotaxis in Dictyostelium have become a paradigm for understanding cell migration in animals. (6) (7) (8) However, the external signals that trigger movement or differentiation are rarely conserved. For instance, growth-factor-like peptides and their tyrosine kinase receptors that play such crucial roles in animal development are not present in Dictyostelium. The homeobox-containing transcription factors that specify segment identity in arthropods and vertebrates have only a minor function in Dictyostelium development. (9) (10) (11) The signaling repertoire of the Dictyostelia is rather different. The model species, D. discoideum makes extensive use of cyclic AMP (cAMP), the ubiquitous intracellular messenger for hormone action in vertebrates. In D. discoideum, cAMP not only mediates the effect of a number of external signals, but is also secreted to act as a chemoattractant and inducer of cell differentiation. (12) Secreted peptides trigger maturation of spores, but are detected by sensor-coupled histidine kinases. (13) Polyketide-based metabolites and adenine-based cytokinins are secreted to regulate cell-type proportioning and spore germination respectively. (14, 15) Animals, plants and Dictyostelia evolved from different unicellular ancestors, supposedly the filter-feeding choanoflagellates for animals, (16) green algae for plants (17) and solitary amoebas for the Dictyostelia. (18) These unicellular progenitors used sensory signaling to monitor their environment and to find food and mates. The different developmental strategies that are now used by their multicellular descendants reflect how evolutionary forces differentially selected, duplicated and adapted these environmental sensing mechanisms for increasingly complex communication between cells. Historical reconstruction as a tool to understand developmental signalling By acting on alleles that are generated by random mutation, organic evolution is intrinsically opportunistic. It does not create optimal design, but selects combinations of traits that provide the highest probability of reproduction in a specific ecological niche. Consequently, there are no unifying schemes to explain organismal development. The underlying molecular mechanisms only truely make sense in the context of their evolutionary history. What mechanisms were used by the ancestors and how were these mechanisms modified to improve functionality in the better adapted descendants? The evolution of novel forms in multicellular organisms requires alteration of the developmental mechanisms that shaped the earlier form. Evo-devo, short for evolutionary developmental biology, is a relatively young discipline that sets out to retrace how gene and genome modifications have altered existing developmental mechanisms to produce novel forms. Evo-devo has been predominantly applied to the development of animals (19) (20) (21) and, to a lesser degree, of higher plants. (22, 23) However, an understanding of development of other multicellular organisms, such as the social amoebas or fungi will equally benefit from this approach. In the model organism D. discoideum, starving amoebas secrete cAMP pulses, which trigger chemotactic movement and aggregation of cells. Once aggregated, the amoebas differentiate into prestalk and prespore cells in a regulated ratio. The organizing tip continues to emit cAMP pulses, which shape the cell mass by coordinating cell movement. The cAMP pulses also cause the prestalk cells, which are chemotactically most responsive, to move towards the front. At the onset of culmination, the cells synthesize a cellulose tube, the apical prestalk cells move into the tube and mature into stalk cells, the remaining prestalk cells form support structures, such as the upper and lower cup and the basal disk. The prespore cells move up the stalk and mature into spores. Moreover, the greater genetic tractability of these organisms will greatly aid in establishing how gene modification caused novel forms to appear. The social amoebas provide other opportunities to retrace the evolution of multicellular development. All known species can be grown under laboratory conditions and complete their multicellular life cycle within a 28 hour period. They show a broad range of different morphologies, with terminal structures varying in size between 0.1 mm and several centimeters. The genome of the model species D. discoideum is completely sequenced (10) and sequencing of four other Dictyostelia genomes is in progress. About 7 years ago, we initiated research into the evolutionary history of developmental signaling in the social amoebas, concentrating on cAMP signalling. A primary requisite for this project was the availability of a family tree that shows the relatedness of social amoeba species relative to the more ancestral solitary amoebazoans. We joined forces with the teams of Sandra Baldauf, an expert in protist molecular phylogeny, Thomas Winckler, who had already prepared an SSU rRNA tree of a subset of Dictyostelia species and two Dictyostelium field biologists, Jim Cavender and Hiromitso Hagiwara, to construct a molecular phylogeny of all known Dictyostelia. (24) We next mapped morphological traits of all species to the tree in order to determine the directionality of morphological evolution in the social amoebas. In parallel studies, the presence, regulation and function of genes that are essential for various aspects of cAMP signalling were investigated in social amoeba species that span the phylogeny. (25) This review presents a synthesis of the outcome of these studies with current insights in developmental signaling in the model species D. discoideum. It highlights major trends in the evolution of multicellular complexity in the social amoebas, and correlates these trends with elaboration of function of deeply conserved cAMP signalling genes. In traditional systematics, the Dictyostelia were grouped with the acrisid amoebas in the division of mycetozoans in the kingdom of fungi. (3, 26) However, recent molecular evidence shows that none of these groups are fungi; the acrasid amoebas are members of the discicristates, while the Dictyostelia and the mycetozoans are members of the amoebozoans. (1, 18, 27, 28) Based on fruiting body architecture, the social amoebas were subdivided into three genera: the dictyostelids with unbranched or laterally branched fruiting structures, the polyspondylids with regular whorls of side branches and the acytostelids with acellular stalks. Comparison of conserved DNA or protein sequences is a more direct and reliable method to establish genetic relationships. Two family trees of the social amoebas were constructed by comparing the DNA sequences of their small subunit ribosomal RNA (SSU rRNA) gene on one hand and the amino-acid sequences of their a-tubulin protein on the other. Both trees show that the similarities in fruiting body architecture only partially reflect an underlying genetic similarity. (24) Instead, both the a-tubulin tree and the SSU rRNA tree shown here (Fig. 3 ) subdivide the 75 known species of social amoebas into four major groups. There are dictyostelids in all four groups. The acytostelids and all white polysphondylids are members of group 2, but the purple polysphondylid P. violaceum occupies a position between groups 3 and 4, and forms a small clade with the dictyostelid D. laterosorum. This indicates that at least two out of the three previously proposed genera are polyphyletic. Multiple origins for the polyspondylids were also predicted by a family tree that was based on 18 combined morphological traits. (29) The DNA-based family tree was subsequently used to investigate trends in the evolution of morphology. The multicellular stages of different species of Dictyostelia show a large variety of shapes and sizes, which have been carefully quantitated and noted in the original species diagnoses, along with differences at the cellular level (see Refs 3, 30 for overviews). Species use different chemoattractants and aggregate as single cells or as inflowing streams (Fig. 4) . Once formed, aggregates may either produce one or several organizing tips, giving rise to solitary or clustered fruiting bodies. Secondary tips may appear in characteristic positions on rising sorogens, giving rise to secondary body axes and a range of different fruiting body architectures. Fruiting body stalks may develop a variety of support structures, such as discs, crampons or triangular supporters, while their tips can vary from thinly pointed to bulbous. Many species form motile slugs, which may optionally form a stalk while migrating. At the cellular level, spores can be round or oblong and, in the latter case, display conspicuous granules at their poles, which are either loosely grouped or consolidated. Some species have retained the ancestral survival strategy of encystation, or display the capacity to mate and form sexual macrocysts. A mapping of all these characters to the molecular tree indicates which characters are shared between close relatives, and yields information about the order in which characters evolved (Fig. 5) . Amoeba size shows no strong group-specific trend, but spores are consistently smaller in the most basal group 1. Of all traits, spore morphology follows the phylogeny most strongly. Spores of the evolutionary youngest group 4 species have no polar granules, in group 2, polar granules are loosely grouped, and in groups 1 and 3, they are consolidated. The acellular stalk evolved only once. The shape of the stalk tip also marks relatedness; species in group 2 usually have pointy or blunt tips, while in groups 1 and 4 stalk tips tend to be club-or head-shaped. Species with similarly coloured stalks and spore heads are often related, but no color is specific for any of the four major groups. Encystation of individual cells is lost from group 4 species, but retained in the evolutionary older groups, while sexual macrocysts are made by species scattered over all four groups. The chemoattractant that is used for aggregation is known for only a few species. It is cAMP for all investigated group 4 species while, in the other groups, at least three other compounds are used. Most species aggregate as inflowing streams of amoeba, but groups 1-3 contain some species that aggregate as individuals. Slug migration also occurs in all groups, but is most common in group 4. Stalkless migration is shown by a small cluster of group 4 species and a single nongroup 4 species, D. polycephalum. Fruiting structures of most but not all species throughout the phylogeny veer towards light (phototropism). Fruiting bodies (sorocarps) tend to be clustered or grouped in groups 1 to 3 and solitary in group 4, while branched structures are also more common in the basal groups. Specific branching patterns do not show strong group-specific trends and laterally branched, rosary-type and whorled morphologies appear, respectively, six, two and five times across the tree. There is a modest trend towards taller sorocarps in the more-derived members of groups 1-3, and a very strong trend towards sorocarps with long thick stalks and large spore heads (sori) in group 4. These large sorocarps are usually buttressed by cellular support structures, such as basal disks and supporters. The crampon-base is almost uniquely associated with a tight cluster of group 3 species. In summary, the most-obvious trend in the evolution of social amoebas appears to be related to size. Evolutionary younger species both have larger sized spores and larger sized fruiting bodies. The latter is particularly evident in group 4 where large stalk and sori size is correlated with a tendency to form solitary and unbranched fruiting bodies. In addition to large size, group 4 displays other distinguishing features, such as formation of cellular support structures, loss of individual encystation, loss of spore granules, and the use of cAMP as attractant. The correlation of the latter two traits was also noted earlier by Traub and Hohl. (31) These workers also associated the presence of polar granules with a tendency to form clustered and/or branched sorocarps and a tendency for those sorocarps to be smaller than in species without polar granules, both of which are borne out by the recent analysis. The adaptive advantage of larger spores and fruiting bodies can be surmised. Larger spores may store more nutrients to survive dormancy, while larger fruiting bodies may aid spore dispersal, both contributing to species propagation. Individual encystation may have become redundant after the sporulation mode of survival became more robust. It is less easy to envisage how loss of spore granules and use of cAMP as attractant improved fitness. In the following paragraphs, we explore how the latter character may have been a means to achieve an end. The appearance of new morphologies in multicellular organisms requires alteration of existing developmental pathways. In the social amoebas, developmental pathways have only been studied in detail in the model organism D. discoideum, where cell-to-cell communication is largely mediated by secreted signaling molecules. cAMP plays a primary role; it is secreted in periodic waves by aggregation centres to mediate the aggregation of starving cells. (32) Later, organizing tips become the sources of cAMP waves (Fig. 1) , which direct the movement of cells in multicellular structures. (33) Secreted cAMP also triggers the differentiation of the prespore cells. (34, 35) In turn, the prespore cells secrete a chlorinated polyketide, DIF, that induces regulated redifferentiation of prespore cells into prestalk cells. (14) Ammonia, which is produced by protein degradation in the starving cells, represses terminal spore and stalk cell maturation during slug migration, (36, 37) and is implicated in cell sorting, slug phototaxis and fruiting body phototropism. (38, 39) Two secreted peptides, conditioned medium factor (CMF) and prestarvation factor (PSF) induce the growth to development transition. (40) Other peptides, the spore differentiation factors (SDFs), trigger the maturation of spores. (13) There is no information on the conservation of the peptide signals in other social amoeba species. Ammonia is always produced by starving amoebas, and at least some of its roles are therefore likely to be conserved. DIF was identified in another group 4 species, D. mucoroides, which also has the DIF-degrading enzyme, DIF dechlorinase. D. minutum and P. violaceum, which reside in group 3 and between groups 3 and 4, respectively, synthesize chlorinated factors that induce stalk cell differentiation in D. discoideum. However, they do not have DIF dechlorinase, indicating that the DIF signaling pathway is at best partially conserved. (41, 42) More information is available on the conservation of cAMP signalling. All tested group 4 species use cAMP to aggregate, but no species outside group 4. However, early biochemical work showed that species that use other attractants to aggregate, nevertheless display cAMP binding sites and cAMP phosphodiesterase on their cell surface after aggregation. (43, 44) The group 3 species, D. minutum, aggregates by continuous release of folic acid, but shows cAMP waves emerging from the tip region after aggregates have formed. (43, 45) This suggested that the role of the tip as a pacemaker of cAMP waves is deeply conserved in the social amoebas. Cell surface cAMP receptors (cARs) mark the use of cAMP as an extracellular signal. D. discoideum has four homologous cARs. cAR1 is expressed shortly after starvation, while cAR3, cAR2 and cAR4 are expressed at progressively later stages. (45, 46) Recent studies show that the cAR1 gene is deeply conserved in social amoeba evolution and is present in all four taxon groups. (25) The gene duplications that gave rise to cAR2, cAR3 and cAR4 only occurred in group 4 (Y. Kawabe and P. Schaap, unpublished results). The basal cAR1s are functionally identical to D. discoideum cAR1, but there is a marked difference in their developmental regulation. The cAR1s from the basal groups are expressed as a single mRNA after aggregation while, in group 4 species, a second cAR1 mRNA is expressed before and during aggregation. (25) Transcription of this early mRNA is driven by a second promoter that is more distal from the cAR1 coding sequence than the promoter that drives expression after aggregation. (47) Also in the gene encoding the extracellular cAMP phosphodiesterase, the promoter that drives late expression is proximal to the coding sequence and the promoters that drive expression before and during aggregation are more distal. (48) This arrangement suggests that the use of cAMP as chemoattractant by the evolutionary younger group 4 species was achieved by addition of distal promoters to existing cAMP signalling genes. Loss of cAR1 function in group 4 species blocks aggregation and further development. In the basal groups 1-3, aggregation is unaffected, but the subsequent formation of slugs and fruiting bodies is disrupted. (25) This indicates that, in the more basal species, extracellular cAMP signalling is required for slug and fruiting body morphogenesis. cAMP signaling and size regulation In addition to using cAMP for aggregation, group 4 species also stand out by having large solitary unbranched fruiting structures, as opposed to the clustered and branched smaller structures that are common to the other groups. Are cAMP signalling and size related? The segmentation of aggregates into clusters of fruiting bodies and the formation of side branches all represent the formation of multiple body axes that are initiated by newly emerging tips (Fig. 4) . Analogous to the phenomenon of apical dominance in plants, where lateral shoots are suppressed by the primary shoot, (49) D. discoideum tips suppress the formation of ancillary tips. D. discoideum tips are selforganizing pacemakers for cAMP waves. The waves are propagated through the cell mass by cAMP-induced cAMP production, also known as cAMP relay. Tip dominance can be established in different manners: (1) higher frequency oscillators entrain cells that oscillate at lower frequency, (50) and (2) tips produce a diffusible inhibitor that reduces the excitability of surrounding cells. (51) The cAMP hydrolysis product adenosine was proposed to fulfill this role. (52, 53) Dominance will break down if there are physical or biochemical barriers that prevent propagation of cAMP waves or diffusion of the inhibitor. Irrespective of the exact mechanism, dominance is intrinsic to oscillatory cAMP signalling. Group 4 species have larger fruiting bodies because their cAMP oscillators are better at suppressing competitors. Two aspects of cAMP signalling are specific to group 4: (1) oscillatory cAMP signalling occurs much earlier in development than in groups 1-3, and (2) the cAMP receptor gene was duplicated three times, and both expression and affinity of the daughter cARs were altered. It is conceivable that either of these novelties may have ''improved'' cAMP signalling to allow it to control larger numbers of cells and make it generally more robust. The plasticity of fruiting body architecture The DNA-based phylogeny of the social amoeba did not reproduce the earlier classification into three genera that was based on fruiting body architecture. In fact, it appeared that many similar fruiting body branching patterns evolved several times independently (Fig. 5 ). This implies that specific architectures cannot be under extensive genetic control. As discussed above, fruiting body branching patterns reflect how and when competing pacemakers for cAMP waves appear on multicellular structures. The production of cAMP waves by D. discoideum cells consists of a positive feedback loop where extracellular cAMP acting on cAR1 stimulates further cAMP production by adenylyl cyclase A (ACA), and a negative feedback loop where cAMP inhibits ACA and stimulates its own hydrolysis. (54) (55) (56) Variation in a range of parameters, such as the relative expression levels of the component proteins, diffusion or cell movement barriers generated by structural components, and the relative motility or cohesiveness of responding cells can potentially affect the dynamics of the signalling process in a such a way as to allow competing pacemakers to arise in a variety of configurations. For instance, D. gloeosporum, which owes its name to its extremely sticky spore matrix, (57) is the single dictyostelid member of the clade of white polysphondylids (Fig. 3) . The branched whorls of the polysphondylids are formed when a group of cells detaches from the rear of a rising cell mass and then forms new tips (Fig. 4) . D. gloeosporum may owe its consolidated single spore head to the fact that, due to the highly adhesive matrix, detachment of cell masses does not occur. Similarly, other fruiting body architectures are likely to result from interaction of the cAMP signaling network with different biophysical environments, rather than being controlled by architecture-specific genes. Does branching have any adaptive value at all? I believe it does. Within groups 1-3 there is a trend towards taller fruiting bodies in the more-derived species. Being carried in the air on tall stalks may not only aid spore dispersal but also contribute to spore preservation, away from the decomposing agents in wet humus. However, the construction of a robust stalk comes at a cost of reducing the spore-to-stalk ratio. Group 4 species resolve this problem by additional cell-type specialization to form support structures. For the basal groups, branching and particularly whorl formation may provide a solution for the problem of building tall well-balanced fruiting bodies without sacrificing too many spores. This review summarizes the recent construction of a systematic framework to study causal relationships between genotypic and phenotypic change during evolution of the social amoebas. The foundation of this framework is the first DNAbased phylogeny for all known species of social amoebas. The phylogeny, which is based on SSU rDNA sequences and confirmed by a-tubulin protein sequences shows subdivision into four major groups and a molecular depth that is equal to that of all animals. (24) A plotting of the most consistently noted species characters onto the phylogeny shows unexpected trends in character evolution with the greatest changes occurring at the transition between the youngest group 4 and the evolutionary older groups 1,2 and 3. Group 4 species are characterized by large solitary and unbranched fruiting structures as opposed to smaller, clustered and branched structures in the other groups. Group 4 species have also lost the ancestral survival strategy of encystation and gained the use of cAMP as chemotactic signal for aggregation. A study into the evolutionary origins of extracellular cAMP signalling revealed that this strategy is used by all social amoebas to coordinate the process of fruiting body formation. Group 4 species have recruited this mechanism to additionally control the aggregation process. This occurred by adding aggregation-specific promoters to existing cAMP signalling genes. Many intriguing questions remain unresolved. Social amoebas are the only known organisms that use cAMP as extracellular signal. How did this role of cAMP originate in the first place? cAMP signals are produced by oscillating pacemakers. Are these dynamics unique for cAMP or are other chemoattractants, such as glorin, also released in an oscillatory manner? Are other D. discoideum signal molecules, such as DIF, ammonia, SDF, PSF and CMF conserved throughout the phylogeny? Thus far, only those features were plotted to the phylogeny that were observed by standard light microscopy. One cellassociated character, the presence of granules in spores proved to be the strongest group-defining determinant. This suggests that there are other characters at the cellular level that define species within groups. More detailed (ultra)microscopic analysis would be required to identify such features. In D. discoideum, the proportion of prespore and prestalk cells in slugs are regulated to the approximate proportions of stalk and spores in the fruiting body. However, in other species such as P. violaceum, P. pallidum and D. lacteum, cells first differentiate into prespore cells only to dedifferentiate into stalk cells at the tip. (58, 59) Apart from stalk cells and spores, D. discoideum has three more cell types, the basal disc, upper cup cells and lower cup cells, that each display specific patterns of gene expression. When and how did cell-type proportioning and greater cell-type specialization evolve? In addition to these development-related aspects, the molecular phylogeny provides a framework to investigate conservation and divergence of any protein with an important function in the cell biology of D. discoideum. This is useful for identification of conserved domains and/or amino-acids in proteins that are thus far not well characterized, but also to outline how protein modification gave rise to novel protein functions. Projects are now in progress to sequence the genomes of at least four group-representative social amoeba species. Combined with detailed information of phenotypic evolution, this information on the evolution of genotype will provide tremendous opportunities to retrace how this particular form of multicellular life evolved.

Plant Plastid Engineering

Genetic material in plants is distributed into nucleus, plastids and mitochondria. Plastid has a central role of carrying out photosynthesis in plant cells. Plastid transformation is becoming more popular and an alternative to nuclear gene transformation because of various advantages like high protein levels, the feasibility of expressing multiple proteins from polycistronic mRNAs, and gene containment through the lack of pollen transmission. Recently, much progress in plastid engineering has been made. In addition to model plant tobacco, many transplastomic crop plants have been generated which possess higher resistance to biotic and abiotic stresses and molecular pharming. In this mini review, we will discuss the features of the plastid DNA and advantages of plastid transformation. We will also present some examples of transplastomic plants developed so far through plastid engineering, and the various applications of plastid transformation.

Genetic material in plants is distributed into nucleus and the chloroplast and mitochondria in the cytoplasm. Each of these three compartments carries its own genome and expresses heritable traits [1, 2] . The chloroplast is one of organelles known as plastids in plant cells and eukaryotic algae [3] . According to Verhounig et al. [4] , plastids and mitochondria are derived from formerly free-living bacteria and have largely prokaryotic gene expression machinery. The plastid (biosynthetic centre of the plant cell) carries out photosynthesis, in plant cells and eukaryotic algae, which provides the primary source of the world's food [3] . There are other important activities that occur in plastids. These include sequestration of carbon, production of starch, evolution of oxygen, synthesis of amino acids, fatty acids, and pigments, and key aspects of sulfur and nitrogen metabolism [5] . In spite of the prokaryotic past of the plastids, their gene expression has very different regulatory mechanisms from those operating in bacteria [6] . There are up to 300 plastids [7] in one plant cell. The plastid genome (plastome or plastid DNA, ptDNA), 1,000-10,000 copies per cell [8] , contrasts strikingly with the nuclear DNA. In most species, plastids are usually strictly maternally inherited [9] in most (80%) angiosperm plant species [10, 11] . It is also not influenced by polyploidy, gene duplication and recombination that are widespread features of the nuclear genomes of plants [12, 13] . Therefore, ptDNA varies little among angiosperms in terms of size, structure and gene content [14] . Currently 170 DNA. The authors, therefore, developed a simple and inexpensive method to obtain plastid DNA from grass species by modifying and extending protocols optimized for the use in eudicots. Plastid engineering involves the targeting of foreign genes to the plastid's double-stranded circular DNA genome instead of chromosomal DNA [31] , and as a consequence the production of the foreign protein of interest (Fig. 1) . The advancement in the particle gun mediated transformation has enabled targeting the plastid genome for developing transgenic plants against many biotic (e.g. insects and pathogens) and abiotic stresses (e.g. drought and salinity), which reduce the plant productivity. Improving the quality of fruits has been another main target in plastid transformation [32] . Plastid transformation was first achieved in a unicellular alga, Chlamydomonas reindhartii [33] . In 1990, Sváb et al. [34] reported the first successful chloroplast transformation of a higher plant (tobacco). This was followed by transformation of the plastid genome in tobacco by many researchers [35, 36] . Recently, tobacco plastid has been engineered to express the E7 HPV type 16 protein, which is an attractive candidate for anticancer vaccine development [37] . Similarly, a protocol for plastid transformation of an elite rapeseed cultivar (Brassica napus L.) has been developed [38] . Brassica napus - [132] Citrus sinensis 155,189 [133] Coffea arabica - [134] Cucumis sativus 155,293 [135] Daucus carota 155,911 [136] Ficus sp. - [137] Glycine max 152,218 [138] Gossypium barbadense 160,317 [139] Gossypium hirsutum 160,301 [140] Helianthus annuus 151,104 [141] Hordeum vulgare, Sorghum bicolor - [142] Lycopersicon esculentum 155,460 [21] Manihot esculenta - [143] Morus indica 156,599 [144] Musa acuminata - [145] Nicotiana tabacum 155,943 [146] Oryza sativa 134,551 [15] Oryza nivara 134,494 [16] Phaseolus vulgaris 150,285 [147] Saccharum officinarum 141,182 [18] Solanum tuberosu 155,298 [148] Spinacia oleracea 150,725 [149] Triticum aestivum 134,545 [19, 20] Vitis vinifera 160,928 [150] Zea mays 22,784 [17] Jatropha curcas 163,856 [151] Vigna radiata 151,271 [22] More recently, a method for plastid transformation in eggplant (Solanum melongena L.) has been reported with pPRV111A plastid expression vector carrying the aadA gene encoding aminoglycoside 300-adenylyltransferase [26] . The authors believe that this may open up exciting possibilities to introduce and express novel genes in the engineered plants via plastid transformation for agronomic or pharmaceutical traits. Up to date, plastid transformation has been extended to many other higher plants, such as Arabidopsis thaliana [39] , potato [40, 41] , tomato [1, 42] , Lesquerella fendleri, a kind of oilseed Brassicaceae [43] , oilseed rape, [38, 44] , petunia [45] , lettuce [46] , soybean [47] , cotton [48] , carrot [49] , rice [50] , poplar [51] , tobacco, [28, 52, 53] , mulberry , [54] and eggplant [26] (see review written by Wang et al. [3] ). Two interesting applications of plastid transformation were carried out by (i) [55] for the construction of a tobacco master line to improve Rubisco engineering in plastids, and (ii), [4] who explored the possibility of engineering riboswitches (natural RNA sensors that regulate gene expression in response to ligand binding) to function as translational regulators of gene and transgene expression in plastids. The dominant trait that attracted the most attention for plastid transformation has been herbicide tolerance [28, 35, 56, 57] . Roudsari et al. [28] revealed the production of high level glyphosate tolerant plants (N. tabacum) through biolistic transformation of plastids by introduction of a mutated herbicide-tolerant gene coding for EPSP synthase. Plastid transformation is routine, however, only in tobacco and the efficiency of transformation is much higher in tobacco than in other plants [58] . Lee et al. [50] discussed the major obstacles to the extension of plastid transformation technology to other crop plants which includes regeneration via somatic embryogenesis. Production of transgenic plants, at laboratory level or commercially, has traditionally been mainly through expression of transgenes in the nucleus [24, 25] . Among the ecological concerns raised about genetically engineered organisms is that transgenes could move ("transgene flow", the process of transgene movement by recurrent hybridisation) via pollen from the crop and into relatives growing in natural or semi-natural communities [59] . Such concerns have led to a new field of transgene containment [60, 61] . Since plastids are inherited maternally in the majority of angiosperm species, they would therefore not be found in pollen grains of corps. Insertion of transgenes, therefore, into the plastid genome has the potential of preventing gene flow via pollen. Hence, genes expressed in the plastome will not be transferred through pollination to weedy or wild relatives of the transgenic crop. Bansal and Sharma, [62] believes that there is little risk of any transgene flow via pollen from transplastomic plants to the neighboring weedy or wild relatives since plastids are almost always maternally transferred to the next progeny. The authors suggested that plastid transformation could be a method of choice for generating improved transgenics in crops that grow along with their weedy or wild relatives in the same geographical region, such as rice, sorghum, cucurbits, solanaceous crops, Vigna and Cajanus species, and various Brassica crops. Therefore, the focus of many researchers has shifted to plastid engineering [26] , rather than nuclear transformation. Singh et al. 2010 [26] reported that engineering of the plastid genome is gaining momentum as an attractive alternative to nuclear transformation. Ruf et al. [60] believe that plastid transformation is considered as a superb tool for ensuring transgene containment and improving the biosafety of transgenic plants. However, they pointed out that plastid transformation would only be effective as a biocontainment measure when applied on a landscape scale if it were combined additional mechanisms such as mitigating genes genetic use restriction technology, and/or male sterility [63] . In a recent study, it has been demonstrated that the use of plastid transformation would provide an imperfect biocontainment for GM oilseed rape (Brassica napus L.) in the United Kingdom [64] . In another study, Allainguillaume et al. [65] revealed that chloroplast transformation may slow transgene recruitment in two settings, but actually accelerate transgene spread in a third. Plastid transformation has become an attractive alternative to nuclear gene transformation due to several other advantages [3] . The high ploidy number of the plastid genome allows high levels (up to 1-40% of total protein) of protein expression or expression of the transgene [28] . Daniell et al. [66] and Hou et al. [44] reported that while nuclear transgenes typically result in 0.5 -3% of total proteins, concentration of proteins expressed by plastid transgenes is much higher; up to 18%. The greater production of the expressed protein is possible because plastid transgenes are present as multiple copies per plant cell, and they are little affected by phenomena like pre-or post-transcriptional silencing. Other advantages of plastid engineering are the capacity to express multiple genes from polycistronic messenger RNA (mRNA) [31] , and the absence of epigenetic effects and gene silencing [40] . Wang et al. [3] believe that transgene stacking in operons and a lack of epigenetic interference allowing stable transgene expression. Added to that, plastid transformation is more environmental friendly than transformation of the nuclear DNA for plant engineering because it eliminates the possibility of toxic transgenic pollen to nontarget insects [67] . Adverse effects of toxic proteins might be minimized by plastid compartmentalization but in case of nuclear transformation, toxic proteins accumulating within the cytosol might result in serious pleiotropic effects. Further, the expression of the transgene in case of plastid transformation is more uniform compared to that of trangenes inserted into the nuclear genome. Although there is a major drawback in the engineering of plastid gene expression, which is the lack of tissue-specific developmentally regulated control mechanisms [3] , the many advantages of plastid engineering stated above attracted researchers to engineer the plastid genome to confer several useful agronomic traits, and hence the number of species whose plastome can be transformed continues to expand [68] . Gene delivery into the plastome was initially done by Agrobacterium mediated method [Block et al.1985 ] [69] . The discovery of biolistic DNA delivery led to plastid trans-formation via particle gun [Sanford 1990 ] [70] . In this method, the Escherichia coli plasmids contain a marker gene and the gene of interest is introduced into plastids. The foreign genes are inserted into plasmid DNA by homologous recombination via the flanking sequences at the insertion site [66] . Polyethylene glycol (PEG) mediated transformation of plastids was also utilized [71, 72] . PEG-mediated transformation of plastids requires enzymatically removing the cell wall to obtain protoplasts, then exposing the protoplasts to purified DNA in the presence of PEG. The protoplasts first shrink in the presence of PEG, then lyse due to disintegration of the cell membrane. Removing PEG before the membrane is irreversibly damaged reverses the process. Biolistic delivery is the routine system for most laboratories, as manipulation of leaves, cotyledons, or cultured cells in tissue culture is a simple practice than the alternative PEG treatment of protoplasts [58] . Recently, particle gun mediated plastid transformation has been demonstrated in rapeseed using cotyledons as explants [38] . The major difficulty in engineering plastid genome for production of transplastomic plants is in generating homoplasmic plants in which all the plastids are uniformly transformed, for that takes a long process of selection, thus hampering the production of genetically stable transplastomic plants (e.g. rice). This is due to the presence of about 10-100 plastids, each of which has up to 100 copies of the plastid genome, in one cell, that does not allow achieving homoplastomic state [73] . It was also stated that getting high level of protein expression, even though the gene copy number is high, is another problem. In 2005, however, Nguyen et al. [41] described the generation of homoplasmic plastid transformants of a commercial cultivar of potato (Solanum tuberosum L.) using two tobacco specific plastid transformation vectors, pZS197 (Prrn/aadA/psbA3 ) and pMSK18 (trc/gfp/Prrn/aadA/psbA3 ). Similarly, Liu et al. [74] were able to develop homoplasmic fertile plants of Brassica oleracea L. var. capitata L. (cabbage). Among other higher plants of which fertile homoplasmic plants with genetically modified plastid genomes have be been produced are Nicotiana tabacum (tobacco), Nicotiana plumbaginifolia (texmex tobacco), Solanum lycopersicum (tomato), Glycine max (soybean), Lesquerella fendleri (bladderpod), Gossypium hirsutum (cotton), Petunia hybrida (petunia), and Lactuca sativa (lettuce) [68] . The amino glycoside 3adenylyltransferase (aadA) gene, which confers dual resistance to spectinomycin-streptomycin antibiotics, is still the selectable marker that is routinely used efficiently for plastid transformation [58, 75, 76] . Since the antibiotic resistant genes used in transformation are not desirable in the final products, different strategies have been developed to eliminate the necessity of using such selectable markers [56, 77] . Apel & Bock, [42] demonstrated the potential of plastids genome engineering for the nutritional enhancement of food crops when they enhanced carotenoid biosynthesis in transplastomic tomatoes by induced lycopene-to-provitamin A conversion. The transplastomic technology could also be useful for engineering agronomic traits including phytoremediation [78] reversible male sterility [79] , and toler-ance/resistance of stresses such as diseases, drought, insect pests, salinity and freezing that can severely limit plant growth and development [3] . Since plastids are transferred mostly through the "maternal inheritance" as identical copies, and hence a female plant transfers identical copies to all the seeds it produces without changes from one generation to the next, an important promise for applying plastid transformation for industry is the stable passing on to the next generation of the foreign DNA [80] . Therefore, the plastid genome has also been utilized for metabolic pathway engineering and in the field of molecular farming (the production of drugs and chemicals through engineered crops) [27, 68] for the expression and production of biomaterials and biopharmaceuticals in plants, human therapeutic proteins, and vaccines for use in humans or animals (reviewed in [26, 27, 50] . Singh et al. [26] believe that for such applications, plastid transformation technology offers solutions to the ecological and technical problems associated with conventional transgenic technologies such as outcrossing and transgene silencing. Transformation of the nuclear genome of plants with genes (e.g. Bt genes) to confer insect resistance gives very low levels of expression unless extensive modifications are carried. Whereas, introduction of the same genes into the plastid genome results in high levels of toxin accumulation as the plastid genome is bacterial in origin [81] . Therefore, the insect resistance genes were investigated for high-level expression from the plastid genome [3] . Hence, when insect resistance genes are expressed into the plastid genome, leaves of these transplastomic plants proved highly toxic to herbivorous insect larvae. One of the major advantages of introducing the Bt toxin into the plastid genome is the high levels of toxin accumulation (3% -5% of total leaf protein as compared to > 0.2% of total soluble protein through nuclear genome transformation) [82] . Achievement of stable transformation of the plastid genome and transforming plastids in species other than tobacco (Nicotiana tabacum) are some of the hurdles for widespread adaptation of this technique [81] . Despite of overwhelming odds, many attempts have been made to produce transplastomic plants expressing Bt toxin for increased resistance against insect pests. [83] generated soybean plastid transformants expressing Bacillus thuringiensis Cry1Ab protoxin. Similarly, Chakrabarti et al. [84] reported the control of potato tuber moth (Phthorimaea operculella) by incorporating a truncated Bacillus thuringiensis cry9Aa2 gene in the plastid genome. The authors observed high-level expression (about 10% of total soluble protein) of the cry gene from the plastid genome which resulted in severe growth retardation. Over-expression of the cry2Aa2 operon in plastids was proved effective in allowing a broad-spectrum of protection against a range of pests [81] . In cabbage, the cry1Ab gene was also successfully transferred into the plastid genome [85] . Expression of cry1Ab protein was detected in the range of 4.8-11.1% of total soluble protein in transgenic mature leaves of the two species. Insecticidal effects on Plutella xylostella were also demonstrated in cry1Ab transformed cabbage. In an attempt to increase insect resistance in trans-genic rice plants, a synthetic truncated cry1Ac gene was linked to the rice rbcS promoter and its transit peptide sequence (tp) for plastid-targeted expression [86] . Use of the rbcS-tp sequence increased the cry1Ac transcript and protein levels by 25-and 100-fold, respectively, with the accumulated protein in plastids comprising up to 2% of the total soluble proteins. The high level of cry1Ac expression resulted in high levels of plant resistance to three common rice pests, rice leaf folder, rice green caterpillar, and rice skipper, as evidenced by insect feeding assays. It was concluded that targeting of cry1Ac protein to the plastid using the rbcS:tp system confers a high level of plant protection to insects. Several other cry proteins have also been expressed in plastids of tobacco [83, 84, 87, 88] and rice [50] (see Table 2 ). Plastid engineering offers a new and effective option in development of plant varieties which are resistant to various bacterial and fungal diseases. In tobacco, introduction of MSI-99 gene, an antimicrobial peptide, into plastids resulted in transplastomic plants resistant to fungal pathogen Colletotrichum destructive [89] . The plastids expressed MSI-99 at high levels and showed 88% (T1) and 96% (T2) inhibition of growth against Pseudomonas syringe, which is a major plant pathogen. In another study, Agrobacterium mediated transformation was used to develop tobacco plants carrying argK gene, which encodes ROCT [90] . Since OCT in plant cells is produced in the plastid, argK was fused to the plastid transit sequence of the pea rubisco small subunit (rbcS) gene for localized expression of the enzyme. The ROCT enzyme produced by the transgenic tobacco showed greater resistance (83-100%) to phaseolotoxin compared to the wild-type OCT (0-22%). When phaseolotoxin was applied exogenously to the leaves of plants, chlorosis was observed in 100% of wildtype tobacco, but not seen in the leaves of the transgenic tobacco plants carrying the argK gene from P. syringae pv. phaseolicola. Transgenic tobacco plants that constitutively expressed both entC and pmsB in the plastid have also been reported [91] where transformation was accomplished through biolistic methods. The transgenic tobacco plants expressing these bacterial genes showed accumulation of salicylic acid that were up to 1000 times higher than that observed in wild-type tobacco. When challenged with the fungus Oidium lycopersicon, the transgenic tobacco plants showed increased levels of resistance compared to the wildtype plants. It was revealed that the transgenic plants generated did not show any adverse effects due to the high level expression of salicylic acid. Thus gene transfer in plastids can provide a significant protection from various bacterial and fungal diseases. Transgenes that confer tolerance to abiotic stress may permanently transfer from transgenic crops to the nuclear genome of their weedy relatives which may result in drought tolerant superweeds [92] when the gene is inserted into the nuclear genome and the transgenic plant outcross with relative weeds. There is a great potential, therefore, for the genetic manipulation of key enzymes involved in stress metabolism in plants within plastids. Because plastid genomes of major crops including cotton and soybean have been Nicotiana tabacum nptII [153] Nicotiana tabacum uidA [154] Nicotiana tabacum Human somatotropin (hST) [155] Nicotiana tabacum cry [88] Nicotiana tabacum cry9Aa2 [84] Nicotiana tabacum Bar & aadA [156] Nicotiana tabacum Cor 15a-FAD7 [157] Nicotiana tabacum rbcL [55] Nicotiana tabacum DXR [158] Nicotiana tabacum aadA & gfp [60] Nicotiana tabacum Delta(9) desaturase [159] Nicotiana tabacum AsA2 [160] Nicotiana tabacum PhaG & PhaC [161] Nicotiana tabacum gfp [4] Nicotiana tabacum A1AT [162] Arabidopsis thaliana aadA [39] Solanum tuberosum aadA & gfp [40] Oryza sativa aadA & gfp [50] Solanum lycopersicon aadA [1] Solanum lycopersicon Lyc [42] Brassica napus aadA & cry1Aa10 [44] Brassica napus aadA [38] Lesquerella fendleri aadA & gfp [43] Daucus carota dehydrogenase (badh) [48] Gossypium hirsutum aphA-6 [49] Glycine max aadA [47] Petunia hybrida aadA & gusA [45] Lactuca sativa gfp [46] Brassica oleracea gus & aadA [163] Lettuce gfp [164] Populus alba gfp [51] Brassica oleracea aadA & uidA [74] Beta vulgaris aadA & uidA [165] Crocus sativus CstLcyB1& CstLcyB2a [166] Solanum melongena aadA [26] Arabidopsis thaliana pre-Tic40-His [167] Zea mays ManA [168] successfully transformed, this offers an exciting new approach to create transgenic plants with abiotic stress tolerance [93] . Therefore, the authors believe that there appears to be tremendous potential for increasing tolerance in plants to a number of stresses by expression of appropriate genes within plastids due to the maternal inheritance of transgenes that confer tolerance to abiotic stress. Plastid engineering had been successfully applied for the development of plants with tolerance to salt, drought [94] and low temperature [reviewed in 3]. Djilianov et al. [95] demonstrated that enhanced tolerance to abiotic stresses has been achieved when the gene was directed to plastid genome. Among many strategies used for development of abiotic stress tolerance in plants, the over-expression of compatible osmolytes like glycinebetaine was found to be successful [24] . Initial attempts for producing transplastomic plants through the introduction of CMO (Choline monooxygenase) and betaine aldehyde dehydrogenase (BADH) pathway were made in tobacco. Tobacco plants were transformed with cDNA for BADH from spinach (Spinacia oleracea) and sugar beet (Beta vulgaris) under the control of CaMV 35 S promoter. The BADH was produced in plastids of tobacco. Betaine aldehyde was converted to betaine by BADH, thus conferring resistance to betaine aldehyde. In another attempt, cDNA for choline monooxygenase from Spinacia oleracea was introduced into tobacco and the enzyme thus synthesized was transported to its functional place i.e., plastids. But the leaves of tobacco accumulated betaine at a very low concentration i.e., 10-100 folds lower [96] . The reason for insufficient synthesis of betaine most probably was the absence of engineered BADH activity in plastids. Therefore, both CMO and BADH need to be present in the plastids for efficient synthesis of betaine in transgenic plants which do not accumulate glycinebetaine. In carrot (Daucus carota), homoplasmic transgenic plants exhibiting high levels of salt tolerance were regenerated from bombarded cell cultures via somatic embryogenesis [48] . BADH enzyme activity was enhanced 8-fold in transgenic carrot cell cultures, grew 7-fold more, and accumulated 50-to 54-fold more betaine than untransformed cells grown in liquid medium containing 100 mM NaCl. Transgenic carrot plants expressing BADH grew in the presence of high concentrations of NaCl (up to 400 mM), the highest level of salt tolerance reported so far among genetically modified crop plants. Further, a gene for CMO, cloned from spinach (Spinacia oleracea) was introduced into rice through Agrobacterium mediated transformation. The level of glycinebetaine in rice was low to the expectations. The author has given several reasons for the low productivity of rice and low glycinebetaine accumulation. Firstly, the position of spinach CMO and endogenous BADH might be different and secondly the catalytic activity of spinach CMO in rice plants might be lower than it was in spinach [97] . Transplastomic plants constitutively expressing BvCMO under the control of the ribosomal RNA operon promoter and a synthetic T7 gene G10 leader were able to accumulate glycinebetaine in leaves, roots and seeds, and exhibited improved tolerance to toxic level of choline and to salt/drought stress when compared to wild type plants. Transplastomic plants showed higher net photosynthetic rate and apparent quantum yield of photosynthesis in the presence of 150 mM NaCl [98] . Thus, it can be concluded that glycinebetaine has a role as compatible solute and its engineering into non-accumulations will be a success only if both CMO and BADH pathways are introduced and if the localization of both CMO and BADH is in plastids. Very recently, George et al. [99] demonstrated how a chloroplastlocalized and auxin-induced glutathione S-transferase from phreatophyte Prosopis juliflora conferred drought tolerance on tobacco. For more examples of conferring tolerance to abiotic stress to plants via plastid engineering, see review by [3] . Mulesky et al. [100] pointed out two reasons that make using crops to produce drugs interesting for industry. These are (i) crops can be employed more efficiently in this process than animals or bacteria, with a larger output achieved with fewer resources, and (ii) the oral delivery of the drugs produced to people and animals is easier. A third reason is the high-level production of antigens for use as vaccines and their tests for immunological efficacy in animal studies [3] . The hyper-expression of vaccine antigens or therapeutic proteins in transgenic chloroplasts (leaves) or chromoplasts (fruits/roots) and antibiotic-free selection systems available in plastid transformation systems made possible the oral delivery of vaccine antigens against cholera, tetanus, anthrax, plague, and canine parvovirus, [101] [102] [103] [104] and reviews of [102, 105, 3 and 106] explained why plastid engineering can be regarded as an attractive strategy and environmentally friendly approach for the production of vaccines, therapeutic proteins, and biomaterials, and provided some examples. Kumar & Daniell [107] described various techniques for creating plastid transgenic plants and their biochemical and molecular characterization. They also provided suitable examples for application of chloroplast genetic engineering in human medicine. Wang et al. [3] also discussed applying plastid transformation for metabolic pathway engineering in plants, the production of biopharmaceuticals, and marker gene excision system and how plastid transformation can be applied to study RNA editing. Similarly, Hefferon [67] , considers plastid engineering as a valuable tool that gives enormous promise for the production of biopharmaceuticals and vaccines, because higher level of the protein expressed by the transgene inserted into the plastid genome can be achieved. Many vaccine antigens, which played a key role for the prevention of infectious diseases, and biopharmaceutical proteins, have been expressed at high levels via the chloroplast genome [108] and they proved to be functional using in vitro assays in cell cultures. [109] stated that production of therapeutic proteins in plastids eliminates the expensive fermentation technology, and that the oral delivery of plastid-derived therapeutic proteins eliminates cold storage, cold transportation, expensive purification steps, and delivery via sterile needles, and hence decrease their cost. The main goal for applications of plastid transformation by the biotech industry is molecular pharming, and food production is considered as only a secondary target [80] . To create an edible vaccine, selected desired genes should be introduced into plants and then inducing these altered plants to manufacture the encoded proteins. Like conventional subunit vaccines, edible vaccines are composed of antigenic proteins and are devoid of pathogenic genes. Plastids of green plants as bioreactors for the production of vaccines and biopharmaceuticals are of great potential as indicated from a number of published studies [110] [111] [112] [113] [114] . The significance of using plants as production platforms for pharmaceuticals is due to the low production and delivery costs, easy scale-up and high safety standards regarding less risk of product contamination with human pathogen [27] . Keeping in view the high efficiency of plastids to express foreign genes, it is meaningful to explore this property of plastids for the production of proteinaceous pharmaceuticals, such as antigens, antibodies and antimicrobials. The candidate subunit vaccine against Clostridium tetani, causing tetanus was the first plastid-produced antigen that proved to be immunologically active in experimental animals [111] . In this initial attempt, fragment C of the tetanus toxin (TetC), a non-toxic protein fragment, was expressed from the tobacco plastid genome which resulted in high levels of antigen protein expression (30% of the plant's total soluble protein (TSP)). Anthrax is an acute infectious disease caused by the spore-forming bacterium Bacillus anthracis. Significant development has been achieved towards the production of plastid-based vaccine for this infectious disease. Expression (14% of the plants TSP) of the pagA gene encoding the protective antigen (PA) from the tobacco plastid genome gave rise to stable antigen protein [112, 113] . The plastid-derived PA was equally effective in cytotoxicity assays as the bacterial protein produced in B. anthracis. The potential of plastid transformation as an alternative tool to produce high levels of HIV-1 Nef and p24 antigens in plant cells have been also demonstrated [115] . Different constructs were designed to express the p27 Nef protein either alone or as p24-Nef or Nef-p24 fusion proteins. All constructs were utilized to transform tobacco (cv. Petite Havana) plastids and the transplastomic lines. Analysis of p24-Nef and Nefp24 fusion proteins showed that both can be expressed to relatively high levels in plastids. As the best results in terms of protein expression levels were obtained with the p24-Nef fusion protein, the correspondent gene was cloned in a new expression vector. This construct was introduced into the tobacco and tomato plastid genomes. Transplastomic tobacco and tomato plants were analyzed and protein accumulation was found to be close to 40% of the leaf's total protein. Transcript and protein accumulation were analyzed in different ripening stage of tomato fruit and green tomatoes accumulated the fusion protein to 2.5% of the TSP [115] . Recently, a strategy for plastid production of antibiotics against pneumonia Streptococcus pneumonia has been outlined [116] . The authors describe it as a new technique for high level expression (to up to 30% of the plant's TSP) of antmicrobial proteins that are toxic to E. coli. It was also shown that the plastid-produced antibiotics efficiently kill pathogenic strains of Streptococcus pneumoniae, the causative agent of pneumonia, thus providing a promising strategy for the production of next-generation antibiotics in plants. In 2007, Chebolu and Daniell [117] achieved a stable expression of Gal/GalNAc lectin of Entamoeba histolytica in transgenic chloroplasts and immunogenicity in mice towards vaccine development for amoebiasis. Shao et al. [118] reported the expression of the structural protein E2 of classical swine fever virus (CSFV), which has been shown to carry critical epitopes CFSV E2 gene in tobacco chloroplasts. In another study on tobacco, plastid transformation of the high-biomass tobacco variety `Maryland Mammoth` has been assessed by McCabe et al. [119] as a production platform for the human immunodeficiency virus type 1 (HIV-1) p24 antigen. Similarly, Meyers et al. 2008 [120] revealed the usefulness of plastid signal peptides in enhancing the production of recombinant proteins meant for use as vaccines. Transgenic plastids were also proved efficient for high-yield production of the vaccinia virus envelope protein A27L in plant cellsdagger by Rigano et al. [121] , who revealed that chloroplasts are an attractive production vehicle for the expression of OPV subunit vaccines. Very recently, Youm et al. [122] were able to produce the human beta-site APP cleaving enzyme (BACE) via transformation of tobacco plastids. The authors argued that the successful production of plastid-based BACE protein has the potential for developing a plant-based vaccine against Alzheimer disease. Because recombinant extra domain A from fibronectin (EDA) could be used as an adjuvant for vaccine development, [104] aimed to express EDA from the tobacco plastome as a promising strategy in molecular farming. Tobacco plastids transformation was also evaluated by Lentz et al. [123] for the production of a highly immunogenic epitope containing amino acid residues 135-160 of the structural protein VP1 of the foot and mouth disease virus (FMDV). The authors concluded that this technology allows the production of large quantities of immunogenic proteins. In spite of this huge success in applying plastid engineering for molecular pharming, Ho & Cummins [73] referred to several risks of plastid engineering in producing GM pharmaceuticals that are associated with its advantages. For more details on the use of plastid engineering for biotechnology applications, see review by Verma & Daniell [5] . Other valuable reviews are available on (a) generation of plants with transgenic plastids, summary of our current understanding of the transformation process and highlights on selected applications of transplastomic technologies in basic and applied research [124] , (b) Progress in expressing proteins that are biomedically relevant, in engineering metabolic pathways, and in manipulating photosynthesis and agronomic traits and the problems of implementing the technology in crops [125] , (c) plastid transformation in higher plants [56] , (d) the characteristics, applications of chloroplast genetic engineering and its promising prospects, [126] , (e) Engineering the chloroplast genome [127] (f) exciting developments in this field and offers directions for further research and development, [128] (g) the expression of resistance traits, the production of biopharmaceuticals and metabolic pathway engineering in plants [27] , (h) chloroplasts as bioreactors, and whether we can replace plants with plants [129] , (i) how plastid transformation played an important role in understanding the RNA editing [3] and (j) comparison of opportunities and challenges between nuclear and plastid genetic engineering of plants [130] . Up to date, many transgenes have been successfully introduced into the plastid genome of model plant tobacco and many other important crop plants for various agronomic traits. Initially, this technology was limited to model plant species, but now it has been extended to some other important crops. Still there are many agronomically important cereals crops in which plastid engineering has not yet been standardized. Plastid transformation has been proved to result in high levels of transgene expression. It has also provided a baseline for production of proteinaceous pharmaceuticals, such as antigens, antibodies and antimicrobials in a cost effective manner. Bock [27] believes that a great progress has been achieved over years in investigating the mechanisms that govern transgene expression from the plastid genome and in using this technology for biotechnological applications. We agree with the author that "the routine use of plastid engineering in biotechnology is still a long way off, but would surely benefit the humanity in the near future". There is no doubt that plastid engineering holds a great potential in plant biotechnology; but like every new technology, there are some challenges which need to be addressed before its widespread adoption. Among other important factors to be solved are the protein purification and expression level control.

Role of glutathione in immunity and inflammation in the lung

Reactive oxygen species and thiol antioxidants, including glutathione (GSH), regulate innate immunity at various levels. This review outlines the redox-sensitive steps of the cellular mechanisms implicated in inflammation and host defense against infection, and describes how GSH is not only important as an antioxidant but also as a signaling molecule. There is an extensive literature of the role of GSH in immunity. Most reviews are biased by an oversimplified picture where “bad” free radicals cause all sorts of diseases and “good” antioxidants protect from them and prevent oxidative stress. While this may be the case in certain fields (eg, toxicology), the role of thiols (the topic of this review) in immunity certainly requires wearing scientist’s goggles and being prepared to accept a more complex picture. This review aims at describing the role of GSH in the lung in the context of immunity and inflammation. The first part summarizes the history and basic concepts of this picture. The second part focuses on GSH metabolism/levels in pathology, the third on the role of GSH in innate immunity and inflammation, and the fourth gives 4 examples describing the importance of GSH in the response to infections.

It is now thought that oxidative stress is implicated in the pathogenesis of several diseases and conditions, from aging to inflammation and carcinogenesis, if not as a primary cause of disease at least as an aggravating mechanism. The reality of oxidative stress is demonstrated by the existence, in all aerobic organisms, of several antioxidant enzymes devoted to ROS detoxification, such as peroxidase, catalase, superoxide dismutase, and peroxiredoxins. A number of small molecules also act as ROS scavengers. Unlike antioxidant enzymes, which catalyze the transformation of ROS into less toxic molecules, the concept of scavenger is that it must be an easily oxidizable target for ROS, and it must be present at concentrations high enough that the probability that a ROS reacts with the scavenger is higher than that of reacting with another target. When dealing with lipid oxidation (such as in rancidification of butter), the most important scavenger is probably vitamin E. As proteins with a sulfydryl group are often target of oxidation (-SH groups can be easily oxidized), small-molecular-weight thiols are potent scavengers. The small thiol present in highest concentration in the cytoplasm is glutathione (GSH), a tripeptide glycine-cysteineglutamic acid ( Figure 2 ): the -SH group of its cysteine is extremely sensitive to oxidation, mainly by peroxides. Its importance is supported by the existence of a molecular machinery that makes it particularly effective. In fact, a scavenger, including any thiol antioxidant such as the common laboratory reagent beta-mercaptoethanol, would normally act as a suicidal decoy, being oxidized and thus becoming useless. However, when GSH is oxidized, it forms GSH disulfide (GSSG), and this can be re-reduced by a specific enzyme, glutathione reductase. Not only can GSH be enzymatically regenerated from GSSG, but also the reaction of GSH with the ROS (a peroxide), which already takes place with high reactivity, is catalyzed by GSH peroxidases that facilitate the inactivation of a wide range of peroxides ( Figure 3 ). The key role of GSH as an antioxidant is demonstrated by many studies showing that experimental depletion of GSH levels -which can be achieved with various chemicals the most used of which is buthionine sulfoximine (BSO) an inhibitor of GSH synthesis -has a worsening effect in GSH in immunity and lung inflammation many disease models. Conversely, repleting GSH levels with precursors of its synthesis such as N-acetyl-cysteine (NAC) or 2-oxothiazolidine-4-carboxylic acid has protective effects. Cysteine precursors, rather than cysteine itself, are used because they are more cell-permeable, and can be given orally. Innate immunity and inflammation, two faces of the same biological coin The first line of immune defense against pathogens, before adaptive immunity (antibodies, T cell responses) develops, is called innate immunity. This is a complex set of responses triggered when specific cells (macrophages, phagocytes, dendritic cells) recognize even a yet unknown pathogen by some characteristics common to most pathogens (these "signatures" are called pathogen-associated molecular patterns) through a family of pathogen recognition receptors. This activates a response that involves production of ROS, a major bactericidal mechanism, and of soluble mediators (cytokines) whose role is to amplify the host response by recruiting and activating other immune system cells. This aspect of the innate immune response is also known, from a different perspective, as the inflammatory response. Basically, the very same mechanisms and mediators of host defense are implicated in the pathogenesis of (noninfectious) inflammatory diseases, and inhibition of these mechanisms is the key to the mechanism of action of antiinflammatory drugs. Many noninfectious diseases of the respiratory system, including asthma, chronic obstructive pulmonary disease (COPD), cystic fibrosis, idiopathic pulmonary fibrosis (IPF), and oxygen toxicity, have an inflammatory component. Inflammation is also implicated in the lung toxicity of ozone, asbestos, silica, cigarette smoke, and particulate matter. An exaggerated inflammatory response is also involved in the pathogenesis or complications of pulmonary infections such as tuberculosis, severe acute respiratory syndrome, influenza, and acute respiratory distress syndrome (ARDS). GSH is synthesized from its 3 amino acids with a biosynthetic pathway shown in Figure 4 . A study in humans, using radioisotopes, 3 has demonstrated that the availability of cysteine,and its precursor methionine, is the rate-limiting factor in GSH synthesis. In general, it is assumed that the two main limiting factors are the levels of cysteine and of the enzyme gamma-glutamyl-cysteine synthetase (also known as glutamate-cysteine ligase). 4 Glutathione synthetase (GS) deficiency (oxoprolinuria) is a rare autosomal recessive disorder affecting about 70 patients in the world. Patients with severe GS deficiency show, among other conditions, an increased susceptibility to bacterial infections. 5, 6 One case report showed that neutrophils from a child with GS deficiency undergo oxidative damage upon phagocytosis, indicating that GSH is important in defending the neutrophils from the ROS they produce as part of their microbicidal armamentarium. 7 Of note, neutrophils from GS-deficient patients, despite normal phagocytosis and increased release of hydrogen peroxide, are less efficient in killing bacteria, indicating a helper role for GSH in the bactericidal activity. 8 Many pathological conditions are associated with decreased GSH levels (Table 1 ). This could be due to several reasons. For instance, oxidative stress could cause GSH loss though oxidation. Another important aspect is nutrition, as it was shown that, even when dietary protein intake is sufficient for maintaining nitrogen balance, it may not be sufficient for maintaining cellular GSH, particularly in conditions of oxidative stress. 9 GSH deficiency could also arise from metabolic problems. For instance, in AIDS patients, a decrease in gamma-cystathionase activity in the liver was reported. 10 The use of stable radioisotopes allowed the characterization of cysteine metabolism and GSH synthesis in septic patients. This study showed that sepsis decreases blood GSH synthesis by 60%. 3 Of note, these patients had an intake of sulfur amino acids below that recommended by the World Health Organization. 3 Studies in animal models have reported an increased requirement for cysteine in sepsis, probably due to an increased turnover of GSH, as GSH synthesis accounts for 40% of the increased cysteine utilization. 11, 12 These and other studies show that even when the dietary intake of cysteine is sufficient for protein synthesis (nitrogen balance), it may not be sufficient to maintain adequate GSH levels. 9 One should also bear in mind that infectious and inflammatory conditions are associated with an increased production of acute-phase proteins by the liver, and it was estimated that they account significantly for the increased utilization of sulfur amino acids, thus competing with GSH. 13, 14 GSH as a regulator of innate immunity Anti-inflammatory role of GSH in lung diseases The idea that GSH may play an anti-inflammatory role became popular in the 1990s with a study by Schreck et al 15 showing that antioxidants inhibit, whereas ROS activate, the transcription factor NF-κB that commands the transcription of several inflammatory genes, 16 a role that has been confirmed in the lung. 17 In fact, a protective role of GSH against inflammatory pathologies of the lung has been demonstrated by the protective effect of different GSH-precursors in various animal models ( Table 2) . Along the same line, depleting endogenous GSH with BSO had a worsening effect in some of the models above, including chemically induced pulmonary edema, 21 cigarette smoke, 31 carrageenan-induced pleurisy, 32 and endotoxininduced pulmonary inflammation. 33 It can bee seen that there is an overlap between the list of pulmonary diseases associated with inflammation and those where GSH repletion is protective, indirectly supporting the hypothesis of an anti-inflammatory role of GSH. An increased susceptibility to sepsis-induced ARDS and lethality 34 was observed in mice lacking the transcription factor Nrf2, which has among its target genes the GSH biosynthetic enzyme gamma-glutamylcysteine synthase. 35 IPF is an example of lung disease in which normalizing the GSH deficit has positive effects. Patients with IPF have a 4-fold lower GSH concentration in the epithelial lining fluid of the normal lower respiratory tract; 36 moreover, the administration of the GSH precursor NAC not only restores GSH levels 37 but, in association with other drugs used for the treatment of IPF, it also improves end points such as vital capacity and single-breath carbon monoxide diffusing capacity. 22 The functions of GSH are not only inhibitory as described above for the inflammatory response. In fact, GSH is essential for some functions of the immune system, both innate and adaptive, including T-lymphocyte proliferation, 38, 39 phagocytic activity of polymorphonuclear neutrophils (PMN), 40 and dendritic cell functions, 41 and is also important for the first step of adaptive immunity, consisting of the antigen presentation by antigen-presenting cells (APC). Indeed, cell-mediated immunity requires that protein antigens be first degraded in the endocytic vesicles of APCs (eg, macrophages, dendritic cells), so that the smaller peptides can be presented on the cell surface by the major histocompatibility complex to activate proliferation of antigen-specific T cells. One of the first steps in antigen degradation and processing is the reduction of disulfide bonds, 42 which requires GSH. 43 It should also be noted that although GSH inhibits the production of most inflammatory cytokines, it is required to maintain an adequate interferongamma (IFN-gamma) production by dendritic cells, 44 which is essential for the host defense against intracellular pathogens such as mycobacteria. 45 This essential role of GSH in immunity might explain why in many diseases, not only AIDS, decreased GSH levels are associated with an increased susceptibility to infection. These include COPD, 46 cystic f ibrosis, 47, 48 influenza infection, 49 and alcoholism, 50,51 as ethanol impairs Th1/Th2 balance via GSH depletion 52 (Table 3) . The role of oxidative stress in the pulmonary damage by influenza virus is well characterized in the mouse model. Mice infected with influenza show pulmonary damage associated with a dramatic decrease in pulmonary GSH levels as well as an increase of oxidative stress markers such as oxidized glutathione (GSSG) and lipid peroxides. 53 In these mice, oxidative stress could be due to the induction of xanthine oxidase, 54 a well-known superoxide-generating enzyme. 55 Furthermore, influenza infection is associated with an induction of inflammatory cytokines, 56,57 which might represent a likely GSH-sensitive step in its pathogenesis. Among the antioxidants shown to be protective in animal models of influenza are the xanthine oxidase inhibitor allopurinol, 54 quercetin, 58 superoxide dismutase, 59 thioredoxin, 60 NAC (alone 61 or with ribavirin 62 or oseltamavir 63 ), and GSH itself. 64 A study reported that influenza virus M2 protein augments ROS production in human airway cells in vitro, resulting in toxic effects that are prevented by the addition of GSH ester. 65 Although these studies indicate a protective role of GSH at the level of the pathogenesis of pulmonary damage, there is a report that BSO increases viral replication, 66 implying a possible antiviral role of endogenous GSH. One clinical study has shown that NAC administration improves parameters of cell-mediated immunity in patients with influenza, 66 suggesting a possible clinical relevance of these observations. Historically, the entire field of the role of GSH in immunity and its effect on NF-κB received the strongest boost by studies on AIDS, particularly by a study from the group of Wulf Droge who showed that AIDS patients have a low concentration Ghezzi of plasma cysteine. 67 The reasons for this are not clear, but AIDS patients have a deficit in the enzyme gamma cystathionase, which synthesizes cysteine from the methionine. 10 Later research showed that AIDS causes a decrease in intracellular GSH in CD4 T cells, and that low GSH is associated with decreased survival. 68 The lower GSH levels in AIDS patients could have various consequences. GSH depletion, at least in vitro, augments HIV replication, 69 while its precursor NAC blocks the stimulatory effect of tumor necrosis factor on HIV replication. 70 Furthermore, the neurotoxic effect of HIV proteins Tat and gp120 is associated with oxidative stress and antagonized by NAC. 71, 72 Example 3: bacterial infectionstuberculosis Mycobacterium tuberculosis is an intracellular pathogen that grows in the phagosomes, where it is protected from immune system effectors such as antibodies and T lymphocytes. Although the literature showing that GSH levels are lower in patients with tuberculosis dates back to the 1950s, it was not until the research of Venketaraman and colleagues that the effect of GSH on M. tuberculosis infection was studied in depth. Using a mouse macrophage cell line, the authors show that IFN-gamma and endotoxin increase both nitric oxide (NO) production and bactericidal activity; the paralleled decrease in GSH suggests that GSH reacts with NO to form S-nitrosoglutathione (GSNO). Under these experimental conditions, GSH depletion with BSO inhibited the microbicidal activity of macrophages while its precursor NAC increased intracellular killing of mycobacteria also from human macrophages, which are normally not very effective in killing mycobacteria. 73, 74 Other investigators have shown that the trans-sulfuration pathway, which converts methionine into cysteine and has a key role in maintaining cysteine, and hence GSH levels, is essential for mycobacterial killing. 75 In that study, it was found that not only are trans-sulfuration pathway enzymes increased in human monocytes as a response to mycobacteria, but their inhibitor propargylglycine lowered GSH levels and inhibited clearance of mycobacteria and phagolysosome fusion, while NAC increased them. 75 Similar results were obtained in whole human blood cultures. 76 In vitro, both GSH and GSNO have direct bactericidal activity against these pathogens. 77 This complex picture has been nicely reviewed by Connell and Venkataraman 78 and is summarized in Figure 5 . Depletion of GSH (by BSO or diethylmaleate) also inhibits macrophage leishmanicidal activity, as well as NO production, while the GSH precursor GSH-ethyl ester restores them, 79 suggesting that the requirement for the GSH/NO pathway might be a common feature of resistance to intracellular pathogens. ARDS is one of the most serious complications in critically ill septic patients. Many studies have pointed out a role for oxidative stress in ARDS and shown a protective effect of GSH precursors. 80, 81 Most of the studies of ARDS in animal models, including those cited above, are based on the administration of lipopolysaccharide (LPS), a bacterial endotoxin. In mice, administration of LPS induces an acute lung injury similar to clinical ARDS, with production of inflammatory cytokines, leukocyte infiltration in the lung, and pulmonary edema. In this model, various thiol-based antioxidants are protective. [80] [81] [82] However, LPS administration is in fact a model of endotoxic shock, involving no live bacteria, to induce a state similar to what was once called septic shock and is now defined as systemic inflammatory response syndrome. [83] [84] [85] However, in septic patients, survival is affected by 2 opposite contributions of innate immunity: innate immunity is detrimental as systemic inflammation, which results in ARDS and shock, but on the other hand it is an essential component of the immune defense against infection. It is difficult to foresee how GSH affects this balance. One more realistic animal model is that induced by cecal ligation and puncture (CLP), where puncturing the cecum causes the release of fecal material in the peritoneum that results in a polymicrobial peritoneal sepsis. This model allows studying the relevance of both arms of innate immunity, the detrimental and the protective one. We have used this model to study the role of endogenous GSH in sepsis. 86 In this model, CLP induced PMN infiltration in the peritoneal cavity, the site of infection, as well as in the lung, ultimately resulting in lung injury and death, with a concomitant decrease in endogenous GSH. 86 Lowering GSH further with BSO worsened the clinical settings. Not only did BSO increase PMN infiltrate in the lung, but it also diminished PMN infiltration in the site of infection, thus increasing bacterial growth. As a result we had more inflammation and less immunity, and survival was dramatically decreased. Repleting GSH with NAC had the opposite effect of reducing PMN infiltration to the lung but increasing that to the site of infection, thus decreasing bacterial colonies. The picture that emerges from these experiments is that endogenous GSH is not just an inhibitor of inflammation, but it is required to allow a proper response to infection, and "direct" the migration of inflammatory PMN away from the lung, where they cause ARDS, and towards the site of infection, where they kill bacteria ( Figure 6 ). The idea, therefore, is that GSH is not just an inhibitor of inflammation but also a regulator of innate immunity in a direction favorable to the host. In parallel to the studies of GSH in immunity, more complex molecular and biochemical studies have pointed out a regulatory role of GSH. This was a result of an evolution from the concept of oxidative stress outlined above to that of redox regulation. The concept of redox regulation implies that some oxidative changes (such as changes in the redox state of protein cysteines) are not necessarily damaging but can have regulatory properties. While this idea shows a more complex picture from the popular one where free radicals and oxidants are bad and antioxidants are good, it implies that GSH is an essential molecule not only in acting as an antioxidant but also in the absence of oxidative stress, as an endogenous signaling molecule. The molecular mechanisms of GSH-mediated redox regulation are being actively investigated and have in part been identified. [87] [88] [89] Although the present review focused on the immuno pathogenesis of pulmonary diseases, the key concepts outlined here may help in interpreting the role of redox in other pathological conditions. The author discloses no conflicts of interest. Submit your manuscript here: http://www.dovepress.com/international-journal-of-general-medicine-journal The International Journal of General Medicine is an international, peer-reviewed open-access journal that focuses on general and internal medicine, pathogenesis, epidemiology, diagnosis, monitoring and treatment protocols. The journal is characterized by the rapid reporting of reviews, original research and clinical studies across all disease areas. A key focus is the elucidation of disease processes and management protocols resulting in improved outcomes for the patient.The manuscript management system is completely online and includes a very quick and fair peer-review system. Visit http://www.dovepress.com/ testimonials.php to read real quotes from published authors.

We should not be complacent about our population-based public health response to the first influenza pandemic of the 21(st )century

BACKGROUND: More than a year after an influenza pandemic was declared in June 2009, the World Health Organization declared the pandemic to be over. Evaluations of the pandemic response are beginning to appear in the public domain. DISCUSSION: We argue that, despite the enormous effort made to control the pandemic, it is now time to acknowledge that many of the population-based public health interventions may not have been well considered. Prior to the pandemic, there was limited scientific evidence to support border control measures. In particular no border screening measures would have detected prodromal or asymptomatic infections, and asymptomatic infections with pandemic influenza were common. School closures, when they were partial or of short duration, would not have interrupted spread of the virus in school-aged children, the group with the highest rate of infection worldwide. In most countries where they were available, neuraminidase inhibitors were not distributed quickly enough to have had an effect at the population level, although they will have benefited individuals, and prophylaxis within closed communities will have been effective. A pandemic specific vaccine will have protected the people who received it, although in most countries only a small minority was vaccinated, and often a small minority of those most at risk. The pandemic vaccine was generally not available early enough to have influenced the shape of the first pandemic wave and it is likely that any future pandemic vaccine manufactured using current technology will also be available too late, at least in one hemisphere. SUMMARY: Border screening, school closure, widespread anti-viral prophylaxis and a pandemic-specific vaccine were unlikely to have been effective during a pandemic which was less severe than anticipated in the pandemic plans of many countries. These were cornerstones of the population-based public health response. Similar responses would be even less likely to be effective in a more severe pandemic. We agree with the recommendation from the World Health Organisation that pandemic preparedness plans need review.

The World Health Organization (WHO) declared that spread of the newly recognised quadruple reassortant influenza A H1N1 virus satisfied the criteria for a pandemic on June 11, 2009 , [1] although technically conditions for declaring a pandemic had been met some weeks earlier. The virus, generally referred to as pandemic influenza H1N1 2009 (pH1N1), had first been recognised in Mexico and the United States in late April 2009. More than a year later, WHO has declared the pandemic to be over and early assessments of the global response have commenced [2] . When the pandemic was declared, Dr Margaret Chan, the Director of WHO, advised member states to implement their pandemic plans [1] and health agencies, other government agencies and businesses worked hard to do this. In most countries it may be correct to conclude, as did an evaluation of the UK response, that the "pandemic and the response it generated have provided confirmation of the value of planning and preparedness" [3] . It is also true that the apparent success of the response in 2009 must not lead to complacency. We now know that the relatively low virulence of pH1N1 meant we did not need to have implemented effective responses to get a good outcome. The response to the pandemic included clinical and public health measures. In developed countries, such as Australia, the clinical response was effective for those whose illnesses were serious [4] . Clinical care will very likely have reduced the number of deaths due to pandemic influenza [5] , although the use of extra-corporeal membrane oxygenation was seen as a last resort and was not supported by the conclusions of a systematic review [6] . In developed and developing countries, the public health response focused on both the individual and the population. Individual responses promoted attention to personal hygiene, with an emphasis on cough hygiene and hand washing, which may not have been optimal [7] , and the use of personal protective equipment for those considered to be at increased risk of infection [8] . Population-based public health responses to the pandemic focused on two major elements: non-pharmaceutical and pharmaceutical interventions. The former comprised border control and various elements of social distancing, while the latter focussed on anti-viral medication for treatment and/or prophylaxis, and the development of a strain-specific vaccine. Australia used the pharmaceutical and non-pharmaceutical interventions detailed in its pandemic plan [9] in an effort to delay entry of the virus into the country, contain the virus to limited areas once it had entered the country, sustain a response when widespread community transmission had been established and to protect the vulnerable [10] -the latter being a new response phase formulated once it was realised that the pandemic was not associated with the high case fatality ratios that had been anticipated [11] . We use Australia's experience to draw attention to issues related to the public health population-based pandemic response. The scope of this perspective does not allow us to consider other categories of response. Now is the time to acknowledge that a number of the strategies used in response to the 2009 pandemic could not control the spread of a novel influenza virus and their place in future pandemic response plans needs to be reconsidered in light of emerging new evidence. We examine four critical cornerstones of Australia's public health population-based response, namely border control, school closure (as an example of social distancing), the use of anti-viral medication and the development and use of a pandemic vaccine. We provide evidence from the pandemic experience in other countries to support our arguments. In a very different world, Australia successfully applied maritime quarantine to delay the entry of a pandemic H1N1 virus into the country in 1918 and 1919. This was in contrast to many of Australia's Pacific neighbours. For instance, the virus reached New Zealand in October 1918 but did not enter Australia until January of the following year. Estimated death rates in countries where the entry of the virus had been delayed were lower than rates where earlier entry was documented [12] . Prior to the 2009 pandemic, modelling studies had suggested a very limited role for border screening, providing an estimated delay of only 1-2 weeks without draconian measures that would be economically unacceptable in most countries [13] . A review of border control in Australia following the SARS epidemic pointed out the opportunity cost of screening. No case was identified despite 1.8 million passengers being screened, 794 referred for further evaluation and four identified as possible cases [14] . A similar lack of success was suggested in a review of the likely success of border control for influenza in other countries [15] . Indeed, it can be argued that prior to the pandemic there was only very limited scientific evidence for border control as an effective intervention. Despite this in 2009, as an island nation with history on its side, Australia, like many other (non-island) countries, embraced the concept. Australia implemented a combination of approaches in an attempt to detect infected arriving passengers at international airports. These comprised notification of health status of passengers by airline staff (the pilot or crew identified passengers with respiratory symptoms), thermal scanning (infra-red cameras were installed in airport terminals), health declaration cards (the passenger reported current symptoms) and nurses at border entry points reviewed and tested passengers detected by one of these screening tools. In the Australian state of Queensland, although the number of passengers screened was not reported, and was likely to have been many tens of thousands, only four cases of confirmed pandemic influenza were found from 780 passengers identified by one or more of these border screening measures [16] . No cases were detected by similar screening at the busy international airport in Perth, Western Australia (unpublished data). It has been suggested that Australia did well in its response to managing the pandemic, specifically in delaying establishment of community transmission [17] . During this time preparations were made to respond to a pandemic that was anticipated to result in many deaths. However, based on epidemiological and modelling evidence, we have demonstrated that community transmission was almost certainly established in the state of Victoria around the time the virus was first recognised in North and Central America in late April 2009 [18] . This followed one or more unrecognised silent importations, and spread of the virus in Australia came substantially from within its borders rather than from overseas. We now know that this is an entirely plausible scenario, given that a significant proportion of pH1N1 infections were afebrile [19] or entirely asymptomatic [20] and therefore impossible to detect at the border -or anywhere else. This should not be surprising, as the finding that a high proportion of influenza infections are asymptomatic or afebrile was not new. Published experimental data from volunteer studies had previously shown that 33% of proven seasonal influenza infections were asymptomatic, but this varied by influenza type and subtype [21] . In particular, as few as 37% of experimental infections with influenza A(H1N1) were associated with fever recorded as >37.8°C, while 30% were completely asymptomatic [21] . Moreover, viral shedding in the presymptomatic phase of influenza infection has recently been confirmed to occur in approximately 1-8% of naturally acquired infections, in a study in which 14% of all influenza infections were asymptomatic and 31% of infections with influenza A (H1N1 or H3N2) did not have fever at the onset of other symptoms [22] . Prodromal, asymptomatic and afebrile infections cannot be detected by temperature measurement, one of the main components of border control, whether by thermal scanning or by core temperature measurement of symptomatic travellers [23] . Moreover, the proportion of afebrile or asymptomatic people is likely to be higher in infected travellers, as more severely unwell people will be less likely to travel. Thermal imaging is therefore even less likely to have been effective at the borders than in other places where more severely ill patients are seen [24] . Indeed, China used intensive thermal screening for pH1N1 at airports and had a positive detection rate of only 14 cases per million passengers screened [25] . The use of border control was evidently not based on a current understanding of influenza epidemiology and was not supported by modelling studies. In particular one modelling study, published two years before the identification of the current pandemic virus, showed that past pandemic patterns could not be adequately modelled without inclusion of asymptomatic infection (as well as varying degrees of pre-existing immunity) [26] . Nonetheless, an early evaluation of the 2009 pandemic, with limitations acknowledged by its authors, suggested that border screening may have led to delays of 7 to 12 days in the establishment of local transmission [27] . We accept that border screening will have detected a limited number of influenza cases, but suggest that many more cases will have been missed than were detected. In Australia, at least, it is likely that border screening was implemented after the virus had entered the country [18] . On balance, we conclude that border screening was as ineffective as it should have been expected to be. It is generally accepted that children, especially children of school age, are responsible for amplification of influenza epidemics [28] . An intervention targeting schools could therefore theoretically be effective in interrupting an epidemic. This assumption is supported by modelling studies, but only when all schools are closed early and remain closed for an unrealistically long period, up to the duration of the pandemic [29] . Modelling also shows that delay in closing schools, or partial closure of schools, are less effective interventions, [29] although, if school closures are timely, they may delay the peak and decrease the peak incidence of the epidemic [30] . As expected, the pH1N1 infection rate was high among school aged children [20, 31] . Of the first 997 cases of confirmed pH1N1 infection in the state of Victoria in Australia, 67% were aged 5-17 years [32] . In Australia, school closure was intended to be associated with voluntary home quarantine. When a school -or class within a school -was closed, members of the class were asked to voluntarily quarantine themselves at home. This meant that parents of young children were frequently required to take time off work to care for children who would otherwise have been at school. Home quarantine has its own risks. We have recently shown that when an entire family was quarantined, the risk of secondary spread within households was increased by approximately 2.5-fold [33] . Moreover, compliance with other social distancing measures needed to have been effective for school closures themselves to have been effective. A survey in Western Australia of parents of school children whose schools were closed at some stage during the pandemic indicated that 74% of home-quarantined children participated in outside activities at least once during the nominal quarantine period, recording an average of 3.7 activities per child. Most commonly reported were attendances at sporting events, parks, beaches and stores, places where it is likely other children would be exposed [34] . Public documentation of school closure during the pH1N1 epidemic in Australia is minimal, but the policy in the early phase of the pH1N1 response was to close only those classes with confirmed cases, escalating to whole schools where multiple classes across different age groups were affected. In the state of Queensland only 2.8% of all schools were closed for short periods [16] . In Western Australia school closures were only for one week and sometimes involved only closure of specific classes [34] . Too limited in scope and time, these strategies could not have been effective in interrupting the spread of the pandemic. On the other hand, experience in Japan confirmed the conclusions from modelling [30] . Early widespread school closures in a defined area were successful in delaying pandemic spread in that area, but when the schools were re-opened, pandemic spread resumed [35] . In Hong Kong closure of kindergartens, pre-schools and primary schools appeared to decrease the attack rate in children aged less than 12 years for the weeks of closure [36] but the effect on the final attack rate in school children is yet to be evaluated. Indeed, it has been argued that the potential benefit of closing schools during a pandemic must be balanced against the enormous social disruption that ensues [37] . Only where schools were closed early and remained closed would there have been any significant interruption of the spread of the pandemic. Countries around the world adopted different approaches to the use of neuraminidase inhibitors (NAIs) in their pandemic plans. In addition to treatment provisions, Australia opted for a stockpile of approximately 10 million courses of NAIs with the intention of implementing widespread prophylaxis, which has been shown in trials to be 58-84% effective in preventing laboratory proven influenza infection if given early following exposure [38] . However, even in the early phases of the response, when numbers of suspected and confirmed pH1N1 cases were low, those with responsibility for contact tracing were rapidly overwhelmed. The logistical difficulties of timely delivery of NAIs to those eligible for treatment or prophylaxis were such that it was likely only a minority received their medication in time for it to be effective. Lateness of NAI availability has been confirmed in a Victorian study of treatment doses. Oseltamivir was prescribed for only 207 (21%) of the first 1,000 confirmed cases. Of 690 cases confirmed not to have received oselatamivir, 670 were not eligible because more than 48 hours had elapsed since symptom onset (Unpublished data, James Fielding, epidemiologist, Victorian Infectious Diseases Reference Laboratory). Other approaches to NAI distribution were used around the world, with varying effectiveness. For example, the UK National Health Service implemented an electronic checklist to allow patients rapid access to NAIs. Bypassing doctors and laboratory testing, this system aimed to speed up NAI availability. However, only 1932/16,560 (17%) of people who received NAIs using the electronic checklist subsequently tested positive for pH1N1 [39] . On the other hand, in four outbreaks in Singapore military camps, when NAIs were able to be delivered effectively in conjunction with isolation of confirmed cases and quarantine of contacts, a beneficial effect could be demonstrated. These measures, which included ring prophylaxis with oseltamivir, resulted in a reduction of the infection rate in the outbreaks from 6.4% to 0.6% [40] . This study demonstrates the potential benefit of NAIs if available early in outbreaks, and when combined with social distancing. However extension of this strategy to large heterogeneous populations remains unproven and it may be feasible only in closed communities, such as boarding schools, military barracks and residential care facilities. Another important consideration in setting out to provide mass treatment and prophylaxis with NAIs is the possibility of development of resistant strains. Surveillance studies during the first wave of the pandemic demonstrated a low frequency of resistance to oseltamivir and no reported resistance to zanamivir. Prior to 2007, it was rare to detect oseltamivir-resistant influenza strains in untreated patients, due to the compromised infectivity and transmissibility of many of the resistant mutants in the absence of drug pressure. But in 2007/2008 an oseltamivir-resistant seasonal A(H1N1) variant emerged that demonstrated viral fitness at least equivalent to the oseltamivir-susceptible strain. The resistant strain spread rapidly around the world and by 2009 had completely replaced the susceptible strain [41] . An oseltamivir-resistant pH1N1 virus might also retain viral fitness and subsequently spread throughout the community. Fortunately, to date, only a low frequency of oseltamivir-resistant pH1N1 strains have been identified [42] . An anti-viral stockpile without a well-developed logistic strategy and resourcing for effective early delivery for treatment of cases and prophylaxis of contacts is not an adequate plan for successful limitation of viral spread in a population, especially considering that the high proportion of cases with asymptomatic and mild infections will not be identified. Moreover, as we have seen with seasonal H1N1 viruses, resistance may develop to NAIs and a resistant virus may retain viral fitness allowing it to become widespread. This would render stockpiles useless. Revised pandemic plans should therefore consider limiting the use of NAIs to treatment of those with more severe influenza infection or medical conditions that make them more vulnerable to complications. Dependent on the availability of NAIs and access to appropriate medical care, treatment should be commenced early in the course of the illness. In Germany the median delay between symptom onset and antiviral treatment was significantly longer in fatal cases than non-fatal cases [43] . Prophylaxis should probably be reserved for closed communities, with any plan for wide-scale use of prophylactic NAIs dependent on a large workforce able to perform contact tracing and a detailed logistics plan for early delivery. After China, Australia was the second country in the world to roll out a population-based pandemic vaccine program, with monovalent pandemic vaccine available by 30 September 2009 [44] . The first wave of the pandemic in Australia had ended by this date. It was not expected that the vaccine would have been available in time to modify the first pandemic wave anywhere in the world. However, even in Australia, it was a case of 'too much too late'. An early estimate of 18% was made for population wide coverage for the vaccine [45] . Most pandemic vaccines in Australia were formulated as multi-dose vials. Given recommendations that the vial contents should be used or discarded within 24 hours of first use, wastage was expected with this formulation. It has been estimated that around 40% of pH1N1 vaccine doses delivered to Australian general practices may have been wasted [46] . There was also concern among some immunisation providers, and within the general community, that the multi-dose vials contained the preservative thiomersal, which had been phased out of paediatric vaccines, and that use of the vials potentially increased the risk of contamination, including with blood-borne viruses [47] . Such concerns, whether ill-founded or not, were likely to have impacted adversely on vaccine uptake, even in identified high risk groups [48] . While it may be reasonable to assume that vaccine uptake would have been higher if the disease had indeed been more severe, future pandemic plans need to include greater flexibility in vaccine purchasing and contracting arrangements, and refinement of vaccine delivery protocols and public messaging, in order to minimise wastage and optimise uptake [49] . The 21st century marks the first time pandemic-specific vaccines have been manufactured on a large scale. A preliminary report from Germany using the screening method estimated pandemic vaccine effectiveness for an adjuvanted pandemic vaccine of 97% in people aged 14-59 years [50] . A similar high level of protection has been reported for children in Canada [51] , although a more modest effectiveness of 72% has subsequently been reported from a pooled case control analysis from a number of European countries [52] . While vaccines were effective in protecting individuals, population coverage in Australia and other countries was unlikely to have been sufficient for the vaccine to have modulated the spread of the pandemic virus. However some European countries, such as Germany, experienced a very modest first pandemic wave [43] and, had they achieved high coverage with pandemic vaccine, may have been able to modify pandemic virus transmission in the next influenza season. Nonetheless, the experience with pH1N1 suggests that a pandemic vaccine will always be too late, at least for one hemisphere, using current vaccine manufacturing technology. Control of pandemic influenza is a critical issue and one on which the world has already spent billions of dollars, both in planning and during the recent response to pH1N1. There are obvious lessons to be learnt from the first pandemic of the 21st century, a pandemic which was much less severe than many plans had anticipated [53] . If we think our response to this pandemic was adequate, we may be falsely reassured. A more severe pandemic may find us wanting. A mild pandemic may find us over reacting. However, with appropriate collection and analysis of data it should be possible to identify the severity of future pandemics early and to make a measured response [54] . The World Health Organization, governments and other agencies around the world are currently involved in reviews of the management of the pandemic [55] . It is vital that these reviews, while not diminishing the commitment and hard work of those who implemented the response plans in 2009, carefully assess the evidence base for those plans. In addition, the widespread implications of the response to the pandemic -for policy makers, health professionals and the public -make it important for these reviews to be in the public domain. In Australia, where pandemic reviews are not yet in the public domain, there were examples where messages appeared to be mixed, and which confused both the public and healthcare professionals [56] . Partially closing some schools for short periods and not implementing other social distancing measures, such as cancelling public gatherings, is just one example. Although we have provided examples from Australia, we believe our arguments will have relevance for many other countries. 'One size fits all', where authorities have only one response strategy for viruses with different infection rates and case fatality ratios, is not an appropriate response to pandemic preparedness. Revised pandemic plans should include different responses for different pandemic severities [57] . All areas of pandemic planning need to be re-examined, but perhaps by alternative processes to those that led to current plans. Certainly, new evidence about the practical difficulties and/or ineffectiveness of control measures, such as border control and school closures, needs to be considered seriously. The inadequacy of many plans has recently been publicly acknowledged by the head of the WHO's global influenza programme. Speaking at a United Kingdom Health Protection Agency conference on the international response to the H1N1 pandemic, Dr Sylvie Briand is reported to have said that the containment strategy during the last pandemic was 'not feasible' and that guidelines might have to be overhauled [58] . We believe this is sound advice. National University for helpful comments on the manuscript. We acknowledge Dr Aeron Hurt from the WHO Collaborating Centre for Reference and Research on Influenza in Melbourne for his expert contribution to the section on NAIs. Geoffry Mercer acknowledges partial funding from an NHMRC strategic influenza grant. We thank Kristina Grant and Francine Cousinery for help with preparation of the manuscript.

The GLEaMviz computational tool, a publicly available software to explore realistic epidemic spreading scenarios at the global scale

BACKGROUND: Computational models play an increasingly important role in the assessment and control of public health crises, as demonstrated during the 2009 H1N1 influenza pandemic. Much research has been done in recent years in the development of sophisticated data-driven models for realistic computer-based simulations of infectious disease spreading. However, only a few computational tools are presently available for assessing scenarios, predicting epidemic evolutions, and managing health emergencies that can benefit a broad audience of users including policy makers and health institutions. RESULTS: We present "GLEaMviz", a publicly available software system that simulates the spread of emerging human-to-human infectious diseases across the world. The GLEaMviz tool comprises three components: the client application, the proxy middleware, and the simulation engine. The latter two components constitute the GLEaMviz server. The simulation engine leverages on the Global Epidemic and Mobility (GLEaM) framework, a stochastic computational scheme that integrates worldwide high-resolution demographic and mobility data to simulate disease spread on the global scale. The GLEaMviz design aims at maximizing flexibility in defining the disease compartmental model and configuring the simulation scenario; it allows the user to set a variety of parameters including: compartment-specific features, transition values, and environmental effects. The output is a dynamic map and a corresponding set of charts that quantitatively describe the geo-temporal evolution of the disease. The software is designed as a client-server system. The multi-platform client, which can be installed on the user's local machine, is used to set up simulations that will be executed on the server, thus avoiding specific requirements for large computational capabilities on the user side. CONCLUSIONS: The user-friendly graphical interface of the GLEaMviz tool, along with its high level of detail and the realism of its embedded modeling approach, opens up the platform to simulate realistic epidemic scenarios. These features make the GLEaMviz computational tool a convenient teaching/training tool as well as a first step toward the development of a computational tool aimed at facilitating the use and exploitation of computational models for the policy making and scenario analysis of infectious disease outbreaks.

The 2009 H1N1 influenza pandemic highlighted the importance of computational epidemic models for the real-time analysis of the health emergency related to the global spreading of new emerging infectious diseases [1] [2] [3] . Realistic computational models are highly complex and sophisticated, integrating substantial amounts of data that characterize the population and geographical context in order to attain superior accuracy, resolution, and predictive power [4] [5] [6] [7] [8] [9] [10] . The challenge consists in developing models that are able to capture the complexity of the real world at various levels by taking advantage of current information technology to provide an in silico framework for testing control scenarios that can anticipate the unfolding of an epidemic. At the same time, these computational approaches should be translated into tools accessible by a broader set of users who are the main actors in the decision-making process of health policy, especially during an emergency like an influenza pandemic. The tradeoff between realistic and accurate descriptions of large-scale dynamics, flexibility, computational feasibility, ease of use, and accessibility of these tools creates a major challenge from both the theoretical and the computational points of view [4, 5, 11, 12, 10, 13] . GLEaMviz is a client-server software system that can model the world-wide spread of epidemics for human transmissible diseases like influenzalike illnesses (ILI), offering extensive flexibility in the design of the compartmental model and scenario setup, including computationally-optimized numerical simulations based on high-resolution global demographic and mobility data. GLEaMviz makes use of a stochastic and discrete computational scheme to model epidemic spread called "GLEaM" -GLobal Epidemic and Mobility model, presented in previously published work [6, 3, 14] which is based on a geo-referenced metapopulation approach that considers 3,362 subpopulations in 220 countries of the world, as well as air travel flow connections and short-range commuting data. The software includes a client application with a graphical user interface (GUI) for setting up and executing simulations, and retrieving and visualizing the results; the client application is publicly downloadable. The server application can be requested by public institutions and research centers; conditions of use and possible restrictions will be evaluated specifically. The tool is currently not suitable for the simulation of vector-borne diseases, infection transmission depending on local contact patterns such as sexually transmitted diseases and diseases with a time scale that would make demographic effects relevant. The tool, however, allows the introduction of mitigation policies at the global level. Localized intervention in space or time can be implemented in the GLEaM model and their introduction in the GLEaMviz computational tool are planned for future releases. Only a few computational tools are currently available to the public for the analysis and modeling of epidemics. These range from very simple spreadsheet-based models aimed at providing quick estimates for the number of patients and hospitalizations during a pandemic (see e.g. FluSurge [15] ) to more complicated tools based on increasingly sophisticated simulation approaches [11, 16, 12, 10, 13, 5] . These tools differ in their underlying modeling approaches and in the implementation, flexibility, and accessibility of the software itself. InfluSim is a tool that provides a visual interface to simulate an epidemic with a deterministic compartmental model in a single population [11] . The model includes age structure and explicit sojourn times with different stages in each compartment, extending an SEIR compartmentalization to include hospitalizations and intervention measures. The software provides the infectious disease dynamics and the user can set parameter values and add or remove interventions. However, no spatial structure or other forms of heterogeneity and stochasticity are considered in the model. On the other hand agent-based models describe the stochastic propagation of a disease at the individual level, thus taking into account the explicit social and spatial structure of the population under consideration. CommunityFlu is a software tool that simulates the spread of influenza in a structured population of approximately 1,000 households with 2,500 persons [13] . User interaction with the software is limited to the spreadsheet portion of the program, where one can choose the type of intervention and other parameters describing the disease and the population. A larger population is considered in FluTe, a publicly available tool for the stochastic simulation of an epidemic in the United States at the level of individuals [10] . The model is based on a synthetic population, structured in a hierarchy of mixing social groups including households, household clusters, neighborhoods, and nation-wide communities. FluTe comes with a configuration file in text format that can be modified by an expert user to set various parameters such as the initiation of the epidemic, the reproductive number, and the interventions considered. No GUI is provided, and the output of the simulations is given in the form of text files that must be analyzed through additional software. EpiFast involves a parallel algorithm implemented using a master-slave approach which allows for scalability on distributed memory systems, from the generation of synthetic population aggregated in mixing groups to the explicit representation of the contact patterns between individuals as they evolve in time [5] . The Epi-Fast tool allows for the detailed representation and simulation of the disease on social contact networks among individuals that dynamically evolve in time and adapt to actions taken by individuals and public health interventions. The algorithm is coupled with a webbased GUI and the middleware system Didactic, which allows users to specify the simulation setup, execute the simulation, and visualize the results via plots. Epidemic models and interventions are pre-configured, and the system can scale up to simulate a population of a large metropolitan area on the order of tens of millions of inhabitants. Another class of models focuses on the global scale, by using a metapopulation approach in which the population is spatially structured into patches or subpopulations (e.g. cities) where individuals mix. These patches are connected by mobility patterns of individuals. In this vein two tools are currently available. The Global Epidemic Model (GEM) uses a metapopulation approach based on an airline network comprised of 155 major metropolitan areas in the world for the stochastic simulation of an influenza-like illness [16] . The tool consists of a Java applet in which the user can simulate a hypothetical H1N1 outbreak and test pre-configured intervention strategies. The compartmentalization is set to an SEIR model, and the parameterization can be modified in the full or stand-alone mode, but not currently in the Java applet. The Spatiotemporal Epidemiological Modeler (STEM) is a modeling system for the simulation of the spread of an infectious disease in a spatially structured population [16] . Contrary to other approaches, STEM is based on an extensible software platform, which promotes the contribution of data and algorithms by users. The resulting framework therefore merges datasets and approaches and its detail and realism depend on continuous developments and contributions. However, these are obtained from a variety of sources and are provided in different formats and standards, thus resulting in possible problems related to the integration and merging of datasets. Such issues are left to the user to resolve. The existing tools described above thus offer the opportunity to use highly sophisticated data-driven approaches at the expense of flexibility and ease of use by non-experts on the one hand, or very simplified models with user-friendly GUIs and no specific computational requirements on the other. Our approach aims at optimizing the balance of complex and sophisticated data-driven epidemic modeling at the global scale while maintaining an accessible computational speed and overall flexibility in the description of the simulation scenario, including the compartmental model, transition rates, intervention measures, and outbreak conditions by means of a user-friendly GUI. In the GLEaMviz tool the setup of the simulations is highly flexible in that the user can design arbitrary disease compartmental models, thus allowing an extensive range of human-to-human infectious diseases and intervention strategies to be considered. The user interface has been designed in order to easily define both features specific to each compartment, such as the mobility of classes of individuals, and general environmental effects, such as seasonality for diseases like influenza. In addition, the user can define the initial settings that characterize the initial geographical and temporal conditions, the immunity profile of the population, and other parameters including but not limited to: the definition of an outbreak condition in a given country; the number of stochastic runs to be performed; and the total duration of each simulation. The tool allows the production of global spreading scenarios with geographical high resolution by just interacting with the graphic user interface. While an expert input would be required to interpret and discuss the results obtained with the software, the present computational platform facilitates the generation and analysis of scenarios from intensive data-driven simulations. The tool can be deployed both in training activities as well as to facilitate the use of large-scale computational modeling of infectious diseases in the discussion between modelers and public health stakeholders. The paper is organized as follows. The "Implementation" section describes the software application architecture and its major components, including the computational model GLEaM. The "Results and discussion" section presents in detail the GLEaMviz client and its components that allow for software-user interaction, including an application of the Simulator to an Influenza-like-illness scenario. The top-level architecture of the GLEaMviz tool comprises three components: the GLEaMviz client application, the GLEaMviz proxy middleware, and the simulation engine. The latter two components constitute the GLEaMviz server, as shown in Figure 1 . Users interact with the GLEaMviz system by means of the client application, which provides graphical userinterfaces for designing and managing the simulations, as well as visualizing the results. The clients, however, do not themselves run the simulations. Instead they establish a connection with the GLEaMviz proxy middleware to request the execution of a simulation by the server. Multiple clients can use the same server concurrently. Upon receipt of requests to run a simulation, the middleware starts the simulation engine instances required to execute the requests and monitors their status. Once the simulations are completed, the GLEaMviz proxy middleware collects and manages the resulting simulation data to be served back to the clients. A schematic diagram of the workflow between client and server is shown in Figure 2 . This client-server model allows for full flexibility in its deployment; the client and server can be installed on the same machine, or on different machines connected by a local area network or the Internet. The two-part decomposition of the server in terms of middleware and engines additionally allows for advanced high-volume setups in which the middleware server distributes the engine instances over a number of machines, such as those in a cluster or cloud. This architecture thus ensures high speed in large-scale simulations and does not rely on the CPU-specific availability accessible by the user. The GLEaMviz simulation engine uses a stochastic metapopulation approach [17] [18] [19] 2, [20] [21] [22] 16 ] that considers data-driven schemes for the short-range and Design the compartmental model of the infectious disease in the Model Builder. 1 Configure the simulation of the world-wide epidemic spreading in the Simulation Wizard. 2 Submit the simulation for execution by the Engine on the server. Inspect the results of a simulation in the interactive Visualization. 5 Inspect all simulations and retrieve results in the Simulations History. long-range mobility of individuals at the inter-population level, coupled with coarse-grained techniques to describe the infection dynamics within each subpopulation [6, 14] . The basic mechanism for epidemic propagation occurs at multiple scales. Individuals interact within each subpopulation and may contract the disease if an outbreak is taking place in that subpopulation. By travelling while infected, individuals can carry the pathogen to a non-infected region of the world, thus starting a new outbreak and shaping the spatial spread of the disease. The basic structure of GLEaM consists of three distinct layers -the population layer, the mobility layer, and the epidemic layer (see Figure 3 ) [6, 14] . The population layer is based on the high-resolution population database of the Gridded Population of the World project by the Socio-Economic Data and Applications Center (SEDAC) [23] that estimates population with a granularity given by a lattice of cells covering the whole planet at a resolution of 15 × 15 minutes of arc. The mobility layer integrates short-range and longrange transportation data. Long-range air travel mobility is based on travel flow data obtained from the International Air Transport Association (IATA [24]) and the Official Airline Guide (OAG [25] ) databases, which contain the list of worldwide airport pairs connected by direct flights and the number of available seats on any given connection [26] . The combination of the population and mobility layers allows for the subdivision of the world into geo-referenced census areas obtained by a Voronoi tessellation procedure around transportation hubs. These census areas define the subpopulations of the metapopulation modeling structure, identifying 3,362 subpopulations centered on IATA airports in 220 different countries. The model simulates the mobility of individuals between these subpopulations using a stochastic procedure defined by the airline transportation data [6] . Short-range mobility considers commuting patterns between adjacent subpopulations based on data collected and analyzed from more than 30 countries in 5 continents across the world [6] . It is modeled with a time-scale separation approach that defines the effective force of infections in connected subpopulations [6, 27, 28] . On top of the population and mobility layers lies the epidemic layer, which defines the disease and population dynamics. The infection dynamics takes place within each subpopulation and assumes a compartmentalization [29] that the user can define according to the infectious disease under study and the intervention measures being considered. All transitions between compartments are modeled through binomial and multinomial processes to preserve the discrete and stochastic nature of the processes. The user can also specify the initial outbreak conditions that characterize the spreading scenario under study, enabling the seeding of the epidemic in any geographical census area in the world and defining the immunity profile of the population at initiation. Seasonality effects are still an open problem in the transmission of ILI diseases. In order to include the effect of seasonality on the observed pattern of ILI diseases, we use a standard empirical approach in which Population layer Short-range mobility layer Long-range mobility layer The short-range mobility layer covers commuting patterns between adjacent subpopulations based on data collected and analyzed from more than 30 countries on 5 continents across the world, modeled with a time-scale separation approach that defines the effective force of infections in connected subpopulations. The long-range mobility layer covers the air travel flow, measured in available seats between worldwide airport pairs connected by direct flights. seasonality is modeled by a forcing that reduces the basic reproductive number by a factor α min ranging from 0.1 to 1 (no seasonality) [20] . The forcing is described by a sinusoidal function of 12 months-period that reaches its peak during Winter time and its minimum during Summer time in each Hemisphere, with the two Hemispheres with opposite phases. Given the population and mobility data, infection dynamics parameters, and initial conditions, GLEaM performs the simulation of stochastic realizations of the worldwide unfolding of the epidemic. From these in silico epidemics a variety of information can be gathered, such as prevalence, morbidity, number of secondary cases, number of imported cases, hospitalized patients, amounts of drugs used, and other quantities for each subpopulation with a time resolution of 1 day. GLEaM has been under continuous development since 2005 and during these years it has been used: to assess the role of short-range and long-range mobility in epidemic spread [30, 31, 6] ; to retrospectively analyze the SARS outbreak of 2002-2003 in order to investigate the predictive power of the model [22] ; to explore global health strategies for controlling an emerging influenza pandemic with pharmaceutical interventions under logistical constraints [21] ; and more recently to estimate the seasonal transmission potential of the 2009 H1N1 influenza pandemic during the early phase of the outbreak to provide predictions for the activity peaks in the Northern Hemisphere [3, 32] . The GLEaMviz simulation engine consists of a core that executes the simulations and a wrapper that prepares the execution based on the configuration relayed from the client by the GLEaMviz proxy middleware. The engine can perform either single-run or multi-run simulations. The single-run involves only a single stochastic realization for a given configuration setup and a random seed. The multi-run simulation involves a number of stochastic realizations as set by the user and performed by the core (see the following Section), each with the same configuration but with a different random seed. The results of the multi-run simulation are then aggregated and statistically analyzed by the wrapper code. The simulation engine writes the results to files and uses lock files to signal its status to the middleware component. The core is written in C++, resulting in a fast and efficient engine that allows the execution of a single stochastic simulation of a 1-year epidemic with a standard SEIR model in a couple of minutes on a high-end desktop computer. The wrapper code is written in Python [33] . The server components can be installed on most UNIX-like operating systems such as Linux, BSD, Mac OS X, etc. The GLEaMviz proxy middleware is the server component that mediates between clients and simulation engines. It accepts TCP connections from clients and handles requests relayed over these connections, providing client authorization management. A basic access control mechanism is implemented that associates a specific client with the simulations it launches by issuing a private simulation identifier key upon submission. Users can only retrieve the results of the simulations they launched, or simulations for which they have obtained the simulation definition file -containing the private simulation identifier key-from the original submitter. Upon receipt of a request to execute a simulation, the middleware sets up the proper system environment and then launches an instance of the simulation engine with the appropriate configuration and parameters according to the instructions received from the client. For singlerun simulations, the daily results are incrementally served back to the client while the simulation is being executed. This allows for the immediate visualization of the spreading pattern, as described in "Visualization interface" Subsection. For multi-run simulations the results are statistically analyzed after all runs are finished, and the client has to explicitly request the retrieval of the results once they become available. The GLEaMviz proxy server component can be configured to keep the simulation data indefinitely or to schedule the cleanup of old simulations after a certain period of time. Multi-run metadata is stored in an internal object that is serialized on a system file, ensuring that authorization information is safely kept after a server shutdown or failure. The GLEaMviz proxy component additionally provides control features such as accepting administrative requests at runtime in order to manage stored simulations or to modify several configuration parameters like the number of simultaneous connections allowed, the number of simultaneous simulations per client, the session timeout, etc. The middleware server is written in Python [33] and uses the Twisted Matrix library suite [34] for its networking functionality. Client and server communicate using a special purpose protocol, which provides commands for session handling and simulation management. Commands and data are binary encoded using Adobe Action Message Format (AMF3) in order to minimize bandwidth needs. The GLEaMviz client is a desktop application by which users interact with the GLEaMviz tool. It provides GUIs for its four main functions: 1) the design of compartmental models that define the infection dynamics; 2) the configuration of the simulation parameters; 3) the visualization of the simulation results; and 4) the management of the user's collection of simulations. In the following Section we describe these components in detail. The client was developed using the Adobe AIR platform [35] and the Flex framework [36] and can thus be deployed on diverse operating systems, including several Windows versions, Mac OS X, and several common Linux distributions. The GLEaMviz client has a built-in updating mechanism to check for the latest updates and developments and prompts the user to automatically download them. It also offers a menu of configuration options of the interface that allows the user to customize preferences about data storage, visualization options, the server connection, and others. The software system presented above is operated through the GLEaMviz client, which provides the user interface: the part of the tool actually experienced on the user side. The GLEaMviz client integrates different modules that allow the management of the entire process flow from the definition of the model to the visualization of the results. In the following we will describe the various components and provide the reader with a user study example. The Model Builder provides a visual modeling tool for designing arbitrary compartmental models, ranging from simple SIR models to complex compartmentalization in which multiple interventions can be considered along with disease-associated complications and other effects. (An example can be found in previous work [37] .) A snapshot of the Model Builder window is shown in Figure 4 . The models are represented as flow diagrams with stylized box shapes that represent compartments and directed edges that represent transitions, which is consistent with standard representations of compartmental models in the literature. Through simple operations like 'click and drag' it is possible to create any structure with full flexibility in the design of the compartmentalization; the user is not restricted to a given set of pre-loaded compartments or transition dynamics. The interactive interface provided by the Model Builder enables the user to define the compartment label, the mobility constraints that apply (e.g. allowed/not allowed to travel by air or by ground), whether the compartment refers to clinical cases, as well as the color and position of their representation in the diagram (see Figure 5 ). This allows the user to model many kinds of human-to-human infectious diseases, in particular respiratory and influenza-like diseases. Transitions individuals is equal to  SI N , where N is the total size of the subpopulation. The GLEaM simulation engine considers discrete individuals. All its transition processes are both stochastic and discrete, and are modeled through binomial and multinomial processes. Transitions can be visually added by dragging a marker from the source to the target compartment. Spontaneous transitions are annotated with their rates, which can be modified interactively. Infection transitions are accompanied with a representation of the infection's source compartment and the applicable rate (i.e. b in the example above), which can also be modified in an interactive way. The rates can be expressed in terms of a constant value or in terms of a variable whose value needs to be specified in the variables table, as shown in Figure 4 . The value can also be expressed by simple algebraic expressions. The client automatically checks for and reports inconsistencies in the model in order to assist the user in the design process (see bottom right window in Figure 4 ). Models can be exported to XML files and stored locally, allowing the user to load a model later, modify it, and share it with other users. The diagram representation can be exported as a PDF or SVG file for use in documentation or publications. A few examples of compartmental models are available for download from the Simulator website. The Simulation Wizard provides a sequence of panels that leads the user through the definition of several configuration parameters that characterize the simulation. Figure 6 shows some of these panels. The consecutive steps of the configuration are as follows: •Choice of the type of the simulation (panel a) The user is prompted with three options: create a new single-run simulation or a new multi-run simulation from scratch, or a new one based on a saved simulation previously stored in a file. •Compartmental model selection and editing The user can design a new compartmental model, modify the current compartmental model (when deriving it from an existing simulation), or load a model compartmentalization from a file. •Definition of the simulation parameters (panel c) The user is asked to specify various settings and parameter values for the simulation, including, e.g., the number of runs to perform (only accessible in the case of a multi-run), the initial date of the simulation, the length of the simulation (in terms of days), whether or not seasonality effects should be considered, the airplane occupancy rate, the commuting time, the conditions for the definition of an outbreak, and others. •Initial assignment of the simulation (panel d) Here the user assigns the initial distribution of the population amongst compartments, defining the immunity profile of the global population on the starting date. •Definition of the outbreak start (panel e) This panel allows the user to define the initial conditions of the epidemic by selecting the city (or cities) seeded with the infection. •Selection of output results (panel f) Here the user selects the compartments that will constitute the output provided by the client at the end of the simulation. The corresponding data will be shown in the Visualization Window and made available for download. When all the above configuration settings are defined, the user can submit the simulation to the GLEaMviz server for execution. This will automatically add the simulation to the user's Simulations History. It is furthermore possible to save the definition of the simulation setup to a local file, which can be imported again later or shared with other users. The Simulations History is the main window of the client and provides an overview of the simulations that the user has designed and/or submitted, in addition to providing access to the Model Builder, the Simulation Wizard, and the Visualization Component. The overview panel shown in Figure 7 lists the simulation identifier, the submission date and time, the simulation type (i.e., single or multi-run), the execution status (i.e., initialized, start pending, started, aborted, complete, failed, or stop pending) and the results status (i.e., none, retrieve pending, retrieving, stop retrieve pending, complete, or stored locally). Additional File 1 provides a detailed explanation of all these values. A number of context-dependent command buttons are available once a simulation from the list is selected. Those buttons allow the user to control the simulation execution, retrieve the results from the server and visualize them, clone and edit the simulation to perform a new execution, save the simulation definition or the output data to the local machine (in order to analyze the obtained data with other tools, for example), and remove the simulation. In addition to exporting the compartmental model (see the "Model Builder" Subsection) the user can export a complete configuration of a simulation that includes the compartmental model and the entire simulation setup to a local file, which can be imported again later or shared with other users. Once the execution of a simulation is finished and the results have been retrieved from the server, the client can display the results in the form of an interactive visualization of the geo-temporal evolution of the epidemic. This visualization consists of a temporal and geographic mapping of the results accompanied by a set of graphs (see Figure 8 ). The geographic mapping involves a zoomable multi-scale map on which the cells of the population layer are colored according to the number of new cases of the quantity that is being displayed. Several visualization features can be customized by clicking on the gear icon and opening the settings widget. It is possible to zoom in and out and pan by means of the interface at the top left of the map. Dragging the map with the mouse (on a location where there are no basin marks) can also pan the visualization. All the widgets and the graphs displayed over the map can be re-positioned according to the user's preferences by clicking and dragging the unused space in the title bar. The color coding of the map represents the number of cases on a particular day. The time evolution of the epidemic can be shown as a movie, or in the form of daily states by moving forward or backward by one day at a time. For single-run simulations it is also possible to show the airline transportation of the 'seeding' individuals by drawing the traveling edge between the origin and destination cities. In the case where the output quantity is a subset of the infectious compartments, the edges show the actual seeding of the infection. Note that the evolution of the epidemic depends strongly on the model definition. For example, it is possible that some basins are infected by a latent individual that later develops the disease. In this case no seeding flight will be shown if only infectious compartments are selected as output. Beside the geographical map, the Visualization Window displays two charts. One chart shows the number of new cases per 1,000 over time (incidence), and the other shows the cumulative number of new cases per 1,000 over time (size). For multi-run simulations, median values and corresponding 95% confidence intervals are shown. The menu above each chart combo lets the user choose the context for which the corresponding charts show incidence and size data. This context is either: global, one of three hemispheres, one continent, one region, one country, or one city. The currently selected day is marked by a vertical line in these plots, and the day number, counted from the initial date selected for the simulation, is shown by side of the time slider. Here we present an example application of the GLEaMviz tool to study a realistic scenario for the mitigation of an emerging influenza pandemic. Disease-control programs foresee the use of antiviral drugs for treatment and shortterm prophylaxis until a vaccine becomes available [38] . The implementation of these interventions rely both on logistical constraints [21, 39] -related, e.g., to the availability of drugs -and on the characteristics of the infection, including the severity of the disease and the virus's potential to develop resistance to the drugs [40] . Here we focus on the mitigation effects of systematic antiviral (AV) treatment in delaying the activity peak and reducing attack rate [41] [42] [43] 7, 8, 39, 40, 3] , and assume that all countries have access to AV stockpiles. We consider a scenario based on the 2009 H1N1 influenza pandemic outbreak and feed the Simulator with the set of parameters and initial conditions that have been estimated for that outbreak through a Maximum Likelihood Estimate by using the GLEaM model [3] . The results provided by the present example are not meant to be compared with those contained in the full analysis carried out with GLEaM [3] due to the fact that in the Figure 8 The simulation results can be inspected in an interactive visualization of the geo-temporal evolution of the epidemic. The map shows the state of the epidemic on a particular day with infected population cells color-coded according to the number of new cases of the quantity that is being displayed. Pop-ups provide more details upon request for each city basin. The zoomable multi-scale map allows the user to get a global overview, or to focus on a part of the world. The media-player-like interface at the bottom is used to select the day of interest, or show the evolution of the epidemic like a movie. Two sets of charts on the right show the incidence curve and the cumulative size of the epidemics for selectable areas of interest. present example we do not consider additional mitigation strategies that were put in place during the early phase of the outbreak, such as the sanitary control measures implemented in Mexico [3, 44] , or the observed reduction in international travel to/from Mexico [45] . Indeed, the current version of GLEaMviz does not allow for interventions that are geographically and/or temporally dependent. However, these features are currently under development and will be available in the next software release. For this reason the simulation scenario that we study in this application of the Simulator does not aim to realistically reproduce the timing of the spreading pattern of the 2009 H1N1 pandemic. The results reported here ought to be considered as an assessment of the mitigating impact of AV treatment alone, based on the initial conditions estimated for the H1N1 outbreak, and assuming the implementation of the same AV protocol in all countries of the world. We adopt a SEIR-like compartmentalization to model influenza-like illnesses [29] in which we include the systematic successful treatment of 30% of the symptomatic infectious individuals (see Figure 9 ). The efficacy of the Figure 9 Compartmental structure in each subpopulation in the intervention scenario. A modified Susceptible-Latent-Infectious-Recovered model is considered, to take into account asymptomatic infections, traveling behavior while ill, and use of antiviral drugs as a pharmaceutical measure. In particular, infectious individuals are subdivided into: asymptomatic (Infectious_a), symptomatic individuals who travel while ill (Infectious_s_t), symptomatic individuals who restrict themselves from travel while ill (Infectious_s_nt), symptomatic individuals who undergo the antiviral treatment (Infectious_AVT). A susceptible individual interacting with an infectious person may contract the illness with rate beta and enter the latent compartment where he/she is infected but not yet infectious. The infection rate is rescaled by a factor ra in case of asymptomatic infection [41, 46] , and by a factor rAVT in case of a treated infection. At the end of the latency period, of average duration equal to eps -1 , each latent individual becomes infectious, showing symptoms with probability 1-p a , whereas becoming asymptomatic with probability p a [41, 46] . Change in travelling behavior after the onset of symptoms is modeled with probability p t set to 50% that individuals would stop travelling when ill [41] . Infectious individuals recover permanently after an average infectious period mu -1 equal to 2.5 days. We assume the antiviral treatment regimen to be administered to a 30% fraction (i.e. pAVT = 0.3) of the symptomatic infectious individuals within one day from the onset of symptoms, reducing the infectiousness and shortening the infectious period of 1 day. [41, 42] . AV treatment is accounted for in the model by a 62% reduction in the transmissibility of the disease by an infected person under AV treatment when AV drugs are administered in a timely fashion [41, 42] . We assume that the drugs are administered within 1 day of the onset of symptoms and that the AV treatment reduces the infectious period by 1 day [41, 42] . The scenario with AV treatment is compared to the baseline case in which no intervention is considered, i.e. the probability of treatment is set equal to 0 in all countries. The GLEaMviz simulation results are shown in Figure 10 where the incidence profiles in two different regions of the world, North America and Western Europe, are shown for both the baseline case and the intervention scenario with AV treatment. The results refer to the median (solid line) and 95% reference range (shaded area) obtained from 100 stochastic realizations of each scenario starting from the same initial conditions. The resulting incidence profiles of the baseline case peak at around mid-November and the end of November 2009 in the US and Western Europe, respectively. These results show an anticipated peak of activity for the Northern Hemisphere with respect to the expected peak time of seasonal influenza. In order to make a more accurate comparison with the surveillance data in these regions, we should rely on the predictions provided by models that can take into account the full spectrum of strategies that were put in place during the 2009 H1N1 outbreak, viz. the predictions obtained by GLEaM [3] . In the case of a rapid and efficient implementation of the AV treatment protocol at the worldwide level, a delay of about 6 weeks would be obtained in the regions under study, a result that could be essential in gaining time to deploy vaccination campaigns targeting high-risk groups and essential services. In addition, the GLEaMviz tool provides simulated results for the number of AV drugs used during the evolution of the outbreak. If we assume treatment delivery and successful administration of the drugs to 30% of the symptomatic cases per day, the number of AV drugs required at the activity peak in Western Europe would be 4.5 courses per 1,000 persons, and the size of the stockpile needed after the first year since the start of the pandemic would be about 18% of the population. Again, we assume a homogeneous treatment protocol for all countries in the world; results may vary from country to country depending on the specific evolution of the pandemic at the national level. Computer-based simulations provide an additional instrument for emerging infectious-disease preparedness and control, allowing the exploration of diverse scenarios and the evaluation of the impact and efficacy of various intervention strategies. Here we have presented a computational tool for the simulation of emerging ILI infectious diseases at the global scale based on a datadriven spatial epidemic and mobility model that offers an innovative solution in terms of flexibility, realism, and computational efficiency, and provides access to sophisticated computational models in teaching/training settings and in the use and exploitation of large-scale simulations in public health scenario analysis. Project name: GLEaMviz Simulator v2. 6 Project homepage: http://www.gleamviz.org/simulator/ Operating systems (client application): Windows (XP, Vista, 7), Mac OS X, Linux. Programming language: C++ (GLEaMsim core), Python (GLEaMproxy, GLEaMsim wrapper), Action-Script (GLEaMviz) Other requirements (client application): Adobe AIR runtime, at least 200 Mb of free disk space. License: SaaS Baseline scenario Scenario with AV Figure 10 Simulated incidence profiles for North America and Western Europe in the baseline case (left panels) and in the AV treatment scenario (right panels). The plots are extracted from the GLEaMviz tool visualization. In the upper plots of each pair the curves and shaded areas correspond to the median and 95% reference range of 100 stochastic runs, respectively. The lower curves show the cumulative size of the infection. The dashed vertical line marks the same date for each scenario, clearly showing the shift in the epidemic spreading due to the AV treatment. Any restrictions to use by non-academics: None. The server application can be requested by public institutions and research centers; conditions of use and possible restrictions will be evaluated specifically. Additional file 1: The GLEaMviz computational tool: Additional File. This file includes information for installing the GLEaMviz Client and details of the features of its various components.

Nursing heroism in the 21(st )Century'

BACKGROUND: The Vivian Bullwinkel Oration honours the life and work of an extraordinary nurse. Given her story and that of her World War II colleagues, the topic of nursing heroism in the 21(st )century could not be more germane. DISCUSSION: Is heroism a legitimate part of nursing, or are nurses simply 'just doing their job' even when facing extreme personal danger? In this paper I explore the place and relevance of heroism in contemporary nursing. I propose that nursing heroism deserves a broader appreciation and that within the term lie many hidden, 'unsung' or 'unrecorded' heroisms. I also challenge the critiques of heroism that would condemn it as part of a 'militarisation' of nursing. Finally, I argue that nursing needs to be more open in celebrating our heroes and the transformative power of nursing achievements. SUMMARY: The language of heroism may sound quaint by 21(st )Century standards but nursing heroism is alive and well in the best of our contemporary nursing ethos and practice.

Today's nursing heroism First, the more traditional concept of heroism as courage and providing service to others in the face of extreme personal danger is undoubtedly alive and well in nursing and in other human services. Firemen still enter burning buildings to save their occupants and nurses still join their health care colleagues in providing care to the hungry, the fearful, the injured and the dying in both natural disaster areas and man made conflict zones. Haitian nursing students and faculty from the Episcopal University of Haiti, were setting up first aid stations to help the victims of their city 30 minutes after that country's massive earthquake [4] . Military nurses and nurses from voluntary organisations such as Red Cross and Medicine Sans Frontiers are found in every war zone, every famine-blighted country, every dictatorial wasteland, every manifestation of 'hell on earth'. We fervently wish that the circumstances that draw them away from their own families and homes to these places did not exist, but they do, and thankfully, these nurses continue to respond. Consider also, nurses' responses to the fear and danger surrounding the emergence of infectious outbreaks such as HIV/AIDS in the 1980s and SARS in 2003. In the early 1980s when first reports were emerging of young gay men in the USA dying of seemingly systemic immune system failure, we could not have realised that this thing called 'GRID' was the start of the AIDS pandemic that has claimed the lives of more than 25 million people worldwide and has left approximately 33.4 million people living with HIV/AIDS [5] . During these times we saw the best and worst of nursing and health care. In an Oral History project: "The AIDS Epidemic in San Francisco: The Response of the Nursing Profession, 1981-1984 [6, 7] " we hear from nurses involved that some nursing and medical staff held the same fears and prejudices that were so widespread in the broader community. They would refuse to care for the AIDS patients and stigmatise them along with the other so-called "4H patients -Homosexuals, Haitians, Hemophiliacs and Heroin users" [8] . Helen Miramontes, who later became one of the world's leading nurse advocates, specialists and educators in HIV/AIDS was then a clinician. She said that: "There was a lot of fear among health care providers about contagion, but there was also significant prejudice and discriminatory behavior because the new disease was identified in a population that was stigmatized by the larger society. Identification of the disease in people of color, especially African Americans and injection drug users, only exacerbated the biases, prejudices, and discriminatory behavior. Many nurses demonstrated the same attitudes, beliefs, and behaviors seen in the larger society. I was a critical care nurse working in an intensive care unit (ICU) in a large teaching facility. In the early years of the epidemic, it was not unusual to have two to three patients with Pneumocystis carinii pneumonia on ventilators in the ICU at any one time. Because some nurses avoided taking care of these patients, several of us volunteered to care for them on a regular basis. There were frequent breaches in confidentiality, not only among nurses but also among other health care workers" [7] . Gary Carr, who was a Nurse Practitioner at the AIDS Clinic at San Francisco General Hospital, described the perverse ambivalence of a wider community that lauds and praises nurses for their 'heroic efforts' in the face of such public health crises. Gary says: "I have no memories of being afraid or being brave. I just wasn't afraid. I just said to myself, this is what I want to do. This is important. The community needs this, and it's what I want to do. I remember there were people who stopped speaking to me. My mother for years didn't tell anybody what I did. My relatives for years thought I still worked in the trauma unit" [6] . When, two decades later, SARS emerged as a potentially lethal viral infection, nurses and health care staff again faced considerable dangers as they strove to treat patients and protect their communities. Dr Dessmon Tai, who led the Singapore efforts against SARS, wrote that: "No other disease had such a phenomenal impact on healthcare workers." [9] But it seems that something in our professional ethos had changed over these two decades. There had been a rediscovery or reaffirmation of our professional ethic and mission as nurses, doctors and health professionals. In Hanoi, during the initial outbreak, rather than look for an 'opt-out' clause in their professional codes, doctors and nurses locked themselves into the hospital in isolation rather than risk spreading the disease [10] . As Emmanuel notes: "More than half of the first 60 reported cases of SARS involved health care workers who had come into contact with SARS patients. Indeed, apart from the very first case, all of the people who died in Vietnam were doctors and nurses. Nearly a quarter of all patients with SARS in Hong Kong were health care workers. In Canada, of the 141 probable cases of SARS diagnosed between 23 February and 14 May 2003, 92 (65%) involved health care workers. Despite deadly peril, physicians and nurses tirelessly cared for patients with SARS". [8] However, the social stigma that surrounded HIV/AIDS twenty years earlier and the associated ambivalence of the community towards health care professionals was not so easily repressed. In Toronto, Canada, Hall and colleagues reported that: "children of nurses were barred from school trips and families were shunned by their neighbours. Other incidents included husbands of nurses being sent home from work, children of nurses shunned at school, nurses refused rides by taxi drivers, and singleparent nurses unable to get babysitters". [11] Similarly, in Singapore, nursing staff were reportedly shunned in public spaces, forbidden to use the lifts in their apartments, found that buses and taxis would not stop for them and as one review reported, "At any packed food court, there would always be a seat for a Tan Tock Seng Hospital nurse. Queue lines would quickly shorten when a nurse joined that queue". [10] These personal travails were compounded by what many saw as the unavoidable violence done to some of the best traditions of the nurse-patient relationship by the nature of the SARS virus and its containment. Nurses working with SARS patients were often isolated from collegial support, asked to eat meals alone and prohibited from attending meetings. Rigorous isolation and anti-infection procedures saw nurses, effectively in spacesuits, caring for patients in enforced cubicle isolation. If this was a terrible way to be ill, it was an even more solitary and disconnected way to die. Yet despite these dangers and demands, nurses and our health care colleagues exemplified the best of who we are and what we do. They worked in a cauldron of contagion, initially unaware of what they were fighting, how the infection spread, how it killed or how it could be treated. As one French doctor commented, "We were not playing with fire. It was playing with us" [9] . Was this heroism and heroic actions on their part, or were they 'just doing their job'? If we accept that heroism is "providing service to others in the face of extreme personal danger", then I have no qualms in considering these nurses as exemplars of 21 st Century Nursing heroism. Heroism and 'militarism': What's in a word? Let me sidetrack slightly at this point to address a concern about the very legitimacy and appropriateness of the term heroism in nursing. For some critics, the very mention of 'wars against disease', 'defeating illness', 'the battle against SARS', or 'nursing heroism', is tainted. The concern is related to the militaristic or combative nature of such language and how this might shape not only our understandings of illness and disease but also our understanding of nursing itself and indeed the content and foci of our education, research and services. The concerns are not new, having been articulated most forcefully and influentially by the late American essayist, critic and author, Susan Sontag in both her books: 'Illness as a Metaphor' and 'AIDS and its Metaphors' [12] . Sontag was deeply critical of the metaphorical language that scaffolds our understandings of, in these cases, Cancer and AIDS. She rejects metaphors of battle, war, magic bullets, invasion, surrender, attack etc and with AIDS is even more scathing of its damning metaphorical encumbrances around divine retribution, plague, sexual contagion and societal decay. "The body is not a battlefield" says Sontag [12] , but when faced with illness or injury, I suspect that it may become one, if for no other reason that nurses, people and patients need something to fight against. People will often accept the most devastating of diagnoses with almost a gratitude that seems completely misplaced, until they explain to us that the previous uncertainty or not knowing had been far, far worse. Author and doctor, Peter Goldsworthy depicts this phenomenon beautifully in his celebrated novella, 'Jesus Wants Me for a Sunbeam' which describes how a family reacts when their young daughter develops leukaemia: "For Rick and Linda there was also, at the end of that terrible week of waiting and worry, an odd feeling of relief that it had happened to them, and theirs. Anything was better than uncertainty; the waiting had been intolerable, the fear of the unmentionable had almost come to be a desire for the unmentionable; its certainty, its mention, was at least a resolution. To finally hear the word (leukaemia) spoken aloud provided a focus for worry, a definite enemy that they could now face, and fight, together, as a family." [13] Before leaving the subject of language, let me touch on a recently articulated concern about the language and discourses of militarism and how these may influence Nursing. In a new paper this year on the Politics of Nursing Knowledge, Perron and her colleagues [14] criticise what they call the 'militarization of nursing', claiming that: "Many accounts provide angelic portrayals of nurses faced with the devastating effects of war. Such nurses are described as loyal, beautiful, peaceful, healing, comforting, reliable, devoted, and courageous in the face of hardship. These romantic descriptions stand in sharp contrast to the organized killing and destruction of warmaking." [14] I would argue that what Perron and her colleagues have almost studiously overlooked are that these allegedly 'romantic descriptions' also stand in sharp contrast to any respectable historical account of the experiences of military nurses [1] . In such histories, and in particular the growing collection of oral histories of nurses' wartime and conflict zone experiences [3, [15] [16] [17] [18] [19] , we will find not saccharine, sentimentalized, spin-doctoring but vivid recollections and narratives from nurses, who, in addition to thankfully being 'loyal, comforting, devoted, healing, reliable and courageous', are also intelligent, skillful, determined, demanding, creative and resourceful. Heather Höpfl, a Professor of Management, takes a quite different, and I believe more coherent and enabling view of women's heroism than Perron and her colleagues. She argues that: "The heroines of history are not impotent women. Quite the contrary, they are women who refuse to be put in their place. The stories of heroines offer a picture of women who were far from meek and far from conciliated into male reality definitions. They were real women not mythological constructions. They shared values rather than common backgrounds and they are, to use an unfashionable term 'indomitable'. (...) The stories of the heroines of 50 years ago and more are stories of recusancy. They are stories of opposition. They are in almost every case stories of gender politics. We are deceived when we are told they are stories of female oppression. (...) These women faced enormous obstacles but squared themselves off against them and responded with integrity and courage." [20] Of other 'quiet' and 'unrecorded' heroisms Let me now speak of other heroisms, for I believe heroism in nursing to be a broad church, a continuum of courage rather than a zero-sum game. The nurse exhibiting what I might call, after Dickens' character, Sydney Carton, a "quiet heroism", or what Patricia Benner calls "unrecorded heroism" [21] , continues to be part of the fabric of health care today. Nursing may not be among the first occupations that springs to mind when the word 'danger' is mentioned but a recent Forbes magazine feature in the USA did identify nurses and nurses' aides as one of the top two groups experiencing among the highest rates of injury at work, albeit usually from lifting and handling incidents [22] . In addition, Hall and colleagues in the US reported that: "Nursing assistants working in long-term care facilities have the highest incidence of workplace violence of any American worker". [23] Nurses in our Emergency Departments experience perhaps even more severe episodes of violence. In Scotland, we used to joke that any particularly rough place was: "Like casualty on a Friday night", but it is no joke now. In a recent Australian survey from WA, Rose Chapman and colleagues found that: "The majority (92%) of nurses said they had been verbally abused, 69% had been physically threatened and 52% had been physically assaulted". "Only 16% of the nurses completed an official incident report. Reasons for not reporting included the view that Work Place Violence is just part of the job, and the perception that management would not be responsive". [24] Would Emergency Department nurses see themselves as 'heroes'? Almost certainly not, as they seem to have internalized a workplace culture where abuse and violence is not an aberration but simply something that "comes with the job" [25] and where reporting such abuse is scarcely worth the trouble. Perhaps if we return to the definition of heroism as 'providing service in the face of extreme personal danger', then our Emergency Department nurses should allow themselves to feel, at least somewhat heroic. When we define heroism as providing service in the face of personal danger, that danger is not always the danger of illness, injury or death. Sometimes, that danger emanates from inside the very organisations that we serve and what is under threat is not nurses' bodily integrity but their professional and personal integrity. I refer of course to the phenomena of whistleblowing in nursing and health care. Nursing's official pronouncements, policy directives, mission and vision statements, strategic plans, nursing philosophies and the like invariably proclaim our commitment to the highest standards (indeed to excellence), to patient safety, to quality care, to collegiality, to mutual respect, to 'patient-centredness', to patient advocacy and more. But for many nurses, there is not merely a gap between these aspirations and clinical reality, there is a Katherine Gorge. Debra Jackson and her research team in Sydney have done some of the best Australian research in this area and it does not make for happy reading. As Debra notes: "Currently, whistleblowing represents a professional dilemma and a personal disaster." [26] This observation does not only apply to nursing. Medical whistleblowers fare equally badly at the hands of their own profession [27] . Indeed in all of the whistleblowing literature, it is difficult, if not impossible to find even one person whose principled actions have NOT resulted in huge personal and professional costs. And yet we know that nurses are not only vital because of our numbers but because we are patient care, safety and quality where the rubber hits the runway. Nursing is not just some inert 'silo', needing to be dismantled. If a hospital or health authority does not understand the difference between nursing as a Silo and nursing as a Sentinel, they are in more trouble than they can possibly imagine. We are the world's best adverse patient experience early warning detection system. We are like the canaries down the mine or the frogs in the ecosystem. If nurses are metaphorically not singing or not croaking, if they are falling off the perch or disappearing from the ponds, then 'Huston -we have a problem"! This problem is a phenomena that blighted not only Bundaberg Hospital [28] , but other hospitals and health organisations across the world. This virulent, vocational virus -I'll call it: MRSA ('Management Resistance to Staff Alerts'), has been implicated in almost every hospital 'scandal' and health system failure inquiry in recent years. This strain of 'MRSA' seems endemic in health care systems and is as dangerous to staff and patients as any hospital acquired infection. In a nutshell, health organisations and their leaders are not only failing to listen to their front line nursing staff, but in the worst cases, they are actively and forcefully trying to silence them. At Bundaberg Hospital, what stopped Dr Jayant Patel was not a new computer system, it was not an updated reporting mechanism, it was not visits from external regulators, it was not another reorganization, it was not a new management theory. It was the bulldog advocacy and persistence of ICU Nurse Toni Hoffmann in the face of managerial inactivity and intimidation [28] . Toni was recognised for her role with an Australian Local Hero Award in 2006 and an Order of Australia in 2007. Is Toni Hoffmann a 21 st Century nurse hero? Without a shadow of doubt. What of the other contemporary meaning of heroism and heroes? Who are our nursing heroes? Who are the nursing giants upon whose shoulders we have stood and who continue to inspire us today? Who are the nursing heroes that we tell our students about so that they can understand the best of Nursing's history? Who are they to emulate if they want to be the best nurse possible? Our hardwired and often confused sense of egalitarianism makes us uncomfortable in even talking in such a way. When I ask nurses to tell me something wonderful that they did recently that made a real difference to a patient, client or family, their embarrassment and discomfort is almost immediately palpable. 'I didn't really do anything', 'it was the team', 'I'm just a nurse', 'I don't like to 'big note' (brag about) myself', 'All I did was..., 'I don't know what you mean'.... Please, let us be more open and unabashed in celebrating our nursing heroes. If these were sportsmen or sportswomen, if they were musicians or actors, if they were business people: we would be lauding their finest performances, dissecting their latest work with relish and handing out awards and trophies by the cabinetful. I thought of my heroes for this paper, the nurses who have inspired me and who continue to do so. So let me celebrate, in no particular order of merit: Possibly the most lucid and coherent writer on nursing ever. I read her little 50 or 60 page book, 'Basic Principles of Nursing Care' as a young student nurse and could recite her definition of nursing, even at parties after a few drinks, as if it were a catechism. It is no exaggeration to say that I understood nursing in a completely different light after Virginia Henderson. As a new PhD student grappling with philosophy, phenomenology, qualitative research approaches and new understandings of practice, reading 'The Primacy of Caring', followed by 'Novice to Expert' was to have the scales fall from one's eyes and to see and understand the world anew. The quality of Patricia Benner's thinking and scholarship is matched only by her graciousness and generosity of spirit. Margaret was Dean and Head of School when I took up my first post-PhD Lecturing position at Glasgow Caledonian University and was simply the Dean from heaven. Her standards and expectations were exhilaratingly exacting and her work rate would have shamed a Chilean mine rescuer. Her enthusiasm for nursing, for education, for research and for innovation and creativity was boundless. Her guiding philosophy seemed to be: 'The answer's yes, now what's the question'. How could you fail to thrive and develop as a new Faculty member under such inspiring and utterly humane leadership? Linda Aiken is undoubtedly the doyenne of contemporary nursing research and one of the most powerful voices in nursing, in that when Linda Aiken's research speaks, the world of healthcare listens. And so it should. Her exemplary international research programme at the Center for Health Outcomes and Policy Research at the University of Pennsylvania demonstrates clearly that nursing is not simply one of many factors involved in improving health outcomes and patient safety, it is THE KEY FACTOR. Get nursing right and you improve safety, quality and patient care. No ifs, no buts. Dodie Bryce was an enrolled nurse at the intellectual disability hospital where I trained as a new nurse in the 1970s. These institutions could be grim places but Dodie Bryce's calm, humane, compassion and exemplary human caring skills made her truly, a light in the darkness. She was not just a great nurse but a presence. Older and wiser, I now understand the difference. Was a ward sister at the Sick Children's Hospital in Edinburgh when I trained in pediatrics. I was simply in awe of her. She managed, as the sole charge nurse, a 30 bed surgical ward, the attached neonatal unit of around 6-10 cots and an attached 2 bed cardiothoracic surgery unit. She knew every child, everything about their condition and its care and seemingly all of their families as well. She was unflappable in any crisis and utterly respected and listened to by every doctor and health professional. That she managed to take time and trouble to help students on the ward like myself made her even more remarkable. Is South Australia' first Nurse Practitioner and heads our Pediatric Palliative Care service. She was also my first clinical research collaborator when I took up my Joint Chair position at Women's & Children's Hospital. Sara is a human dynamo, possessed with vision, drive and absolute determination. Give a hospital half a dozen Saras in clinical leadership positions and they could rule the world. Debra was my PhD student and the PhD student of every supervisor's dreams. Debra has a fierce intelligence matched only by her unstinting work ethic. When you combine these with a heart and personality that draws the best out of everyone around her, you can see why she now heads what is easily one of Australia's best research centers. Who are the heroes on your list and why? The business of Nursing? We are told constantly that health care is a business and that nursing should follow more business-like principles. As a health service or hospital is indeed a multi-million dollar organisation that needs to be well managed, there are no arguments from me on that score. If we are in business however, then let us be absolutely clear about the nature of our business as nurses. We are in the transformation business and the 'making a difference' business. Nurses don't just make the tea and coffee they make decisions. We need to appreciate the importance of processes and structures, but more importantly, we need a laser focus and a near-reverence for tangible and valued outcomes that improve patients' experiences. We are in the transformation business. As a clinician, you are not in the injections business or the dressingchanging business or the putting up IVs business or the bathing business. Instead, as Kerfoot notes, we transform [29] "We transform a frightened 4-year-old girl in the emergency room into a little person who now feels she has some measure of control and can stop crying. We transform a 50-year-old father with out-of-control diabetes into a person who has the confidence to manage his condition. And we transform the frightening and painful experience of childbirth into a beautiful memory of ecstasy for a family that has created a new person. When life ends, we transform those final moments of life into sacred, beautiful transitions of passage for families to complete the circle of life." As a nurse educator, you are not in the business of 'lecturing', 'marking', 'supporting students', or 'writing curricula'. You are in the transformation business. [30] We transform students into safe, skilled and self-confident practitioners. We transform apathy and cynicism into enthusiasm and robust idealism. We transform clinical, interpersonal and ethical problems from potential career-ending setbacks, into opportunities for deep learning and personal and professional mastery. We transform patient and client experiences from everyday anecdote into the bedrock of clinical judgement and service quality. We foster and build confidence and self-belief where this been eroded, damaged or has never developed while also challenging an equally dangerous overconfidence, arrogance or narcissism. As a nurse researcher, you are not in the business of interviewing, administering surveys or managing data. We transform the glib stereotype of the 'ivory-tower' academic by our meaningful, productive and mutually advantageous collaborations with clinical colleagues and service areas. We challenge the prejudice that academics and their research has little relevance or use in the 'real world' of health policy and politics by our focus on knowledge translation, transfer and research impact and by the demonstrable profile and presence that our work has in numerous key areas of health policy and politics. We are in the transformation business and the 'making a difference' business. All over the world, nurses are making rhetorical notions of 'The Patient Experience', Quality & Safety and Improved Outcomes very, very real: Somewhere a nurse is helping a struggling and despairing new mum to learn all of the messages and nuances that her new baby is signalling, Somewhere a nurse is bearing witness to another mother's dying, and comforting her during her last moments on this earth, Somewhere a nurse is inserting a child's IV and helping them and their family begin their journey into the world of chemotherapy, Somewhere a nurse is listening to an Alzheimer's patient tell a story and trying to help them piece together who they really are, Somewhere a nurse is helping a new student learn from a patient encounter and is passing on the wisdom of our art, Somewhere a nurse is turning a hunch or a problem into a question that will eventually be researched and provide new knowledge and understanding, Somewhere a nurse is managing a service with the passion and enthusiasm that enables her staff to thrive and to appreciate why they wanted to become nurses in the first place, And somewhere a nurse is working in a war zone, helping service personnel and villagers alike. These 'quiet' or 'unrecorded' heroisms surely deserve our acknowledgement and appreciation. At the end of her classic novel 'Middlemarch' [31] , George Eliot writes an epitaph for her heroine Dorothea: "But we insignificant people with our daily words and acts are preparing the lives of many Dorotheas...Her finely-touched spirit had still its fine issues, though they were not widely visible. Her full nature, like that river of which Cyrus broke the strength, spent itself in channels which had no great name on the earth. But the effect of her being on those around her was incalculably diffusive: for the growing good of the world is partly dependent on unhistoric acts; and that things are not so ill with you and me as they might have been, is half owing to the number who lived faithfully a hidden life, and rest in unvisited tombs." So too, the health, wellbeing, safety and experiences of patients, clients and families are dependent upon the often invisible and overlooked caring practices of nurses. Today in the 21 st Century, they are worthy of sharing the term 'heroism' and I like to think that Sister Vivian Bullwinkel would agree.

Implications of copy number variation in people with chromosomal abnormalities: potential for greater variation in copy number state may contribute to variability of phenotype

Copy number variation is common in the human genome with many regions, overlapping thousands of genes, now known to be deleted or amplified. Aneuploidies and other forms of chromosomal imbalance have a wide range of adverse phenotypes and are a common cause of birth defects resulting in significant morbidity and mortality. “Normal” copy number variants (CNVs) embedded within the regions of chromosome imbalance may affect the clinical outcomes by altering the local copy number of important genes or regulatory regions: this could alleviate or exacerbate certain phenotypes. In this way CNVs may contribute to the clinical variability seen in many disorders caused by chromosomal abnormalities, such as the congenital heart defects (CHD) seen in ~40% of Down’s syndrome (DS) patients. Investigation of CNVs may therefore help to pinpoint critical genes or regulatory elements, elucidating the molecular mechanisms underlying these conditions, also shedding light on the aetiology of such phenotypes in people without major chromosome imbalances, and ultimately leading to their improved detection and treatment.

One of the fastest-growing research areas in genetics in the past few years has been the investigation of structural variation in the human genome, with copy number variants (CNVs) found to be much more common than previously imagined. As well as contributing to interindividual phenotypic variation in the general population, these genomic variants may hold the key to understanding the differences in severity seen in people with chromosomal abnormalities. Aneuploidy, for example, can be regarded as a state of large-scale changed copy number resulting from loss or gain of an entire chromosome, arising due to non-disjunction during cell division in either meiosis I or II. Aneuploidies and other chromosomal abnormalities are a common cause of birth defects and are associated with significant morbidity and mortality. The spectrum of phenotypes seen in each affected individual is clearly dependant on the particular chromosomal region concerned, although there is still a great deal of variability in the presentation of phenotypes. Individuals with Down's syndrome (DS), for example, like those with other aneuploidies or with chromosomal deletions or duplications, vary greatly in their clinical features, capabilities, disabilities and prognosis. Although many of these people, with the necessary social and clinical support, can and do lead active and fulfilling lives, some DS sub-phenotypes are associated with increased morbidity and mortality in infancy. Congenital heart defects (CHD), for example, occur in *40% of DS; 50% of these require surgical correction within the first year of life and survival to one year is only *76% compared with *91% for non-CHD DS patients (Yang et al. 2002 ). An appreciation of the molecular mechanisms behind this phenotypic variation and the identification of susceptibility loci may provide diagnostic and prognostic markers to enable us to predict the occurrence of clinically serious phenotypes, such as CHD, in DS. Additionally, analysis of these loci might help elucidate the mechanisms of CHD in non-DS children. To assess the implications of genomic copy number variants in the context of chromosomal abnormalities, it is helpful to consider the evidence that they affect phenotype in euploid individuals. Until recently, it was assumed that there are two copies of every gene in euploid individuals, with one on the maternally-inherited chromosome and the other inherited paternally. Any two human genomes were thought to be 99.9% similar, with the main source of genetic variation due to single nucleotide polymorphisms (SNPs). Technological advances often result in paradigm shifts, however, and an unexpected amount of copy number variation was revealed through the advent of array comparative genomic hybridisation (array CGH) and, more recently, of second-generation sequencing technologies. There are currently over 14,000 CNV regions listed on the TCAG Database of Genomic Variants (DGV, updated March 2010) , predicted to overlap a significant proportion of the genome. Approximately 11,500 genes, including over 2600 listed in the Online Mendelian Inheritance in Man database, are reportedly overlapped by CNVs (DGV March 2010). To understand the role of CNVs in human disease, it is important to consider how these genomic variants may affect phenotype. Deletion or duplication of dosage-sensitive genes, or their regulatory regions, could have adverse phenotypic effects, and copy number changes correlate with expression of affected genes (Henrichsen et al. 2009; Stranger et al. 2007 ). The majority of common CNVs are much smaller than the subset investigated by Stranger et al. (2007) , leading them to propose that a substantial proportion of heritable variation in gene expression may be explained by copy number variation. In addition to direct copy number effects, CNVs may also exert positional effects, for example by shifting them closer to or further away from heterochromatin. Furthermore, adverse phenotypic effects may result from interactions between CNVs and other variants, for example a deletion on one chromosome may unmask a harmful recessive allele on the other. Conversely, a genomic duplication that incorporates a harmful mutation may have a double gain-of-function effect: a duplication of the PRSS1 and PRSS2 genes along with a missense mutation was recently shown to cause hereditary pancreatitis (Masson et al. 2008) . It is likely, therefore, that CNVs do significantly affect human phenotypes, and this is reflected by an increasing number of associations reported between CNVs and disease (Tables 1, 2). Several autoimmune disorders have been associated with CNVs, including systemic lupus erythematosus (Fanciulli et al. 2007 ), Crohn's disease (Fellermann et al. 2006; McCarroll et al. 2008 ) and psoriasis (Hollox et al. 2008) , and a range of other diseases have now been associated with CNVs (reviewed in Zhang et al. 2009 ). It has been postulated that CNVs could account for a large proportion of the missing heritability of common diseases that has emerged after the recent plethora of genome-wide Diskin et al. (2009) association studies (GWAS) (reviewed in Manolio et al. 2009 ), but this is the subject of heated controversy. In a recent large-scale genotyping survey of common CNVs, Conrad et al. (2010) found that the CNVs that they were able to genotype easily were only in linkage disequilibrium with 34 out of 1,554 trait-associated SNPs from GWAS, thus they suggest that common CNVs are unlikely to account for much of the missing heritability. There is, however, a particular challenge in analysis of amplified sequences (which may be remote from the original sequence copy (Conrad et al. 2010) ), multi-allelic variants (particularly VNTRs) and recurrent events, so further analysis of those CNV types is required before the full contribution of common CNVs to human disease can be assessed. As an alternative to the common disease-common variants hypothesis (Chakravarti 1999; Lander 1996) , the rare variants hypothesis postulates that a collection of many, individually less frequent copy number changes may collectively significantly contribute to disease susceptibility. Although many CNVs are present in an appreciable proportion of the population, the majority are likely to be less frequent: for example, in our array CGH investigation of 50 apparently healthy French male samples, 809 out of 1469 (*55%) multi-probe CNV regions were identified in only one individual (de Smith et al. 2007 ). Rare copy number changes have been implicated in neurodevelopmental disorders, including autism (Marshall et al. 2008; Sebat et al. 2007 ) and schizophrenia (Wilson et al. 2006; Xu et al. 2008 ). Furthermore, a number of rare genomic structural variants have recently been associated with obesity (Bochukova et al. 2010; Walters et al. 2010) , and many more low frequency CNVs with strong effects may contribute to common disease phenotypes: large sample cohorts and different populations will need to be investigated to uncover these rare variants. Thus, it is now clear that genomic structural variants may significantly impact phenotype in euploid individuals: these effects may be even more pronounced in people with chromosomal abnormalities. For example, a phenotype associated with a particular monosomy or sub-chromosomal deletion may be ameliorated by the presence of an amplification CNV on the unaffected chromosome, as gene expression within the CNV region could remain at a normal, euploid level. Conversely, a deletion CNV could cause a more severe phenotype. An example could be Turner syndrome, which is the only whole chromosome monosomy (45, X) that is viable in humans, with approximately 1 in 2,000 female births having one copy of the X chromosome (Nielsen and Wohlert 1990) . In addition to the main features of short stature and ovarian failure, present in almost all cases, there are many other phenotypes that may present, including a short webbed neck, kidney malformations, hearing problems and various learning difficulties, as well as increased risk of type 1 diabetes (Gravholt et al. 1998) . It is possible that the extensive copy number variation on the X chromosome-37.8% according to the DGV (March 2010)-may contribute to phenotypic variability in this disorder. For example, a critical region for the neurocognitive deficits in Turner syndrome was mapped to Xp22.3, containing 31 genes (Ross et al. 2000) , and several CNV loci have been reported in this region. Similarly, phenotypic variation in genomic disorders could also be a reflection of underlying copy number variation within the relevant genomic region. The most common microdeletion in humans is DiGeorge syndrome (or velo-cardio-facial syndrome) caused by deletion of a 1.5-3 Mb region at chromosome 22q11.2 (Scambler et al. Triplication (4 copies 1992) and occurring in 1 in 5,000 births (Botto et al. 2003) . This disorder is also associated with variable phenotypes, such as cleft palate, congenital heart disease (Shprintzen et al. 1978) , renal anomalies (Czarnecki et al. 1998 ) and increased risk of schizophrenia (Murphy et al. 1999) . Several CNV loci have been identified within the 22q11.2 deletion region in apparently healthy individuals: of interest, two of the candidate genes for the schizophrenia phenotype, PRODH (Li et al. 2004 ) and GNB1L (Williams et al. 2008) , are overlapped by a number of CNVs, and these could potentially modify this phenotype in DiGeorge patients. In polyploidy or sub-chromosomal duplications where abnormalities result from extra copies of the chromosome regions, deletion CNVs may ''normalise'' copy number. Alternatively, there is potential for even greater amplification of copy number where amplification CNVs are present. A simple duplication CNV on the non-disjoining chromosome may lead to the presence of up to six copies of a gene in a trisomic individual (Fig. 1 ) and the subsequent phenotypic effects associated with increased gene expression levels. Conversely, a gene deletion on the non-disjoining chromosome may alleviate the expected effect of trisomy by reducing the local copy number state to one or two copies. Thus, the compounded effect of aneuploidy and copy number variation has the potential to generate a much wider range of phenotypes than would be expected on the basis of chromosome number alone. It has been suggested, therefore, that CNVs are likely to act as 'modifiers of the phenotypic variability of trisomies' (Beckmann et al. 2007 ) and the same is probably also true for smaller chromosomal imbalances. Using Down's Syndrome as an example, an extra copy of chromosome 21 is not sufficient to cause the full range of phenotypes associated with this disorder, as individuals can present with a range of sub-phenotypes. Some features are almost always present, such as the characteristic facial appearance and mental retardation, although these can vary widely in the severity of their presentation (Kallen et al. e illustrates how a deletion on parent B and MII NDJ in parent A, however, that leads to a total of two copies of the gene in the trisomic child which could potentially ameliorate the pathological effects of trisomy 1996). Other features, however, are only present in a fraction of DS cases: for example, congenital heart defects (CHD) are present in *40% of cases (Park et al. 1977 ) and gastrointestinal (GI) defects in *8% of DS (Epstein et al. 1991) . Leukaemia is also common in DS, with an *20% increased risk of acute lymphoblastic leukaemia (ALL) (Hasle et al. 2000) , and transient myeloproliferative disorder occurring in *10% of DS newborns, of which 10-20% develop acute megakaryoblastic leukaemia (AMKL) before the age of 4 (reviewed in Hasle 2001) . A detailed knowledge of the genes on the affected chromosome is required to understand the phenotypic effects of aneuploidy. In addition, it is important to determine the overall effects of gene dosage imbalance, which may also be complex. Individuals with partial trisomy 21 resulting from unbalanced chromosomal translocations will only exhibit those features associated with the extra genomic material present. Molecular analysis of these patients enables 'phenotypic mapping' by defining the critical genomic regions that harbour genes associated with various phenotypes (Epstein et al. 1991) . As with all regions of the genome, many common CNVs and rarer genomic structural variants are found on chromosome 21. Two recent reports describe structural variations in which the presence of increased copies of genomic regions on chromosome 21 contributes to human phenotype. In one study, duplication of a region at 21q21 that included the amyloid precursor protein (APP) gene was found to cause familial Alzheimer's disease in five separate families (Rovelet-Lecrux et al. 2006) . The duplication varied in size from 0.58 to 6.37 Mb and contained from 5 to 12 annotated genes including APP. Despite the largest of these duplications being over 6 Mb, none of the families exhibited any clinical evidence of DS (Cabrejo et al. 2006) . A second duplication of a 4.3 Mb region at 21q22.13-q22.2, containing just over 30 genes, did help refine the map of the DS critical region (DCR) as it caused a DS phenotype in three family members (Ronan et al. 2007 ). These individuals had the facial characteristics of DS and mild cognitive disability indicating that this region includes part of the DCR but not all of it. It is apparent, therefore, that different sub-phenotypes of trisomy 21 are caused by increased copy number at different regions across chromosome 21. It is also evident that some DS individuals who are trisomic for a particular critical region do not always exhibit the associated phenotype, or display a much milder form. A complex interplay of molecular factors is likely to be responsible for the trisomic phenotypes, possibly involving non-chromosome 21 genes. Furthermore, interaction of trisomy with the presence of certain embedded deleterious alleles or haplotypes may contribute to the variable presentation of the different phenotypes (including degree of mental retardation, congenital heart defects, Hirschprung and other gut diseases, and leukaemia). The increasing amount of copy number variation that has been discovered in the human genome, specifically on chromosome 21, adds another dimension to the molecular consequences of trisomy: it is possible that the phenotypic variability seen in DS may be due to CNVs on this chromosome. On the DGV, there are 507 reported CNV calls covering 35.0% of chromosome 21, which is similar to the CNV coverage of other chromosomes (33.9%) (DGV March 2010). Many of these are common in the population. For example, our investigation of only 50 healthy French male samples revealed 20 multi-probe CNVs, and an additional 57 single probe CNV signals, on chromosome 21 (de Smith et al. 2007 ): almost half were detected in multiple samples (47%), and 34% had a frequency of [5% (Fig. 2 ). In addition, these CNVs overlapped 38 known genes. There were several CNVs within TIAM1 (T-lymphoma invasion and metastasis-inducing protein 1), for example, which is expressed in almost all analysed tumour cell lines, including B-and T-lymphomas, melanomas and carcinomas (Habets et al. 1995) . A CNV was also found within DSCAM¸which is a good candidate gene for the CHD phenotype seen in DS, as it lies within the predicted minimum critical region and is expressed in the heart during cardiac development (Korbel et al. 2009 ). Several other CNVs have since been discovered in this gene, as listed on the DGV, including a common deletion that incorporates 4 exons (Matsuzaki et al. 2009 ). Other genes within the CHD critical region also overlap CNVs: a common intronic deletion (10%) was identified in C2CD2 (Conrad et al. 2010) ; 6 CNVs have been reported within PRDM15, including one deletion overlapping 5 exons that was found in 3/90 Yoruban individuals (Matsuzaki et al. 2009 ); and one deletion was identified in BACE2 overlapping the last exon of this gene (Mills et al. 2006) . CNVs also overlap RUNX1, with one deletion removing the promoter region and first exon of this gene (Gusev et al. 2009 ). RUNX1 codes for a transcription factor involved in haematopoiesis (North et al. 2002) and is associated with AMKL, which is estimated to be 500-fold more frequent in children with DS than in the general population (reviewed in Zipursky et al. 1992) : CNVs in this gene may, therefore, protect against or increase susceptibility to leukaemia in DS. These CNV loci and other regions that could potentially impact variable phenotypes of trisomy 21 are listed in Table 3 . Analysis of copy number variation on chromosome 21 may, therefore, lead to a better understanding of the etiology of DS sub-phenotypes and could possibly shed light on the origin of these conditions in the wider non-trisomic population. Knowledge of which genes are involved in the generation of CHD in DS, for example, and how they interact with other gene products, could have many implications for managing CHD in the non-trisomic population: for example, adult onset CHD may be prevented by therapeutic agents that target critical pathways controlling cardiogenesis, and new approaches may be developed for stem-cell guided cardiac repair (reviewed in Passier et al. 2008) . Since the discovery of trisomy 21, it has been hypothesised that genes present in three copies are over-expressed Shaikh et al. (2009) by 1.5-fold relative to the euploid state, but this has subsequently been shown to not always be the case. Some genes are over expressed, some are expressed at euploid levels while other genes appear to be down-regulated (Li et al. 2006; Prandini et al. 2007 ). This may, of course, reflect adaptive regulatory control, but may also reflect the effects of local copy number variation on chromosome 21, which will result in deviations from the predicted three copies of each gene, as shown in Fig. 1 . Copy number variation may, therefore, make a significant contribution to phenotypic variation in aneuploidy and other chromosomal abnormalities. Targeted analysis of such variation might, therefore, represent an effective strategy for refining critical regions and pinpointing genes central to key pathways (i.e. of cardiogenesis or oncogenesis). This would also help to enable the prediction of which individuals may develop serious phenotypes, such as CHD in DS, which may add an extra dimension to antenatal screening programmes in the future by contributing to the process of 'informed choice' by the potential parents of a DS child, as well as well as aiding the long term management of these patients. Additional benefits may also accrue from increased understanding of pathogenic mechanisms relevant to patients with similar phenotypes outside the context of chromosomal imbalance. In summary, recent technological advances, such as high-resolution array CGH and second-generation sequencing, have vastly and rapidly increased our knowledge of copy number variation in the human genome. As these techniques continue to improve, and greater numbers of samples and populations are investigated, so our understanding of how copy number correlates with phenotype will grow, yielding information not just relevant to individuals with chromosomal imbalances, but to similar conditions in the wider population.

Hepatitis Associated Aplastic Anemia: A review

Hepatitis-associated aplastic anemia (HAAA) is an uncommon but distinct variant of aplastic anemia in which pancytopenia appears two to three months after an acute attack of hepatitis. HAAA occurs most frequently in young male children and is lethal if leave untreated. The etiology of this syndrome is proposed to be attributed to various hepatitis and non hepatitis viruses. Several hepatitis viruses such as HAV, HBV, HCV, HDV, HEV and HGV have been associated with this set of symptoms. Viruses other than the hepatitis viruses such as parvovirus B19, Cytomegalovirus, Epstein bar virus, Transfusion Transmitted virus (TTV) and non-A-E hepatitis virus (unknown viruses) has also been documented to develop the syndrome. Considerable evidences including the clinical features, severe imbalance of the T cell immune system and effective response to immunosuppressive therapy strongly present HAAA as an immune mediated mechanism. However, no association of HAAA has been found with blood transfusions, drugs and toxins. Besides hepatitis and non hepatitis viruses and immunopathogenesis phenomenon as causative agents of the disorder, telomerase mutation, a genetic factor has also been predisposed for the development of aplastic anemia. Diagnosis includes clinical manifestations, blood profiling, viral serological markers testing, immune functioning and bone marrow hypocellularity examination. Patients presenting the features of HAAA have been mostly treated with bone marrow or hematopoietic cell transplantation from HLA matched donor, and if not available then by immunosuppressive therapy. New therapeutic approaches involve the administration of steroids especially the glucocorticoids to augment the immunosuppressive therapy response. Pancytopenia following an episode of acute hepatitis response better to hematopoietic cell transplantation than immunosuppressive therapy.

Aplastic anemia, acquire or congenital anemia associated with hypoplastic "fatty or empty" bone marrow and global dyshematopoiesis, has been first described by Paul Ehrlich in year 1888 [1] . The pathophysiology is believed to be idiopathic [2] or immune-mediated phenomenon with active destruction of haematopoietic stem cells [3] . The abnormal immune response may be elicited by environmental exposures, such as to chemicals, drugs, viral infections and endogenous antigens generated by genetically altered bone marrow cells [4] . A small fraction of the genes involved in pancytopenia has been represented by the congenital BM failure syndromes (relatively rare) which lately develop in clinical syndromes as Fanconi's anemia, Dyskeratosis Congenita, and Shwachman-Diamond syndrome [5, 6] . Hepatitis-associated aplastic anemia (HAAA) is a well recognized and distinct variant of clinical syndrome, acquired aplastic anemia, in which an acute attack of hepatitis leads to the marrow failure and pancytopenia [7] [8] [9] . HAAA has been first reported in two cases by Lorenz and Quaiser in 1955 [8] and the number of the cases increase up to value of 200 by the year 1975 [10] [11] [12] . However, this syndrome has been reported in 2-5% cases of west and 4-10% in area of more prevalent to hepatitis and Human Immunodeficiency Viruses (HIV) in the Far East [12, 13] , it belongs to the area of low socioeconomic status [14, 15] . HAAA is not considered relative to age, sex and severity of hepatitis [14] , predominantly it has been found in children [13] , adolescent boys and in young aged men [10, 16] . The onset of syndrome, pancytopenia, usually takes two to three months [62 days: ranging from 14 to 225) after attack of acute hepatitis [12, 14] . Hepatitis associated with aplastic anemia may be acute and chronic [7] , mild and transient [16] , self-limiting and fulminant and the development of AA is always fatal if not treated on time [7, 10] . Majority of the cases have been found as fulminant where the mortality rate reaches up to 85% [17] . Aetiology of the syndrome has been attributed to various agents and factors [7] which may include pathogenic viruses, autoimmune responses, liver transplantation procedure [18] bone marrow transplantation, radiation [19] and drugs administered to control the viral replication [14] . Several hepatitis viruses such as Hepatitis A [20] , B [21, 22] , C [23] , and E, G [24] have been anticipated to be associated with this set of symptoms [7] . No association has been found with blood transfusion, toxins and drugs [17] . In one study Safadi and co-workers (2001) found out that sera of eight of the patients had the existence of Hepatitis Bc IgG and/or anti-HBs antibodies which was suggested due to past exposure and immunizing effect and they were unable to establish any direct relation with acute hepatitis B [14] . As non-A, non-B hepatitis agents are usually responsible for the hepatitis associated aplastic anemia; the prevalence of anti-hepatitis C virus antibodies is similar in HAAA and aplasia of other origins [25] . HCV seropositivity has been observed in the patients developing cytopenia following non-A non-B hepatitis (NANBH). HCV viremia has been frequently observed without detecting anti-HCV antibodies in patients' blood reflecting the transfusion associated HCV infection [25, 26] . However, it has also been reported that HCV is not generally implicated as a causative agent of hepatitis preceding aplastic anemia [27] . Hepatitis G virus (HGV) has been reported as a possible etiological agent of acute hepatitis, chronic liver dysfuntioning, and fulminant hepatitis and hepatitis associated aplastic anemia. A relation of Hepatitis G virus with hepatitis subsequently developing in the aplastic anemia has been seen in a 24 years old person by measuring the hematological, biochemical, serological and virological parameters and detecting HGV RNA in his serum by PCR and electro-immunoassay [24] . The development of aplastic anemia generally found to occur in hepatitis not caused by the hepatitis A and hepatitis B viruses commonly known as the non-A and non-B hepatitis associated aplastic anemia which were reported in above than 80% of the cases of hepatitis preceding severe cytopenia [28, 29] . Viruses other than the hepatitis viruses have also been implicated as a causative agent of AA [1, 5] which include parvovirus B19 [19, 30, 31] , Cytomegalovirus, Epstein bar virus [19, 31, 32] , Echovirus 3 [33] , GB virus-C [34] , Transfusion Transmitted virus (TTV) [35] , SEN virus and non-A-E hepatitis virus (unknown viruses) [7] . Parvovirus B19, an under recognized hepatotrophic virus, is documented as an offending agent of HAAA. Its infection cause hepatic manifestation ranging from abnormal liver functioning to Fulminant Hepatic failure and aplastic anemia [30] . The primary site of infection of this virus is erythroid progenitor cell in which it halts the erythropoises and leads to the anemia in immunocompromised hosts. The DNA of this virus has been detected in liver of fulminant hepatic failure manifested with bone marrow aplapsia and in the serum of fulminant hepatitis children of unknown origin [36, 37] . Association of Torque Teno virus, single stranded circular DNA has liver as a susceptible host as well as various other tissues including bone marrow. It has been firstly reported in 12 years old Japanese boy suffered from cytopenia following acute hepatitis by detecting Torque Teno virus DNA of genotype 1a and IgM antibodies against this virus in peripheral blood and bone marrow mononuclear cells which precludes that acute bone marrow failure majorly concerns with infection of TTV virus to haemopoietic progenitor cells. However, other studies have also been done on assessing the TTV as an etiological agent of HAAA [33, 38] . In a study a 6-year-old boy was experienced aplastic anemia two months after the onset of acute hepatitis associated with echovirus-3 [33] . As idiopathic aplastic anemia is associated with the increased level of secretion of INF-γ and TNF-α from T cells which inhibit the hematopoietic cell proliferation [39] , similar increased level of serum INF-γ and TNF-α was found in the patient four weeks after the onset of aplastic anemia [33] . Similarly varying degrees of cytopenia has been related to the HIV infection and severe aplastic anemic conditions develop after subsequent attack of HSV-6 [17, 40] . Epstein Bar virus infection, involved in hepatitis, manifests the pathogenesis of marrow aplasia [40] . The pathogenic mechanism involve in the EBV infection is direct cytotoxity or mediates the immune response of host [7, 17, 14, 41] . Aplastic anemia can be acquired or congenital [1, 42] . As the aplastic anemia following the hepatitis has been elucidated as a severe bone marrow failure with an episode of acute hepatitis, following lymphocyte variations occur during the course of the syndrome: activation of circulating cytotoxic T cells increase, tend to accumulate in the liver, broad skewing patter of T cell reportrie in peripheral blood of the patient forms, a large number of T cell infiltration from liver parenchyma occurs [13, 26, 43, 44] defective monocyte to macrophage differentiation [45] and decreased circulating level of interleukin-1 occur [46] . Various Immunological abnormalities have been accountable for the development of aplastic anemia following hepatitis. The immunological abnormalities with HAAA show that CD8+ kupffer cells detecting by liver biopsies appear as a mediator of this syndrome. In a study it has been reported that patient showed a decreased ratio of CD4/CD8 cells and a high percentage of CD8 cells which can be cytotoxic and myleopoietic during the in vitro study of aplastic anemia [10, 20] . The residing of CD8 cells in bone marrow during HAAA produces a high level of interferon gamma (INF-γ) and cells derived from bone marrow locating in liver may activate these cytotoxic T cells causing their intrahepatic accumulation strongly affected by the Tumor necrosis factor alpha (TNF-α) and interferon gamma (INF-γ) causing the onlooker damage to liver cell of genetically modified mouse model. However, it has also been shown in several studies that increased level of soluble IL-2 receptor forms the major reason of non specific inflammation of HAAA [47, 48] . The pathogenesis of HAAA in children has been suggested to relate with the interruption in balance of lymphocyte sub-populations and T lymphocyte activation [49] . Being an acquired disease, severe HAAA has also been presented as a Familial Bone Marrow Failure Syndrome (FBMFS) in a study while finding the family donor of HSCs transplant for treatment of idiopathic fulminant liver failure patient who has developed myelodysplastic syndrome after onset of severe aplastic anemia. The donor sibling also found to be developed the acute lymphoblastic leukemia after diagnosing the hypocelluarity of bone marrow. The event of finding these two familial cases shows the bone marrow failure syndrome to be inherited [50] . HAAA has not been clarified with any genetic tendency [18] . Mutation in genes of the telomere repair complex, TERC (the gene for the RNA component of telomerase) and TERT (the gene for the telomerase reverse transcriptase catalytic enzyme), reduce the marrow regenerative capacity, making genes mutation carriers susceptible to the development of aplastic anemia once it has been started [42] . Most of the clinical features relating to aplastic anemia following the hepatitis include: Pallor and multiple skin bleeding [11] , lymphocytopenia, hypogammaglobulin [51] low number of CD8/T cell ratio [12] and increased number of cytotoxic cells [48] Neutropenia, fever [17] . Bacterial and fungal infection may emerge as secondary in presenting the disease [2] . Later complications may develop especially involving myelodysplasia [4] . The victims of severe aplastic anemia following the hepatitis experience a severe immune deficiency that might be either due to the hepatitis or aplastic anemia that is yet to be discovered [51] . On a course of HAAA, hepatitis can be detected on some of the following parameters: subsequent increase in serum Alanine Trasnaminase (ALT), Aspartate Transaminase (AST), by at least three times above the normal values which are 6 to 41U/l, 9-34U/l, 5-58U/L for ALT and AST respectively [12, 18, 30, 35, 36, 43, 52, 53] , increase in serum Alkaline phosphatase (ALP), gamma glutaryl transferase (GGT) and billirubin (39-117U/l, 5-58 U/l, and 2-7 micromol/L, respectively). Peripheral blood count can be determined by Flow cytometry analysis with directly conjugated monoclonal antibodies for CD2, CD3, CD4, CD8, CD19 and HLA-DR, whereas haematopoietic failure with bone marrow hypocellularity can be elucidated in terms of absolute neutrophil count (less than 500 per mm 3 ), Platelet count (less than 20,000 per mm3), Reticulocyte count (less than 60,000 per mm 3 ) [53] and Protrombin Index (%): normal value 70-100% [24] . To establish the onset of pancytopenia following hepatitis, hypocelluarity of bone marrow below 50% might be obtained by bone marrow aspiration [14] and trephine biopsy [11] . Various virological and serological markers are available for the detection of hepatitis A, B, C, D, E, G, TTV and parvovirus. Among these tests, anti-HAV Ig total antibodies, HBsAg, HB core antigen, HBsIgG antibodies, various HCV recombinant antigens and hepatitis E virus IgM and IgG are being in use. However, to determine causative nature of all hepatitis viruses, RNA genome of RNA containing viruses such as HCV, HDV, HEV and HGV can be qualitatively detected by RT-PCR reaction and DNA of parvovirus B19 and TTV can be detected by Nested PCR [14, 53] . IgG antibody for Cytomegalovirus, EBV and parvovirus has been found a useful tool for the diagnostic purposes [18] . However, serological and virological parameters for hepatitis A, B, and C were found negative in majority of the HAAA cases reported in several studies [17, 23, 27] . The standard therapy which is employed for the treatment of HAAA is allogenic bone marrow (BM) transplantation treatment from HLA matched siblings [13, 54] . HAAA is mostly occurring in children and it would be easier to find the HLA matched donor. As HAAA shows poor prognosis, most often it has been treated by hematopoietic stem cell transplantation [26, 55] . Immunosuppressive therapy has proved effective after BM transplantation. Various immunosuppressive drugs named Antithymocyte Globulin (ATG) and Cyclosporine have been administered without eliciting any acute side effects. Steroids such as Glucocorticoids have also been employed in combination with immunosuppressive medications for the treatment of the HAA patients [26] . A durable remission from HAA has been achieved by the administrating high dose of Cyclophosphamide (CY), a highly immunosuppressive, which elicits its effect by readily destroying the lymphocytes and committed myeloid cells. The haemopoietic stem cells are not susceptible to the toxic effects of the CY due to releasing the aldehyde dehydrogenase enzyme which inactivates the drug. However, restoration of haematopoesis process may achieved by high dose of CY which mediates autoimmune attack on haematopoiectic stem cells HSCs [13] . Patients irresponsive to the IST are prone to be cured by unrelated donor bone marrow transplantation [18] . Immunosuppressive therapy might have proved as a safe and alternative treatment for HAAA after of bone marrow or haemotopoietic stem cell transplantation [55] . Several studies showed that Parvovius induced aplastic anemia improves by the administration of retroviral therapy [21, 56] . Antiviral therapy for treating hepatitis B associated HAAA has been unknown yet; however it has been tested by administrating the nucleoside analogs, lamiviudine, against the aplastic anemic secondary to hepatitis B virus infection and remission occurs from the severe aplastic anemia accompanied with the hepatitis B viral infection. Interferon, an effective therapy for the hepatitis B and C viral infections, cannot be employed as a potential approach for the HAAA because of its mylosuppressive effects [21] . Acyclovir has been used for the treatment of aplastic anemia caused by the Epstein Bar virus [32] . The blood count comes to the normal range after five years of treatment [55] , however, chances of recovery from hepatitis associated acquired aplastic anemia is rare [57] . Growth factors deficiencies have been found to be responsible for the majority of the aplastic anemic cases [1] . As these growth factors released by stromal cells are essential for the survival, proliferation and differentiation of hematopoietic stem cells [58, 59] , a transient increase in granulocytes has been found effective in most aplastic anaemic trials by administrating the erythropoietin, growth factors, granulocyte colony-stimulating factor, granulocyte macrophage colony-stimulating factor, interleukin-3 [1] and androgens [5] . The limiting factors in success of immunosuppressive therapy are found to be the extent to which organ has been destructed, tissue regeneration capacity and most importantly pharmacology effect of drugs that is not sufficient for uncontrolled potent immune response [1] . The survival of the patients treated with hematopoietic cell transplantation and response rate to immunosuppressive therapy found to be 85% and 70% respectively [7] . Children response better than the adults to bone marrow transplantation and survival rate with Bone marrow transplantation from HLA matched donors is found to be similar as that for the non hepatitis associated aplastic anemia [14] . It has also been reported that parameters of liver dysfunctioning tend to improve when pancytopenia starts presenting itself [18, 28] . Although majority of the patients survives after aplastic anaemia tends to have complete recovery, the mortality rate is yet very high [52] . The mean survival rate after developing the severe bone marrow aplasia has been 2 months and fatality rate ranges from 78-88% [28, 60, 61] . HAAA is a well documented and diverse variant of clinical syndrome of aplastic anemia, in which an acute attack of hepatitis leads to the marrow failure and pancytopenia that may be acute or chronic. This disorder has been reported in 2-5% cases in West, 4-10% in Far East and high in area of low socioeconomic status. HAAA is not related to age, sex and severity of hepatitis, predominantly it has been found in children, adolesecent boys and in young aged men. A number of hepatitis viruses manifest the disease symptoms. Amongst the hepatitis viruses, HBV, HCV and HGV seropositivity has been mostly commonly observed in reported cases of HAAA. The causative agent of HAAA can be detected using hematological, biochemical, immunological and virological markers. The clinical features relating to HAAA are pallor and multiple skin bleeding, lymphocytopenia, hypogammaglobulin, low number of CD8/T cell ratio and increased number of cytotoxic cells, neutropenia and fever etc. For HAAA, immunosuppressive therapy is more effective after BM transplantation.

Factors Affecting Intention to Receive and Self-Reported Receipt of 2009 Pandemic (H1N1) Vaccine in Hong Kong: A Longitudinal Study

BACKGROUND: Vaccination was a core component for mitigating the 2009 influenza pandemic (pH1N1). However, a vaccination program's efficacy largely depends on population compliance. We examined general population decision-making for pH1N1 vaccination using a modified Theory of Planned Behaviour (TBP). METHODOLOGY: We conducted a longitudinal study, collecting data before and after the introduction of pH1N1 vaccine in Hong Kong. Structural equation modeling (SEM) tested if a modified TPB had explanatory utility for vaccine uptake among adults. PRINCIPAL FINDINGS: Among 896 subjects who completed both the baseline and the follow-up surveys, 7% (67/896) reported being “likely/very likely/certain” to be vaccinated (intent) but two months later only 0.8% (7/896) reported having received pH1N1 vaccination. Perception of low risk from pH1N1 (60%) and concerns regarding adverse effects of the vaccine (37%) were primary justifications for avoiding pH1N1 vaccination. Greater perceived vaccine benefits (β = 0.15), less concerns regarding vaccine side-effects (β = −0.20), greater adherence to social norms of vaccination (β = 0.39), anticipated higher regret if not vaccinated (β = 0.47), perceived higher self-efficacy for vaccination (β = 0.12) and history of seasonal influenza vaccination (β = 0.12) were associated with higher intention to receive the pH1N1 vaccine, which in turn predicted self-reported vaccination uptake (β = 0.30). Social norm (β = 0.70), anticipated regret (β = 0.19) and vaccination intention (β = 0.31) were positively associated with, and accounted for 70% of variance in vaccination planning, which, in turn subsequently predicted self-reported vaccination uptake (β = 0.36) accounting for 36% of variance in reported vaccination behaviour. CONCLUSIONS/SIGNIFICANCE: Perceived low risk from pH1N1 and perceived high risk from pH1N1 vaccine inhibited pH1N1 vaccine uptake. Both the TPB and the additional components contributed to intended vaccination uptake but social norms and anticipated regret predominantly associated with vaccination intention and planning. Vaccination planning is a more significant proximal determinant of uptake of pH1N1 vaccine than is intention. Intention alone is an unreliable predictor of future vaccine uptake.

Influenza contributes significantly to worldwide morbidity and mortality [1] . Periodically, influenza viruses mutate into antigenically-different strains leading to global pandemics [2] . The 2009 influenza pandemic (pH1N1) was caused by a triple reassortment of human, swine and avian influenza viruses [3] . Vaccination is the most effective intervention for preventing influenza [4] and a core part of national pandemic plans for pandemic mitigation. Lead times of at least 6 months in producing a vaccine against a novel strain means that while vaccines may be unavailable in time to prevent the first wave of a pandemic [5, 6] , effective public uptake of a vaccine may mitigate subsequent waves [7] . Significant health promotion activities regarding influenza prevention have been prominent in Hong Kong since well before the onset of pH1N1, arising largely from the Severe Acute Respiratory Infection (SARS) epidemic and A/H5N1 Bird Flu outbreaks. Seasonal influenza vaccination is widely promoted each year. Hong Kong's pH1N1 epidemic started on 11 June 2009, peaking in September, and by early November had petered out ( Figure 1 ). By the end of December 2009, the Hong Kong government had recorded 37,174 human pH1N1 cases [8] in a population of ,7 million. To minimize any potential second wave, significant televised and other publicity was given to the launch of a pH1N1 vaccination programme on 21 December 2009, initially for five priority groups: healthcare workers, persons with chronic illness and pregnant women, children aged 6 months to 6 years, adults aged 65 years or above, and pig farmers and slaughtering industry personnel [9] . On 26 January 2010 pH1N1 vaccination was extended to the general public. The vaccination was free for priority group members [10] , but cost HK$100-150 (US$13-20, 1-1.5% of Hong Kong's median monthly income of HK$10,000/ US$1,286/J991) per dose for the general population. A study in July 2009 of 301 respondents projected that vaccine uptake would be influenced by end-user cost, with 45%, of Hong Kong's general population being ''highly likely'' to take up pH1N1 vaccine if free, and 24% if costing HK$100-200 (US$ [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [11, 12] . From November 2009 onwards, television, radio, newspaper and official websites strongly encouraged priority groups to have pH1N1 vaccination [13] . However, the Hong Kong government did not make recommendations for the general population, who were asked to judge for themselves whether to be vaccinated or not. Shortly after the vaccine launch for priority groups, local media prominently attributed several adverse events to pH1N1 vaccination, including, a case of Guillain-Barre Syndrome (GBS) diagnosed a week after pH1N1 vaccination, reported on 6th January 2010, and an intrauterine death (IUD) 3 weeks following the mother's vaccination, reported on 20th January 2010 ( Figure 2 ). In both cases local health agencies presented convincing evidence challenging the link between vaccination and the two adverse events but were largely ignored. Retrospectively, a drop in pH1N1 vaccination uptake among priority groups was observed [14] . It seems probable that the adverse media reports had impeded vaccination uptake among general population. We collected baseline data between 12-25 January, 2010, immediately before pH1N1 vaccine was made available to the general population and then two months later (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) (30) March 2010) we recorded their reported vaccination status ( Figure 2 ) with the intention of modelling how general population decision-making regarding pH1N1 vaccination might predict subsequent vaccine uptake. Empirical studies have found that history of seasonal influenza vaccination [12, [15] [16] [17] [18] , perceived risk of pandemic influenza [17, [19] [20] [21] [22] [23] [24] [25] [26] , worry [17, 22, 26, 27] , and attitudes towards vaccine, such as vaccine efficacy and side-effects [12, 15, 20, [24] [25] [26] were significantly associated with intention to receive a vaccine against the influenza pandemic. This is consistent with the findings related to determinants of vaccination against seasonal influenza [28] [29] [30] [31] [32] . However, there are some common and significant limitations to these empirical studies. First, all except one [24] relied on vaccination intention to predict the actual vaccination uptake. In one study, since only a few respondents reported having received the pH1N1 vaccine, the authors combined those intending to get vaccinated with those who had already received the vaccine into one ''intending'' group and examined factors associated with this 'vaccination intention' [20] . This is problematic because factors associated with vaccination intention and actual vaccination receipt probably differ. Moreover, the reliability of intention as a predictor of actual behavior remains controversial. Harris et al. found that only about half of ''intending'' recipients of seasonal influenza vaccination actually take it and almost all those who do not intend to take it remained unvaccinated [33] . Moreover, most studies conducted before the pandemic occurred or before the vaccine was available [11, 12, [16] [17] [18] [19] 21, 23] of Dutch respondents reported intending to take pH1N1 vaccination prior to or at the onset of the (potential) pandemic phase, respectively [23] . Similarly, in Hong Kong 45% of 301 respondents in July 2009 reported being ''highly likely'' to receive pH1N1 vaccine if offered for free [11, 12] . However, by the time vaccination became available intention appeared much lower with only 10-15% of study respondents in France and in Turkey intending to take the pH1N1 vaccine [20, 24] . Second, all the studies are cross-sectional rather than longitudinal; none assessed subsequent actual vaccination status. Thus, although associations have been identified, there is no way to infer causality. Third, most of the studies are atheoretical. Although some of the studies developed their study questions based on theoretical framework such as HBM [17, 23, 24] , none have conducted model analysis and evaluated the model fit. Therefore, due to these three reasons, there remains a significant concern about how valid such results are and a significant knowledge gap about how the observed pattern of influences could be explained. A major limitation of previous empirical studies [12, [15] [16] [17] [18] [19] [20] [21] [22] [23] 25, 26] is failure to accommodate the intention-behaviour gap. Although several behavioral theories such as Protection Motivation Theory (PMT) [34, 35] , Theory of Reasoned Action (TRA) [36, 37] and Theory of Planned Behaviour (TPB) [38, 39] propose that intention is the proximal determinant of behaviour, intention does not necessarily translate into actual behaviour. Empirical studies of the intention-behavior relationship showed that intention had a medium effect (a correlation of ,0.4-0.5) on behavior [40] [41] [42] , but a recent review including 47 experimental studies found that a medium-to-large change in intention induced by manipulated interventions caused only a small-to-medium change in behavior [43] , where an effect size of 0.5 is medium and one of 0.2 is small [44] . Sheeran found that about 47% of those intending to take action fail to act [42] , consistent with Harris et al's findings [33] . Factors that are prime contenders to moderate/ mediate the relationship between intention and behaviour include behavioural control/efficacy, action planning and anticipation of consequences [41] [42] [43] . Perceived behavioural control/self-efficacy. The TPB is an extension of the TRA incorporating the concept of perceived behavioural control (PBC) as an intervening variable predicting both intention and also actual behavioural change directly [38, 39] . The direct effect of PBC on actual behavioural change partly explains why not all intention translates into behaviour. Previous reviews suggested that intention-behaviour relationships could be moderated by perceived behavioral control, with higher levels of perceived behavioural control improving prediction of intention on behaviour [42, 43] . Although some researchers suggested that PBC differs from self-efficacy because selfefficacy emphasized perceived internal control more while PBC also considers external control factors [45] , a systemic review on the efficacy of TBP found that PBC and self-efficacy had comparable effects on intention and behaviour [41] . Despite being a dominant theory of behavioural change, because the TPB is limited in predicting behaviour we sought to enhance its predictive power by replacing PBC with self-efficacy and incorporating enhanced social effects to accommodate external control factors. Implementation of intention/planning. Implementation of intention, termed ''planning'', is a potentially important factor facilitating translation of intention into behaviour [42, 43, 46, 47] . Planning is specific to situations (e.g., when, where, and how) within which one will perform the behaviour [46] . It activates the situational context for goal attainment and thereby makes the goal become more accessible [46, 47] . A meta-analytic review showed that implementation of intention as planning consistently caused a medium-to-large effect on behavioural change [47] . Anticipated regret. Anticipated regret is the expectation of feeling regret or upset if one does or does not conduct certain behaviours. Anticipated regret has been found to be a strong predictor of vaccine uptake against seasonal influenza [31, 32] , playing the lottery [48] and exercise [49] . Anticipated regret might also moderate the intention-behavior relationship: the higher anticipated regret for inaction, the better the prediction of intention on behaviour [48, 49] . A robust theoretical framework comprehensively explaining behavior change that elucidates population decision-making for health protective and promoting behaviour has long been sought. As the main contender, the TPB explains ,34% of variance in health behavioural change related to addictive behaviour, automobile-related behaviours, clinical and screening behaviour, eating behaviour, exercising behaivour, HIV/AIDS-related behaviour and oral hygiene behaivour [40] . The standard version of TPB proposes that attitudes towards the behaviour, subjective norm and PBC predict behavioral intention while intention and PBC predict the actual behavioural change [38, 39] . Additional predictors that significantly improve the model's predictive power are needed [39] . Two previous studies have examined modified versions of TPB to predict vaccination uptake against seasonal influenza [50, 51] . One study used TPB plus two additional factors: influenza vaccination history and anticipated regret, to predict intention to receive vaccine against seasonal influenza among elderly from social clubs [50] : vaccination history and anticipated regret respectively accounted for an additional 10.7% and 13.7% of total variance in influenza vaccination intention [50] . However, again the study was cross-sectional and actual vaccination uptake was not assessed. A second study of healthcare workers [51] adopted an extended version of TPB that included additional elements of anticipated regret, moral norm, descriptive norm and professional norm. The study found that controlling for the original TPB variables, moral norm and anticipated regret were significant determinants of actual receipt of seasonal influenza vaccine [51] . The study provides useful information for future application of the extended version of TPB. However, since the study was conducted among healthcare workers, some of the variables such as moral norm and professional norm which emphasize obligation and professional convictions may not be applicable among the general population. Factors influencing pH1N1 vaccine uptake at the later stage of a pandemic might be more cognitively driven unlike behavioral responses during the early stage of a pandemic which might be more affect driven [52] . Therefore, taking into account prior work on seasonal influenza vaccination uptake [50, 51] , extending the TPB could provide theoretical utility for understanding public decision on taking pH1N1 vaccination. Starting with TPB and existing literature, we therefore built a conceptual model of public decision-making for pH1N1 vaccination ( Figure 3 ). In addition to the original TPB components, seasonal influenza vaccination history, anticipated regret and vaccination planning were included in the model. The model proposed that attitudes towards vaccination (perceived benefits of pH1N1 vaccination and concerns regarding possible adverse effects of pH1N1 vaccination), perceived social pressures from significant others and other people around regarding pH1N1 vaccination (social norms regarding pH1N1 vaccination), perceived self-efficacy in taking vaccination (perceived self-efficacy), anticipated regret for not taking the pH1N1 vaccination (anticipated regret) and seasonal influenza vaccination history would predict vaccination intention, which in turn predicts vaccination planning and future vaccination uptake; anticipated regret and perceived self-efficacy could also predict vaccination status directly; finally, vaccination planning was proposed to bridge the intention-behavior gap and predict vaccination status directly ( Figure 3 ). We conducted a longitudinal study of influences on pH1N1 vaccination behaviour in Hong Kong to test this model ( Figure 3) , and subsequently followed up participants to record their selfreported receipt of pH1N1 vaccine. In this study, we aimed to answer the following research questions: How well does intention predict future uptake of pH1N1 vaccine? Does vaccination planning mediate the relation between intention and future vaccination uptake? And do the original TPB components and the additional components (extended social norms, anticipated regret and seasonal influenza vaccination history) contribute to peoples' decisions on vaccination uptake? The study obtained ethics approval from the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster. Written informed consent was waived by the IRB because all the data were analyzed anonymously, but verbal consent was obtained from all the subjects before the interview started. Hong Kong has 99% landline telephone penetration, local calls are free and telephone interviews are common and representative methods of survey data collection [53] . We conducted 13 main cross-sectional telephone surveys of psychological and behavioural responses to the first wave of the 2009 influenza A/H1N1 pandemic in Hong Kong from April through November 2009 (the parent study) [53] in order to monitor these variables. As an extension, the present study re-contacted subjects from some of these surveys and sought to understand public decision-making regarding pH1N1 vaccine uptake for mitigating the potential second wave of the pandemic. Between 12-25 January, 2010 a baseline assessment for the present study was performed, immediately prior the local pH1N1 vaccination campaign extending to the general community (vaccination for high risk groups started from December 21, 2009), and we again contacted participants for follow-up two months later, between 15-30 March 2010. Sample size determination. We estimated that a sample of at least 500 was required to achieve 80% power at an a = 0.05 to reject a model of the specified complexity ( Figure 3 ) if the model fit index Root Mean Square Error of Approximation (RMSEA) exceeded 0.08 [54, 55] . To allow for a response rate ,60% in the follow-up and the baseline surveys, we need to target at least 1,389 subjects in the baseline survey. Subject selection and inclusion criteria. A flow chart showing subject selection is provided in Figure 4 . A total of 12,965 subjects participated in the parent study [53] . All these subjects were Cantonese-speaking adults (aged$18) selected within households using a Kish Grid methodology, who were capable of and willing to answer a telephone interview. Additional details about inclusion criteria are available elsewhere [53] . Respondents in the 7 th , 9-12 th surveys of the 13 surveys comprising the parent study who, in the parent study agreed to be re-contacted and who had not received pH1N1 vaccine were invited to complete the baseline assessment for the present study. These five surveys (the 7 th , 9-12 th surveys) were selected because participants in these surveys had not had any follow-up contact either in the parent study or otherwise. This minimizes interview fatigue thereby improving response rates. These surveys were all of a comparable sample size, between 1,000-1,007 [53] . The five selected surveys were conducted between 21 July and October 23, 2009, and generated a representative [53] pool of 5,014 respondents of whom 61.4% (3,079/5,014) gave consent for further contact. From a list of the 3,079 subjects who agreed to be re-contacted, 1,648 calls were randomly selected and successfully made by a university telephone polling organization. Unanswered calls were tried at least four times at different hours and weekdays before being replaced by new numbers. Finally, a total of 1,511 (92%, 1,511/1,648) respondents agreed to participate in the baseline survey. Of these 78 (5%, 78/1,511) reported already having received pH1N1 vaccination and were therefore excluded as ineligible, leaving 1,433 respondents who completed baseline interviews. The interview questionnaire for the baseline survey was derived from literature review, our previous cross-sectional surveys [53] and the theoretical framework constructed for this study (Figure 3 ). Specialists in health psychology, statistics, infectious disease and public health jointly determined the measures comprising the final questionnaire, guided by the need to maintain low assessment load and parsimony to ensure good response rates. The finalized questionnaire consisted of five sections: Section 1 addressed respondents' self-rated health and their experience of influenzalike illness in the past six months; Section 2 addressed risk perceptions regarding pH1N1; Section 3 addressed perceived trust in information related to pH1N1 and pH1N1 vaccination from different information sources; Section 4 addressed attitudes, beliefs and social norms regarding pH1N1 vaccine/vaccination, vaccination intention and planning; Section 5 addressed key respondent demographics. Overall, the baseline assessment consisted of 44 questions, which took less than 15 minutes to complete. Other demographic data were obtained from the parent study [53] . Prior to baseline assessment for the present study, subjects were reminded of their prior participation and that they had agreed to participate in a further study. The study was introduced as a survey of attitudes towards swine flu vaccination. We sought their willingness to participate. Those agreeing were asked about their vaccination status. Subjects who reported that they had already received pH1N1 vaccination were excluded. The remaining interview was performed. A follow-up survey was conducted 2 months later wherein respondents were reminded of the earlier survey and asked about their vaccination status and reasons for having had or not having vaccination. All the data were collected through telephone interview in both Baseline and Follow-up surveys. The measures comprising the study instruments were used to build the conceptual model ( Figure 3 ) and are described below and in Table 1 . Perceived benefits of pH1N1 vaccination, and, Concerns regarding adverse effects of pH1N1 vaccination. These two constructs assessed attitudes towards pH1N1 vaccination. Perceived benefits of pH1N1 vaccination was assessed by measuring agreement on five-point ordinal scales (from 1 ''strongly disagree'' to 5 ''strongly agree'') with three statements (Table 1) . A Cronbach's alpha (a) of 0.71 indicated an acceptable internal consistency for this scale and these two items were treated as the indicators of a latent scale (Perceived benefits of pH1N1 vaccination). Concerns regarding adverse effects of pH1N1 vaccination were assessed by measuring agreement, using fivepoint scales, with two statements. The Cronbach's a for these two items was 0.64, considered acceptable by some researchers [56] , though clearly less than desirable. We therefore treated the items as reflecting a latent variable (Concerns regarding adverse effects of pH1N1 vaccination). Social norms regarding pH1N1 vaccination. While TBP considers the influence of solely coercive social pressure from significant others to perform a behaviour, previous studies suggest that it is also important to consider the generalized tendency to adopt behaviours demonstrated by others encountered in daily life for imitative reasons [48, 57] . We use the term Social norms rather than subjective norm to represent these broader coercive and imitative social influences. Social norms were assessed by agreement on a 5-point scale with two statements. The internal consistency for these two items was weaker, with a = 0.53, which suggests each item appropriately measures different social influences. We initially incorporated these items separately in the structural equation model but except for the path weights dividing almost equally between the two items, no difference was otherwise seen. We therefore retained them as indicators of a combined latent construct in the model for purposes of model parsimony [58] . Anticipated regret. Anticipated regret was assessed with two statements asking about respondents' likelihood of feeling regret. Responses of these two items were on a 7-point categorical scale (from 1 ''definitely not'' to 7 ''certain''). The internal consistency a for these two items was 0.68. The two items were used to indicate the latent variable ''anticipated regret'' in the modeling analysis. Perceived self-efficacy. One item was used to measure selfefficacy, asking about respondents' agreement on a 5-point scale with the statement ''I am confident that I can go independently to get human swine flu vaccination''. A standard scale of self-efficacy was not adopted to minimize assessment load. However, a single item for self-efficacy has been shown elsewhere to have validity in predicting behavioural change [59, 60] . Seasonal influenza vaccination history. Respondents were asked whether they had received any seasonal influenza vaccination in the past three years (Yes/no/don't know). Vaccination intention. Respondents were asked how likely it was that they would get vaccinated against pH1N1 during the winter flu season, using a 7-point Likert scale (from 1 ''definitely not'' to 7 ''certain''). Vaccination planning. We measured vaccination planning by assessing respondents' agreement on a 5-point scale with three statement items, such as ''I have planned when and where to get my human swine flu vaccination this winter''. The internal consistency a for these three items was 0.59, though less than the most common acceptable level of above 0.7, remaining at the minimal acceptable level (a ranged between 0.5-0.6) of reliability for preliminary research [56] . These items were also treated as indicators of a latent variable for modeling purposes. Reported vaccination uptake. In the follow-up survey, respondents were asked to confirm if they had received pH1N1 vaccine within the past three months. Respondents were also asked to indicate their major reasons for having or not having taken the pH1N1 vaccination using open-ended questions. Multiple reasons could be given by each respondent. We first compared demographic differences between follow-up and lost-to-follow-up respondents with Pearson chi-square test while demographic differences of the respondents who completed both the baseline and follow-up survey and the general population [61] were assessed using Cohen's effect sizes [44] . Proportions were calculated to describe patterns of vaccination intention, reported vaccination uptake, and major reasons for taking or not taking pH1N1 vaccination. Structural equation modeling was then applied to examine the determinants of pH1N1 vaccination, vaccination intention and vaccination planning based on the extension of TBP. Mplus 6.0 for Windows (Muthén & Muthén, 1998 -2010 was employed because the model comprised dichotomous (vaccination status) and ordinal (vaccination intention) outcome variables. Before testing the full structural model, zeroorder correlations between the measures of related constructs were calculated. Confirmatory factor analysis was performed to assess the adequacy of the measurement model including perceived benefits of pH1N1 vaccination, concerns regarding adverse effects of pH1N1 vaccination, social norms regarding pH1N1 vaccination, anticipated regret and vaccination planning. To test the full structural model, all variables were entered into the model simultaneously. Mean and variance adjusted weighted least squares estimation was applied to evaluate the standardized parameters (beta, b). Since chi-square test is very sensitive to sample size and non-normally distributed data, several other model fit indices were evaluated including the Comparative Fit index (CFI), the Tucker Lewis index (TLI), and RMSEA. A CFI.0.90 and TLI.0.90 indicates a good fit. RMSEA less than 0.05 and one ranging between 0.05-0.08 respectively indicate a good and acceptable model fit [55] . Misfitting models were respecified guided by theoretical soundness and modification indices [55] . Missing proportions ranged from 0.1% for seasonal flu vaccination history to 5.5% for the item ''I have planned when and where to get my pH1N1 vaccination this winter''. There was no missing data for reported vaccination uptake. Missing data were handled with multiple imputation [62] . Of the 1433 respondents who completed the baseline assessment, 896/1433 (63%) respondents agreed to participate and completed the March follow-up survey (Figure 4 ). Demographic characteristics of respondents in the baseline and followup surveys are shown in Table 2 . Compared to respondents completing both baseline and follow-up surveys, respondents lost to follow-up were younger (x 2 = 14.24, p = 0.001) and more likely to be single (x 2 = 20.26, p,0.001). Overall, the low Cohen effect sizes (,0.3) showed that the demographics of respondents who completed both the baseline and follow-up surveys were comparable to those of the general population of Hong Kong [61] . Of the 1,433 respondents who completed the baseline survey, 36% (510/1,433) reported that they would ''definitely not'' take pH1N1 vaccination during the winter flu season; 36% (521/1,433) reported being ''very unlikely/unlikely'' to take it; 19% (278/ 1,433) reported their pH1N1 vaccination likelihood as ''evens'' (50:50/equal likelihood); and only 8% (119/1,433) reported vaccination likelihood as ''likely/very likely/certain''. Within the subset of 896/1,433 respondents who completed both baseline and follow-up surveys, 7% (67/896) had reported at baseline that they would be ''likely/very likely/certain'' to receive pH1N1 vaccination. However, in the follow-up survey, only 7/896 (0.8%) respondents reported having received pH1N1 vaccination in the intervening period, 4 of whom had reported being ''likely/very likely/certain'' to receive pH1N1 vaccination at baseline. Reporting higher intention to receive pH1N1 vaccination in the baseline was associated with greater likelihood to vaccinate by follow-up (Fisher's exact test, x 2 = 24.24, p,0.001). The 7 respondents who reported taking pH1N1 vaccination gave the major reasons for deciding on vaccination as follows: Three choose vaccination because of the 'high risk of swine influenza' characterized by statements like ''swine flu is serious'', ''I am worried that swine flu will become more serious'', and ''I feel vulnerable to swine flu''; two reported that their decision was due to 'doctors' advice' and two reported 'belief of the vaccine efficacy'. Other reasons provided by one respondent only were 'belief in the vaccine's safety', 'government recommendation', 'convenient availability', and 'protection of patients'. Reasons for not having vaccination given by the 889 respondents who did not receive pH1N1 vaccination ( Figure 5) were, most frequently 'low risk of or from swine influenza' (529/889, 60%) and 'concerns regarding adverse effects of the vaccine' (328/889, 37%). Around 11% (100/889) of the respondents reported both 'low risk of/from swine influenza' and 'concerns regarding adverse effects of the vaccine'. Table 3 For the final full structural model (Figure 6 ), two additional paths were added and estimated based on the modification indices including a path from social norms to vaccination planning and path from anticipated regret to vaccination planning while the path from perceived self-efficacy to vaccination and the path from anticipated regret to vaccination were removed, coefficients for these two paths being nonsignificant and too small to be meaningful. The final model indicated a good fit with CFI = 0.96, TLI = 0.93 and RMSEA = 0.06 ( Figure 6) . The model showed that respondents perceiving greater pH1N1vaccination benefits (b = 0.15), less concerns regarding vaccine adverse effects (b = 20.20), greater sensitivity to social norms b = 0.39), higher anticipated regret if not vaccinated (b = 0.47), higher perceived self-efficacy in taking pH1N1 vaccination (b = 0.12) and receiving seasonal influenza vaccination in the past three years (b = 0.12) reported greater intention to take pH1N1 vaccination, and accounted for 59% of variance in vaccination intention scores. Greater adherence to social norms (b = 0.70), higher vaccination intention (b = 0.31) and higher anticipated regret (b = 0.19) were associated with more vaccination planning, together accounting for 67% of variance in vaccination planning. Both vaccination intention (b = 0.30) and vaccination planning (b = 0.36) significantly predicted actual pH1N1 vaccination, accounting for 36% of variance in pH1N1 vaccination ( Figure 6 ). The World Health Organization recommended a stepwise use of pH1N1 vaccines for protecting people against the pH1N1 influenza pandemic in July 2009 [63] . However, a vaccination program's efficacy largely depends on the public's compliance. Our study found that only 5% of 1,511 subjects reported having received pH1N1vaccination and of 1,433 subjects remaining unvaccinated, only 8% reported intending (being likely/very likely/certain) to take the pH1N1 vaccine. Two months later in the follow-up survey, an even smaller proportion, 0.8% of the respondents who completed both the baseline and follow-up survey reported having been vaccinated against pH1N1. Perceived low risk of pH1N1 and concerns regarding vaccine-related adverse effects were the two most frequently cited reasons for refusing the vaccination. The extended TPB model suggests that both the original TPB components and the additional components contribute to people's decisions on vaccination uptake but that social norms and anticipated regret for not taking vaccination were the strongest determinants of vaccination intention and vaccination planning. Finally vaccination planning partially-mediated the relation between intention and reported vaccination uptake. Compared to previous studies, vaccination intention was much lower in our study than that found in surveys conducted prior to the influenza pandemic [23] or before the vaccine was available [11] , but was comparable to the findings of surveys conducted in France [20] and Turkey [24] after pH1N1 vaccination programmes were launched there. An earlier Hong Kong study that relied on expressed intent to predict vaccination uptake [11] failed to accurately predict the subsequent meager population uptake of pH1N1 vaccination by, at best, an order of magnitude [64] , suggesting that intention alone is insufficient for predicting future vaccination uptake, consistent with empirical findings in other areas [33, 42] . Despite predictions that intended pH1N1 vaccination uptake would decline if there was insufficient data on novel vaccine safety and efficacy [11] , safety issues were not the predominant barrier to vaccination in the present study. While 37% of our study respondents who remained unvaccinated cited vaccine safety concerns, despite good evidence that the vaccine is effective with a risk profile similar to that of seasonal influenza vaccine [65] , almost twice as many, 60%, cited 'low risk of/from swine influenza' as their reason for not getting vaccinated, suggesting that these respondents felt no advantage would be gained by vaccination. Around 11% of respondents, cited both 'low risk of/ from swine influenza' and 'concerns regarding adverse effects of vaccine' as the reasons for not getting vaccinated, seemingly adopting a risk-benefit approach to vaccination decision-making. However, in the setting of low influenza risk, with the reports of vaccine related adverse events in the media after the vaccine was available for the priority groups (Figure 2) , people may shift their perceived risks away from influenza and towards vaccination, suggestive of availability bias (risk distortion by easily recalled events) [66] . We believe that perceived vaccine risk would become progressively less of a barrier to vaccination as perceived influenza risk increases, and vice versa. Moreover, despite recent reports that Hong Kong residents would be sensitive to vaccination pricing when considering whether to vaccinate [11, 12] , only 2.5% of our respondents cited high vaccine cost as the reason for rejecting vaccination. Major reasons for taking pH1N1 vaccination corresponded to reasons for not taking it, with perception of pH1N1 risk most frequently cited. However, the few respondents receiving pH1N1 vaccination prohibited meaningful comparison. The extended version of TPB model fits well to the survey data. The model showed that an expanded social norms and anticipated regret accounted for most of the variance in vaccination intention, rather than the more core elements of TPB. In turn, social norms independently accounted for more than twice the variance in vaccination planning than did intention, and vaccination planning accounted for more variance in vaccination uptake than did intention. Thus it seems that social norms comprise the major influences on vaccination uptake through modifying vaccination intention and planning. A meta-analytic review of TPB efficacy concluded that the TPB variable subjective norm (perceived coercive social pressure from significant others) weakly predicted intention compared to other TPB components, mainly due to poor measurement [41] . ''Descriptive norm'' (perception of what other people do, imitation or conformity behaviour) is reportedly a more important predictor for intention [48, 57] . Here we combined Table 3 . Correlations, means, standard deviations, and standardized factor loadings for the measurement model. measures of subjective and descriptive norms, treated as a latent variable (social norms), because they were found to have much the same predictive direction and weight. Multiple item measures of norms should have better predictive power than single item measures [41] . This model importantly informs public health approaches to population behaviour during respiratory epidemics. First, information uncertainty or untrustworthiness, for example regarding vaccine safety, is likely to prompt people look to others for their cues to action: the social environment, namely what other people believe and do powerfully influences decisions to action [45, 67] . People often tend to imitate others, so establishing a ''vaccination trend'' may help uptake. For example, it could be effective to encourage those who remain unvaccinated with feedback from vaccinated peers and by providing an updated total of numbers vaccinated. What the general public think and do may prove to be as influential as information from scientists or health professional [68, 69] . Second, encouraging uptake of a new vaccine will be problematic if the associated threat element is low, irrespective of vaccine pricing, particularly for novel and untested vaccines. Vaccine safety and efficacy data should be provided wherever possible at all levels including through health-care providers, media and the general public. To effectively communicate the risk and benefit of a novel vaccine, it is important to establish an effective surveillance system to monitor vaccination progammes and rapidly respond to any reported adverse events [70] . The media have an important influence and both reactionary and opinionated news items should be recognized as potentially detrimental to vaccination uptake. In particular, the need to develop stories that generate revenue increasingly overrides balanced reporting in contemporary media. Hence risk amplification remains a problem. Public health agencies need to improve their liaison with influential media outlets to minimize this, where possible. Third, omission bias, a phenomenon where people view vaccination as more risky than remaining unvaccinated, could be a barrier for vaccination [71] . Omission bias arises when there is anticipation of greater regret about adverse effects of vaccination, if taken, than the regret about being infected with influenza if vaccination is rejected [72] . Therefore, social marketing emphasizing the far greater likelihood of regret for consequences due to refusing vaccination than the regret over an improbably low adverse event due to taking vaccination may help to reduce this bias. For example, previous studies found that simply asking two questions about feeling regret for inaction could increase respondents' intention to play a lottery or do exercise [48, 49] . Finally, vaccination planning is a key intervening variable between vaccination intention and actual vaccination. This is to be expected given that it is more proximal to actual behaviour than intention is. In those who may be undecided, interventions facilitating planning may prompt action. This could include suggesting where, when and how to get vaccination, improving and publicizing accessibility of vaccination centres and opening times. Even so, intention and planning explained only 36% of the variance in the reported vaccination behaviour, suggesting that other factors, such as intention stability [42] , influencing vaccination behaviour await identification. Study limitations include baseline attitudes/beliefs, vaccination intention and planning being measured at the same time point, prohibiting exploration of causality in observed associations. Some study measures were constrained due to length of telephone interviewing, and while sub-optimal were necessary methodological compromises. Although most researchers recommended Cronbach's a of 0.7 as the minimal acceptable for internal consistency of multi-item scales, others accept 0.6 or 0.5-0.6 for preliminary research as the cut-off point [56] . Other than dimensionality concerns, lower a can reflect too few items comprising the putative scale [73] . This is more likely for complex variables, such as social norms which have a broad spectrum of elements. Though less than perfect, measurement errors can be reduced by incorporating the items as a latent variable in SEM [55] , an approach we adopted. Additionally, collinearity between exogenous indicators, such as social norms and perceived benefit can be potentially problematic, perhaps lowering the accuracy of SEM estimation. However, since high associations between measures of the constructs were not observed (Table 3) then collinearity-related error is probably small [74] . Despite being randomly selected for the parent study, subjects of this study were not randomly selected from the general population, although demographics suggest the current sample is comparable to the Hong Kong general population [61] (Table 2 ). Moreover, subject recruitment was based on voluntariness and all data were selfreported. All could cause social desirability and selection bias, so caution is needed before extrapolation to the general population. Also refusal at follow-up could have influenced patterns of responses. Our study examined public decision-making regarding a novel influenza pandemic vaccine. Our findings may not apply to vaccination against seasonal influenza due to numerous differences in beliefs towards the vaccination. For example, although perceived low risk remains the major reasons for refusing vaccination against seasonal influenza as in our study, vaccine safety is seldom cited as a barrier [28, 29] whereas we found that about one third of respondents had vaccine safety concerns. Cultural differences in influenza and vaccination-related beliefs are possible [75] , but these differences may gradually diminish with the increasing identical news information available through the three dominant news agencies and common public health strategies being increasingly universal. Related stories, such as use of preservatives and adjuvants in vaccine manufacture may enhance knowledge and reduce trust in product safety [76] . The role of media remains much under-researched in this regard. Finally, data was insufficient to reliably report the reasons for pH1N1 vaccination uptake among the population. Nonetheless, compared with other cross-sectional studies [12, [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] , the longitudinal design of this study strengthens understanding of influences on population decision-making for pandemic influenza vaccination uptake and represents a step forward in this area of research. This study is novel in linking theoretically derived, vaccination-related cognitions to subsequent influenza vaccination behaviour, and exemplifies that within the Hong Kong Chinese culture, social norms and action planning are far more influential than intention in predicting vaccination behaviour.

Dengue Virus Virulence and Transmission Determinants

The mechanisms of dengue virus (DENV) pathogenesis are little understood because we have no models of disease; only humans develop symptoms (dengue fever, DF, or dengue hemorrhagic fever, DHF) and research has been limited to studies involving patients. DENV is very diverse: there are four antigenic groups (serotypes) and three to five genetic groups (genotypes) within each serotype. Thus, it has been difficult to evaluate the relative virulence or transmissibility of each DENV genotype; both of these factors are important determinants of epidemiology and their measurement is complex because the natural cycle of this disease involves human-mosquito-human transmission. Although epidemiological and evolutionary studies have pointed to viral factors in determining disease outcome, only recently developed models could prove the importance of specific viral genotypes in causing severe epidemics and their potential to spread to other continents. These new models involve infection of primary human cell cultures, “humanized” mice and field-collected mosquitoes; also, new mathematical models can estimate the impact of viral replication, human immunity and mosquito transmission on epidemic behavior. DENV evolution does not seem to be rapid and the transmission and dispersal of stable, replication-fit genotypes has been more important in the causation of more severe epidemics. Controversy regarding viral determinants of DENV pathogenesis and epidemiology will continue until virulence and transmissibility can be measured under various conditions.

models can estimate the impact of viral replication, human immunity and mosquito transmission on epidemic behavior. DENV evolution does not seem to be rapid and the transmission and dispersal of stable, replication-fit genotypes has been more important in the causation of more severe epidemics. Controversy regarding viral determinants of DENV pathogenesis and epidemiology will continue until virulence and transmissibility can be measured under various conditions. Dengue virus (DENV) pathogenesis seems to be determined by numerous, interacting factors: viral virulence, host immunity and immune status, host genetics and possibly others (e.g., preexisting diseases). Because we have no models of severe dengue disease (DHF), all associations of viruses with increased pathogenesis have been indirect and painstakingly slow in being developed. Transmissibility has also been measured indirectly: the successful isolation of viruses from patients and the preponderance of one serotype over another have been documented in numerous countries but this has also introduced biases in our sampling. We do not have available a fully representative set of DENV genomes to study and understand what truly constitutes the natural range of DENV variation; many samples from mosquitoes or less-ill human infections are missing, in addition to those from countries lacking the laboratory and public health infrastructure necessary for detecting and isolating viruses. Virus-mosquito interactions also add a layer of complexity to the determination of which genotype is being transmitted and we are only beginning to measure the effects of this selection. However, many new methods have been applied to the study of DENV genetic variation, replication fitness and their effects on transmissibility and pathogenesis in humans. None of these methods are perfect and we must still regard them as surrogates for measurements of the natural viral determinants of disease. Thus, understanding this complex system will probably require multidisciplinary approaches to solving the mysteries of the interaction of the many factors that contribute to DENV epidemiology. Other, nonviral factors contributing to DENV pathogenicity and transmission are discussed in accompanying chapters. The first descriptions of DENV virulence differences came from epidemiologic and entomologic studies done in the South Pacific by Rosen and Gubler, in the 1970s (Gubler et al. 1978; Rosen 1977) . It was noted that some outbreaks in this region had fewer or no cases of DHF and the transmitted viruses were considered of low virulence; other outbreaks had many cases of DHF, after primary infection and these viruses were therefore more virulent. However, it took the development of RNA nucleotide sequencing techniques and the use of these sequences to generate phylogenetic trees of evolutionary relationships among viruses to discover that specific variant groups, or genotypes, were more frequently associated with dengue epidemics and severe disease (Chungue et al. 1995; Lanciotti et al. 1997; Lanciotti et al. 1994; Messer et al. 2003; Rico-Hesse 1990; Rico-Hesse et al. 1997 ). More recently it has been shown that some genotypes associated with DHF have been introduced and become established (endemic) in other continents, sometimes displacing the less-virulent DENV already being transmitted in those regions (causing DF only). "Virulent" genotypes have been described for serotypes 2 and 3 and it remains to be seen if further evolutionary studies will pinpoint similar groups in serotypes 1 and 4 (Rico-Hesse 2003) . In the case of DENV, there is no evidence for rapid evolution and selection as in HIV, influenza or SARS viruses; for DENV, man-made ecologic disruption or increments in the number of mosquitoes or hosts are more important than evolution towards more virulent genotypes. There has yet to be evidence for the circulation of a recombinant DENV and those recombinant genomes described to date have been the product of enzymatic amplification techniques and are thus probably lab artifacts (Aaskov et al. 2007; Holmes and Twiddy 2003; Worobey et al. 1999) ; no one has isolated and fully characterized a recombinant DENV that is being transmitted in nature, causing disease. Although there is evidence for recombination in other, positive-strand viruses, specific steps in DENV replication might keep this event from occurring, although ample opportunities seem to exist in multiplyinfected humans and mosquitoes (Monath et al. 2005) . Also, there is no evidence for the epidemic transmission of the sylvatic genotype viruses from West Africa or Malaysia. These older, seemingly less-virulent viruses are transmitted mainly by canopy-dwelling mosquitoes to monkeys and they do not cause outbreaks in the human populations inhabiting those areas. The viruses isolated from dengue patients during outbreaks belong to genotypes imported from other continents (in the case of serotype 2, a genotype originating in the Indian subcontinent was introduced to Africa) (Rico-Hesse 2003) . Some researchers, with ample field experience in tropical areas, believe that these zoonotic cycles will eventually disappear, because of the constant reduction of natural forests; thus, these cycles have practically no importance as reservoirs of human dengue (Rodhain 1991) . Although we have not detected increases in replication fitness for any given genotype, some of the evolutionary events leading to more virulent strains seem to have already occurred and by the time these viruses rapidly spread from Southeast Asia to other areas of the world (1940s), they were already virulent or replicatively fit (Gubler 2002) . We have yet to measure any specific genetic changes that are fixed in the viral population as virulent genotypes are successfully dispersed to other continents and we do not know whether there have been any changes imposed by selection in their new environments (e.g., for serotype 2, Southeast Asian genotype introduced into the Americas; for serotype 3, Sri Lankan genotype III introduced to the Americas). That is, these viruses had increased fitness at their origin, in that they are directly linked to the appearance of DHF and they are transmitted more efficiently by mosquitoes. However, some researchers believe differences in clinical presentation and severity of epidemics are a function of only immunologic and genetic differences between the human populations in both continents (Southeast Asia and Americas) (Halstead 2006) or that nonneutralizing antibodies formed during DENV infection play a role in gradually selecting for more pathogenic viruses in humans (Morens and Fauci 2008) . Of special concern lately has been the effect of global warming on the incidence and spread of dengue disease. Although there are probably no effects on DENV replication in humans, if environmental temperatures rise, many investigators presume there might be an effect on virus transmission by mosquitoes. This is derived from the fact that increases in temperature (along with increases in rainfall) directly affect mosquito development (from larval to adult stages) and their populations can increase dramatically. Also, increases in average temperatures in new climes might make conditions favorable for mosquito breeding and the establishment of new populations. This could surely lead to more chances of exposure to mosquito bites for the human population and thus for infection by mosquito-borne viruses. However, the reason why many mosquito-borne diseases have yet to affect large populations in developed nations seems to be a lack of exposure to mosquito bites; air-conditioning and human behavior have been shown to reduce DENV transmission in the southern United States (Reiter et al. 2003) . That is, human activities and their impact on local ecology have generally been more significant in increasing dengue prevalence; thus, dengue disease seems to be influenced more by economic than climatic factors (Gubler et al. 2001; Reiter 2001 ). Also, research described below has shown that increases in temperature might have more of an impact on selecting for those virulent genotypes that are already being transmitted by mosquitoes. Mosquito survival rates and the time it takes for a virus to infect and be transmitted by the mosquito seem to be more important in this context. Another subject that has received renewed attention is the possibility of human modification or management of viral virulence by impacting transmission dynamics. This is important for the application of rapid public health measures in the event of the emergence of new strains of parasites, bacteria or viruses (Lipsitch and Moxon 1997) . We presume we can influence virulence by the application of changes in human habits or by the use of control factors such as vaccines -and there are numerous precedents of how public health measures have changed the population dynamics of microorganisms (e.g., there are now more cases of vaccine-induced polio or yellow fever in some countries). Efforts to understand the relationship between parasite adaptation to hosts, virulence and transmission have developed into a small industry in evolutionary biology (Bull and Dykhuizen 2003) . Although most discussion is still theoretical, the relationships between virulence and transmission have been weighed, presuming that there is an evolutionary trade-off for optimizing either factor within an organism (Ebert and Bull 2003) . That is, to increase its chances of transmission to another host, an organism will limit its replication or virulence in so far as to not kill its host. For example, highly virulent, zoonotic viruses such as Ebola and rabies are less transmissible than measles or common cold viruses, which rarely kill their human host. However, this is overly simplistic when it comes to the evolution of viruses that have multiple strains (multiple infections, with some cross-protection), where virulence involves immunopathology, or where there is another host or an amplifying vector involved in transmission (Day et al. 2007 ). Thus, the main goal for disease control is to understand how we can reduce virulence in a virus population without making its level of transmission higher. However, this seems a daunting task if we include the effects of evolution by individual versus group selection, bottlenecks and changes in fitness trade-offs. The concern here is whether we will shift DENV evolution and population dynamics by applying incomplete control strategies (e.g., nonsterilizing vaccination or vector transformation). The controversies mentioned above point to factors we should consider or understand in the control of dengue disease: evolutionary studies of DENV have helped us concentrate on detection (diagnosis and sampling), analysis (genetic and phenotype) and control (transmission dynamics in host and mosquitoes) of those genotypes already shown to be the culprits (Rico-Hesse 2007). The methods described below could help us reach these goals. The determination of viral RNA sequences from different areas of the genome has now become routine and numerous laboratories around the world have this capability; this has added exponentially to the number of DENV samples available for comparison in GenBank. The comparison of these nucleotides and their encoded amino acids can be done with sophisticated computer algorithms that can tell us much about the rates and sites of mutation or evolution in the viral genome. These data can then be matched with patient viral loads, diagnoses, outbreak characteristics and transmission distribution, to look for specific associations. The RT-PCR technique has allowed for the enzymatic amplification of these sequences from very small amounts of viral RNA from almost any type of tissue but many researchers have now avoided virus isolation and characterization, thus introducing mistakes in some of the banked information (e.g., from amplification, sequencing, or cloning artifacts) and without information on viability or antigenicity. Therefore, it is also important to have access to classic virology techniques, especially if one is to derive information about virus phenotype. The comparisons of full genome sequences of many DENV have helped pinpoint differences that could be involved in virulence: the comparison of viruses from two different genotypes associated with DHF or DF only (Southeast Asia and Americas, respectively) showed that there were consistent differences in the 5 0 -untranslated region (UTR), one envelope protein site (aa390) and the 3 0 -UTR of the DENV serotype 2 genome (Leitmeyer et al. 1999) . Comparisons within the Southeast Asian genotype did not identify any specific nucleotides associated with producing DHF (Mangada and Igarashi 1998; Pandey and Igarashi 2000) , so we assume all viruses of this genotype have the potential to produce severe disease. This has also been the case with other DENV, where recent studies have suggested that differences in 5 0 -and 3 0 -UTR in the genome can alter levels of replication (Miagostovich et al. 2006; Sirigulpanit et al. 2007; Tajima et al. 2007) , which can be extrapolated to viral load in blood or disease presentation (Wang et al. 2006) . Another chapter in this volume describes how these influences may occur. The first targets of DENV replication, after mosquito bite, were postulated to be monocytes or macrophages and numerous studies focused on these cell types. However, more recent studies, using newer technologies for cell identification (mainly flow cytometry) have shown that DENV infects human monocytes poorly compared to dendritic cells, including Langerhans cells and monocyte-derived dendritic cells (Marovich et al. 2001; Wu et al. 2000) . Primary dendritic cell cultures can be derived from human peripheral blood donations to banks and this is the usual source for studies to compare the replication of low-passage DENV from patients. Because these samples are obtained anonymously, we must be careful to obtain them from blood banks in areas where there is no DENV transmission (i.e., no antigenic priming of cells) and we cannot determine the human genetic background that might lead to differences in virus replication. However, studies reported in 2003 (Cologna et al. 2005 Cologna and Rico-Hesse 2003) were able to show consistent differences in replication of DENV of two genotypes of serotype 2, demonstrating that there is an ex vivo correlation to the virulent phenotype derived from evolutionary studies. Although there seem to be innate, probably genetic differences in the yields of virus produced by cells from individual donors, this variation could be accounted for statistically and the correlations with virulence of patient-derived viruses were established (and note that these differences occur in the absence of antibodies). These primary cell cultures were also used to test recombinant viruses, to determine the influence of specific genome regions on virus replication and yields from human cell targets. These studies confirmed that the exchange of three genomic regions (5 0 and 3 0 UTRs, and E390) could reduce the levels of replication and virus yields of a Southeast Asian virus to those of wild-type, less virulent viruses of the American genotype (Cologna and Rico-Hesse 2003) . Other uses for cultured primary human cells include the identification of specific cells that are producing more virus (tropism) and whether virus replication is even required for pathologic effects. Another step in determining if a virulent genotype had an increased transmission fitness phenotype was to study differences in replication and dissemination (i.e., the possibility of transmitting virus by bite) in the natural mosquito vector, Aedes aegypti. Laboratory-reared colonies of mosquitoes (e.g., Rexville or Rockefeller strains) seem to have lost their selectivity for infection and it is recommended that mosquitoes used in these experiments be from the F4 generation or lower (F0 ¼ field-collected eggs). The virulent, Southeast Asian strains of serotype 2 were shown to infect a larger proportion of mosquitoes than the less virulent, American genotype strains, after feeding mosquitoes on blood containing the same titer of virus (Armstrong and Rico-Hesse 2001) ; also, a greater proportion of mosquitoes develop disseminated infections with the virulent genotype (Armstrong and Rico-Hesse 2003) . If mosquitoes were fed both genotypes simultaneously, they were much more likely (sevenfold) to develop an infection with the virulent strains (Cologna et al. 2005) . When the dynamics of virus replication and dissemination were compared for both genotypes, the virulent strains had reached the salivary glands up to 7 days earlier than the less virulent viruses (Anderson and Rico-Hesse 2006) . This means that virulent strains may replicate and be transmitted much sooner to human hosts, outcompeting the less virulent viruses and causing many more cases of disease, thus ecologically displacing those that cause less severe dengue. This efficiency of transmission by the vector could explain how certain genotypes have displaced others, shifting the evolution of dengue disease towards more virulence (i.e., more DHF). During this decade, major advances have been made in the development of new mouse breeds and their transplantation with human stem cells (from umbilical cord blood cells) that may effectively mimic the human immune system or show human signs of disease upon infection. The combination of studies in these mice with those in mice that are defective in interferon production or receptors have led to insights into the mechanisms of dengue pathogenesis (Bente et al. 2005; Kuruvilla et al. 2007; Kyle et al. 2007; Shresta et al. 2006) . Thus far, none of these models develop DHF and the production of DENV-specific antibodies has been low or undetectable. However, others have shown that mice engrafted with human hematopoietic cells can be effectively used to study pathogenesis of viruses for which no other models exist (Melkus et al. 2006; Watanabe et al. 2007) . It is anticipated that after adaptation of this system to DENV infection by multiple strains, with the acquisition of DENV-specific, functional B and T cells, that the signs of DHF might appear in these "humanized" mice. This would finally allow for the measurement of the many effects of immunopathogenesis, including the protective cross-immunity created by serial infection and the evaluation of many basic questions, such as dosedependence of infection, the relevance of mosquito factors to infection (e.g., salivary gland proteins) and the role of other cells as primary targets of infection (e.g., endothelial cells). Most importantly, this model could allow for the immediate testing of antivirals and vaccine candidates, where effective systems for testing products before human use have been lacking. Another promising new field has been the development of mathematical models of DENV transmission, including "evolutionary epidemiology" and virulence management. These models are being used to estimate the effect of changing host immunity or mosquito transmission on the amount of virus circulating and the risks to a hypothetical human population and epidemic "topology." These analyzes have suggested that DENV cross-serotype immunity and mosquito demographics, rather than immune enhancement, are the most important determinants in the dynamics of specific serotype cycles or genotype replacement during epidemics and that the application of incomplete control strategies might actually increase the incidence of severe disease (Adams et al. 2006; Cummings et al. 2005; Nagao and Koelle 2008; Wearing and Rohani 2006) . Although these models are complex and still require many basic measurements for their refinement, some of the details are being added as they become available from laboratory or ecological studies (e.g., quantification of cross-protection by various DENV strains and many different measurements of mosquito transmission dynamics, including vector genetics and their effect on competence and capacity, etc.). Other applications of computer modeling have involved measuring the importance of host genetics over parasite contributions to virulence. For virology, the importance of host genetics in disease pathogenesis has been discussed for many years but recent technologies have allowed researchers to weigh the influence of virus virulence over human host genetics, in the case of pandemic influenza A (Gottfredsson et al. 2008; Pitzer et al. 2007 ). However, these approaches are extremely controversial at this point, as other investigators have reached opposing conclusions when using similar methods and results (Albright et al. 2008 ). Only recently have public health officials in developed countries become concerned about the increased transmission and geographic spread of DENV. In many cases this is due to the increase in cases imported by tourists from less-developed tropical regions. For the United States, there has been a marked increase in imported cases and autochthonous transmission in Texas and Hawaii during this decade (Morens and Fauci 2008) and for the first time we have been able to document the introduction of a virulent DENV genotype into this country (CDC 2007; Rico-Hesse 2007) . This has added a sense of urgency to research using some of the models described here and a hope for additional support for studies of this long-neglected tropical disease.

Tylosema esculentum (Marama) Tuber and Bean Extracts Are Strong Antiviral Agents against Rotavirus Infection

Tylosema esculentum (marama) beans and tubers are used as food, and traditional medicine against diarrhoea in Southern Africa. Rotaviruses (RVs) are a major cause of diarrhoea among infants, young children, immunocompromised people, and domesticated animals. Our work is first to determine anti-RV activity of marama bean and tuber ethanol and water extracts; in this case on intestinal enterocyte cells of human infant (H4), adult pig (CLAB) and adult bovine (CIEB) origin. Marama cotyledon ethanolic extract (MCE) and cotyledon water extract (MCW) without RV were not cytotoxic to all cells tested, while seed coat and tuber extracts showed variable levels of cytotoxicity. Marama cotyledon ethanolic and water extracts (MCE and MCW, resp.) (≥0.1 mg/mL), seed coat extract (MSCE) and seed coat water extract (MSCW) (0.01 to 0.001 mg/mL), especially ethanolic, significantly increased cell survival and enhanced survival to cytopathic effects of RV by at least 100% after in vitro co- and pre-incubation treatments. All marama extracts used significantly enhanced nitric oxide release from H4 cells and enhanced TER (Ω/cm(2)) of enterocyte barriers after coincubation with RV. Marama cotyledon and seed coat extracts inhibited virion infectivity possibly through interference with replication due to accumulation of nitric oxide. Marama extracts are therefore promising microbicides against RV.

Tylosema esculentum (Burch.) (marama) A. Schreib. (family Caesalpiniaceae or Leguminosae) [1] , also known as "the Green Gold of Africa," is a creeping plant found in the southern parts of Africa, namely South Africa, Namibia, and Botswana. Tylosema esculentum bean and tuber extracts have been used in traditional African medicine to treat diarrhoea and for general upkeep of human health [2] . Only little chemical characterization has been done to date on the T. esculentum plant. Research on chemical and health benefits of T. esculentum plant is part of an ongoing research project under EU-INCO Marama II FP6 programme (contract number 032059). Various bioactive constituents are expected to be present in T. esculentum plant. These may include phenolic constituents, carbohydrates, and certain fatty acids, among many components. Large amounts of gallic acid have been detected in T. esculentum plant [3] . Gallic acid esterifies with glucose, the resultant hydrolysable tannins (HTs), secondary metabolites widely distributed in the plant kingdom, are known to be effective antagonists against viruses [3] . Tylosema esculentum tubers contain yet to be determined phenolics [3] . High amounts of phytosterols have also been detected in T. esculentum oil (about 75% of all phytosterols being 4-desmethylsterols and about 15.72% of the total being 4,4-dimethylsterols) [4] . Rotavirus (RV) infections are a major cause of acute gastroenteritis in infants and young children, accounting for about 611,000 deaths each year worldwide, mostly in developing countries [5] . Rotavirus is also an important pathogen in many agricultural species, inducing serious diarrhoeal diseases in neonatal and postweaning pigs and calves [6] [7] [8] . Rotavirus infection is limited to the mature enterocytes of the tips of the intestinal villi of humans, domestic animals, and mice, leading to severe gastroenteritis in the young [9] . Some strains of bovine and human RV have high sequence homology and interspecific infectivity, both implying phylogenetic relatedness [10, 11] . Previous in vitro tests of the bovine RF strain of RV on human, pig, chicken, and goat intestine epithelial cells in our laboratory have shown interspecific infectivity of rotavirus [12] , and, therefore, the bovine strain of rotavirus was used to infect human and pig intestinal cells in the current study. It is known that infection of intestinal epithelial cells with RV causes disruption of tight junctions and loss of transepithelial resistance (TER), even in the absence of cell death [13] . Loss of TER involves alteration of tight junction proteins, mainly claudin-1, occludin, and ZO-1 as determined in human small intestine carcinoma cells (Caco2) [13] , although use of cancer cells for such mechanistic studies is not appropriate. Rotavirus infection is also associated with increased production of lactate, decreased mitochondrial oxygen consumption, and reduced cellular ATP, conditions known to reduce the integrity of epithelial tight junctions [13, 14] . While vaccines such as RataTeq and Rotatrix have been made available for prevention of RV infections [15, 16] , their effectiveness remains to be verified [17] . Use of a few synthetic compounds against simian RV, such as ribavirin [18] and isoprinosine [19] , and natural products against human and bovine RV [20] , such as theaflavins, has been reported. Unfortunately, these compounds are not available for human use, which necessitates alternative methods to control RV infection [21] . A promising alternative strategy to reducing the burden of diarrhoea caused by RV may lie in identifying and developing costeffective nutritional or phytomedical solutions; which can be applicable especially in children and immunocompromised persons. With this information, and related empirical supporting information from other legume plants, it was hypothesised that T. esculentum bean and tuber extracts can inhibit RV infection. The proposed mechanisms of antiviral effects of T. esculentum are summarised in Figure 1 . It was herein projected that extracts from T. esculentum beans and tubers can be included in various medicine mixtures or whole beans and tubers be used as functional foods, with the aim to combat diarrhoea caused by RV. In this study we sought to examine the antirotaviral activity of water and ethanol extracts from bean cotyledons and bean seed coats and water extracts from T. esculentum tubers by investigating their ability to increase survival of human foetal, pig, and calf intestinal epithelial cells. All research on the plant was done with full authorisation granted to Marama II INCO (EU-FP6) research programme. T. esculentum (marama) bean samples were obtained in Botswana in 2005 and 2007, and the T. esculentum tuber sample was obtained in Botswana in 2007. Seeds were stored in plastic bags at room temperature until time for extraction. The T. esculentum tuber (approximately 20 kg wet mass) was dug out from a site in Jwaneng, then ground and air-dried at University of Botswana, and then further ground to fine powder before the extraction was done at University of Maribor, Slovenia. A voucher specimen of T. esculentum plant is kept in the Namibian National Herbarium (collecting number: GM 1063 and herbarium number: 59520). The following water extracts from T. esculentum beans and tuber were obtained: tuber water extract (MTW), cotyledon water extract (MCW), and seed coat water extract (MSCW) ( Table 1 ). Procedure. Crude bean ethanolic extracts were prepared as described by Bolling and Parkin [22] . Briefly, 1 litre of ethanol was used to extract 100 g of shade-dried T. esculentum bean at 78 • C using a Soxhlet apparatus (Carl Roth WHLG2/ER-Serie, Karlsruhe, Germany) for extraction. The evaporation procedure ensured the exclusion of ethanol from the extracts; the resultant extract was a thick paste; these extracts were labelled MSCE (seed coat ethanolic extract) and MCE (cotyledon ethanolic extract) and stored at −20 • C until use. Ethanolic extracts are known to contain mainly lipids and other ethanol soluble contents such as isoflavones [22] [23] [24] . The following extracts from T. esculentum beans and tuber were obtained: cotyledon ethanolic extract (MCE) and seed coat ethanolic extract (MSCE) ( Table 1) . Below are the yields of each extract (mg) per g of T. esculentum bean and tuber material used. Cell cultures. The following cell lines were used in the antiviral tests: human foetal small intestine epithelial cells (H4) [25] (a generous gift from Dr. Tor Savidge, USA), pig small intestine epithelial cells (CLAB), isolated in our laboratory [26] and maintained by University of Maribor, Slovenia, and bovine calf small intestine epithelial cells (CIEB), isolated in our laboratory [26] and maintained by University of Maribor, Slovenia. The cells were grown in advanced Dulbecco's Modified Eagle's Medium (DMEM) (Sigma-Aldrich, Grand Island, USA), supplemented with 10% foetal calf serum (Cambrex, Verviers, Belgium), L-glutamine (2 mmol/L, Sigma), penicillin (100 units/mL, Sigma), and streptomycin (1 mg/mL, Fluka, Buchs, Switzerland). Cell lines were routinely grown in 25 cm 2 culture flasks (Corning, New York, USA) at 37 • C in a humidified atmosphere of 5% CO 2 and 95% air until confluent monolayers attained. Culture medium was changed routinely. Propagation. An RF RV strain, as previously described in [27, 28] , was used in all tests. Confluent cultures (in 25 cm 2 Corning flasks) of H4, CLAB, and CIEB cells were washed twice in single strength phosphate buffer (1 × PBS) to remove the foetal bovine serum (FBS) before infection with RV (previously passaged 5 times on H4, CLAB, and CIEB cells in rotation; at 1 × 10 7.5 /mL TCID 50 in the presence of 1 μg of trypsin (Sigma Chemical Co., St. Louis, Mo) per mL. After a 1-hour adsorption period at 37 • C, DMEM, containing 0.5% FBS (Sigma Chemical Co., St. Louis, Mo.) and 1 μg of trypsin per mL, was added to each flask. The flasks with cells were incubated for 24 to 48 h at 37 • C, under 10% CO 2 until cytopathic effect (CPE) was observed by microscopy. When the cells showed extensive cytopathic degeneration, the flasks with cells and virus were subjected to two freeze-thaw cycles to detach and break the cells. The supernatant fluids containing detached cells and extracellular virus were removed, and then the cellular debris was removed by centrifugation at 3500 g for 10 min. Virus was stored at −70 • C until time for tests. Tissue culture infective dose (TCID 50 ) was determined using the method of Reed and Muench [29] , as previously described in [30] . Clarified supernatants of RV titrating 1 × 10 7.5 TCID 50 /mL (tissue culture infectious dose 50%, TCID 50 ) were subsequently used for the antiviral tests with T. esculentum extracts. Similar TCID 50 values were obtained by Bae et al. [31] (1.27 × 10 6 per mL) when they titrated RV on Macacus Rhesus monkey kidney cells (MA104). Cells. Tylosema esculentum extracts were dissolved in DMEM (Sigma-Aldrich, Grand Island, USA), supplemented with 10% foetal calf serum (Cambrex, Verviers, Belgium), L-glutamine (2 mmol/L, Sigma), penicillin (100 units/mL, Sigma), and streptomycin (1 mg/mL, Fluka, Buchs, Switzerland) using a rotary magnetic stirrer (Yellowline, BigSquid, Kika-Werke GMBH & Co, Germany). As control, DMEM (Sigma-Aldrich, Grand Island, USA), supplemented with 10% foetal calf serum (Cambrex, Verviers, Belgium), L-glutamine (2 mmol/L, Sigma), penicillin (100 units/mL, Sigma), and streptomycin (1 mg/mL, Fluka, Buchs, Switzerland), was used. The extracts were added diluted 10-fold to monolayers of cells in 96 well plates, then maintained at 37 • C in a humidified atmosphere of 4 Evidence-Based Complementary and Alternative Medicine 5% CO 2 and 95% air for 24 h. Following incubation, cells were carefully rinsed with PBS to remove T. esculentum extracts and other debris. Upon washing, plates were stained with 0.01% crystal violet for 5 minutes and then rinsed with water. Plates were then dried, and then crystal violet incorporated in viable cells was resuspended with 10% acetic acid (100 μL per well). Photometric quantification of crystal violet previously retained in living cells was done at 595 nm with a microplate reader (Multiscan, Finland) as described by Agelis et al. [32] . After this test for cytotoxicity, certain concentrations of T. esculentum extracts were selected for antiviral tests as shown in Table 2 . RV with T. esculentum Extracts. Tylosema esculentum ethanolic extracts, namely, seed coat ethanolic extract (MSCE) and cotyledon ethanolic extract (MCE) were dissolved using a rotary magnetic stirrer (Yellowline, BigSquid, Kika-Werke GMBH & Co, Germany) in DMEM (Sigma-Aldrich, Grand Island, USA), supplemented with 10% foetal calf serum (Cambrex, Verviers, Belgium), L-glutamine (2 mmol/L, Sigma), penicillin (100 units/mL, Sigma), and streptomycin (1 mg/mL, Fluka, Buchs, Switzerland). Rotavirus at 1 × 10 7.5 /mL TCID 50 was combined with concentrations of T. esculentum extracts (see Table 2 ). As controls, DMEM (Sigma-Aldrich, Grand Island, USA), supplemented with 10% foetal calf serum (Cambrex, Verviers, Belgium), Lglutamine (2 mmol/L, Sigma), penicillin (100 units/mL, Sigma), and streptomycin (1 mg/mL, Fluka, Buchs, Switzerland), dilutions of Combivir GlaxoSmithKline-150 mg lamivudine and 300 mg zidovudine (AZT) as antiviral agent as described by Agelis et al. [32, 33] and T. esculentum extract at below IC 50 concentrations and RV solutions at 1 × 10 7.5 /mL TCID 50 were separately used. The plant extract and virus, Combivir, and virus mixtures were added to monolayers of cells in 96-well plates and then maintained at 37 • C in atmosphere of 5% CO 2 for 24 h. Cytopathic effects of RV alone and in coculture with T. esculentum extracts or Combivir added at 0.75 mg/mL {modified from Agelis et al. [33] } on cell monolayers after treatment as listed above were measured after 24-hour coincubation of cell monolayers (1 × 10 6 cells/mL) at 37 • C in atmosphere of 5% CO 2 . It was previously shown that T. esculentum extracts alone were not cytotoxic at the concentrations used in these tests; hence the observed reduction in survival of cells upon coincubation of extract RV solutions was attributed mainly to the virus. Following incubation, relative viability of cells was determined photometrically in terms of amount of crystal violet incorporated in viable cells as described previously. Human small intestinal epithelial cells (H4) were introduced into inserts in 12-well plates (Corning Transwell, 0.4 μm pore size, New York, USA) at concentration of 1 × 10 6 cells/mL. Transepithelial resistance (TER-Ω/cm 2 ) was measured periodically using Millicell-ERS Electrical Resistance System (Millipore, Bedford, USA) until values of about 1000 Ω/cm 2 were attained, such high values indicating maximum attainable polarity in H4 cells, as previously determined (results not shown). Rotavirus suspensions (at 1 × 10 7.5 /mL TCID 50 ) were mixed with T. esculentum extracts at below their predetermined IC 50 concentrations ( Table 2 ) and then introduced onto H4 monolayers in insert wells. Tylosema esculentum extracts diluted in DMEM, DMEM alone, and Combivir diluted in DMEM were introduced as respective controls. The effect of the extracts on the cell polarity was evaluated by measurement of transepithelial electrical resistance (TER) as previously described in [34] . TER values of monolayers exposed to combinations of marama/Combivir treatments, DMEM, and RV controls were plotted. Rotavirus (at 1 × 10 7.5 /mL TCID 50 ) was incubated with T. esculentum water and ethanolic extracts at room temperature for 1 h. Following the incubation, dilutions of the incubated virus were added to CLAB, CIEB, and H4 cell monolayers in 96well plates and then incubated for 1 h at 37 • C in a humidified atmosphere of 5% CO 2 and 95% air, after which the unattached virus and the extracts were washed off with PBS. DMEM (Sigma-Aldrich, Grand Island, USA), supplemented with 10% foetal calf serum (Cambrex, Verviers, Belgium), L-glutamine (2 mmol/L, Sigma), penicillin (100 units/mL, Sigma), and streptomycin (1 mg/mL, Fluka, Buchs, Switzerland), was then introduced on the cells and then incubated for 24 h at 37 • C in atmosphere of 5% CO 2 until viral cytopathic effects were observed in control wells. Cell viability was then determined with quantification of crystal violet as described previously. As control, viral only, T. esculentum only and cell culture medium only treatments were plated onto the epithelial cells. Relative protection by T. esculentum extracts against RV infection was determined by crystal Tests were done in triplicate wells. Tylosema esculentum extracts used were as follows: cotyledon ethanolic extract (MCE), seed coat ethanolic extract (MSCE), tuber water extract (MTW), cotyledon water extract (MCW), and seed coat water extract (MSCW). Tylosema esculentum extracts were introduced at and below IC 50 or at 1 and 0.1 mg/mL of MCE (the highest concentration of MCE was not cytotoxic). Cells after Exposure to T. esculentum Extracts. Release of NO after cell exposure to T. esculentum extracts was detected using a microplate reader (Multiscan, Finland) at 540 nm after introduction of Griess reagent (Sigma, Germany) into supernatants after 24-hour incubation. Results were expressed as percent difference between nitric oxide release of test from control. Nitric oxide is proportional to the amount of nitrate accumulating in supernatants as shown by the change of colour at 540 nm. Correlation coefficients for data obtained on level of protection, nitric oxide release, and transepithelial resistance (TER) were calculated using Statsoft Statistica version 7 software. Intestinal Epithelial Cells. To determine concentration of RV in the stock solution, the method of Reed and Muench [29] was adopted with modifications. Titration results showed that the RV strain had high and variable but not cell-typespecific cytopathic effects on all the cells used ( Figure 2) . Rotavirus, when incubated without T. esculentum extracts on cells, showed dose-dependent cytopathic effect on human small intestine epithelial cells (H4 cells), pig small intestine epithelial cells (CLAB), and bovine calf small intestine epithelial cells (CIEB) (Figure 2 ). For subsequent antiviral testing an RV concentration at and/or below 1 × 10 7.5 /mL TCID 50 was used to avoid cotoxicity with extracts. Cells. Tylosema esculentum bean water and ethanolic extracts and T. esculentum tuber water extract were added on cell monolayers of CLAB, CIEB, and H4 cells at concentrations as shown in Table 2 and Figures 3 and 6 . Tylosema esculentum extracts, at the concentrations used, resulted in dose-dependent enhancement of the three types of cells. Exposure to Tylosema esculentum seed coat water and ethanolic extracts (MSCW and MSCE, resp.) generally led to enhanced survival of CLAB, CIEB, and H4 cells. This result could show that exposure to cotyledon extracts, besides the projected inhibition of RV cytopathic effects, could ensure cell survival. Figure 3 . All T. esculentum extracts, especially cotyledon ethanolic extract (MCE), led to significant protection of CIEB, CLAB, and H4 cells from cytopathic effects of rotavirus. Tylosema esculentum seed coat water extract (MSCW) offered significant protection at the lowest concentration (0.001 mg/mL), making it the most potent extract. Tylosema esculentum cotyledon water extract (MCW) also led to highly significant protection of CLAB cells. There was largely highly significant protection of H4 cells exposed to MSCE, MSCW, MTW, and MCE. Combivir (control) was shown to have highly significant protection of cells against RV. Exposure to MCE was shown to enhance survival of CIEB cells (56%) and H4 cells (275%) more than that to Combivir. On H4, there was higher protection of cells with all extracts (91%-275%) depending on concentration than Combivir alone (Figure 3 ). Some extracts, mainly seed coat and cotyledon water and ethanolic, led to higher protection at concentrations lower than the determined IC 50 . (% Ω/cm 2 ) of H4 Cells over Time. Figure 4 shows changes in TER of H4 cells exposed to T. esculentum extracts, Combivir (AZT), or DMEM with and without RV over time. Over the whole 67 hours, coincubation of RV with bean seed coat extracts (ethanolic extract (MSCE), and water extract (MSCW) or AZT led to higher TER across the cells than when the virus was coincubated with cotyledon extracts (MCE and MCW). Incubation of H4 cells with individual extracts alone or AZT alone resulted in lower TER than extract/AZT-RV coincubation treatments. Incubation of cells with RV alone from the start of the experiment until about 40 hours resulted in TER lower than the rest of the treatments. However between 40 and 67 hours of exposure, wells with RV only treatments generally had higher TER (though sporadic) than the rest of the treatments. All T. esculentum extracts generally had better enhancement of TER than AZT over the initial 67 hours. NO 2 release was significant from human intestinal epithelial cells (H4) exposed to all extracts used {seed coat water and ethanolic extracts (MSCW and MSCE, resp.), cotyledon water and ethanolic extracts (MCW and MCE, resp.), and tuber water extract (MTW)}, while only MCE and MTW significantly increased NO in CLAB cells ( Figure 5 ). Results in Figure 6 show that pre-exposure to T. esculentum extracts can significantly reduce infectivity of RV. This finding broadly points to inhibition of RV cytopathic effect by T. esculentum extracts prior to or during virus entry (specific mechanisms were not determined). Exposure of RV (at 1 × 10 7 Table 3 shows correlation of antiviral effects of different treatments with T. esculentum extracts on different cell cultures. There was high positive correlation between CLAB and H4 cells (r = 0.94), moderately positive correlation between CLAB and CIEB cells (r = 0.57) in survival from rotavirus cytopathic effects after coincubation with T. esculentum extracts, while a weak negative correlation between H4 and CIEB cells (r = −0.02) was observed after coincubation. Highly positive correlation between survival of CLAB cells after coincubation compared to preincubation with T. esculentum extracts (r = 0.91) and moderately positive correlation between CIEB cells pre-and coincubated with the extracts (r = 0.59) were observed. Poorly positive correlation between H4 cells co-and pretreated with the extracts was observed. Moderate positive correlations were also observed between survival of and nitric oxide release from CLAB (r = 0.43) and H4 cells (r = 0.79), as well as highly positive correlation in levels of nitric oxide release from H4 and CLAB cells (r = 0.9). Negative correlation was observed between survival rates and TER of cells (H4, CLAB and CIEB) after coincubation of rotavirus with T. esculentum extracts. We show herein that the RV strain used (RF strain) could be adapted to pig and human cells, additional to bovine cells. T. esculentum bean cotyledons had low or no cytotoxicity, as shown by high IC 50 concentrations ( Table 2; Figures 3 and 6) , while seed coat and tuber extracts, especially seed coat ethanolic extract (MSCE) and tuber water extract (MTW), had elevated cytotoxic effects (low IC 50 concentrations lying between 0.01 and 0.001 mg/mL) on all the cells used (Table 2 ; Figures 3 and 6) . This study has identified T. esculentum bean ethanolic cotyledon extract (MCE) and bean seed coat extract (MSCE) as having highly significant inhibitory effects against cytopathic effects of RV, both with coincubation ( Figure 3 ) and preincubation ( Figure 6 ) of the extracts and RV on human, pig, and bovine intestinal epithelial cells. Tylosema esculentum seed coat ethanolic extract (MSCE) showed the best antirotaviral activity, as shown by the lowest IC 50 concentrations (1 μg/mL). Marama cotyledon ethanolic extract (MCE) offered similar and/or higher protection against cytopathic effects of RV than that offered by antiviral drug Combivir on all intestinal epithelial cells, with higher effect on CIEB and H4 cells. Exposure of CLAB cells to seed coat and cotyledon water and ethanolic extracts at concentrations lower than IC 50 resulted in higher survival of the cells. This could be due to potentiated toxic effects due to combined presence of RV and extract (at IC 50 concentration) and diluted cytotoxicity at lower extract concentrations. This effect was cell culture dependent, implying possible different interspecies responses. There was positive correlation between levels of protection of intestinal cells from cytopathic effects of RV (crystal violet assay) with coincubation and preincubation and enhancement of polarity (TER) of H4, CLAB, and CIEB cells by T. esculentum extracts and Combivir ( Table 3) . Survival of CLAB and H4 cells was moderately to highly correlated to release of nitric oxide (r = 0.79 and r = 0.43, resp.); similarly very high positive correlation was observed in levels of nitric oxide release from H4 and CLAB cells (r = 0.9) exposed to T. esculentum extracts (Table 3 ). These findings show that there is a positive link between effects of T. esculentum extracts (pre-and coincubated) against rotavirus on the different species of cells and the release of nitric oxide. Coincubation of RV and marama seed coat ethanolic extract (MSCE), cotyledon ethanolic extract (MCE), or tuber water extract (MTW) on H4 cells resulted in generally higher enhancement of TER than Combivir throughout 67hour test period. Enhancement of polarity across intestinal epithelial cells can be an important mechanism to control entry of RV across the intestinal epithelial barrier. Rotaviruses are known to disrupt tight junctions, resulting in loss of transepithelial resistance (TER) [13] , but without cell death during viral replication [35] ; hence T. esculentum bean seed coat and cotyledon ethanolic extracts (MSCE and MCE, resp.) and T. esculentum tuber water extract (MTW) can reduce entry of RV across the intestinal barrier in vitro. Effects of T. esculentum extracts on transepithelial resistance of H4 cells, however, could not be linked to survival of H4 cells, as shown by negative correlation between survival of cells and TER in presence of the extracts. Another mechanism of inhibition of RV by T. esculentum extracts, especially in seed coats, may lie in the inactivation of RV, as shown with enhancement of cell survival after preincubation of rotavirus with T. esculentum extracts. It is herein suggested that the observed inactivation of RV may be through interference with either viral replication or capacity to bind to permissive cells. Clark et al. [36] have shown that some phenolic acids, which were expected to dominate the ethanolic extracts, have inactivation effect against RV. Similar observations were also made with extracts from peppermint, sage, and lemon balm leaves, which were highly inhibitive against HIV [37] . The inhibitive effects of T. esculentum extracts, especially seed coat and cotyledon ethanolic extract (MSCE and MCE) and tuber water extract (MTW) (≥100%), could additionally be due to enhanced release of nitric oxide (NO) from intestinal cells. Nitric oxide has been shown to inactivate RV, mainly by inhibiting replication of the virus [38, 39] . The observed enhancement of nitric oxide release from the human and pig cells was viewed as one of the mechanisms T. esculentum crude extracts can induce a nonspecific immune response in intestinal epithelium. Certain plant phytochemicals have been shown to inhibit RV activity through inhibition of viral penetration and viral replication. Some flavonoids were shown to inhibit penetration of RV in rhesus monkey epithelial cell line (MA104) [31] , while others have an effect on the viruses or on the enzymes responsible for their replication [40] . These effects could not be singly demonstrated in this study; further work on these possible mechanisms of antiviral effects of T. esculentum extracts is, therefore, encouraged. The observed antirotaviral effects can be attributed to the high phenolic acid content (especially gallic acid), in bean cotyledons, among other components (Marama II report, unpublished). Gallic acid was previously shown to prevent viral replication, inhibition of virus attachment to and penetration into cells, and virucidal effects [41] . Interference with attachment and penetration into cells could not be proved in this study. Tylosema esculentum tubers may contain some yet to be determined phenolics (Marama II report, unpublished). Ethanolic extracts had notably higher antiviral activity than aqueous bean and tuber extracts (Figures 3 and 6 ). Some essential oils as found in T. esculentum beans may contribute towards the observed antiviral effects; similar findings were reported by Astani et al. [42] using essential oils extracted from Melaleuca alternifolia. Moreover, recently, a Kunitz-type inhibitor, distinct from other known plant serine protease inhibitors and shown to be specific for elastase, was isolated in T. esculentum beans [43] . The quantity of this elastase inhibitor present in T. esculentum beans is many times greater than in soybean or any other bean or nut source reported to date [43] . Certain protease inhibitors have been shown to impart high antienteroviral effects through interference with proteolytic cleavage during the replication process [44] . The high phytosterol content of T. esculentum bean oil 4-desmethylsterols (75%) and 4.4-dimethylsterols and 4-monomethylsterols (15.72%) [4] could also contribute to the high antiviral activity in bean extracts. Phytosterols have been determined to have antihuman cytomegalovirus (HCMV) and antiherpes simplex virus (HSV) effects [45] . The antiviral activity of phytosterols is through the blocking effect on immediate-early antigen expression in fibroblast cells and blocking of virus-cell interaction and/or virus multiplication [45] . Our results have shown high in vitro inhibition of RV by T. esculentum extracts, especially seed coat ethanolic extracts (0.01 to 0.001 mg/mL) and cotyledon ethanolic extracts (≥0.1 mg/mL), and tuber water extract (0.1 to 0.01 mg/mL). Mechanisms of antiviral action may include release of nitric oxide and may also interfere with viral cytopathic effects (as shown by moderate to high positive correlation between survival from rotavirus and release of nitric oxide), inactivation of RV (inhibition of virus replication or entry into cells may be responsible), possibly through interference with replication, and enhancement of tight junctions (though not directly related to levels of survival) (Figure 1 ). Our findings suggest that T. esculentum beans can be an important source of microbicides against RV. Further phytochemical characterization of T. esculentum extracts, identification of the responsible bioactive compounds, and the elucidation of other modes of action and quality standard studies are essential. MCE: Tylosema esculentum (marama) cotyledon ethanolic extract MSCE: Tylosema esculentum (marama) seed coat ethanolic extract MTW: Tylosema esculentum (marama) tuber water extract

Autonomous Targeting of Infectious Superspreaders Using Engineered Transmissible Therapies

Infectious disease treatments, both pharmaceutical and vaccine, face three universal challenges: the difficulty of targeting treatments to high-risk ‘superspreader’ populations who drive the great majority of disease spread, behavioral barriers in the host population (such as poor compliance and risk disinhibition), and the evolution of pathogen resistance. Here, we describe a proposed intervention that would overcome these challenges by capitalizing upon Therapeutic Interfering Particles (TIPs) that are engineered to replicate conditionally in the presence of the pathogen and spread between individuals — analogous to ‘transmissible immunization’ that occurs with live-attenuated vaccines (but without the potential for reversion to virulence). Building on analyses of HIV field data from sub-Saharan Africa, we construct a multi-scale model, beginning at the single-cell level, to predict the effect of TIPs on individual patient viral loads and ultimately population-level disease prevalence. Our results show that a TIP, engineered with properties based on a recent HIV gene-therapy trial, could stably lower HIV/AIDS prevalence by ∼30-fold within 50 years and could complement current therapies. In contrast, optimistic antiretroviral therapy or vaccination campaigns alone could only lower HIV/AIDS prevalence by <2-fold over 50 years. The TIP's efficacy arises from its exploitation of the same risk factors as the pathogen, allowing it to autonomously penetrate superspreader populations, maintain efficacy despite behavioral disinhibition, and limit viral resistance. While demonstrated here for HIV, the TIP concept could apply broadly to many viral infectious diseases and would represent a new paradigm for disease control, away from pathogen eradication but toward robust disease suppression.

Similarly, the subscript j denotes SAC j, c j is the average number of sexual partners per year in SAC j, and N j is the sum of all sexually active individuals in SAC j. In this Because a TIP shares all the same viral coat proteins as HIV-1, the TIP transmission probability per partnership is expressed as a function of identical form with the same parameter values as . ' [ &%.(9-'&&'()* 3+R* '(* &@'9* >2%'L.&'$(4* * I$%* 9'-#<'1'&=7* &@2* %'9`* 9&%"1&"%2* '(* &@2* -$>2<* '9* %2#<.12>* B=* $(2* !MO* D'&@* 5# P* T4[X* #.%&(2%9b=2.%* U&@2* D2')@&2>* .L2%.)2* $,* &@2* ,$"%* !MO9V4* * g$&2* &@.&* @2&2%$)2(2'&=* '(* 1$(&.1&* %.&29* '9*`($D(* &$* '(1%2.92* &@2* %2#%$>"1&'L2* ("-B2%* '(* .* ($(<'(2.%* ,.9@'$(* hW[i7* 9$* &@2* 29&'-.&29* >2%'L2>* B=* &@'9* .##%$.1@* D'<<* B2*

Screening of Random Peptide Library of Hemagglutinin from Pandemic 2009 A(H1N1) Influenza Virus Reveals Unexpected Antigenically Important Regions

The antigenic structure of the membrane protein hemagglutinin (HA) from the 2009 A(H1N1) influenza virus was dissected with a high-throughput screening method using complex antisera. The approach involves generating yeast cell libraries displaying a pool of random peptides of controllable lengths on the cell surface, followed by one round of fluorescence-activated cell sorting (FACS) against antisera from mouse, goat and human, respectively. The amino acid residue frequency appearing in the antigenic peptides at both the primary sequence and structural level was determined and used to identify “hot spots” or antigenically important regions. Unexpectedly, different antigenic structures were seen for different antisera. Moreover, five antigenic regions were identified, of which all but one are located in the conserved HA stem region that is responsible for membrane fusion. Our findings are corroborated by several recent studies on cross-neutralizing H1 subtype antibodies that recognize the HA stem region. The antigenic peptides identified may provide clues for creating peptide vaccines with better accessibility to memory B cells and better induction of cross-neutralizing antibodies than the whole HA protein. The scheme used in this study enables a direct mapping of the antigenic regions of viral proteins recognized by antisera, and may be useful for dissecting the antigenic structures of other viral proteins.

The 2009 A(H1N1) influenza virus, which is also referred to as the swine-origin influenza virus (S-OIV), caused the first global influenza pandemic in recent decades [1] . Given continuous antigenic drift and reassortment of heterotypic influenza viruses circulating in human and animal reservoirs, global concerns have been raised regarding an increasing threat of an influenza pandemic [2] . Current treatment strategies for influenza-A viruses, such as vaccines and drugs, have not provided broad and lasting protection, partly due to the constantly evolving nature of the viral surface glycoprotein, hemagglutinin (HA) that allows it to avoid host immune attack. HA is the key viral antigen in determining host specificity and inducing neutralizing antibody since it plays a major role in binding to host cell receptors and fusing with host cell membranes [3] . For many challenging diseases caused by viruses, the recognition of certain neutralizing epitopes by the immune system can indeed provide broad and potent protection [4, 5] . The antigenic structure of HA and the corresponding antibody response are not fully understood, complicating rational design of vaccines aimed at modulating antibody responses for targeting key epitopes. Our previous knowledge of viral antigenic structure was based mainly on the structure of antibody-antigen complexes or mutational analysis of related antigenic drifts [6] [7] [8] . Recently, many new approaches have emerged for the rapid extraction of monoclonal antibodies toward target antigens from antibody phage display libraries [5, 9, 10] , using direct sorting of memory B cells [11] , or by immortalization of IgG expressing B cells that reflect the antibody repertoire [12, 13] . These methods provide much more information on viral epitopes, although they are often laborious. The major alternative approaches for epitope mapping are derived from scanning with antigenic peptides displayed on the surface of bacteriophage, bacteria and yeast [14] [15] [16] [17] [18] [19] [20] [21] , which in recent years have increasingly incorporated fluorescence-activated cell sorting (FACS). In these efforts, short defined or random peptides with a narrow length range (,50 amino acid residues (aa)) are used to screen against monoclonal or polyclonal antibodies. Inspired by advances in cell surface display technology [22, 23] and peptide fragment library construction [24] , we devised a highthroughput scheme that utilizes yeast display and FACS that allows for direct screening of random viral peptide libraries against complex antisera instead of isolated antibodies (Zuo T, Shi XL, Xu WH, Lin Z, and Zhang LQ, unpublished results). The peptide library is generated by random digestion of the gene encoding the target viral protein, followed by a PCR-based reassembly step that results in fragments with controllable lengths (normally in the range of 100-500 bp) [24] . These peptides are then expressed on the yeast cell surface by fusion to the yeast adhesion receptor AGA 2 protein, which has previously been used to display defined short antigen fragments of various viral and non-viral proteins [19] [20] [21] . From the sequences of the antigenic peptides after screening, it is possible to determine the relative frequency of each amino acid residue involved in recognition by antisera, and how these residues are distributed in the three-dimensional structure of HA. Using this approach, we analyzed the 2009 A(H1N1) influenza virus HA protein using mouse, goat and human antisera. Unexpectedly, different antigenic profiles were seen for different antisera. Moreover, five antigenic regions were identified, four of which are located in the conserved HA stem region responsible for membrane fusion. ELISA binding assays and absorption experiments using peptides that encompass these regions have confirmed their antigenic activities. The procedure for construction, expression and screening of random viral peptide libraries using yeast surface display is shown in Fig. S1 (see also Materials and Methods for details). In step 1, the full length HA gene of the pandemic 2009 A(H1N1) influenza virus (including the core HA protein of 519 aa, the signal peptide of 16 aa, and the inter membrane domain of 36 aa) was digested and re-assembled by PCR to form a pool of fragments enriched in 100-500 bp [24] . In step 2, the fragment library was ligated into the yeast display vector pCTCON-T (derived from the yeast surface display vector pCTCON-2 [23] shown in Fig. 1 ), transformed into yeast EBY100 cells and induced for expression, generating a typical library of 1610 5 -10 6 clones to cover most of the possible fragments [25] . Because there are 3 stop codons on pCTCON-T downstream of the peptide inserts to terminate every ORF, the theoretical possibility for in-frame fusions is 1/3 for the N-terminal end of the peptide inserts. Thus, the theoretical yield for cells expressing in-frame peptides is 1/6 (the fragments can be inserted in forward and reverse directions). Subsequently in step 3, the cell library was incubated with antisera, labeled with secondary fluorescent antibody and subjected to FACS in step 4, after which the cells harboring antibody-binding peptides were distinguished from those with non-binding peptides. In step 5, we isolated plasmids directly from the pool of yeast cells containing positive peptide sequences, which were retransformed into E. coli for sequencing. This reduced laborious plasmid extraction from individual yeast clones. Plasmids containing in-frame sequences were transformed back into yeast for individual flow cytometry (FCM) verification. Retransformation and FCM were also performed for plasmids containing out of frame sequences (false positive). Antigenic peptide sequences were aligned with the HA sequence in step 6 (see also Figs. 2 and 3). In step 7, statistical analyses both on the sequence and structural levels were then performed to map out the antigenically important regions of the HA protein, by summarizing the frequency of each amino acid residue appearing in the antigenic peptides (see also Fig. 4) . The HA polypeptide is cleaved into two subunits linked by a disulfide bond, HA1 and HA2, to form the mature HA trimer. The majority of the HA1 subunit forms the viral membrane-distal globular head responsible for receptor binding, whereas the HA2 subunit together with the N-and C-termini of HA1 forms the membrane-proximal stem region that plays a major role in membrane fusion. Thus two plasmids, pCTCON-HA1, pCTCON-HA2, respectively, were constructed to display the HA1 subunit of 328 aa (residues 17-344 of the HA protein) and the HA2 ecto-domain of 192 aa (residues 345-536 of the HA protein) as positive controls for FCM (Fig. 1) . The HA1 and HA2 displayed on the yeast surface were recognized by the mouse, goat and human antisera, and confirmed by FCM (Fig. S2 , panels B, E, and, H; and C, F, and I, respectively). Screening of antigenic peptides recognized by mouse antisera immunized with HA protein We first screened the yeast library displaying the H1N1 HA peptides to study the murine immune response to recombinant HA protein of the 2009 A(H1N1) Influenza virus. Freshly induced library cells were incubated with mixed sera from immunized mice and subsequently labeled with fluorescent secondary antibodies. To avoid potential bias introduced by growth advantage of the yeast cells, only one round of sorting was performed. Plasmids isolated from positive yeast cells were transformed into E. coli cells and plated. Single clones (100) were picked randomly and sequenced, of which 82 (82%) were shown to contain in-frame inserts. These results indicate that the sorting process is able to distinguish in-frame sequences from the random library that contained only 1/6 (16.7%) in-frame inserts efficiently. After verification of individual peptides by FCM detection, the ratio of recombinant antigen expressed on the yeast surface (RAYS) was determined by dividing the mean fluorescence intensity of peptide expressing yeast cells by the mean fluorescent intensity of the nonexpressing (pCTCON-2) yeast cells [26] . 56 positive clones with a RAYS ratio $2 were considered as antigenic and analyzed at both the sequence and structural level [26] . Alignment of positive peptides with the original HA sequence shows that 87% are in the HA2 region (95627 aa), with just 13% in the HA1 region (259626 aa) which cover nearly the entire HA1 protein (Fig. 2, panel A). The FCM histograms of several representative yeast clones are shown in Fig. 3 , panel A. Clones M-5 (corresponding to residues 55-313 of HA, or 70 aa shorter than HA1) and M-7 (residues 69-275, or 120 aa shorter than HA1) were strongly positive (RAYS ratio .5), and contained the entire receptor binding domain (RBD) and the vestigial esterase domain in the globular HA head. Clones M-33 (residues 387-489) and M-52 (residues 423-492) displayed much shorter peptides but also showed strong fluorescent FCM signals (Fig. 3 , panel A), albeit with relatively lower RAYS ratios of 3-4. Judged from the mean fluorescence intensities, the binding affinities of these displayed peptides were generally weaker than the whole HA1 (Fig. S2, panel B) . The sequencing results of the antigenic peptides alone are less informative (Fig. 2) . Nevertheless, when we perform a statistical analysis by summarizing the normalized frequency of each amino acid residue in all positive antigenic peptides (defined as the residue frequency map), a major peak with the full width at half maximum (FWHM) from residues 387-480 is revealed (Fig. 4 , panel A). We also modify the residue frequency map by taking into account the respective RAYS ratios of positive peptides, which however results in an overall rather similar frequency map (Fig. S3, panel B) . Furthermore, even if we combine all the in-frame sequences from sorted yeast cells, a similar map (Fig. S3 , panel C) is again obtained, implying that the noises from the in-frame but The residue frequency map is then overlaid onto the crystal model of the trimeric HA of the 2009 A(H1N1) influenza virus (PDB ID: 3LZG) [27] , and annotated by colors from red (high frequency) to blue (low frequency) (Fig. 4, panel B ). This structural frequency map reveals that the major peak (corresponding to residues 387-480 within the FWHM) on the residue frequency map can be divided into two dominant antigenic regions since they have independent structural features. One corresponds to the central coiled-coil helix CD in the HA2 subunit responsible for membrane insertion, designated as R1 (residues 424-480, indicated by red, see also Fig. 5 ), while the other contains helix A in the HA2 subunit that is important for conformational change upon low pH exposure, designated as R2 (residues 387-423, indicated by faint red and white, see also Fig. 5) [28] . R1 is present in a relatively recessed surface area while R2 is located near the exterior surface area of the stem region. For the HA1 subunit, no distinct antigenic region is identified, because of the limited number of peptides (7) with relatively longer lengths (all .200 aa), which are insufficient to form peak areas. We also screened the same yeast library against serum samples from goats immunized by the 2009 A(H1N1) influenza virus to examine whether different animal models would yield a similar profile of antigenic peptides. Following sorting and sequencing, 78/100 clones were confirmed as in-frame, among which 55 tested positive by the goat anitsera in FCM (RAYS ratio $2). Surprisingly, alignment of the positive antigenic peptides with the full HA protein shows a distribution strikingly different from the alignment obtained using the mouse antisera in several aspects. First, 78% of the antigenic peptides are located in the HA1 region and have shorter lengths (54632 aa), while only 9% are in the HA2 region (96619 aa). Second, 13% of the peptides are in the region across HA1 and HA2 (66631 aa) (Fig. 2, Representative antigenic clones were subjected to FCM. As shown in Fig. 3 , panel B, clones G-15 (residues 38-136 of HA) and G-29 (residues 70-162) in the HA1 region, G-46 (residues 308-339) near the HA1-HA2 cross region, and G-55 (residues 413-479) in the HA2 region, all shorter than 100 aa, demonstrated strong antigenicity (RAYS ratio .5). Judged from the mean fluorescence intensities, the binding affinities of these displayed peptides were generally stronger than the whole HA1 (Fig. S2 , panel E). By analyzing the residue frequency map, more peaks are revealed than those obtained from the mouse anitsera, with the largest one with its FWHM spanning residues 22-125 with a small split around residue 60, and a medium one with its FWHM at residues 303-350 (Fig. 4 , panel C, and Fig. S4 ). In the structural frequency map (Fig. 4 , panel D), the largest peak is composed of two independent regions: residues 22-60 at the N-terminus of HA1 adjacent to helix A on HA2, designated as R3; and residues 61-125 located in the vestigial esterase domain of the globular head, designated as R5 (indicated by two red regions in the globular head and stem region, respectively, see also In order from most membrane proximal to distal: R1 (orange, residues 424-480 of HA) and R2 (yellow, residues 387-423) in HA2 were determined by screening against mouse and human antisera; R3 (purple, residues 22-60), R4 (green, residues 303-350), and R5 (cyan, residues 61-125) in HA1 were determined by screening against goat antisera. The HA monomer cartoon view is shown on the right and follows the same coloring scheme, with the third monomer shown in the back and colored in grey. doi:10.1371/journal.pone.0018016.g005 should be noticed that R1 and R2 identified in the mouse serum screening also appear in this residue frequency map, although in a much less dominant manner as illustrated by the small plateau between the end of R4 and residue 479 (Fig. 4, panel C) . Taken together, the antigenic peptides recognized by the goat antisera are different from those recognized by the mouse antisera, and in particular present a more diverse picture of antigenic sites in the HA1 region. The screening against human plasma immunized by the 2009 A(H1N1) influenza vaccine was carried out essentially as described above. Among the 100 sequences obtained after FACS analyses, 74 peptides were found to be in-frame, and 51 were confirmed to be antigenic (RAYS ratio $2). It is interesting to see that the distribution of theses peptides is similar to that obtained using the mouse antisera, with only 4% of peptides in the HA1 region (87 and 37 aa in length), 4% across HA1 and HA2 (96 and 39 aa in length), and 92% in the HA2 region (100630 aa) (Fig. 2 , panel C). Using FCM, three representative clones, H-11 (residues 359-479 of HA), H-26 (residues 390-480) and H-45 (residues 423-480), all in the HA2 region, showed strong signals (RAYS ratio .5). Although only 2 antigenic peptides are in the HA1 region, a clone expressing H-1 (residues 108-194), which harbored the Sa antigenic site in RBD of the HA protein, yielded a RAYS ratio greater than 3 (Fig. 3, panel C) . Again, judged from the mean fluorescence intensities, the binding affinities of these displayed peptides were generally weaker than the whole HA1 (Fig. S2, panel H) . The residue frequency map also reveals a major peak (FWHM at residues 392 to 480). This peak position is similar to that obtained using the mouse antisera (FWHM at residues 387-480), resulting in an almost identical structural frequency map (Fig. 4 , panels E and F, and Fig. S5 ). Thus, these antigenic regions are considered identical to R1 and R2 obtained from the mouse serum screening (Fig. 5 ). The accessibility of short antigenic peptides displayed on the yeast cell surface [19] [20] [21] was confirmed in this study with fluorescence confocal microscopy, as exemplified by the antigenic clones G-29 (93 aa) and G-46 (32 aa), from the screening against the goat antisera (Fig. S6) . As shown in the figure, the antigenic peptides were localized on the outer surface of the membrane, and well accessible to the antibodies in the complex antisera. Three representative yeast clones displaying antigenic peptides G-15, G-55, and H-11, obtained from the screening against the goat antisera and human plasma samples, were characterized for neutralization titers (IC 50 ) by testing protection efficacy to MDCK cells from the A/reassortant/NYMC X-179A/California/07/ 2009(H1N1) exposure. Cells containing pCTCON2 were used as absorbent controls for evaluation of background IC 50 values. HA1 and HA2 were also incorporated to determine the ability of the globular head and stem regions to absorb antibodies in the antisera. As summarized in Table S1 , all single clones reduced the IC 50 of the corresponding antisera, demonstrating qualitatively the neutralization efficacy of these antigenic peptides. Binding affinities of purified peptides to mouse, goat and human antisera To further confirm the antigenic activities of the identified R1-R5 regions, five newly designed short peptides encompassing regions R1-R5 (Fig. 6, panel A) were expressed as C-terminal fusions to thioredoxin (Trx) and purified from E. coli [29, 30] . The binding affinities of these peptides to the mouse, goat and human antisera were characterized by the method of ELISA, with thioredoxin as the control (Fig. 6, panels B and D) . As can been seen from the figure, the ELISA results generally correspond well to the screening results shown in Figs. 4 and 5 . Specifically, the peptide P1 (corresponding to the R1 region identified in all three screenings, Figs. 4 and 5 ) was reactive to all three antisera. The peptides P3, P4 and P5, corresponding to the R3-R5 regions identified only in the screening against the goat antisera (Fig. 4 , panels C, D, and Fig. 5) were only reactive to the goat antisera (Fig. 6, panel C) . There is one exception in that the peptide P2 (corresponding to the R2 region identified in the screenings using the mouse and human antisera, Figs. 4 and 5) showed no measurable affinity to either mouse or human antisera, suggesting that the R2 region might be less conformation-independent than the other four regions. It should be noted that, after subtraction of the background binding of the negative control protein (Trx), the binding affinity of the peptide P1 to the human antisera after vaccination increased only about one fold compared with the antisera before vaccination (Fig. 6, panel D) . This might be contributed to cross-reactive antibodies to other influenza subtypes pre-existing in the human serum subjects, given the conservative nature of the P1 sequence or the R1 region [13] . We report here the identification of antigenic peptides of the 2009 A(H1N1) influenza HA protein from a combinational library of viral protein fragments displayed on the surface of yeast cells and sorted by FACS. The peptide fragments were constructed in a way to have a broad but limited range of lengths [24] , and screened using antisera from mouse, goat and human. We then used a novel statistical approach to identify antigenically important regions of the viral protein based on the frequency of each residue appearing in the antigenic peptides at both the primary sequence and structural level. This approach reveals interesting antigenic features of the HA protein. First, the antibody responses to the 2009 A(H1N1) viral protein HA appear to vary significantly depending on the species, with the goat response being strikingly different from those of mouse and human (Fig. 4) . These results imply possibly different immune responses among animal species, and in the case of human, the specific immunization background may also play a role [13] . Moreover, the antigenic peptides obtained from the screening against the mouse or human antisera are predominantly located in the HA2 region, whereas those from the screening against the goat antisera are predominantly in the HA1 region. Second, five antigenically important regions, R1 to R5 in the order from most membrane proximal to distal (Fig. 5) , are identified. Moreover, except for R5 which is located in the vestigial esterase domain of the HA globular head, all of the other regions are in the HA stem region. Among them, R1 and R2 come together to form a major peak with a FWHM range from residues 387 to 480 (Fig. 4, panels A and E) . Although the peptide P2 corresponding to the R2 region showed no measurable affinity to either mouse or human antisera (Fig. 6) , which suggests that this region might be more conformation-dependent than the rest four regions, it is PLoS ONE | www.plosone.org assigned as a separate region for the following reasons: (1) structurally it is distinguishable from the R1 region (Fig. 5) , and (2) several recent studies have reported the epitopes of monoclonal antibodies targeting H1 subtype HA proteins, which fall into the R2 region of this study [10, 31, 32] (see below). Previously most influenza HA antibodies have been found to recognize epitopes in the globular head and interfere with binding to target cells [6] [7] [8] [33] [34] [35] [36] , for example, the classical five antigenic sites (Sa, Sb, Ca 1 , Ca 2 , and Cb) [6, 7] . However, technically the monoclonal antibodies used in these previous studies were obtained from murine hybridoma cells by radioimmunoassay (RIA) or hemagglutination inhibition (HI) assay. The nature of these assays favored the isolation of antibodies that bound predominantly to the globular head of the virus and inhibited binding of the virus to the host cells. In this current study, however, complex antisera were used, which should contain antibodies recognizing other regions of the HA protein. On the other hand, only one antigenic region (R5) is identified in the globular head in this study, possibly because the epitopes in this head region are more conformation-dependent and may require longer peptides or even a complete domain of HA1 to be displayed for full antigenicity (at least for the mouse serum and human plasma samples). It is then noteworthy that the head region, especially the area around the receptor binding site, contains mostly beta structures, which more likely require stabilization by long-range interactions than the helical structures in other regions of HA (Fig. 5) . However, the peptide library constructed in this study was enriched at 100-500 bp (or less than 170 aa), which was likely insufficient to map the conformational epitopes. Nonetheless, the current screening method is a valuable complementation to the more classical approaches that are biased toward full HA1 or HA2 domain. Several recent studies support our findings regarding antigenic regions R1-R4 in the stem region of HA. In efforts aimed at characterizing antibodies with cross-neutralizing activity for various influenza virus subtypes [10, 13, 31] , it was found that most of these antibodies bind to the HA stem region and interfere with conformational changes of the protein critical for membrane fusion. For example, in the case of antibody CR6261 (binding to HA of the SC1919 virus, H1 type), two antigenic regions on HA stem were revealed: helix A in the HA2 region, and the adjacent HA1 region [32] . The critical sites in helix A (corresponding to residues 390-391, 393-395, 398, 401-402, 405 in H1N1 HA) fall within the R2 region identified in this study, whereas the hydrophobic interaction sites in the adjacent HA1 region (corresponding to residues 38-42, 306-307 in H1N1 HA) fall into regions R3 and R4 of this study, respectively. More recently, it was reported that for a different cross-neutralizing antibody that targets a broad range of H3 subtype virus strains, 12D1, the dominant contacts on the antigen (corresponding to residues 425-455 in H1N1 HA) [37] correlate well with the R1 region found in this study. Furthermore, vaccination in mice using a HA2-based synthetic peptide mimicking the 12D1 epitope was shown to provide protection against influenza viruses of H3N2, H5N1, and H1N1 subtypes [38] . Taken together, our approach should be useful for dissecting antigenic regions on the HA stem, and providing clues for designing more potent peptide vaccines in inducing cross-neutralizing antibodies than the entire HA protein [32] . Compared with other labor-intensive epitope identification methods that rely on identifying critical residues in escape mutants and crystal structures, the method used here gives a direct and panoramic mapping of the antigenic regions of viral proteins recognized by antisera in a facile and high-throughput way. Our approach likely contains an inherent bias for shorter peptides, but nonetheless complements the more classical approaches that favor longer antigenic peptides or full antigenic domains. The scanning of antigenic peptides based on phage, bacterial or yeast surface display methods has been increasingly used in recent years. Our approach presents two advantages. First, technically we adopt an easy-to-implement scheme for generating a controllable size of peptides and our screening typically involves only one round of FACS-based sorting, in contrast to multiple rounds of panning or sorting used in the literature [15, 16, 18] . Although the latter results in many repeat sequences, we argue that much information is lost in the enrichment process. Second and more importantly, since simple alignment of antigenic peptides with the corresponding viral protein yields only limited information [15, 17, 39] , we employ a novel statistical means of summarizing the frequency for each residue appearing in the antigenic peptides at both the primary sequence and structural level, which enables a direct mapping of the antigenic structure of a viral protein. Along this line, it is interesting to note that, using our statistical approach, we reevaluate the highly repeated antigenic peptides identified for the H5N1 HA protein in a recent report [15] . Several major peaks are clearly revealed (Fig. S7) , which should be useful for further characterization of the antigenic peptides for H5N1 HA. (H1N1) influenza vaccine, and a mixture of 9 samples with the highest activity among 110 samples as judged by ELISA. These samples were provided by Dr. Boping Zhou of Shenzhen East Lake Hospital (Shenzhen, China), and written informed consent was obtained from all study participants. All samples were de-identified prior to analysis, and the protocols were approved by the Institutional Review Board (IRB) of Shenzhen East Lake Hospital. All viral strains are closely related. Restriction enzymes and DNA polymerases were purchased from New England Biolabs (Beverly, MA) or Takara (Dalian, China). DNase I was obtained from Worthington (Lakewood, NJ). Fluorescence-labeled secondary antibodies were purchased from Invitrogen (Shanghai, China) or Santa Cruz Biotechnology (Santa P5 (blue arrows), in relation to the five antigenic regions R1-R5. The coordinate of the HA protein is indicated by the grey bar, with the five antigenic regions represented in the same color as in Fig. 5 . The peptides P1-P5 were expressed as C-terminal fusions to the thioredoxin (Trx) tag. Binding activities of the peptides against the mouse (B), goat (C) and human (D) antisera before and after immunization were characterized by the method of ELISA, with the Trx protein as the negative control. The x-axis shows the dilution ratios of corresponding antiserum samples. The y-axis shows the absorbance at 450 nm after development with the substrate 3,39,5,59-tetramethylbenzidine (TMB). doi:10.1371/journal.pone.0018016.g006 Cruz, USA). Oligonucleotides were synthesized by Invitrogen or Takara. The kits for DNA purification, gel recovery, plasmid miniprep and A-Tailing modification were obtained from Tiangen (Beijing, China) or Takara (Dalian, China). The kit for yeast plasmid DNA isolation was from Omega Biotech (Victoria BC, Canada). Sequencing was performed by Invitrogen or by SinoGenoMax (Beijing, China). Escherichia. coli DH5a was obtained from Takara. Yeast strain Saccharomyces cerevisiae EBY100 was obtained from Invitrogen. The pCTCON-2 yeast display vector was kindly provided by Prof. Dane Wittrup [23] . The DNA fragments coding the HA1 and HA2 proteins were amplified from plasmid CMV-R-Cali Two Xcm I restriction sites (CCANNNNN/NNNNTGG) were introduced between Nhe I and BamH I sites of pCTCON-2 to create pCTCON-T. Digestion of pCTCON-T by Xcm I restriction nuclease and gel extraction yielded the T-vector with two 39 T overhangs. Random fragments with 39 A tails were inserted between the two Xcm I sites by T-A ligation. The gene coding for the full HA protein of A/California/04/ 2009(H1N1) virus was amplified from plasmid CMV-R-Cali-04-09 with the forward primer 59-CGCCACCATGAAGGC-TATCC-39 and reverse primer 59-TTAGATGCAGATTCTG-CACTG-39. Detailed procedures for fragmentation and reassembly were described previously, except that PhusionH high-fidelity polymerase (NEB) was used in reassembly [24] . The reassembled DNA samples were purified and modified to add 59 A overhangs by an A-Tailing Kit (Takara). The backbone vector pCTCON-T was digested with Xcm I and purified with a Tiangen gel purification kit to reduce self-ligation. The gene fragments and backbone vector were ligated at 16uC for 16 h and then electroporated into E. coli DH5a competent cells for propagation. The plasmid library was then isolated and used to transform yeast EBY100 cells by the Li-Ac method [40] . Expression of peptide libraries on yeast EBY100 cells was performed as reported by Wittrup's group [23] . Transformed yeast cells were grown for 24 h at 30uC with shaking in SDCAA medium (yeast nitrogen-based casamino acid medium containing 20 g?L 21 glucose) and passed one time into fresh medium to eliminate dead cells. The library cells were then centrifuged, resuspended in SGCAA medium (yeast nitrogen-based casamino acid medium containing 20 g?L 21 galactose) to an OD 600 value of 0.5-1 and induced at 20uC for 36-48 h with shaking. Prior to yeast library screening, nonspecific interactions of serum samples (i.e., mouse, goat or human antisera) with yeast proteins were removed by incubation with induced yeast cells carrying the control vector pCTCON-2. To label cells for FACS, 1610 7 (For yeast EBY1001 cells, OD 600 <1610 7 cells/mL) freshly induced library cells were pelleted at 8,000 g for 1 min in a 1.5 mL microcentrifuge tube and washed two times with 1 mL of 16Phosphate Buffered Saline (PBS). Cells were then incubated with 100 mL of pre-absorbed serum (1:50 diluted in PBS) at 4uC for 1 h. Unbound antibodies were removed by washing two times with 1 mL of PBS. The library cells were incubated with 1-2 mg of fluorescein isothiocyanate (FITC) or phycoerythrin (PE) labeled secondary antibodies in 100 mL at 4uC for 45 min. After washing two times with PBS, cells were resuspended in 1 mL of PBS and loaded onto a BD-FACS AriaII TM machine (Shanghai, China) for sorting. Yeast cells harboring the pCTCON-2 vector were processed as described above and used as a negative control. The positive gate was set to exclude all negative control cells (,0.1% leakage). The total amount of cells used for sorting was adjusted to ensure that more than 1610 5 cells could be harvested, which were then collected in 5 mL of SDCAA medium and grown overnight at 30uC with shaking. After one passage of library cells into fresh medium, heterogeneous plasmid DNA was isolated from the library and transformed into E. coli DH5a cells. For each library, a total of 100 colonies were randomly picked from the plates and sequenced using a pair of primers (pCTCON2-Seq-For: 59-GTTCCAGAC-TACGCTCTGCAGG-39 and pCTCON2-Seq-Rev: 59-GTTC-CAGACTACGCTCTGCAGG-39). The plasmids carrying gene fragments inserted in-frame with the original HA gene were retransformed into yeast EBY100 cells, induced, and labeled as described for FCM detection on a BD-FACS Calibur cytometer (Shanghai, China). The ratio of recombinant antigen expressed on the yeast surface (RAYS) was determined by dividing the mean fluorescence intensity of peptide expressing yeast cells by the mean fluorescence intensity of the non-expressing (pCTCON-2) yeast cells. Clones with a RAYS ratio $ 2 were considered positive and the corresponding peptide sequences were selected for further analysis [26] . Yeast cells containing plasmids for displaying antigenic peptides were grown at 30uC for 24 h in SDCAA medium and induced at 20uC for 36 h in SGCAA medium. 1610 7 cells were harvested by centrifugation, incubated with goat antisera (1:100 diluted in 100 mL of PBS) and labeled with 1 mg of FITC conjugated antigoat IgG secondary antibody as described above. Labeled cells were resuspended to 5610 7 cells/mL, fixed with 4% paraformaldehyde and photographed on a Zeiss 710 laser scanning confocal microscopy (Carl Zeiss, Germany) at the Center of Biomedical Analysis, Tsinghua University. Yeast cells containing nonexpressing vector pCTCON-2 and HA1-expressing vector pCTCON-HA1 were treated in the same way as negative and positive controls, respectively. ELISA assays for purified representative peptides against three antisera Five peptides newly designed to encompass the regions R1-R5 were expressed as fusions to the C-terminus of thioredoxin (Trx) and purified from E. coli, using a modification of the self-cleaving elastin-like polypeptide tag method (ELP) (with purity over 80% as judged by SDS-PAGE) [29, 30] . For comparison, the Trx tag was used as the negative control. When mouse and goat antisera were employed as the first antibody, the wells were coated with 100 ng of peptides or Trx. But for human antisera, 10 ng of peptides or Trx control was used to decrease the background signal caused by unspecific interactions between the human antisera and the trace impurities from E. coli cells. After blocking with 10% Fetal Bovine Serum (FBS) diluted in 16PBS with 0.25% Tween-20 (PBST), serial dilutions were added to each well and incubated for 1 h at 37uC, followed by addition of 1:5,000 diluted HRP-conjugated secondary antibody. Assays were developed by adding 100 mL of 3,39,5,59-tetramethylbenzidine (TMB) substrate solution and the reactions were stopped with 50 mL of H 2 SO 4 (1 M). The assays were carried out in duplicates and each well was washed 4 times with PBST between steps. The absorbance at 450 nm was recorded on a SpectroMAX 190 Microtiter reader (Molecular Devices, CA). Prior to the Neutralization assay, 500 mL antisera (diluted 1:50, and pre-absorbed with induced yeast cells carrying the control vector pCTCON-2) were mixed with freshly induced yeast cells (1610 8 ) displaying specific antigenic peptides on surface. After incubation at 4uC for 12 h, the mixtures were centrifuged at 8,000 g for 1 min and supernatants mixed with another batch of freshly induced yeast cells. This process was repeated four times, and then the supernatants were used for determining the neutralization titers, using a microneutralization assay with A/reassortants/NYMC X-179A/California/07/ 2009(H1N1) virus and MDCK cells according to standard procedures [41] . Briefly, triplicate serial dilutions (1:50 to 1:1600) of antiserum samples were incubated with 100 50% tissue culture infective dose (TCID 50 ) of virus for 2 h at 35uC prior to adding MDCK cells. Cells were incubated at 35uC for 72 h, and the half maximal inhibitory concentration (IC 50 ) was determined. All the neutralization assays were performed at the AIDS Research Center, School of Medicine, Tsinghua University following standard procedures. Figure S1 Schematic outline of the screening approach. In Step 1, the gene encoding the viral protein H1N1 HA was amplified, digested and re-assembled to generate the fragment library (the red and grey segments indicate the antigenic and nonantigenic peptides, respectively). In Step 2, the fragment library was ligated into the display vector pCTCON-T, transformed into yeast cells and induced for expression. In Step 3, the yeast cells expressing random peptides were incubated with antisera and fluorescence-labeled second antibodies and subjected to FACS in Step 4. Afterward in Step 5, the gene fragments were isolated from the sorted cells and sequenced. The in-frame sequences were retransformed into yeast cells and verified by FCM detection individually. In Step 6, an antigenic peptide profile was extracted for the sequences corresponding to the antigenic peptides, based on which the residue frequency and structural frequency maps were derived in Step 7. (TIF) [15] . The x-axis represents all H5N1 HA amino acid residues. The y-axis shows the normalized frequency of individual residue appearing in the 784 antigenic peptides (39 unique sequences) obtained from panning against H5N1 avian influenza convalescent sera. The six clusters (I-VI) defined by Khurana et al. are graphically represented below the x-axis. Several representative antigenic peptides are also shown as green arrows (numbered according to Khurana et al.) . Even though the antigenic peptides were enriched multiple times during the screening process and thus these peptides are less diverse and might be biased in sequences, several peaks are clearly identifiable, and predominantly in the HA2 region. (TIF)

New Insights of an Old Defense System: Structure, Function, and Clinical Relevance of the Complement System

The complement system was discovered a century ago as a potent defense cascade of innate immunity. After its first description, continuous experimental and clinical research was performed, and three canonical pathways of activation were established. Upon activation by traumatic or surgical tissue damage, complement reveals beneficial functions of pathogen and danger defense by sensing and clearing injured cells. However, the latest research efforts have provided a more distinct insight into the complement system and its clinical subsequences. Complement has been shown to play a significant role in the pathogenesis of various inflammatory processes such as sepsis, multiorgan dysfunction, ischemia/reperfusion, cardiovascular diseases and many others. The three well-known activation pathways of the complement system have been challenged by newer findings that demonstrate direct production of central complement effectors (for example, C5a) by serine proteases of the coagulation cascade. In particular, thrombin is capable of producing C5a, which not only plays a decisive role on pathogens and infected/damaged tissues, but also acts systemically. In the case of uncontrolled complement activation, “friendly fire” is generated, resulting in the destruction of healthy host tissue. Therefore, the traditional research that focuses on a mainly positive-acting cascade has now shifted to the negative effects and how tissue damage originated by the activation of the complement can be contained. In a translational approach including structure-function relations of this ancient defense system, this review provides new insights of complement-mediated clinical relevant diseases and the development of complement modulation strategies and current research aspects.

The complement system was first recognized in the late 19th century when leading microbiologists such as Paul Ehrlich, Jules Bordet and George Nuttall discovered a bactericidal function of blood on anthrax bacilli (1) (2) (3) (4) . They noted that this bactericidal function was inactivated when blood was heated up to 55°C or kept at room temperature and named it "alexin." Research on guinea pigs demonstrated that the bactericidal activity of blood not only depended on the already described heat-labile alexin, but also on a heat-stable bactericidal factor. In 1899, Paul Ehrlich renamed alexin as complement and called the heat-stable substance amboceptor (3) . By 1920, four components of complement (C1, C2, C3 and C4) had already been detected, each factor being assigned a number in the order in which it had been discovered. Although the order of their discovery did not represent their activation sequence, the names were kept to avoid confusion. The antibodydependent pathway of complement activation was named the "classical pathway." Although it had already been discovered in 1913 that some bacteria and yeast as well as cobra venom factor could induce the complement system independently of antibodies, it was not until 1954 that Pillemer discovered the "properdin pathway." Now known as the "alternative pathway," it is able to induce the complement cascade independently of antibody interaction by binding directly to bacteria and yeast (5) . Two decades ago, the mannosebinding lectin (MBL), or "lectin activation pathway," was discovered. Kawasaki et al. (6) found the MBL protein in 1978, but its function remained unclear until 1989, when Super et al. (7) recognized that reduced serum levels of MBL correlated with an opsonic defect in children. Matsushita et al. then detected the proteolytic activity of the MBLassociated serine proteases (MASP-1 and MASP-2), leading to the formation of the classical C3 convertase (8) (9) (10) (11) . antibody-independent binding of danger signals such as bacteria, yeast and virusinfected cells, but also protein A, C-reactive protein, cobra venom factor, polysaccharides and damaged tissue (14, 16) . Because constant activation of the alternative pathway is due to spontaneous hydrolysis of the highly reactive C3, constant control by complement regulators is required (17) . Healthy cells are capable of various control mechanisms that prevent the spontaneous activation of C3 and protect the host from undesirable complement activation. These control mechanisms exist both in the fluid phase and membrane bound (see below). The spontaneous hydrolysis of C3 produces C3(H 2 O), which functionally resembles C3b. C3(H 2 O) associates reversibly with factor B, while plasmatic protease factor D cleaves factor B. This event, as well as the small fragment Ba, produces the C3 convertase of the alternative pathway C3(H 2 O)Bb. Binding of the protein properdin stabilizes the fragment, extending the half-life 10-fold. The C3 convertase then splits C3 into C3a and C3b, with C3b being capable of creating a new C3 convertase with the aid of factor B and D. This amplification loop is highly important not only for the alternative pathway but also for the two other activation pathways. Binding of more C3b to the C3 convertase now creates the C5 convertase C3(H2O)BbP3b. This initiates the terminal enzymatic cascade of the lytic membrane attack complex (13) (14) (15) 18) . The lectin pathway. The lectin activation pathway has been rather less intensely studied. Activation takes place when MBL binds mannose-containing surface proteins on pathogenic surfaces. In its ultra-structure, MBL closely resembles C1q, and along with the serine proteases MASP-1 and -2 (which themselves resemble C1r and C1s, respectively), forms a potent multi-enzyme complex ( Figure 1 ). Upon activation, MASP-2 catalyzes the cleavage of C2 and C4 in a similar manner to the classical pathway and forms a C3 convertase named C4b2a. MASP-1 is capable of C2 and C3 cleavage, although to a much lesser extent (11, 19) . Subsequently, C3 is cleaved into C3a and C3b, and by accretion of C3b to the C3 convertase, the C5 convertase is formed. A third serine protease MASP-3 has a distinct function compared with MASP-1 and -2, exerting inhibitory actions against MASP-2 (20) . In addition to the established activation of the lectin pathway via MBL and MASPs, it was demonstrated that ficolins were also capable of initiating the lectin pathway by forming active complexes with MASPs. There are three distinct ficolins named ficolin-1 (M-ficolin), ficolin-2 (L-ficolin) and ficolin-3 (H-ficolin or Hakata antigen). Structurally homologous to the collectin MBL as well as C1q, ficolins are soluble collagen-like proteins that bind to sugar structures presented on microorganisms and dying host cells and consequently activate the innate immune system (8, 10, (21) (22) (23) . Surfactant protein A and D (SP-A, SP-D), such as MBL, belong to the collectin family (24) , but unlike MBL, they are not able to activate the complement directly. The impact of the MBL pathway remains to be completely elucidated. It is suspected that its major role takes place during early childhood and in particular during the translational period from the passive immunity provided by the mother's antibodies to the development of the body's own mature immunity (25) . The lytic membrane attack complex. The final stage of all three activation pathways is the formation of the lytic membrane attack complex (MAC). In contrast to the three different upstream paths forming a C5 convertase, only the cleavage of C5 into the anaphylatoxin C5a and the active C5b represents an enzymatic step, while the rest of the cascade is solely an accretion of stable proteins. In detail, C5b remains bound to the target cell followed by association of C6, resulting in a hydrophilic complex. By accretion of C7, a conformational change occurs-facilitating a stable linkage by exposure of lipophilic groups. Attachment of C8 with its binding component, C8b, induces the penetration of C8a-g into the lipid double layer of the target cell membrane. The final step toward formation of a stable transmembrane pore with a diameter of 10 angstrom is the binding of 10-15 C9 proteins, which generate a cylindrical structure ( Figure 2) . Assembly of such a pore may lead to osmotic imbalance through the constant flow of ions, small molecules and water along their concentration gradient, resulting in the lysis of the target cell (26) . It is noteworthy that the importance of these transmembrane pores should not be overrated, since, for instance, the blockage of the MAC only leads to a small increase in bacterial Neisseria infection (25, 27) . The upstream effects of the complement system, such as the anaphylactoid reaction and the opsonization, appear to play a more important role (14, 28) . Effects of the complement system. The main effect of the complement system is the induction of a pathogen-associated and modulated enzymatic cascade that, once triggered, ends with the lysis of the target cell and protects the host from infection. In addition to this apparent effect, the complement system also displays crucial additional activities that appear to be even more relevant. One effect is the opsonization of the pathogen. Cleavage products such as C3b and C4b as well as C5b opsonize the surface of recognized pathogenic substances and therefore facilitate phagocytosis. Additionally, opsonization is also important for the clearance of soluble, circulating antigen-antibody complexes. After the attachment of C3b and C4b to these complexes, they are bound to com-plement receptor 1 (CR1) on erythrocytes and are subsequently transported to the spleen and liver, where the immune complexes are eliminated. C3 cleavage products also bridge the innate and the adaptive immune systems. Opsonized antigens are bound to the complement receptor 2 (CR2) on B-cells via the C3-fragment C3d, initiating the production of specific antibodies as well as the differentiation of B-memory cells. Presumably, the most important function is the induction of an anaphylactoid reaction. The small activation products C3a, C4a and particularly C5a are potent anaphylatoxins, capable of inducing the migration of phagocytes (29) , smooth muscle relaxation, degranulation of mast cells and basophile granulocytes and therefore unleashing vasoactive substances such as histamine, prostaglandins, kinins and serotonin. All can cause vasodilation and capillary leakage (30) and induce the migrated cells to release eicosanoids, oxygen radicals and lysosomal enzymes, which cause damage to the pathogens (31) (32) (33) . It is noteworthy, that complement acts far beyond "inflammation," as indicated by its close interaction with the coagulation cascade (34) (35) (36) and its involvement in the regulation of apoptosis (37) (38) (39) (40) and cellular growth (41) . The anaphylactoid functions are mediated by the interaction of C3a and C5a with their corresponding seventransmembrane-spanning receptors C3aR and C5aR (CD88), respectively (see Figure 2 ). The role of the second C5a receptor named C5L2 is not fully understood and is still controversially discussed. However, there is increasing evidence that C5L2 represents a functional receptor acting as a negative regulator of the inflammatory response. For example, it was shown that inflammation in C5L2 knock-out mice was amplified (42) and that blockage of C5L2 increased serum interleukin (IL)-6 (43). In summary, complement is highly capable of inducing all classical signs of inflammation, with the occurrence of pain, swelling, reddening, hyperthermia and impaired function. In addition to the established pathways, new pathways of complement activation were recently discovered ( Figure 3 ). During the last decade, more and more interaction sites between the two major serine protease systems of the human body, namely the coagulation and the complement cascades, were found. The potent serine protease thrombin is able to directly cleave C3 as well as C5 in a dose-and time-dependent manner, leading to biologically active C3a/C5a (44) . In addition to thrombin, investigations by our group indicated proteolytic cleavage of C3 and C5 also by FXa, FXIa and plasmin (35) . Interestingly, FVIII and tissue factor failed to directly interact with C3 and C5 (35) . Furthermore, FXIIa activates the classical complement pathway via C1q. Crosstalk between the lectin pathway and the coagulation cascade has only recently been ascertained by the observation that the complex of FVIII and von Willebrand factor possesses lectin activity (45) . Vice versa, complement factors also interact with the coagulation system. C1 inhibitor not only blocks all three established complement pathways but also the endogenous coagulation path (46). Ikeda et al. (47) found evidence that C5a induces tissue factor activity on endothelial cells. Furthermore, crosstalk between the anaphylatoxin receptor C5aR and tissue factor was recently found (36, 48) . Another possibility for complement activation exists via direct cellular interactions. Huber-Lang et al. demonstrated that phagocytic cells (macrophages, polymorphonuclear leucocytes [PMNs]) are able to cleave C5 into biologically active C5a. This cleavage was conducted by a cell-bound serine protease that was inducible for alveolar macrophages, being more constitutively active on PMNs (49) . The complement system can exert manifold detrimental effects not only on pathogenic or damaged tissue but also on healthy host tissue. To protect against a complement attack, the human body has developed various strategies. Principally, there are both membrane-bound and fluid phase complement regulators that are briefly covered ( Figure 4 ). The best known regulatory protein is the C1 inhibitor (C1-INH). C1-INH controls the activity of the classical pathway by binding to the C1 complex and initiating the diffusion of the fragments C1r and C1s. This process leads to an irreversible inactivation of the initiating serine protease. As the classical and the lectin pathway resemble each other in many ways, the C1-INH also inactivates MASP-1 and -2, thereby also inhibiting the lectin pathway. As well as inactivating complement components, C1-INH also blocks certain parts of the kinin, fibrinolytic and coagulation systems, such as coagulation factors XII and IX. Factor I is a serine protease catalyzing the cleavage of the α-chain of C3b and C4b, leading to their permanent inactivation. Cofactors for this enzymatic activity are factor H, C4-binding protein (C4-bp), CD35 and CD46. Factor H is a protein that hinders the formation of the C3 convertase by competing for the binding site with factor B. Additionally, it facilitates the dissociation of already active C3 convertases and supports the proteolytic cleavage of C3b by factor I. C4-bp also facilitates the proteolytic cleavage of the α-chain of C4b by factor I in a complex together with protein S. Serumprotein S (Vitronectin) and clusterin (Sp-40, 40) hinder the formation of the lytic membrane attack complex by adhesion to the lipophilic groups of C7, therefore leading to impaired anchorage in the cell membrane. Carboxypeptidase N inactivates anaphylatoxins of the complement system as well as other factors such as kinins and creatinine kinase MM through cleavage of terminal arginine and lysine residues of the peptides (17, 50) . CD35 (complement receptor 1 [CR1]) is found on the surface of erythrocytes as well as on leukocytes and on podocytes in the glomerula of the kidney. CD35 facilitates the decay of the C3/C5 convertase and also acts as a cofactor for factor I. CD46 (membrane cofactor protein [MCP]) also acts as a cofactor for factor I-mediated cleavage of C3b and is broadly expressed, except on erythrocytes. CD55 (decay accelerating factor [DAF]) is widely expressed, except on natural killer cells and on a special subgroup of T-cells. The protein is glycosylphosphatidyl-inositol (GPI) anchored in the cell membrane and accelerates the decay of the classical as well as the alternative C3 convertases by replacing C2a/Bb in these complexes. CD59 (protectin) is expressed ubiquitously and is similarly integrated into the cell membrane by GPI anchors. It regulates the formation of the terminal lytic membrane attack complex by inhibiting the interaction of the C8α-subchain and the first molecule of C9 so that integration into the cell membrane and the creation of a transmembrane pore is prevented (50) (51) (52) (53) . In contrast to all the beneficial effects for the host organism, the complement system can also be detrimental for the host tissue ( Figure 5 ). Many distinct pathogenetic mechanisms may lead to the expression of an excessive and uncontrolled immune response. Depending on the individual's immune status, this immune response leads to a proinflammatory systemic immune response syndrome or to compensated antiinflammatory response syndrome. Clinical complications of these reactions can be progressive sepsis and the development of multiple organ dysfunction syndrome, with enhanced susceptibility to infections. Sepsis is defined as a systemic immune response syndrome with signs of infection. Excessive inflammation is induced by the recognition of pathogen-associated molecular patterns on invading microorganisms or danger-associated molecular patterns of damaged tissue by the cellular "first line of defense" and the complement system. This result consequently leads to the robust release of cytokines from phagocytes ("cytokine storm") to fight the infection. Although of benefit to the organism when acting locally, this pattern leads to a dramatic life-challenging event when occurring systemically (54) (55) (56) (57) (58) . Various studies proposed excessive complement activation during sepsis in humans (59) . Because of its potent inflammatory profile, C5a appears to be the most detrimental molecule and has been described as "too much of a good thing" (54) . When activated, C5a may lead to immune paralysis, multiorgan dysfunction and thymocyte apoptosis (37, 40) as well as disturbance of the coagulation and fibrinolytic cascades (34) . In accordance, some protection of septic mice has been shown by the application of a C5a receptor antagonist (60) . Furthermore, C5aR antagonism during sepsis led to a changed cytokine profile, such as decreased levels of tumor necrosis factor-α and IL-6, suggesting a direct or indirect role in the synthesis of these factors (40) . Czermak et al. (61) showed an enhanced survival rate of septic rats treated with a C5a antibody. Additionally, these authors found that C5a binds to neutrophils, which led to inactivation of their functions. In a study by Flierl et al. (62) , the effects of complement on sepsis using C3 -/and C5 -/deficient mice were examined. In the absence of either C3 or C5, a reduced production of proinflammatory mediators was found. On a cellular level, it has been shown that C5a effectively interacts with cells and modulates their apoptosis rate. Interestingly, the effects on programmed cell death seem to be cell dependent, with a higher rate of apoptosis in thymocytes (37,40) but decreased apoptosis in neutrophils (39, 63, 64) . Overall, the C5ainduced changes point toward an enhanced susceptibility toward infections, as well as to a prolonged presence of neutrophils resulting in an exaggerated inflammatory response and host damage. In the emerging field of osteoimmunology, the role of complement in bone biology in general and fracture healing in particular has started to raise interest. Although some direct interactions between the immune system and bone cells have been found (65, 66) , few studies demonstrated the presence of complement components in bone cells. In particular, the expression of various complement factors might depend on the cell differentiation state. Murine osteoblastic cells were shown to produce C3 in response to vitamin D3 (67, 68) . During osteoblastic differentiation in murine osteoblasts and in human cell lines, the complement components C1q, C4, C1 inhibitor, C3a receptor (C3aR), properdin and the complement factor H were upregulated (69), whereas the expression of the subcomponents C1r and C1s together with factor H was decreased (70) . The expression of functionally active C5a receptor (C5aR; CD88) that modulated IL-6 production was described in a human osteoblastic cell line (71, 72) . Recently, mRNA and protein expression of C3aR and C5aR in human MSC were reported (73) . Immunohistochemical studies clearly indicated that complement was activated during enchondral ossification, possibly for the modulation of apoptosis (74, 75) . These studies indicated that some interaction of the complement system with bone cells exists. However, the resulting function to date has been minimally investigated. With the exception of our unpublished data and the immunohistochemical studies in enchondral ossification, in vivo data on the expression of complement components in bone are lacking, particularly with regard to fracture healing. Recently, a major role of an excessively stimulated complement system in the posttraumatic inflammatory response was postulated (16, 76, 77) . Prospective studies with polytraumatized patients showed a consumption and massive activation of complement products, positively correlating with the mortality rate (78) (79) (80) . Fracture healing is delayed in particular when additional severe injuries stress the organism (81) . The massive trauma-induced inflammation correlates with consequent organ dysfunction (79) and may possibly be involved in the delayed fracture healing. Studies with experimental blunt thoracic trauma found that the complement system contributed to the inflammatory reaction, and additionally successful inhibition of various inflammatory mediators occurred upon application of an antibody against C5a (76). Ganter et al. reported an earlier complement activation after major trauma, for which magnitude correlated with the mortality rate (16) . Unpublished data from our group underscore the role of complement in fracture healing, reflected by the enhanced C5aR immunostaining of bone sections. The anaphylatoxins C3a and C5a have been reported to exert both protective and detrimental effects in the central nervous system (82, 83) . It is noteworthy that many cells of the central nervous system are more susceptible to a complement attack, since they lack important complement regulators such as CD59 (84) . Complement products were demonstrated to be dramatically upregulated in models of cerebral ischemia in rats (85) , and evidence for deposition of C3d and C9 after experimental cerebral contusion was found (86) . In agreement with these experimental findings, systemic complement activation has been shown in stroke patients (87) . A negative role for complement has been proposed in traumatic brain injury, since systemic depletion of complement by infusion of the cobra venom factor improved blood flow and neurological function and reduced cerebral edema during experimental intracerebral hemorrhage (88, 89) . Moreover, C1 deficiency (90) and complement inhibition by infusion of the C1 inhibitor revealed some neuroprotective effects (91, 92) . In contradiction to these results for traumatic/ischemic injuries, C1q has a strong neuroprotective effect in neu-rodegenerative diseases (93) . Traumatic brain injury induces C5aR upregulation in an experimental model in rats with enhanced C5a serum levels for >1 week. The functional consequences of traumatic brain injury-induced systemic C5a generation have still to be elucidated (94) . After ischemia/reperfusion injury, complement is activated via all three established pathways (classical, alternative and MBL) (95) (96) (97) . Endothelium damaged by hypoxic stress activates complement, leading to elevated vascular permeability, release of anaphylatoxins and the accumulation of neutrophils (98) . Experimental studies indicated complement activation after ischemia/reperfusion in several organs, including lung, liver, gut, kidney, myocardium and skeletal muscle (99) (100) (101) (102) (103) (104) (105) . Cardiac surgery and cardiopulmonary bypass surgery are known to cause a considerable immunologic impact to the body, resulting in a systemic inflammatory reaction (106, 107) . Various factors contribute to the extent of the immunological impact. In addition to general surgical trauma, hypothermia and blood loss, the main mechanisms are exposure of blood to foreign surfaces of the cardiopulmonary bypass, the ischemia/reperfusion injury by aortic cross-clamping and splanchnic hypoperfusion leading to mucosal damage and consequently endotoxemia (108). An important pathogenetic role for complement in ischemia/reperfusion of the myocardium was indicated experimentally, where inhibition of the complement cascade greatly reduced myocardial damage after myocardial infarction (109) (110) (111) . In translational studies, C5 inhibitor pexelizumab was capable of reducing mortality in patients undergoing coronary artery bypass surgery (112) but not myocardial infarction (113) . Recently, serum levels of C3a and C5a were found to be significantly elevated in patients suffering from stent restenosis after im-plantation of drug-eluting stents in patients with coronary heart disease (114) . It was recently found that high MBL levels and low plasma levels of sC5b-9 were associated with an increased risk of dysfunctional cardiac performance after myocardial infarction, possibly owing to the increased complement activity during ischemia/reperfusion, which generally triggers an inflammatory response (115) . However, activation of the complement cascade after ischemia/reperfusion injury or exposure to foreign surfaces is not the only cause of harmful effects of the complement system. Several studies alluded to a role for complement in cardiovascular diseases such as atherosclerosis or vasculitis. Kawasaki disease, a systemic vasculitis in childhood causing coronary artery aneurysmata, appears to be related to MBL deficiency according to genetic family studies by Biezeveld et al. (116) . Rugonfalvi-Kiss et al. (117) found significantly lower restenosis rates in females with low MBL levels undergoing thromboendarteryectomy of the carotid artery because of atherosclerotical stenosis. Additionally, patients with MBL deficiency had a greater risk of myocardial infarction than individuals with normal MBL serum levels (118, 119) . The proposed protective effect of complement in the pathogenesis of atherosclerosis was experimentally underlined by C3 -/mice exhibiting accelerated development of atherosclerosis (120) . Clinical analysis of C2-deficient humans revealed a significant increase in cardiovascular diseases (121) . In particular, classical pathway involvement was demonstrated by Bhatia et al. (122) when C1q protected against the development of atherosclerosis in combined C1q and LDLr (low-density-lipoprotein receptor) knockout mice. In agreement with these results, deficiency of complement inhibitors such as CD59 promoted atherogenesis (123) . However, it is noteworthy that both downstream complement defects and activation of the terminal pathway were as-sociated with inherited proatherogenic effects and an increased risk of cardiovascular disease (124) . In conclusion, tightly regulated complement activation seems to protect against atherogenesis, possibly through the clearance of apoptotic cells and other debris, whereas inhibition of the terminal pathway by complement regulators appears to hinder proinflammatory effects (123, 125) . Once an atherogenic plaque has been formed, increasingly leading to ischemia and reperfusion injury, complement is excessively activated and may lead to harmful tissue damage. Generally, a major complication of complement insufficiency is an enhanced susceptibility toward infection. Several studies have shown low MBL serum levels (genetically derived) are correlated with enhanced development of systemic immune response syndrome/sepsis in patients admitted to the intensive care unit (126, 127) . Low MBL serum concentrations were also associated with the development of pneumonia after surgery in colorectal cancer patients (128) . Similarly, MASP-2 deficiency caused increased infection rates with Streptococcus pneumoniae (129) . Furthermore, there is evidence for a higher rate of chorioamnionitis in patients with low MBL levels and preterm birth (130) . Low MBL serum levels appear to be associated with recurrent spontaneous abortion based on in utero infection. This was proposed by Kruse et al. (131) , who demonstrated that patients with MBL levels <100 ng/mL had a higher abortion rate. Some urogenital infections such as vulvovaginitis, herpes simplex virus-2 infection, vulvovaginal candidiasis and vestibulitis appear to be linked to low MBL levels in comparison to healthy controls (132) (133) (134) . Downstream complement deficiencies are typically associated with recurrent invasive infections of Neisseria meningitidis and gonorrhea (135) . In defects further upstream, such as factor D or properdin, N. meningitidis also appear to be involved in most infections (136) . Deficiencies of the classical pathway may also cause enhanced infection rates, since invasive infection was the predominant clinical manifestation in patients with C2 deficiency, especially with S. pneumoniae (121) . Many efforts have been undertaken to effectively modulate the activity of complement, thereby trying to ameliorate or completely abolish the symptoms of complement-mediated diseases. It is tempting to speculate that the effective modulation of the broad spectrum of internal diseases (as listed below) may also reflect an effective future target of various complement-dependent surgical diseases. C1 inhibitor provided protective effects in myocardial cell injury (137) , transplantation (138) and various other diseases (139) . C1-INH was effective in a randomized, double-blind clinical study in septic intensive care unit patients. Whereas the renal dysfunction was reduced, the mortality rate remained unchanged in both groups (140) . The only clinical application for C1 inhibitor so far is for C1 inhibitor deficiency leading to hereditary angioedema (141) . Nafamostat/FUT-175 is an unspecific serine protease inhibitor that, besides complement activation, also blocks abundant serine protease of the coagulation system (142) . It is currently used clinically for the treatment of acute pancreatitis and for the prevention of thrombosis in disseminated intravascular coagulation and extracorporal circulation (143, 144) , underscoring the crosstalk of the complement and coagulation cascades (35) . Because of a lack of specificity and short half-life, other serine protease inhibitors, such as factor D, were not transferred to clinical trials, and further development was stopped (145, 146) . sCR1/TP10 is a soluble complement regulator inhibiting both the classical and alternative pathways and therefore is rather promising regarding ischemia/ reperfusion injury and various other conditions. The substance revealed some protective effects in male patients undergoing coronary artery bypass grafting but not in females (145, 147) . Therefore, clinical trials have been stopped (145) . Similar substances such as sCR1-sLe x / TP20 with an improved structure (148) and Microcept/APT070 (149) are currently under investigation for the treatment of diseases such as acute myocardial infarction, stroke and inflammatory diseases but have not entered clinical evaluation (150) . A hybrid form of the complement regulators DAF and MCP was designed, named complement activity blocker 2 (CAB2), and entered clinical trials under the name MLN-2222 for the treatment of coronary artery bypass grafting (151) . The complement regulator CD55 (DAF) was recently shown to ameliorate the hepatic inflammation in experimental chronic hepatitis C (152) and autoimmune posterior uveitis (153) . The complement regulator CD59 was investigated in the context of cancer research. Neuroblastoma cells are known to escape cell lysis by abundantly expressing CD59. Therefore, a peptide was generated that was capable of suppressing CD59 expression that resulted in complement-mediated killing of the neuroblastoma cells (154) . Experimental studies for the treatment of paroxysmal nocturnal hemoglobinuria with the substitution of CD59 have recently been conducted (155) , but definitive results are pending. C4BP as well as factor H have both been experimentally used to successfully abrogate complement activation on artificial surfaces and to inhibit the development of arthritis. However, these substances have not yet been clinically evaluated (156) (157) (158) . Because of their specificity, antibodies against various complement factors appear to be a more reliable treatment strategy. Eculizumab is a monoclonal antibody against C5 investigated for the treatment of paroxysmal nocturnal hemoglobinuria (159) and is currently in a phase I clinical trial for the treatment of systemic lupus erythematosus (160) . The first clinical studies in patients suffering from atypical hemolytic uremic syndrome were conducted and were partly promising, with the optimal dose and timing still to be determined (161) . Pexelizumab is also a monoclonal antibody against C5 and has revealed beneficial effects after cardiopulmonary bypass surgery, but not after myocardial infarction in clinical studies (112, 113) . Various other antibodies are currently in developmental progress. Neutrazumab and TNX-558 are both directed against C5aR. TNX-234 is an antibody against factor D. TA106 blocks factor B from targeting, for example, in age-dependent macular dystrophia and asthma (162) . At least one antibody directed against properdin is currently undergoing testing (145) . In experimental studies, this antiproperdin monoclonal antibody was shown to be beneficial during coronary artery bypass grafting, reducing the activation of platelets and neutrophils (163) . Ofatumumab is a monoclonal antibody against CD20 and exerts its effects not by inhibition, like most other complement therapeutics, but by stimulation of complement-dependent cytotoxicity. It is currently in clinical trials for the treatment of rheumatoid arthritis, B-cell chronic lymphocytic leukemia and follicular lymphoma (164, 165) . A similar approach that has not yet been clinically tested is the efficacy of the related substances HuMax-CD38 for multiple myeloma and HuMax-ZP3 for the treatment of colon, pancreatic and prostate cancer. PMX-53 is a peptidic C5aR antagonist and has proven to be advantageous in experimental animal studies for neurodegenerative diseases, rheumatoid arthritis, ischemia/reperfusion and inflammatory bowel disease (102, 166, 167) . It is currently being evaluated in clinical trials, but a recent study in humans failed to show significant effects on synovial inflammation (168) . Compstatin, a peptidic C3 inhibitor, is considered a promising drug and is currently in clinical trials for the treatment of age-dependent macular dystrophia (169) . Since compstatin has proven to be effective in various animal models for other conditions (170) , its application does not appear to be limited to agedependent macular dystrophia. It is estimated that 30% of the human population present decreased plasma levels of MBL owing to genetic mutations. Therefore, a recombinant human form of MBL as a substitution therapy for MBL-deficient people was developed and found to be safe (171) . The substance is currently under clinical investigation for people suffering from multiple myeloma and undergoing high-dose chemotherapy (145, 172, 173) . ARC1905 is an aptamer-based C5 inhibitor that inhibits the cleavage of C5 into C5a and C5b. It is intended for intravitreal application in age-dependent macular dystrophia and is currently undergoing phase I clinical trials (174) . With JPE-1375 and JSM-7717, there now exists even more C5aR antagonists that are in preclinical evaluation for the treatment of inflammatory, renal and ocular diseases. Although complement research has been in the center of interest for many years, our understanding and insight into the cascade mechanisms and their complex interaction with other protein cascades such as the coagulation cascade is still in its nascent phase. Currently, complement activation is not divisable into three canonical pathways with separate activation patterns. The exact role of complement in many diseases needs to be further clarified because successful complement interventions need to be matched to the individual patient and be as specific as possible. The authors declare that they have no competing interests as defined by Molecular Medicine, or other interests that might be perceived to influence the results and discussion reported in this paper.

Targeting vaccination against novel infections: risk, age and spatial structure for pandemic influenza in Great Britain

The emergence of a novel strain of H1N1 influenza virus in Mexico in 2009, and its subsequent worldwide spread, has focused attention to the question of optimal deployment of mass vaccination campaigns. Here, we use three relatively simple models to address three issues of primary concern in the targeting of any vaccine. The advantages of such simple models are that the underlying assumptions and effects of individual parameters are relatively clear, and the impact of uncertainty in the parametrization can be readily assessed in the early stages of an outbreak. In particular, we examine whether targeting risk-groups, age-groups or spatial regions could be optimal in terms of reducing the predicted number of cases or severe effects; and how these targeted strategies vary as the epidemic progresses. We examine the conditions under which it is optimal to initially target vaccination towards those individuals within the population who are most at risk of severe effects of infection. Using age-structured mixing matrices, we show that targeting vaccination towards the more epidemiologically important age groups (5–14 year olds and then 15–24 year olds) leads to the greatest reduction in the epidemic growth and hence reduces the total number of cases. Finally, we consider how spatially targeting the vaccine towards regions of country worst affected could provide an advantage. We discuss how all three of these priorities change as both the speed at which vaccination can be deployed and the start of the vaccination programme is varied.

Vaccination has long been viewed as a vital tool in the armoury against infectious diseases. However, this perspective is largely based on our experience with endemic infections ( [1, 2] , where a routine policy of vaccination can be used to increase the level of herd immunity [3] ) and hence reduce or even eliminate the infection [4] . Examples of such approaches abound, from the eradication of smallpox and the virtual eradication of polio, to the long-running campaigns against childhood diseases such as measles, mumps and rubella [5, 6] , to the recently introduced schemes such as vaccination against human papillomavirus [7] . However, the recent experience with novel infections, such as SARS in 2003 and H1N1 pandemic influenza in 2009, have illustrated how vaccines are not a panacea due to the time needed to develop, manufacture and deploy the vaccine [8] . The UK, for example, had contracts to provide up to 132 million doses in the case of an influenza pandemic being declared, and 90 million doses were ordered in May 2009; however immunization did not begin until October. In total, only around five million doses [9] (approx. 400 000 to healthcare workers, about 37% of the 11 million people deemed in risk groups and about 20% of the three million children between 6 months and 5 years [10] ) had been administered to people in priority groups by the end of February 2010, when the pandemic had effectively died out; therefore in February and April 2010 the orders were substantially reduced. Such statistics mean that we must carefully assess whether mass or targeted vaccination has a role in controlling future pandemics or large-scale outbreaks, and develop a range of robust models that can rapidly assess the benefits of vaccination and the best ways by which it can be targeted [8, [11] [12] [13] [14] [15] . Here, we extend a relatively simple model for vaccination (of previously unvaccinated individuals) at a constant rate in three main directions to consider how different forms of heterogeneity can impact optimal vaccination. In particular, we focus on the trade-off between vaccinating those at risk of severe complications (if they become infected) compared with vaccinating individuals who are more epidemiologically active; we consider who this epidemiologically active set are in terms of age groups within the population; and we consider whether vaccination should be deployed randomly or if there are benefits from geographical targeting. Throughout, our aim is to develop relatively simple models, where assumptions and parameters are transparent, and where it is feasible to rapidly perform large sweeps over parameter space. Therefore, while all results are formulated based on the UK experience of the 2009 H1N1 pandemic (assuming R 0 ¼ 1.4 and a doubling time of around one week [16] ), the model structure is sufficiently flexible that the qualitative results could pertain to a range of novel infections. ( We believe that the methodology and results outlined here are likely to hold for any rapidly transmitted infection, with a short infected period and life-long immunity, and where a vaccine can be rapidly developed and manufactured.) Obviously, once a new infectious disease is identified, specialist models are required that can accurately capture the known dynamics and can incorporate the appropriate economic and logistical facets [8] . Producing accurate results from all types of model (including the very complex and relatively simple) relies on the availability of high-quality surveillance and detailed case records. In the initial stages of an epidemic, before these data are available, policy-makers need to know what range of scenarios are consistent with the initial data. Additionally, for many parts of the world, detailed data are never available. For simple models with fewer basic parameters, a more comprehensive sweep of parameter space is feasible allowing the rapid assessment of various targeting vaccination strategies for ranges of parameters that are broadly consistent with early qualitative information. We therefore believe that the results developed here provide generic insights into the optimization of vaccine deployment during a novel outbreak of a directly transmitted pathogen with lifelong immunity. We first introduce the most basic model of vaccination in response to an infection that conforms to the simple SIR-type (susceptible -infectious -recovered) paradigm ( [17] [18] [19] ); this model will form the basic template for all the work that follows. We assume that the uncontrolled epidemic obeys the simple SIR model dynamics, such that susceptible individuals (of which there are S) can become infected and infectious by interaction with infected individuals (of which there are I), infected individuals recover at a constant rate and enter the recovered class (of which there are R) after which they are assumed immune for life. (We stress that all results presented are qualitatively invariant to the precise model formulation, in particular using a model with gamma-distributed exposed and infectious classes more reminiscent of the 2009 H1N1 pandemic [16] .) In all the models that follow, we assume frequency-dependent mixing (such that the number of epidemiologically relevant contacts is independent of population size) and ignore the demographics of birth and death [17] -a reasonable assumption given the rapid time-frame of an epidemic. To this simple model, we add vaccination at a constant rate, v; we assume that individuals are vaccinated independently of their disease status but individuals are only vaccinated once, vaccination begins at time T, and a proportion p of vaccinated individuals are successfully immunized. where N is the total population size. The precise way in which vaccination is implemented within the model ensures that a fixed number of individuals (v) are vaccinated per day, of which a fraction p are completely protected, and assumes that each person only receives one course of vaccine. If multiple doses of vaccine are required, or if protection only develops some time after vaccination, these can be included by delaying the time, T, at which immunization begins to take effect. Even this simplest of models confirms two simple rules-of-thumb regarding successful vaccination campaigns that seek to minimize the number of cases ( figure 1) . Firstly, that vaccination should begin as early as possible, so that susceptibles are depleted by vaccination before many cases arise; and secondly, that vaccination should be performed as rapidly as possible-both of which have been discussed before for a range of control measures [20, 21] . From figure 1 it is also clear that an early start to a vaccination campaign is far more beneficial than faster vaccination of the population. It should be stressed that when considering an ongoing epidemic, the critical vaccination threshold for the elimination of an endemic infection or prevention of epidemic invasion (¼1 2 1/R 0 ) [17] no longer plays such a clear role. Instead, the primary aim should be to immunize many people in as short a time as possible, subject to trade-offs from economic costs or adverse effects of vaccination (such as that observed for smallpox [12, 20] ). Here, and in all the figures that follow, we have considered a wide range of vaccination speeds ( y-axis in figure 1, line colours in figure 2, x-axes in figure 4 ); while some of these are extremely rapid and may be practically unachievable, the associated results are shown to provide a clearer picture of the most optimistic control scenario. Given this absence of clear epidemiological trade-offs in this simple model, we need to consider a range of more structured models in which we can consider the trade-offs involved with the prioritization of different groups for vaccination. When targeting a vaccination campaign (especially against the 2009 H1N1 influenza strain), there are often two competing priorities: minimization of transmission by immunizing those individuals that are most epidemiologically important; and minimization of the effects of the disease by immunizing those individuals that have the most severe health consequences when infected. (We note that for other infections these two groups may strongly overlap, in which case there is no conflict of priorities to resolve.) To tease apart these conflicting ideals, we extend the simple model above by having three groups [18, 19, 22, 23] : the dominant transmitter group (denoted by a subscript D), the group at highest risk of severe health complications if they become infected (denoted by a subscript H ) and the rest of the general population (denoted by a subscript G). These groups obey the basic equation (2.1) with two main modifications: firstly, the transmission dynamics are coupled through a 'who acquires infection from whom' matrix (b); and secondly, vaccination is prioritized so that either the dominant transmitter group (D) or the severe health risk group (H ) is vaccinated first, followed by the other group (either H or D), finally followed by the general population (G), (see electronic supplementary material). Here we have ignored the possibility that there is a group that are both dominant transmitters and at high risk of complications-obviously if such a group exists then it should be prioritized for vaccination before all others. Obviously, an epidemiological model with three interacting groups has a large number of associated parameters, making a comprehensive sweep of the entire parameter space impractical and difficult to visualize. Instead, we show results from a relatively restricted scenario, but comment that these results are representative of all plausible scenarios that have been considered. In particular, we constrain the number of individuals in the three groups to be N D ¼ N H ¼ 0.1 N and H G ¼ 0.8 N, and constrain the transmission rates between all groups, except within the epidemiologically important group, to be equal (b XY ¼ g, except when X ¼ Y ¼ D). We note that different forms and parameters within this transmission matrix can potentially lead to different optimizations of vaccine [22] . We now consider the optimal prioritization (either group D first or group H first), as four key parameters are varied: the transmission rate within the dominant transmitter group, b DD ; the relative adverse consequences of infection for the three groups, s H . s D ¼ s G ; the timing for the start of the vaccination programme, T; and the speed with which the population is vaccinated, v. Here the consequences of infection could capture a variety of measures, from risk of symptoms if infected, to concepts such as loss of QALYs (auality-adjusted life years), to risk of hospitalization, to risk of mortality associated with infection. The curves shown in figure 2 separate regions of parameter space where one form of prioritization is optimal, in terms of minimizing the total consequences of infection over the entire epidemic and across all three groups; regions above and to the left of the curves are where it is best to initially target vaccination towards the group with potentially severe health complications. Two clear conclusions can be drawn. More rapid vaccination (larger v) generally favours prioritizing vaccination towards the group with potential health consequences, as does a later onset of vaccination (larger T ). However, it should be noted that for either extremely rapid or extremely slow vaccination (or extremely late start of the vaccination programme), the differences between the two prioritization schemes will be minimal. We can . The sequential deployment of vaccination that for each dose targets the age class that offers the greatest reduction in the epidemic growth rate. As such, at each point in the vaccinate delivery schedule that the distribution is optimal in terms of minimizing the rate at which new cases are produced. The surrounding seven sub-graphs show this deployment of vaccine at different points during the epidemic (as indicated on the central epidemic curve); the shaded grey areas represent the vaccine levels where the reproductive ratio (R) is less than one, and therefore the epidemic is in decline. The population is broken into eight age groups, mirroring those used by the Health Protection Agency to report age-structured case reports. percentages, together with the fact that only 10 per cent of the general population are considered to have health problems, lead to estimates of s H : s G of atleast 9 : 1. Therefore, while there are a range of scenarios in which it would be optimal to target the dominant transmitters first, these tend to be in relatively extreme portions of parameter space, when the transmission rate b DD is very high and vaccination begins very early in the epidemic; for the vast majority of realistic scenarios, it is generally optimal to target vaccination towards those members of the population with underlying health problems first, before tackling the dominant transmitters and the rest of the general population. Analysis of the 2009 pandemic to date in Britain, and elsewhere, indicates that there are some strong agedependent signatures. Most notably, school children have suffered the greatest per capita burden of infection as recorded by surveillance systems, whereas pre-school children have experienced the greatest per capita level of severe infection (as measured by hospital admissions), while the over 65 age group were most likely to suffer severe problems if they became infected [9, 16] . These different age-dependent effects are due to several interacting and conflicting factors: the highly structured mixing between age groups, the age-related I I I I I I I I I I I I I I I I I I I I R I I I I I I I I I I I I I R R R I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I R R I I I I I I I I I I I I R R R I I I I I I I I I I I I R R R I I I I I I I I I I I I I R R I I I I I I I I I I I I I districts I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I R I I I I I I I I I I I I I R R R I I I I I I I I I I I I R R R I I I I I I I I I I I I I R R I I I I I I I I I I I I I districts I I I I I I I I I I I I S S S S I I I I I I I I I I I I S S S S I R I I I I I I I I I I I S S S I R I I I I I I I I I I I I S S I R I I I I I I I I I I I I I S I R I I I I I I I I I I I I I S I R I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I time to complete vaccination (days) I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I susceptibility to infection and the age-dependent risk of severe symptoms following infection. To combine these factors require a mathematical model based on the available age-structured information. Here we use data from the POLYMOD study [24] to parametrize agerelated mixing patterns, where P a,b captures the estimated contact rate between individuals of ages a and b. In the electronic supplementary material we also show that an age-dependent susceptibility vector (q a ) can be estimated such that the early dynamics, as predicted by the dominant eigenvector of the transmission matrix, agree with the observed early age-structured distribution of infection. However, for greater generality, we set q ¼ 1 in figure 3 , although even using the estimated age-dependent susceptibility from the 2009 pandemic, together with the impact of school holidays, does not dramatically change the predictions (see electronic supplementary material). We use the age-dependent transmission matrix b to determine an optimal priority for a rapid age-dependent vaccination programme [23, [25] [26] [27] (figure 3). The methodology is as follows: for each single dose of vaccine, we consider which age class should be immunized such that the resultant growth-rate (as predicted by the dominant eigenvalue) is minimized; repeating this process successively generates a vaccination strategy that should rapidly control the epidemic for any given level of vaccine coverage. (We note that [27] provide an alternative, more analytical method of minimizing the eigenvalue, which is equivalent to our approach once the total level of vaccine exceeds a threshold.) The vaccination strategies given in figure 3 therefore inform about the instantaneous epidemiological significance of each age group at a particular point during an epidemic. We do not claim that these strategies are truly optimal (in terms of minimizing the predicted total number of cases across all possible distributions of vaccine), nor that such strategies are entirely relevant if vaccination is slow relative to the epidemic timescales (owing to the changes in the priorities we observe as the epidemic progresses, as shown in the sub-graphs). However, these age-specific vaccination profiles do provide an intuitive means of sequentially and efficiently increasing the vaccination coverage at any given point in the epidemic and have been found to agree with the optimal distribution of a fixed quantity of vaccine that minimizes the dominant eigenvalue [27] . What is crucial to note in these plots is that they represent a theoretical ideal when vaccine supply is limited rather than an achievable goal. If vaccine is not in short supply then it is clearly always better (both in terms of reducing growth rate and total epidemic size) to vaccinate someone than not, even if this leads to substantial deviation away from the optimal age profile. Therefore, these plots inform about possible prioritization of the vaccine campaign. In the early stages of the epidemic, before there has been significant depletion of susceptibles, the predicted vaccination strategy initially targets the 5 -14 and 15-24 year old age groups; vaccination should then begin in the 25 -44 age group, with older ages (greater than 45) and the younger ages (under 5) not being targeted until vaccine coverage exceeds 50 per cent. What is somewhat counterintuitive is that the optimal deployment of vaccine could be partial in many age classes. For example, if 50 per cent of the population can be vaccinated, then the optimal strategy would be to attain highest levels of coverage in the 25-44 year old age group and less in the ages 5 -24, despite the fact that the 5 -24 year old age groups are favoured for early vaccination before the 25-44 year old age group. A second feature emerges as vaccination is begun later in the epidemic. Because the epidemic process has already depleted much of the susceptible population in the most epidemiologically active age classes (namely the 5 -14 year olds and 15-24 year olds), there is a decreased benefit from large-scale vaccination of these age groups. It should be stressed that both the distribution of optimal vaccination at a given time, and the way this changes as the epidemic progresses, are critically dependent on the number of individuals in each age group, the mixing matrix (b a,b ) and hence the age-dependent susceptibility (q a ). This means that precise details will depend on the detailed epidemiology of the infection under investigation, and are therefore likely to vary between different strains of influenza-for example, with models parametrized to match the 2009 pandemic in England (see electronic supplementary material), the 1 -4 year old age group plays a far more dominant role. The final important question to address when delivering a vaccination programme, is whether there are any advantages in spatially targeting its deployment [28, 29] . Obvious choices could be to target regions with a high immediate burden (targeting based on current proportion of infectious cases), or to target regions that are likely to have many cases in the future (targeting based on current proportion of susceptibles), or simply to vaccinate randomly with a fixed per captia rate. Determining the optimal spatial targeting of vaccination is difficult because the impact of vaccination is long-lasting, cumulative and nonlinear, meaning that a generic understanding cannot be generated by considering the impact of low vaccination levels, nor can the action of vaccination be considered piece-wise as it was above. The only viable option is to simulate the dynamics with a variety of strategies and ascertain which performs the best numerically, however the issue now becomes the number of possible ways in which the vaccine could be distributed. We use a deterministic metapopulation model of the progress of an influenza-like infection in the 408 districts of Great Britain, linked by the commuter movements recorded in the 2001 census, and consider a wide range of epidemiological scenarios; with different onset times for the start of vaccination (T ), different numbers of districts initially infected and different levels of transmission heterogeneity, as captured by different b d in each district ( see electronic supplementary material). (Because of the added complexity of dealing with an explicitly spatial model, other forms of heterogeneity such as age or risk structure have been ignored, although their inclusion could be formulated in a similar manner to that described above. We have focused on modelling at the district scale as targeting could be practically achieved as such a resolution; finer scale targeting such as at the ward level would lead to a enhanced vaccination scheme but is unlikely to be practical.) The vaccination level in district d is given as a function of the current epidemiological conditions in that district: As with equation (2.1), this formulation ensures the vaccination of a constant number (v) of ( previously unvaccinated) individuals each day, targeting unvaccinated individuals in each subpopulation at a per capita rate proportional to f. We have adopted the convention that variables and parameters with subscripts refer to specific districts, while those without subscripts refer to the entire population. The initial assessment was to consider targeting ( f ) that was proportional to one of the standard model variables figure 4 . (Alternative more complex methods of targeting are shown in the see electronic supplementary material.) Each square is colour-coded and labelled according to which targeting consistently generates the lowest number of cases (from 100 replicate simulations with random initial distribution of infection). When there is no regional heterogeneity in epidemiological parameters (b d ¼ b, top row), then the best strategy depends on the speed of vaccination (x-axis), the time when the vaccination campaign begins (column) and to a lesser extent the number of districts initially infected ( y-axis). Early vaccination (begun as soon as cases are detected) generally favours targeting of vaccination proportional to the current proportion of infected cases ( f d ¼ I d /N d ); if vaccination takes more than three weeks to complete, then a later start to the vaccination programme (begun when 10% of the population have been infected and therefore approx. 20% of the way through the uncontrolled epidemic) favours targeting proportional to the proportion of remaining susceptibles in the population ( f d ¼ S d /N d )-therefore, areas that have seen relatively less infection are favoured. Finally, if vaccination is implemented, even later (begun when 20% of the population have been infected and therefore approx. 40% of the way through the uncontrolled epidemic) targeting according to the proportion of susceptibles is again favoured when the epidemic is relatively dispersed and vaccination is relatively prompt. The top row of figure 4 corresponds to the unlikely scenario when each spatial district has identical parameters; in reality, there is likely to be significant heterogeneity in transmission between different areas owing to social and demographic factors. In the 2009 influenza pandemic in the UK, it was observed that the epidemic grew more rapidly in London and the West Midlands, possibly owing to such a mixture of socio-demographic factors. Unfortunately, the degree of such heterogeneity is difficult to estimate and is likely to vary between outbreaks; we therefore force the transmission rate within each district to be lognormally distributed about the mean, with the variance of the distribution controlled by the spatial heterogeneity (H ). (In particular, the transmission rate is given where j d is a random variable, normally distributed with mean zero and a variance of one.) When such heterogeneity is added, in the overwhelming majority of scenarios (and certainly when H ! 0.2), targeting in terms of the proportion of individuals currently infectious ( f d ¼ I d /N d ) is the best strategy. This advantage for targeting in terms of current infectious cases is due to two main reasons: firstly, when there is significant heterogeneity in growth rates it will naturally target vaccination at those regions likely to suffer larger epidemics; secondly, by controlling infection in the worst affected area, the spread of infection to other areas is reduced and therefore more time is gained to vaccinate the remaining areas. (As a further refinement to targeting in the electronic supplementary material, we consider f ¼ (I d /N d ) a and seek the exponent, a*, that minimizes the predicted number of cases.) The mass use of an effective vaccine clearly has the potential to provide major health benefits in terms of a reduction in the total number of infected cases, and therefore a reduction in the total number of adverse effects [1, 3, 7] . However, when used against an ongoing epidemic, the logistical constraints in terms of the speed with which vaccine can be manufactured and administered means that its deployment must be carefully targeted [8, [11] [12] [13] [14] [15] 23, 26, 27] . Many of these have focused on particular aspects of targeted vaccination (such as age-or risk-based) or have developed approximate [26] or analytical [27] methods for optimal targeting. Here, using standard differential equation models augmented to reflect particular heterogeneities, we have attempted to provide a relatively general framework that will be familiar with public-health epidemiologists and other non-specialist modellers. In general, our results agree with those of earlier studies although we have often attempted to consider a farwider parameter space, with the aim of spanning much of the parameter uncertainty that is likely to arise during the early stages of an epidemic. Three forms of targeting have been considered, in terms of risk groups, age structure and spatial location, to derive general insights into the benefits of targeted vaccination. Throughout we have focussed on the development of highly parsimonious models, where assumptions and parameterization are relatively transparent; however, even for these models the parameter space is often too large to be visualized, in such cases model parameters are based on observations from the 2009 influenza pandemic in Great Britain. Despite Targeted vaccination M. J. Keeling and P. J. White 667 this large parameter space, several generic conclusions can be drawn about the optimal deployment of vaccine: -as predicted by even the simplest models, vaccination campaigns are most effective when they are applied rapidly and early in an epidemic before many cases arise. In fact, figure 1 shows that an early start to the vaccination campaign (and therefore rapid development of the vaccine) is of more advantage than administering the vaccine quickly; -owing to the natural levels of heterogeneity displayed within the population, it is generally better to initially target vaccination towards those groups likely to have severe symptoms rather than those groups that are most responsible for transmission ( figure 2 ). To some extent, this is due to the ease with which those groups that are likely to suffer complications can be identified. The parameter regime in which it is better to initially vaccinate the severe risk groups, is extended when the vaccination campaign begins later, or when the population can be vaccinated more rapidly; -once the groups with severe health risks have been protected (or it has been deemed better to target other groups), attention naturally focuses on which elements of the population are the most epidemiologically active and the benefits associated from vaccinating these individuals. Two main results emerge ( figure 3) ; firstly, any one group should not be targeted to the exclusion of all others, the best strategy is generally a mixed strategy. Therefore, while it is generally predicted (conditional upon age-dependent susceptibility) that school-age children should be the initial focus of vaccination, the optimal strategy does not concentrate on achieving complete coverage of this group, instead it is best to target other groups simultaneously; -in addition, the best groups to target for vaccination vary as the epidemic progresses. Most notably, age groups that play the dominant epidemiological role are rapidly depleted and therefore their importance wanes as the epidemic progresses, and hence the advantages of targeted vaccination compared with random vaccination also decline; and -although it is difficult (if not completely impractical) to calculate the true optimal spatially targeted vaccination policy, instigating at least some measure of targeting has significant advantages. Depending on the level of spatial heterogeneity in transmission within the population (which could reflect underlying demographic heterogeneity), targeting of vaccination towards regions that are currently experiencing high levels of infection generally reduces the total number of cases. The intuitive reason for this targeting is that by concentrating on centres of infections (and reducing the immediate growth rate), it buys extra time to vaccinate other regions. Achieving such a targeting in practice would require health services to be able to rapidly shift resources around the country, and only applies when there is a national limit to the deployment of vaccine. However, these results indicate two important points: firstly, that even when vaccination schemes are administered locally, there are likely to be strong advantages in spatially targeting any additional national resources; and secondly that the likely human reaction for there to be a greater demand for vaccine in regions with a higher proportion of cases would assist in control. Because of the relatively simplistic nature of the models developed in this paper, several caveats should be made regarding the results. The first is that vaccination campaigns are unlikely to achieve a constant level of uptake over time; a more realistic assumption is that vaccination initially begins slowly due to logistical issues, builds to a plateau, but may finally decline if the epidemic begins to wane during the period of the vaccination programme and there is less incentive for individuals to be vaccinated. Associated with this is the fact that not all individuals are prepared to be vaccinated, and in particular parents in the UK are often anxious about vaccinating children; therefore, the optimal targeting of vaccination is unlikely to be possible and it may simply be better to vaccinate any individuals who wish to be vaccinated. The second issue, which applies to figures 2 and 3, is that it may not always be possible to identify or target relevant risk groups. For example, there may be considerable overlap between the age groups used in figure 3 , and the groups at risk of severe symptoms defined in figure 2 ; this was undoubtedly the case in 2009 (as well as previous pandemics), where age was often a key contributing factor to both the risk of infection and the risk of complications. Throughout, we have assumed a perfect vaccine, which offers 100 per cent protection to all those vaccinated. In practice, this is never realized, all vaccines fail to some degree; however, there are two ways in which this failure can be modelled. The first is an all or nothing approach in which a fraction of all vaccinations fails to generate any protection, while the remaining offers full protection; in this case, the results of our model natural extrapolate based on considering the numbers protected. The second approachis to consider 'leaky' vaccines that offer partial protection reducing either susceptibility or onward transmission; given the relatively low reproductive ratios considered within this paper, we believe that our results are likely to generalize to this case. A further issue related to figure 2 is the ethics of vaccinating the epidemiologically important groups in order to protect the group with severe health risks. While for influenza the vaccine has little associated risk and therefore it may be argued that the health benefits to the epidemiologically important (but healthy) groups outweigh the dangers, the same may not be true for other pathogens and the associated vaccine. Additionally, we have generally used deterministic models and hence assumed that vaccine would be used to mitigate the impact of infection rather than to prevent an epidemic occurring. This is particularly pertinent to the metapopulation model (figure 4) when we neglect the possibility of using vaccine to eradicate infection in the early stages and prevent its spread to the remainder of the country as exemplified in Ferguson et al. [11] . We feel this is a reasonable assumption given that most novel infections will be seeded continually by imports from abroad, and vaccine is unlikely to be available at the start of the epidemic. Finally, throughout we have assumed that there is a strong public demand for vaccine. Clearly, demand will vary both with disease severity and public perception of the infection; for example, the relatively mild nature of the 2009 pandemic meant that demand and uptake of the vaccine was low. However, if the infection has severe health implications, and therefore there is a clear need to optimize control measures, then demand for vaccination is likely to be substantial. The models in this paper, and therefore the results generated, are not designed to replace very detailed simulations parametrized to match epidemiological data from a given outbreak; however, they do provide a high degree of generality that is difficult to obtain with more case-specific simulations and hence provide a rapid assessment of conflicting methods of targeting vaccination. Often public-health action has to be taken in the absence of critical information: levels of prior immunity in the population affect the transmission patterns, but it takes time to develop the appropriate serological test; parameters such as casefatality ratios or the probability of hospitalization are difficult to estimate in real time as an epidemic progresses; and although clinical trials can measure the theoretical vaccine efficacy, vaccine effectiveness in practice can only be determined once the vaccination programme has begun. Given this range of uncertainties, models can be best used to assist policy-makers by examining a range of scenarios ahead of time; so decisions can be taken based on the range of scenarios that are consistent with the available data at the time that the decision has to be taken. Simple models often allow us to partition parameter space into clearly defined regions where a particular strategy is optimal; and while precise parameters may be difficult to estimate early in an epidemic, there may be sufficient evidence to suggest that the infection parameters lie within one of these prescribed regions. As such, the simple models developed here provide useful policy guidance before or during the early stages of an epidemic before there are sufficient data to parametrize more detailed simulations.

Networks and the Epidemiology of Infectious Disease

The science of networks has revolutionised research into the dynamics of interacting elements. It could be argued that epidemiology in particular has embraced the potential of network theory more than any other discipline. Here we review the growing body of research concerning the spread of infectious diseases on networks, focusing on the interplay between network theory and epidemiology. The review is split into four main sections, which examine: the types of network relevant to epidemiology; the multitude of ways these networks can be characterised; the statistical methods that can be applied to infer the epidemiological parameters on a realised network; and finally simulation and analytical methods to determine epidemic dynamics on a given network. Given the breadth of areas covered and the ever-expanding number of publications, a comprehensive review of all work is impossible. Instead, we provide a personalised overview into the areas of network epidemiology that have seen the greatest progress in recent years or have the greatest potential to provide novel insights. As such, considerable importance is placed on analytical approaches and statistical methods which are both rapidly expanding fields. Throughout this review we restrict our attention to epidemiological issues.

The science of networks has revolutionised research into the dynamics of interacting elements. The associated techniques have had a huge impact in a range of fields, from computer science to neurology, from social science to statistical physics. However, it could be argued that epidemiology has embraced the potential of network theory more than any other discipline. There is an extremely close relationship between epidemiology and network theory that dates back to the mid-1980s [1, 2] . This is because the connections between individuals (or groups of individuals) that allow an infectious disease to propagate naturally define a network, while the network that is generated provides insights into the epidemiological dynamics. In particular, an understanding of the structure of the transmission network allows us to improve predictions of the likely distribution of infection and the early growth of infection (following invasion), as well as allowing the simulation of the full dynamics. However the interplay between networks and epidemiology goes further; because the network defines potential transmission routes, knowledge of its structure can be used as part of disease control. For example, contact tracing aims to identify likely transmission network connections from known infected cases and hence treat or contain their contacts thereby reducing the spread of infection. Contact tracing is a highly effective public health measure as it uses the underlying transmission dynamics to target control efforts and does not rely on a detailed understanding of the etiology of the infection. It is clear, therefore, that the study of networks and how they relate to the propagation of infectious diseases is a vital tool to understanding disease spread and, therefore, informing disease control. Here, we review the growing body of research concerning the spread of infectious diseases on networks, focusing on the interplay between network theory and epidemiology. The paper is split into four main sections which examine the types of network relevant to epidemiology, the multitude of ways these networks can be characterised, the statistical methods that can be applied to either infer the likely network structure or the epidemiological parameters on a realised network, and finally simulation and analytical methods to determine epidemic dynamics on a given network. Given the breadth of areas covered and the ever-expanding number of publications (over seven thousand papers have been published concerning infectious diseases and networks) a comprehensive review of all work is impossible. Instead, we provide a personalised overview into the areas of network epidemiology that have seen the greatest progress in recent years or have the greatest potential to provide novel insights. As such considerable importance is placed on analytical approaches and statistical methods which are both rapidly expanding fields. We note that a range of other networkbased processes (such as the spread of ideas or panic) can be modelled in a similar manner to the spread of infection; however, in these contexts, the transmission process is far less clear; therefore, throughout this paper, we restrict our attention to epidemiological issues. There are a wide number of network structures and types that have been utilised when considering the spread of infectious diseases. Here, we consider the most common forms and explain their uses and limitations. Later, we review the implications of these structures for the spread and control of infectious diseases. We start our examination of network forms by considering the ideal network that would allow us to completely describe the spread of any infectious pathogen. Such a network would be derived from an omniscient knowledge of individual behaviour. We define G i, j (t) to be a time-varying, real, and high-dimensional variable that informs about the strength of all potential transmission routes from individual i to individual j at time t. Any particular infectious disease can then be represented as a function ( f pathogen ) translating this high-dimensional variable into an instantaneous probabilistic transmission rate (a single real variable). In this ideal, G subsumes all possible transmission networks, from sexual relations to close physical contact, face-to-face conversations, or brief encounters, and quantifies the time-varying strength of this contact. The disease function then picks out (and combines) those elements of G that are relevant for transmission of this pathogen, delivering a new (single-valued) timevarying infection-specific matrix (T i, j (t) = f pathogen (G i, j (t))). This infection-specific matrix then allows us to define the stochastic dynamics of the infection process for a given pathogen. (For even greater generality, we may want to let the pathogen-specific function f also depend on the time since an individual was infected, such that time-varying infectivity or even time-varying transmission routes can be accommodated.) Obviously, the reality of transmission networks is far from this ideal. Information on the potential transmission routes within a population tends to be limited in a number of aspects. Firstly, it is rare to have information on the entire population; most networks rely on obtaining personal information on participants, and therefore participation is often limited. Secondly, information is generally only recorded on a single transmission route (e.g., face-to-face conversation or sexual partnership) and often this is merely recorded as the presence or absence of a contact rather than attempting to quantify the strength or frequency of the interaction. Finally, data on contact networks are rarely dynamic; what is generally recorded is whether a contact was present during a particular period with little consideration given to how this pattern may change over time. In the light of these departures from the ideal, it is important to consider the specifics of different networks that have been recorded or generated and understand their structure, uses, and limitations. One of the few examples of where many of the potential transmission routes within a population have been documented comes from the spread of sexually transmitted infections (STIs). In contrast with airborne infections, STIs have very obvious transmission routes-sex acts (or sharing needles during intravenous drug use)-and as such these potential transmission routes should be easily remembered (Figure 1(a) ). Generally the methodology replicates that adopted during contact tracing, getting an individual to name all their sexual partners over a given period, these partners are then traced and asked for their partners, and the process is repeated-this is known as snowball sampling [4] (Figure 1(b) ). A related methodology is respondent-driven sampling, where individuals are paid both for their participation and the participation of their contacts while protecting each individual's anonymity [5] . This approach, while suitable for hidden and hard to reach populations, has a number of limitations, both practical and theoretical: recruiting people into the study, getting them to disclose such highly personal information, imperfect recall from participants, the inability to find all partners, and the clustering of contacts. In addition, there is the theoretical issue that this algorithm will only find a single connected component within the population, and it is quite likely that multiple disjoint networks exist [6] . Despite these problems, and motivated by the desire to better understand the spread of HIV and other STIs, several pioneering studies were performed. Probably the earliest is discussed by Klovdahl [1] and utilises data collected by the Center for Disease Control from 19 patients in California suffering from AIDS, leading to a network of 40 individuals. Other larger-scale studies have been performed in Winnipeg, Manitoba, Canada [7] and Colorado Springs, Colorado, USA [8] . In both of these studies, participants were tested for STIs, and the distribution of infection compared to the underlying network structure. Work done on both of these networks has generally focused on network properties and the degree to which these can explain the observed cases; no attempt was made to use these networks predictively in simulations. In addition, in the Colorado Springs study, Interdisciplinary Perspectives on Infectious Diseases [3] ; squares refer to primary contacts. Given that the identity of contacts is known, they can be interlinked. (b) Caricature of a snowball sampling algorithm, squares are primary contacts, diamonds are secondary, and circles are tertiary contacts. Given that the identity of contacts is known they can be linked. (c) Example of a configuration model network. Each individual has a prescribed degree distribution, which gives rise to "halflinks" that are connected at random. (d) A household configuration network, consisting of completely interconnected households (cliques) with each individual also having one random link to another household. (e) Map showing Great Britain, together with the movements of cattle from six farms (each represented in a separate colour). Notice the heterogeneity between farms and the generally localised nature of movements. (f) Example of a small-world model based on a 2D lattice with nearest neighbor connections. The small-world property is given by the presence of rare random links that can connect distant parts of the network. tracing was generally only performed for a single iteration although many initial participants in high-risk groups were enrolled, while in the Manitoba study, tracing was performed as part of the routine information gathered by public health nurses. Therefore, while both provide a vast amount of information on sexual contacts, it is not clear if the results are truly a comprehensive picture of the network and sampling biases may corrupt the resulting network [9] . In addition, compared to the ideal network, these sexual contact networks lack any form of temporal information; instead, they provide an integration of the network over a fixed time period and generally lack information on the potential strength of a contact between individuals. Despite these difficulties, they continue to provide an invaluable source of information on human sexual networks and the potential transmission routes of STIs. In particular, they point to the extreme levels of heterogeneity in the number of sexual contacts over a given period-and the variance in the number of contacts has been shown to play a significant role in early transmission dynamics [10] . One of the few early examples of the simulation of disease transmission on an observed network comes from a study of a small network of 22 injection drug users and their sexual partners [3] (Figure 1(a) ). In this work, the risk of transmission between two individuals in the network was imputed based on the frequency and types of risk behaviour connecting those two individuals. HIV transmission was modelled using a monthly time step and single index case, and simulations were run for varying lengths of (simulated) time. This enabled a node's position in the network (as characterised by a variety of measures) to be compared with how frequently it was infected during simulations, and how many other nodes it was typically responsible for infecting. A different approach to gathering social network and behavioural data was initiated by the Human Dynamics group at MIT and illustrates how modern technology can assist in the process of determining transmission networks. One of the first approaches was to take advantage of the fact that most people carry mobile phones [11] . In 2004, 100 Nokia 6600 smart-phones preinstalled with software were given to MIT students to use over the course of the 2004-2005 academic year. Amongst other things, data were collected using Bluetooth to sense other mobile phones in the vicinity. These data gave a highly detailed account of individuals behaviour and contact patterns. However, a limitation of this work was that Bluetooth has a range of up to 25 meters, and as such networks inferred from these data may not be epidemiological meaningful. A more recent study into the encounters between wild Tasmanian devils in the Narawntapu National Park in northern Tasmania utilised a similar technological approach [12] . In this work, 46 Tasmanian devils were fitted with proximity loggers that could detect and record the presence of other loggers within a 30 cm range. As such, these loggers were able to provide detailed temporal information on the potential interaction between these 46 animals. This study was initiated to understand the spread of Tasmanian devil facial tumour disease, which causes usually fatal tumours that can be transmitted between devils if they fight and bite each other. Although only 27 loggers with complete data were recovered, and although the methodology only recorded interaction between the 46 devils in the study, the results were highly informative (generating a network that was far from random, heterogeneous, and of detailed temporal resolution). Analyses based on the structure of this network suggested that targeted measures, that focus on the most highly connected ages or sex, were unlikely to curtail the spread of this infection. Of perhaps greater relevance is the potential this method illustrates for determining the contact networks of other species (including humans)-the only limitation being the deployment of a suitable number of proximity loggers. Given the huge logistical difficulties of capturing the full network of interactions between individuals within a population, a variety of methods have been developed to generate synthetic networks from known attributes. Generally, such methods fall into two classes: those that utilise egocentric information and those that attempt to simulate the behaviour of individuals. Egocentric data generally consists of information on a number of individuals (the egos) and their contacts (the alters). As such the information gathered is very similar to that collected in the sexual contact network studies in Manitoba and Colorado Springs, but with only the initial step of the snowball sampling was performed; the difference is that for the majority of egocentric data the identity of partners (alters) is unknown and therefore connections between egos cannot be inferred (Figure 1(c) ). The data, therefore exists as multiple independent "stars" linking the egos to the alters, which in itself provides valuable information on heterogeneities within the network. Two major studies have attempted to gather such egocentric information: the NATSAL studies of sexual contacts in the UK [13] [14] [15] [16] , and the POLYMOD study of social interactions within 8 European countries [17] . The key to generating a network from such data is to probabilistically assign each alter a set of contacts drawn from the information available from egos; in essence, using the ego data to perform the next step in the snowball sampling algorithm. The simplest way to do this is to generate multiple copies of all the egos and to consider the contacts from each ego to be "half-links"; the half-links within the network can then be connected at random generating a configuration network [18] [19] [20] ; if more information is available on the status (age, gender, etc.) of the egos and alters then this can also be included and will reduce the set of half-links that can be joined together. However, in the vast majority of modelling studies, the egocentric data have simply been used to construct WAIFW (who-acquiresinfection-from-whom) matrices [15, 17, 21] that inform about the relative levels of transmission between different groups (e.g., based on sexual activity or age) but neglect the implicit network properties. This matrix-based approach is often reliable: for STIs it is the extreme heterogeneity in the number of contacts (which are close to being powerlaw or scale-free distributed; see Section 3.2) that drives the infection dynamics [22] although larger-scale structure does play a role [23] ; for social interactions, it is the assortativity between (age-) groups that controls the behaviour, with the number of contacts being distributed as a negative binomial [17] . The POLYMOD matrices have therefore been extensively used in the study of the H1N1 pandemic in 2009, providing important information about the cost-effective vaccination of different age-classes [21, 24] . The general configuration model approach of randomly linking together "half-links" from each ego [18, 19] has been adopted and modified to consider the spread of STIs. In particular, simulations have been used to consider the importance of concurrency in sexual networks [25, 26] , where concurrency is defined as being in two active sexual partnerships at the same time. A dynamic sexual network was simulated, with partnerships being broken and reformed such that the network density remained constant over time. The likelihood of two nodes forming a partnership depended on their degree, but this relationship could be tuned to make concurrency more or less common and to make the mixing assortative or disassortative based on the degrees of the two nodes. Transmission of an STI (such as gonorrhoea and chlamydia [25] or HIV [26] ) was then simulated upon this dynamic network, showing that increasing concurrency substantially increased the growth rate during the early phase of an epidemic (and, therefore, its size after a given period of time). This greater growth rate was related to the increase in giant component size (see Section 3.1) that was caused by increased concurrency. Interdisciplinary Perspectives on Infectious Diseases 5 A slightly more general approach to the generation of model sexual networks was employed by Ghani et al. [27] . In their network model, individuals had a preferred number of concurrent partners and duration of partnerships, and their level of assortativity was tunable. A gonorrhoea-like infection was simulated on the resulting dynamic network. Regression models were used to consider the association between network structures (either snapshots of the state of the network at the end of simulation or accumulated over the last 90 days of simulation) and prevalence of infection. These simulations showed that increasing levels of concurrent partnerships made invasion of the network more likely and also that the mixing patterns of the most sexually active nodes were most important in determining the final prevalence of infection within the population [27] . The same model was later used to consider the importance of different structural measures and sampling strategies, showing that it was important to endeavour to identify infected individuals with a high number of sexual partners in order to correctly define the high-risk group for interventions [23] . The alternative approach of simulating the behaviour of individuals is obviously highly complex and fraught with a great deal of uncertainty. Despite these problems, three groups have attempted just such an approach: Longini's group at Emory [28] [29] [30] [31] , Ferguson's group at Imperial [32, 33] , and Eubank's group at Los Alamos/Virginia Tech [34, 35] . The models of both Longini and Ferguson are primarily agent-based models, where individuals are assigned a home and work location within which they have frequent infection-relevant contacts together with more random transmission in their local neighbourhood. The Longini models separate the entire population into subunits of 2000 individuals (for the USA) or 13000 individuals (for South-East Asia) who constitute the local population where random transmission can operate; in contrast, the Ferguson models assign each individual a spatial location and random transmission occurs via a spatial kernel. In principle, both of these models could be used to generate an explicit network model of possible contacts. The Eubank model is also agentbased aiming to capture the movements of 1.5 million people in Portland, Oregon, USA; but these movements are then used to define a network based on whether two individuals occur in the same place (there are 180 thousand places represented in the model) at the same time. It is this network that is then used to simulate the spread of infection. While in principle this Eubank model could be used to define a temporally varying and real-valued network (where the strength of connection would be related to the type of mixing in a location and the number of people in the location); in the epidemiological publications [35] , the network is considered as a static contact network in which extreme heterogeneity in numbers of contacts is again predicted, and the network has "small world" like properties (see below). A similar approach of generating artificial networks of individuals for stochastic simulations of respiratory disease has been recently applied to influenza at the scale of the United States, and the software made generally available [36] . This software took a more realistic dynamic network approach and incorporated flight data within the United States, but was sufficiently resource-intensive to require specialist computing facilities (a single simulation taking around 192 hours of CPU time). All three models have been used to consider optimal control strategies, determining the best deployment of resources in terms of limiting transmission associated with different routes. The predicted success of various control strategies, therefore, critically depends on the strength of contacts within home, at work, within social groups, and that occuring at random. Whilst smallpox has been eradicated, concern remains about the possibility of a deliberate release of the disease. The stochastic simulation models of the Longini group have predominantly focused on methods of controlling this infection [28, 31] . Their early work utilised networks of two thousand people with realistic age, household size, and school attendance distributions, with the likelihood of each individual becoming infected being derived from the number and type of contacts with infectious individuals [28] . This paper focused on the use of vaccination to contain a smallscale outbreak of smallpox and concluded that early massvaccination of the entire population was more effective than targeted vaccination if there was little or no immunity in the population. Later models [31] combined these subnetworks of two thousand people into a larger network of fifty thousand people (with one hospital), and the adult population were able to contact each other through workplaces and high schools. Here, the focus was on surveillance and containment which were generally concluded to be sufficient to control an outbreak. The epidemiological work of the Eubank group has also focused on a release of smallpox although these simulations showed that encouraging people to stay at home as soon as they began to feel unwell was more important than choice of vaccination protocol [35] ; this may in part be attributed to the scale-free structure of the network and hence the superspreading nature of some individuals. The Ferguson models have primarily been used to consider the spread and control of pandemic influenza, examining its potential spread from an initial source in South-East Asia [32] and its spread in mainland USA and Great Britain [33] . The models of South-East Asia were primarily based on Thailand, and included demographic information and satellite-based spatial measures of population density. It focused on containment by the targeted use of antiviral drugs and suggested that as long as the reproductive ratio (R 0 ) of a novel strain was below 1.8, it could be contained by the rapid use of targeted antivirals and social distancing. However, such a strategy could require a stockpile of around 3 million antiviral doses. The models based on the USA and Great Britain, considered a wider range of control measures, including school closures, household prophylaxis using antiviral drugs, and vaccination, and predicted the likely impact of different policies. Networks. An alternative source of network information comes from the recorded movements of individuals. Such data frequently describe a relatively large network as information on movements is often collected by national or international bodies. The network of movements, therefore, has nodes representing locations (rather than individuals) and edges weighted to capture the number of movements from one location to another-as such the network is rarely symmetric. Four main forms of movement network have played important roles in understanding the spread of infectious diseases: the airline transportation network [37, 38] , the movement of individuals to and from work [39, 40] , the movement of dollar bills (from which the movement of people can be inferred) [41] , and the movement of livestock (especially cattle) [30, 42] . While the structure of these networks has been analysed in some detail, to develop an epidemiological model requires a fundamental assumption about how the epidemic progresses within each locations. All the examples considered in this section make the simplifying assumption that the epidemic dynamics within each location are defined by random (mean-field) interactions, with the network only informing about the flow of individuals or just simply the flow of infection between populations-such a formulation is known as a metapopulation model [43] . Probably the earliest work using detailed movement data to drive simulations comes from the spread of 1918 pandemic influenza in the Canadian Subarctic, based on records kept by the Hudson's Bay Company [44] . A conventional SIR metapopulation model was combined with a network model (the nodes being three fur trading posts in the region: God's Lake, Norway House, and Oxford House), where some individuals remained in their home locations whilst others moved between locations, based on records of arrivals and departures recorded in the post journals. Whilst this model described only a small population, it was able to be parameterised in considerable detail due to the quality of demographic and historical data available and showed that the movement patterns observed interacted with the starting location of a simulated epidemic to change the relative timings of the epidemics in the three communities, but not the overall impact of the disease. The movement of passenger aircraft as collated by the International Air Transport Association (IATA) provides very useful information about the long-distance movement of individuals and hence how rapidly infection is likely to travel around the globe [37, 45, 46] . Unlike many other network models which are stochastic individual-level simulations, the work of Hufnagel et al. [37] and Colizza et al. [45] was based on stochastic Langevin equations (effectively differential equations with noise included). The early work by Hufnagel et al. [37] focused on the spread of SARS and showed a remarkable degree of similarity between predictions and the global spread of this disease. This work also showed that extreme sensitivity to initial conditions arises from the structure of the network, with outbreaks starting in different locations generating very different spatial distributions of infection. The work of Colizza was more focused towards the spread of H5N1 pandemic influenza arising in South-East Asia and its potential containment using antiviral drugs. However, it was H1N1 influenza from Mexico that initiated the 2009 pandemic, but again, the IATA flight data provided a useful prediction of the early spread [47, 48] . While such global movement networks are obviously highly important in understanding the early spread of pathogens, they unfortunately neglect more localised movements [49] and individual-level transmission networks. However, recent work has aimed to overcome this first issue by including other forms of local movement between populations [40, 50] . This work has again focused on the spread of influenza, mixing long-distance air travel with shorter range commuter movements and with the model predictions by Viboud et al. [40] showing good agreement with the observed patterns of seasonal influenza. An alternative form of movement network has been inferred from the "Where's George" study of the circulation of dollar bills in the USA [38] ; this provided far more information about short-range movements, but again did not really inform about the interaction of individuals. A wide variety (and in practice the vast majority) of movements are not made by aircraft but are regular commuter movements to and from work. The network of such movements has also been studied in some detail for both the UK and USA [39, 40, 51] . The approaches adopted parallel the work done using the network of passenger aircraft, but operate at a much smaller scale, and again, influenza and smallpox have been the considered pathogens. As with the aircraft network certain locations act as major hubs attracting lots of commuters every day; however, unlike the aircraft network, there is the tendency for the network to have a strong daily signature with commuters moving to work during the day but travelling home again in the evening [52] . As such the commuter network can be thought of as heterogeneous, locally clustered, temporal, and with each contact having different strengths (according to the number of commuters making each journey); however, to provide a complete description of population movement, and hence disease transmission requires other causes of movement to be included [51] and requires strong assumptions to be made about individual-level interactions. The key question that can be readily addressed from these commuter-movement models is whether a localised outbreak can be contained within a region or whether it is likely to spread to other nodes on the network [39] . Undoubtedly, one of the largest and most comprehensive data sets of movements between locations comes from the livestock tracing schemes run in Great Britain and being adopted in other European countries. The Cattle Tracing Scheme in particular is spectacularly detailed, containing information of the movements of all cattle between farms in Great Britain; as such, this scheme generates daily networks of contacts between over 30,000 working farms in Great Britain [42, [53] [54] [55] [56] (Figure 1 (f)). Similar data also exist for the movement of batches of sheep and pigs [57] although here the identity of individual animals making each movement is not recorded. This data source has several key advantages over other movement networks: it is dynamic, in that movements are recorded daily; the movement of livestock is one of the major mechanisms by which many infections are transferred between farms, and the metapopulation assumption that cattle mix homogeneously within a farm is highly plausible. In principle, the information in the Cattle Tracing Scheme can be used to form an even more comprehensive network, treating each cow as a node and creating an edge if two cows occur within the same farm on the same day-this would generate an individual-level network for each day which can then be used to simulate the spread of infection [52] . The early spread of foot and mouth disease (FMD) in 2001 was primarily due to livestock movements, particularly of sheep [58] . Motivated by this epidemic, Kiss et al. [57] conducted short simulated outbreaks of FMD on both the sheep movement network based on 4 weeks' movements starting on 8 September 2004 and simulated synthetic networks with the same degree distribution. Due to the short time-scales considered (the aim being to model spread of FMD before it had been detected), nodes were susceptible, exposed or infected but never recovered, and network connections remained static. Simulated epidemics were smaller on the sheep movement network than the random networks, most likely due to disassortative mixing in the sheep movement network. Similarly, Natale et al. [59] employed a static network simulation of Italian cattle farms. Here, farms were not merely represented as nodes, but a deterministic SI system of ODEs was used to model infection on each node essentially generating a metapopulation model. The only stochastic part of the model was the number of infectious individuals moved between connected farms in each time step. This simulation model highlighted the impact of the centrality of seed nodes (measured in several different ways) upon the subsequent epidemics' course. The use of static networks to model the very dynamic movement of livestock is questionable. Expanding on earlier work, Green et al. [53] simulated the early spread of FMD through movement of cattle, sheep, and pigs. Here, the livestock network was treated dynamically, with infection only able to propagate along edges on the day when that edge occurred; additional to this network spread, local transmission could also occur. These simulations enabled regional patterns of risk to a new FMD incursion to be assessed, as well as identifying markets as suitable targets for enhanced surveillance. Vernon and Keeling [55] considered the relationship between epidemics predicted from dynamic cattle networks and their static counterparts in more detail. They compared different network representations of cattle movement in the UK in 2004, simulating epidemics across a range of infectivity and infectious period parameters on the different network representations. They concluded that network representations other than the fully dynamic one (where the movement network changes every day) fail to reproduce the dynamics of simulated epidemics on the fully dynamic network. Contact tracing and hence the networks generated by this method can take two distinct forms. The first is when contact-tracing is used to initiate proactive control. This is often the case for STIs, where identified cases are asked about their recent sexual partners, and these individuals are traced and tested; if found to be infected, then contact tracing is repeated for these secondary cases. Such a process is related to the snowball sampling that was discussed earlier, with the notable exception that tracing is only performed from known cases. Similar contact-tracing may operate for the early stages of an airborne epidemic (as was seen for the 2009 H1N1 pandemic), but here, the tracing is not generally iterative as contacts are generally traced and treated so rapidly that they are unlikely to have generated secondary cases. An alternative form of contacttracing is when a transmission pathway is sought between all identified cases [1, 60, 61] . This form of contact tracing is likely to become of ever-increasing importance in the future when improved molecular techniques and statistical inference allow infection trees to be determined from genetic differences between samples of the infecting pathogen [62] . These forms of network have two main advantages but one major disadvantage. The network is often accompanied by test results for the individuals within the network, as such we not only have information on the contact process but also on the resultant transmission of infection. In addition, when contact tracing is only performed to define an infection tree, there is the added advantage that the infection process itself defines the network of contacts, and hence there is no need for human interpretation of which forms of contact may be relevant. Unfortunately, the reliance on the infection process to drive the tracing means that the network only reflects one realisation of the epidemic process and, therefore, may ignore contacts that are of potential importance and would be needed if the epidemic was to be simulated; therefore, while they can inform about past outbreaks, they have little predictive power. Obtaining large-scale and reliable information on who contacts whom is obviously very difficult; therefore, there is a temptation to rely on alternative data sets, where network information can be extracted far more easily, and where the data is already collected. As such the movement networks and contact tracing networks discussed above are examples of such surrogate networks although their connection to the physical processes of infection transmission are far more clear. Other examples of networks abound [22, [63] [64] [65] ; while these are not directly relevant for the spread of infection, they do provide insights into how networks form and grow-structures that are commonly seen in surrogate networks are likely to arise in the types of network associated with disease transmission. One source of network information that would be fantastically rich and also highly informative (if not immediately relevant) is the network of friendships and contacts on social networking sites (such as Facebook); some sites have made data on their social networks available, and these data have been used to examine a range of sociological questions about online interactions [66] . Given the huge complexity involved in obtaining large-scale and reliable data on realtransmission networks many researchers have instead relied on theoretically constructed networks. These networks are usually highly simplified but aim to capture some of the known (or postulated) features of real-transmission networks-often the simplifications are so extreme that some analytical traction can be gained. Here, we briefly outline some of the commonly used theoretical networks and identify which features they capture; some of the results of how infection spreads on such networks are discussed more fully in Section 4.2. One of the simplest forms of network is to allow each individual to have a set of contacts that it wishes to make (in more formal language each node has a set of half-links), these contacts are then made at random with other individuals based on the number of contacts that they wish to make (half-links are randomly connected) [19] . This obviously creates a network of contacts ( Figure 1 (c)). The advantage of these configuration networks is that because they are formed from many randomly connected individuals, there are no short loops within the network and a range of theoretical results can be proved ranging from conditions for invasion [18, 67, 68] to descriptions of the temporal dynamics [69] . Unfortunately, the elements that make these networks amenable to theoretical analysis-the lack of assortativity, short loops or clusteringare precisely factors that are thought to be important features of real networks. An alternative formulation that offers a compromise between tractability and realism occurs when individuals that exist in fully interconnected cliques have randomly assigned links within the entire population [69, 70] ( Figure 1(d) ). As such, these networks mimic the strong interactions within families and the weaker contacts between them. While such models offer a significant improvement over configuration networks and capture the known importance of the household in transmission, they make no allowance for clustering between households due to spatial proximity. Hierarchical metapopulation models [71] allow for this form of additional structure, where households (or other groupings) are themselves grouped in an ascending hierarchy of clustering. Worlds. Both lattice networks and small world networks begin with the same formulation: individuals are regularly spaced on a grid (usually in just one or two dimensions), and each individual is connected to their k nearest neighbours-these connections define a lattice. The advantage of such networks is that they retain many elements of the initial spatial arrangement of points, and hence contain both many short loops as well as the property that infection tends to spread locally. There is a clear link between such lattice-based networks and the field of probabilistic cellular automata [72, 73] . The fundamental difficulty with such lattice models is that the presence of short loops and localised spread means that is it difficult (if not impossible) to prove exact results, and hence large-scale multiple simulations are required. Small world networks improve upon the rigid structure of the lattice by allowing a low number of random contacts across the entire space (Figure 1(e) ). Such long range contacts allow infection to spread rapidly though the population and vastly reduce the shortest path length between individuals [74] -this is popularly known as six degrees of separation from the concept that any two individuals on the planet are linked through at most six friends or contacts [75] . Therefore, small world networks offer a step towards reality, capturing the local nature of transmission and the potential for long-range contacts [76, 77] ; however, they suffer from neglecting heterogeneity in the number of contacts and the tight clustering of contacts within households or social settings. Networks. Spatial networks, as the name suggests, are generated using the spatial location of all individuals in the population, as such lattices and small worlds are a particular form of spatial network. The general methodology initially positions each individual i at a specific location x i , usually; these locations are chosen at random, but clustered spatial distributions have also been used [78] . Two individuals (say i and j) are then probabilistically connected based upon the distance between them; the probability is given by a connection kernel which usually decays with distance such that connections are predominantly localised. These spatial networks (especially when the underlying distribution of points is clustered) have many features that we expect from disease networks although it is unclear if such simple formulations can be truly representative. In recent years, there has been growing interest in exponential random graph models (ERGMs) for social networks, also called the p * class of models. ERGMs were first introduced in the early 1980s by Holland and Leinhardt [79] based on the work of Besag [80] . More recently, Frank and Strauss studied a subset of those that have the simple property that the probability of connection between two nodes is independent of the connection between any other pair of distinct nodes. [81] . This allows the likelihood of any nodes being connected to be calculated conditional on the graph having certain network properties. Techniques such as Markov Chain Monte Carlo can then be used to create a range of plausible networks that agree with a wide variety of information collected on network structures even if the complete network is unknown [82, 83] . Due to their simplicity, ERGMs are widely used by statisticians and social network analysts [84] . Despite significant advances in recent years (e.g., [85] ), ERGMs still suffer from problems of degeneracy and computational intractability for large network sizes, which has limited their use in epidemic modelling. Here, we have shown that a wide variety of network structures have been measured or synthesised to understand the spread of infectious diseases. Clearly, with such a range of networks, no clear consensus can be drawn on the types of underlying network structures that are generally present; in part, this is because different studies have focused on different infectious diseases and different diseases require different transmission routes. However, three factors emerge that are key components of epidemiological networks: heterogeneity in the number of contacts such that some individuals are at a higher risk of both catching and transmitting infection, clustering of contacts such that groups of individuals are often highly interconnected, and some reflection of spatial separation such that contacts usually form locally, but occasional longrange connections do occur. Three fundamental problems still exist in the study of networks. Firstly, are there relatively low-dimensional ways of capturing key aspects of a network's structure? What constitutes a key aspect will vary with the problem being studied, but for epidemiological applications, it should be hoped that a universal set of network characteristics may emerge. There is then the task of assessing reasonable and realistic ranges for these key variables based on values computed for known transmission networks-unfortunately very few transmission networks have been recorded in any degree of detail although modern electronic devices may simplify the process in the future. Secondly, there is the related statistical problem of inferring plausible complete networks from the partial information collected by methods such as contact tracing. This is equivalent to seeking an underlying model for the network connections that is consistent with the known partial information, and hence, has strong resonance with the more mechanistically motivated models in Section 2.3. Even when the network is fully realised (and an epidemic observed), there is considerable statistical difficulty in attributing risk to particular contact types. Finally, there are the key questions of predicting the dynamics of infection on any given network-and while for many complex networks, direct simulation is the only approach, for other simplified networks some analytical traction can be achieved, which helps to provide more generic insights into which elements of network structure are most important. These three key areas are discussed below. Real networks can exhibit staggering levels of complexity. The challenge faced by researchers is to try and make sense of these structures and reduce the complexity in a meaningful way. In order to make any sense of the complexities present, researchers over several decades have defined a large variety of measurable properties that can be used to characterise certain key aspects [63, 65, 86] . Here, we describe the definitions of the most important characterisations of complex networks (in our view), and outline their impact on disease transmission models. In general, networks are not necessarily connected; in other words, all parts of the network are not reachable from all others. The component to which a node belongs is that set of nodes that can be reached from it by paths running along edges of the network. A network is said to have a giant component if a single component contains the majority of nodes in the network. In directed networks (one in which each edge has an associated direction), a node has both an in-component from which the node can be reached and an out-component that can be reached from that node. A strongly connected component (SCC) is the set of nodes in the network in which each node is reachable from every other node in the component. The concept of a giant component is central when considering disease propagation in networks. The extent of the epidemic is necessarily limited to the number of nodes in the component that it begins in, since there are no paths to nodes in other components. In directed networks, in the case of a single initial infected individual, only the out-component of that node is at risk from infection. More generally, the strongly connected component contains those nodes that can be reached from each other. Members of the strongly connected component are most at risk from infection imported at a random node, since a single introduction of infection will be able to reach all nodes in the component. The degree is defined as the number of neighbours that a node has and is most often denoted as k. In directed graphs, the degree has two components, the number of incoming edges k in , (in-degree), and the number of outgoing edges k out , (outdegree). The degree distribution is defined as the set of probabilities, P(k), that a node chosen at random will have degree k. Plotting the distribution of degrees of nodes is one of the most basic and important ways of characterising a given network (Figure 2 ). In addition, useful characterisations are obtained by calculating the moments of the degree distribution. The nth moment of P(k) is defined as with the first moment, k , being the average degree, the second, k 2 allowing us to calculate the variance k 2 − k 2 , and so on. The degree distribution is one of the most important ways of characterising a network as it naturally captures the heterogeneity in individuals' potential to become infected as well as cause further infection. Intuitively, the higher the number of edges a node has, the more likely it is to be a neighbour of an already infected node. Also, the more neighbours a node has, the more likely it is to cause a large number of onward cases. Thus, knowing the form of P(k) is crucial for the understanding of the spread of disease. In random networks of the type studied by Erdös and Rényi, P(k) follows a binomial distribution, which is effectively Poisson in the case of large networks. Most real social networks have distributions that are significantly different from the random case. For the extreme case of P(k) following an unbounded power law and assuming equal transmission across all edges, Pastor-Satorras and Vespignani [87] showed that the classic result of the epidemic threshold from mean field theory [10] breaks down. In real-transmission networks, the distribution of degree is often heavily skewed, and occasionally follows a power law [22] , but is always bounded, leading to the recovery of epidemic threshold, but one which is much lower than expected in evenly mixed populations [88] . The degree distribution provides very useful information on uncorrelated networks such as those produced by configuration models. However, real networks are in general correlated with respect to degree; that is, the probability of finding a node with given degree, k, is dependent on the degree of the neighbours of that node, k , which is captured by the conditional probability P(k | k). To characterise this behaviour, several measurements have been proposed. The most straightforward, and probably most useful measure, is to consider the average degree of the neighbours of a node where the sum of degrees is made over the neighbours (Nbrs) of i. One can then calculate the average of k nn over all nodes with degree k which is a direct measure of the conditional probability P(k | k), since When k nn (k) increases with k, the network is said to be assortative on the degree; that is, high-degree nodes have a tendency to link to other high degree nodes, a behaviour often observed in social networks. Other types of networks, such as the internet at router level, show the converse behaviour; that is, nodes of high degree tend to link to nodes with low degree [63, 89] . Characterising degree correlations is important for understanding disease spread. The classic example is the existence of strong correlations in sexual networks which were shown to be a key factor in understanding HIV spread [90] . More recently, mean field solutions of the SIS model on networks have shown that both the speed and extent of an epidemic are dependent on the correlation pattern of the substrate network [91, 92] . In a network, the shortest path between two nodes i and j, is the path requiring the smallest number of steps to reach j from i, following edges in the network. There may be (and often there is) more than one shortest path between a pair of nodes. The distance between any pair of nodes d i, j is the minimal number of steps required to reach j from i, that is, the number of steps in the shortest path. The average distance, d is the mean of the distances between all pairs of nodes and measures the typical distance between nodes where N is the number of nodes in the network. The diameter of the network is defined as the maximum shortest path distance between a pair of nodes in the network, max(d i, j ), which measures the most extreme separation of any two nodes in the network. Characterising networks in terms of the number of steps needed to reach any node from any other is also important. Real networks frequently display the small-world property; that is, the vast majority of nodes are reachable in a small number of steps. This has clear implications for disease spread and its control. Percolation approaches have shown that the effects of the small-world phenomenon can be profound [93] . If it only takes a short number of steps to reach everyone in the population, diseases are able to spread much more rapidly. The notion of shortest distance through a network can be used to quantify how central a given node is in the network. Many measures have been used [94] , but the most relevant of these is betweenness centrality. Betweenness captures the idea that the more shortest paths pass through a node, the more central it is in the network. So, betweenness is simply defined as the proportion of shortest paths that pass through a single node where N is the number of nodes in the network and the denominator quantifies the total number of shortest paths in the network. In terms of disease spread, identifying those nodes with high betweenness will be important. Central nodes are likely to become infected early on in the epidemic, and are also key targets for intervention [3] . Clustering. An important example of an observable property of any network is the clustering coefficient, φ, a measure of the local density of a graph. In social network terms, this quantifies the likelihood that the friend of your friend is also your friend. It is defined as the probability that two neighbours of a node will also be neighbours of each other and can be expressed as follows: where a connected triple means a single node with edges to a pair of others. φ measures the fraction of triples that also form part of a triangle. The factor of three accounts for the fact that each triangle is found in three triples and guarantees that 0 ≤ φ ≤ 1 (and its inclusion depends on the way that triangles in the network are counted). Locally, the clustering coefficient for each node, i, can be defined as the fraction of triangles formed through the immediate neighbours of i [74] φ i = # triangles centered on i # triples centered on i . The clustering property of networks is essential to the understanding of transmission processes. In clustered networks, rapid local depletion of susceptible individuals plays a hugely important role in the dynamics of spread [95, 96] ; for a more analytic treatment of this, see Section 4.2 below. Degree and clustering characterise some aspects of network structure at an individual level. Considering distances between nodes provides information about the global organisation of the network. Intermediate scales are also present, and characterising these can help in our understanding of network structure and therefore the dynamics of spread. At the simplest level, networks can be thought of being comprised of a collection of subgraphs. The simplest subgraph, the clique, is defined as a group of more than two nodes where all the nodes are connected to each other by means of edges in both directions. In other words, a clique is a fully connected subgraph, with the smallest example being a triangle. This is a strong definition and one which is only fulfilled in a limited number of cases, most notably households (see Figure 1 (d), Section 4.2 and House and Keeling [70] ). n-cliques relax the above constraint while retaining its basic premise. The shortest path between all the nodes in a clique is one. Allowing this distance to take higher values, one arrives at the definition of n-cliques, which are defined as a subgroups of the graph containing more than two nodes where the maximum shortest path distance between any two nodes in the group is n. Over the years, many variants of these basic ideas have been formalised in the social network literature and a good summary can be found in Wasserman and Faust [94] . Considering higher order structures can be very informative but is more involved. Milo and coworkers began by looking for specific patterns of connections between nodes in small subgraphs, dubbed motifs. Given a connected subgraph of size 3, for example, there are 13 possible motifs. Statistically, some of these appear more often and are found to be overrepresented in certain real networks compared to random networks [97] . Understanding the motif composition of a complex network has been shown to improve the predictive power of deterministic models of transmission when motifs are explicitly modelled (see Section 4.2 and House et al. [98] ). In the above definitions, a subgraph has been defined only in reference to itself. A different approach is to compare the number of internal edges to the number of external edges, arising from the intuitive notion that a community will be denser in terms of edges than its surroundings. One such definition, the definition of community in the strong sense, is defined as a subgraph in which each node has more edges to other nodes within the subgraph than to any other nodes in the network. Again, this definition is quite restrictive, and in order to relax these constraints, the most commonly used (and most intuitive) definition of communities is groups of nodes that have a high density of edges within them and a lower density of edges between groups. This intuitive definition is behind the most widely used approach for studying community structure in networks. Newman and Girvan formalised this in terms of the modularity measure Q [99] . Given a particular network which is partitioned into communities, the modularity measure compares the expected number of edges within communities to the actual number of edges within communities. Although the impact of communities in transmission processes has not been fully explored, a few studies have shown it can have a profound impact on disease dynamics [100, 101] . An alternative measure of how "well-knit" a graph is, named conductance [102] , most widely used in the computer science literature has also been found to be important in a range of networks [103] . 3.6. Higher Dimensional Networks. All of the above definitions have concentrated on networks where the edges remain unchanged over time and all edges have equal weight. Both of these constraints can naturally be relaxed, but generally, this calls for a higher-dimensional characterisation of the edges within the network. It is a matter of common experience that social interactions which can lead to infection do change, with some contacts being repeated regularly, while others are more sporadic. The frequency, intensity, and duration of contacts are all time-varying. How these inherently dynamic networks are represented for the purposes of modelling can have a significant impact on the model outcomes [55, 104] . However, capturing the structure of such dynamic networks in a parsimonious manner remains a substantial challenge. More work has been done on weighted networks, as these are a more straightforward extension of the classical presenceabsence networks [105, 106] . In terms of disease spread, the movement networks discussed in Section 2.4 are often considered as weighted [37, 40, 107] . In the sections that follow, we discuss how these network properties can be inferred statistically and the improvements in our understanding of the transmission of infection in networks that have come as a result. One of the key advantages of the simulation of disease processes on networks is that it enables the study of systems that are too complex for analytical approaches to be tractable. With that in mind, it is worth briefly considering efficient approaches to disease simulation on networks. There are two main types of simulation model for infectious diseases on networks: discrete-time and continuoustime models; of these, discrete-time simulations are more common, so we discuss them first. In a discrete-time simulation, at every time step, disease may be transmitted along every edge from an infectious node to a susceptible node with a particular probability (which may be the same for all extant edges or may vary according to properties of the two nodes or the edge). Also, nodes may recover (becoming immune, or reverting to being susceptible) during each timestep. Within a time-step, every infection and recovery event is assumed to occur simultaneously. In a dynamic network simulation, the network is typically updated every time step-for example, in a livestock movement network, during time-step x, infection could only transmit down edges that occurred during time-step x. Clearly, in a directed network, infection may only transmit in the direction of an edge. Whilst algorithms for discrete-time simulations are not complex, some simple implementation techniques (arising from the observation that most networks of epidemiological interest are sparse) can significantly enhance software performance. In a directed network with N nodes, there are N(N −1) possible edges; in a sparse network with mean node degree k, there are Nk N(N − 1) edges. Accordingly, rather than representing the network as an N by N array, where the element in each array is 0 if the edge is absent, nonzero otherwise, it is usually more efficient to maintain a list of the neighbours of each node. Then, if a list of infected nodes is maintained during a simulation run, it is straightforward to consider each susceptible neighbour of an infected node in turn and test if infection is transmitted to that node. Additionally, a fast high-quality pseudorandom number generator such as the Mersenne Twister should be used [108] . The "contagion" software package implements these techniques (amongst others) and is freely available [109] . The alternative approach to simulating disease processes on networks is to simulate a series of stochastic Markovian events-the continuous-time approach. Essentially, given the state of the system, it is possible to calculate the probability distributions of when possible subsequent events (i.e., recovery of an infectious node or infection of a susceptible node) will occur. Random draws from these distributions are then made to determine which event occurs next, the state of the system updated, and the process repeated. This approach was pioneered by Gillespie to study the dynamics of chemical reactions [110] ; it is, however, computationally intensive, so approximations have been developed. The τleap method [111] , where multiple events are allowed to occur during a time period τ, is clearly related to the discretetime formulation discussed above. However, the ability to allow τ to vary during a simulation to account for the processes involved [112] has potential benefits. The continuous-time approach is clearly in closer agreement with the ideal of standard disease models; however, utilising this method may be computationally prohibitive especially when large networks are involved. Discrete-time models may provide a viable alternative for three main reasons. Firstly, as the time steps involved in the discretetime model become sufficiently small, we would expect the two models to converge. Secondly, inaccuracies due to the discrete-time formulation are likely to be less substantial in network models compared to random-mixing models, providing two events do not occur in the same neighbourhood during the same time step. Finally, the daily cycle of contacts that regulate most of our lives means that using time steps of less than 24 hours may falsely represent the temporal accuracy that can be attributed to any simulation of the real world. In this section, we use the word "analytic" broadly, to imply models that are directly numerically integrable, without the use of Monte Carlo simulation methods, rather than systems for which all results can be written in terms of fundamental functions, of which there are very few in epidemiology. Analytic approaches to transmission of infection on networks fall into three broad categories. Firstly, there are approaches that calculate exact invasion thresholds and final sizes for special networks. Secondly, there are approaches for calculating exact transient dynamics, including epidemic peak heights and times, but again, these only hold in special networks. Finally, there are approaches based on moment closure that are give approximately correct dynamics for a wide class of networks. Before considering these approaches on networks, it is worth considering what is meant by nonnetwork mixing and showing explicitly how this can derive the standard transmission terms from familiar differential equation models. Nonnetwork mixing can be taken to have one of two meanings: either that every individual in the population is weakly connected to every other (the mean-field assumption), or that an Erdös-Rényi random graph defines the transmission network, depending on context. To see how this determines the epidemic dynamics, we consider a population of N individuals, with a homogeneous independent probability q that any pair of individuals is linked on the network, which gives each individual a mean number of edges n = q(N − 1). We then assume that the transmission rate for infection across an edge is τ and that the proportion of the population infectious at a time t is I(t); then, the force of infection experienced by an average susceptible in the population is nτI(t) ≡ βI(t). The quantity β, therefore, defines a population-level transmission rate that can be interpreted in one of two ways as N → ∞. In the case where the population is assumed to be fully connected, the limit is that q is held at unity, and so τ is reduced to as N is increased to hold q(N − 1)τ constant. In the case where the population is connected on a random graph, q is reduced as N is increased to hold n constant. In either case, having defined an appropriate populationlevel transmission rate, a stochastic susceptible-infectious model of transmission is defined through a Markov chain, in which a population with X susceptible individuals and Y infectious individuals transitions stochastically to a population with X − 1 susceptible individuals and Y + 1 infectious individuals at rate βXY/(N − 1). Then, the exact mean behaviour of such a system in the limit N → ∞ then has its transmission behaviour captured bẏ where S, I are the proportion of individuals susceptible and infectious, respectively. The mathematical formalism behind deriving such sets of ordinary differential equations from Markov chains is given by Kurtz [113] , and a summary of the application of this methodology to infectious disease modelling is given in Diekmann and Heesterbeek [114] . However, it should be clear that (8) is familiar as the basis of all random-mixing epidemiological models. In the case of exponentially distributed infectious periods and recovery from infection offering long-lasting immunity, the standard SIR equations provide an exact description of the mean behaviour of this system. Nevertheless, the existence of waning immunity, a latent period between an individual becoming infected and being able to transmit infection, and nonexponentially distributed recovery periods are also important for epidemiological applications [10, 42, 115] . These can often be incorporated into analytical approaches through the addition of extra disease compartments, which necessitates extra algebraic and computational effort but typically does not require a fundamental conceptual reevaluation. Sometimes, significant additional complexity does not even modify quantitative epidemiological results-for example, regardless of the rate of waning immunity, length of latent period, or infectious period distribution, if the mean infectious period is T, then the basic reproductive ratio is The estimation of this quantity for complex disease histories, from data likely to be available, is considered by Wallinga and Lipsitch [116] . We, therefore, focus on the transmission process, since this is most affected by network structure, and other elements of biological realism typically act at the individual level. An important caveat to this, however, is when an infected individual's level of transmissibility varies over the course of their infectious period, which sets up correlations between the processes of transmission and recovery that pose a particular challenge for analytic work, especially in structured populations, as noted by for example Ball et al. [117] . For nonnetwork mixing, the threshold for invasion is given by the basic reproductive ratio R 0 , defined as the expected number of secondary infectious cases created by an average primary infectious case in an otherwise wholly susceptible population. In structured populations, this verbal definition is typically altered to be the secondary cases caused by a typical primary case once the dynamical system has settled into its early asymptotic behaviour. As such, the threshold for invasion is R 0 = 1: for values above this, an infection can grow in the population and the disease can successfully invade; for values below it, each chain of infection is doomed to eventual extinction. Values of R 0 can be measured directly during the course of an epidemic by detailed contact tracing; however, there are considerable statistical issues concerning censoring and data quality. Provided there are no short closed loops in the network, R 0 can be defined through a next-generation matrix where K km defines the number of cases in individuals with k contacts from an individual with m contacts during the early stages of the epidemic. Here and elsewhere in this section we use square brackets to represent the numbers of different types on the network; hence, [m] is the number of individuals with m edges in the network and [km] is the number of edges between individuals with k and m contacts, respectively. In addition, p is the probability of infection eventually passing across the edge between a susceptible-infectious pair (for Markovian recovery rate γ and transmission rate τ this is given by p = τ/(τ + γ)). The basic reproductive ratio is given by the dominant eigenvalue of the next-generation matrix This quantity corresponds to the standard verbal definition of the basic reproductive ratio, and correspondingly the invasion threshold is at R 0 = 1. Once an appreciable number of short closed loops are present in the network, exact threshold parameters can still sometimes be defined, but these typically depart from the standard verbal definition of R 0 . For example, Ball et al. [117] consider a branching process on cliques (households) connected to each other through configuration-model edgescliques are connected to each other at random (Figure 1(d) ). By considering the number of secondary cliques infected by a clique with one initial infected individual, a threshold called R * can be defined. (For the configuration-model of households where each household is of the same size and each individual has the same number of random connections outside the household, the threshold R * is given later as (20) ; however, the methodology is far more general). The calculation of the invasion threshold for the recently defined triangular configuration model [118, 119] involves calculating both the expected number of secondary infectious individuals and triangles rather than just working at the individual level. Trapman [120] deals with how these sort of results can be related to more general networks through bounding. A general feature of clustered networks for which exact thresholds have been derived so far is that there is a local-global distinction in transmission routes, with a general theory of this given by Ball and Neal [121] , where an "overlapping groups" and "great circle" model are also analysed. Nevertheless, care still has to be taken in which threshold parameters are mathematically well behaved and easily calculated (e.g [122] ). Size. The most sophisticated and general way to obtain exact results for the expected final size of a major outbreak on a network is called the susceptibility set argument and the most general version is currently given by Ball et al. [117] . We give an example of these kind of arguments from Diekmann et al. [123] , who consider the simpler case of a network in which each individual has n contacts. Where there is a probability p of infection passing across a given network link (so for transmission and recovery at rates τ and γ, resp., p = τ/(τ + γ)), the probability that an individual avoids infection is given by Here, a two-step process is needed because in an unclustered, regular graph two generations of infection are needed to stabilise the network correlations and so the auxiliary variable S must also be solved for. Once this and S ∞ are known, the expected attack rate is R ∞ = 1 − S ∞ . The main way to calculate approximate final sizes is given by percolation-based methods. These were reviewed by Bansal et al. [124] and also in [125] . Suppose that we remove a fraction ϕ of links from the network and can derive an expression for the fraction of nodes remaining in the giant component of the network, f (ϕ). Then, and an invasion threshold is given by the value of p for which this final size becomes nonzero in the "thermodynamic limit" of very large network size. This approach is not exact for clustered graphs, but for unclustered graphs exact results like (12) are reproduced. (where the probability of the population being in each possible configuration is calculated) on small, fully connected graphs as summarised in Bailey [126] . The rate at which the complexity of the system of master equations grows means that these equations quickly become too complex to integrate for the most general network. The presence of symmetries in the network, however, does mean that automorphism-driven lumping is one way to manipulate the master equations (whilst preserving the full stochastic information about the system) for solution [127] . At present, this technique has only been applied to relatively simple networks; however, there are no other highly general methods of deriving exact lower-dimensional systems of equations from the master equations. Nevertheless, other specific routes do exist that allow exact systems of equations of lower dimensionality to be derived for special networks. For static networks constructed using the configuration model (where individuals have heterogeneous degree but connections are made at random such that the presence of short loops can be ignored in a large network, see Figure 1 (c)), an exact system of equations for SIR dynamics in the limit of large network size was provided by Ball and Neal [69] . This construction involves attributing to each node an "effective degree", which starts the epidemic at its actual degree, and measures connections still available as routes of infection and is, therefore, reduced by transmission and recovery. Using notation consistent with elsewhere in this paper (and ignoring the global infection terms that were included by Ball and coworkers) this yields the relatively parsimonious set of equationṡ S k = −ρ τ + γ kS k − γ(k + 1)S k+1 , . (14) Here, S k , I k are the proportion of effective degree k susceptible and infectious individuals, respectively. Hence, for a configuration-network where the maximum degree is K, we require just 2K equations to retrieve the exact dynamics. While R 0 can be derived using expressions like (11), calculation of the asymptotic early growth rate r requires systems of ODEs like (14) . If we assume that transmission and recovery are Markovian processes with rates τ and γ, respectively, two measures of early behaviour are Interdisciplinary Perspectives on Infectious Diseases where · informs about the average over the degree distribution. These quantities tell us that the susceptibility to invasion of a network increases with both the mean and the variance of the degree distribution. This closely echos the results for risk-structured models [10] but with an extra term of −1 due to the network, representing the fact that the route through which an individual acquired infection is closed off for future transmission events. For more structured networks with a local-global distinction, there are two limits in which exact dynamics can also be derived. If the network is composed of m communities of size n 1 , . . . , n m , with the between-community (global) mixing determined by a Poisson process with rate n G and the within-community (local) mixing determined by a Poisson process with rate n L , then in the limit as the communities become large, n i → ∞, the epidemic dynamics on the system areṠ where S a and I a are the proportion of individuals susceptible and infectious in community a, and Hence, we have a classic metapopulation model [43] , defined in terms of Poisson local and global connections and large local community sizes. In the limit where n L → (n − 1) and m → ∞-such that there are infinitely many communities of equal size and each community forms a fully interconnected clique-"selfconsistent" equations such as in Ghoshal et al. [128] and House and Keeling [70] are exact. These equations evolve the proportion of cliques with x susceptibles and y infecteds, P x,y , as well as the proportion of infecteds in the population, I, as follows: where β G = n G τ. Both of these two local-global models, the metapopulation model (16) and the small cliques model (18) , are reasonably numerically tractable for modern computational resources, provided the relevant finite number (m or n, resp.) is not too large. The basic reproduction number for the first system is clearly while for the second, household model, invasion is determined by where Z n ∞ (τ, γ) is the expected final size of an epidemic in a household of size n with one initial infected. Of course, the within-and between-community mixing for real networks is likely to be much more complex than may be captured by a Poisson process, but these two extremes can provide useful insights. These models show that network structure of the form of communities reduces the potential for an infectious disease to spread, and hence, greater transmission rates are required for the disease to exceed the invasion threshold. Dynamics. While all the exact results above are an important guide to intuition, they only hold for very specialised networks. A large class of models exists that form a bridge between "mean-field" models and simulation by using spatial or network moment closure equations. These are highly versatile models. In general, invasion thresholds and final sizes can be calculated rigorously, but exact calculation of transient dynamics is only possible for very special networks. If one wants to calculate transient effects in general network models-most importantly, peak heights and times-then moment closure is really the only versatile way of calculating desired quantities without relying on full numerical simulation. It is also worth noting that there are many results derived through these "approximate" approaches that are the same as exact results or are numerically indistinguishable from exact results and simulation. We give some examples below and also note that the dynamical PGF approach [129] is numerically indistinguishable from the exact model (14) above for certain parameter values [130] . What is currently lacking is a rigorous mathematical proof of exactness for ODE models other than those outlined in Section 4.2.4 above. While for many practical purposes the absence of such a proof will not matter, we preserve here the conceptual distinction between results that are provably exact, and those that are numerically exact in all cases tested so far. The idea of moment closure is to start with an exact but unclosed set of equations for the time evolution of different units of structure. Here, we show how these can be derived by considering the rates of change of both types of individual and types of connected pair. Such pairwise moment closure model are a natural extension to the standard (random-mixing) models, given that infection is passed between pairs of infected individuals Here, we use square brackets to represent the prevalence of different species within the network. We also use some nonstandard notation to present several diverse approaches in a unified framework: generalised indices κ, λ represent any property of a node (such as its degree), while arrows represent the direction of infection (and so for a directed network, the necessity that an edge in the appropriate direction be present), see Table 1 . Clearly, the system (21) is not closed as it relies on the number of connected triples, and so some form of approximate closure must be introduced to relate the triples to pairs and nodes, which will depend on underlying properties of the network. Most commonly, these closure assumptions deal with heterogeneity in node degree, assortativity, and clustering at the level of triangles. Examples include Keeling [95] and Eames and Keeling [96] , where the generalised variables κ, λ above stand for node degrees (k, l), the triple closure is symmetric with respect to the direction of infection, and the network is assumed to be static and nondirected. A general way to write the closure assumption is where n ≡ n is again the average degree distribution and φ measures the ratio of triangles to triples as a means of capturing clustering within the network (see Section 3.4). The typical way to analyse the closed system is direct numerical integration; however, some analytic traction can be gained. One example is the use of a linearising Ansatz to derive the early asymptotic behaviour of the dynamical system. Interestingly, when this is done for φ = 0 (such that there are no triangular loops in the network) as in Eames and Keeling [96] , the result for the early asymptotic growth rate agrees with the exact result of (10). In [95] , the differential equations for an n-regular graph were also manipulated to give an expression for final size that agreed with the exact result (12) Equation (22), however, is not the only possible network moment closure regime: Boots and Sasaki [131] and Bauch [132] considered regimes in which closure depended on the disease state (i.e., triples composed of different arrangements of susceptibles and infecteds close differently) to deal with spatial lattice-based systems and early disease invasion respectively. For example Boots and Sasaki [131] where O represents empty sites within the network that are not currently occupied by individuals, and the parameter ε = 0.8093 accounts for the clustering within lattice-based networks. House and Keeling [133] considered a model of infection transmission and contact tracing on a network, where the closure scheme for [ABC] triples was asymmetric in A and C-this allowed the natural conservation of quantities in a highly clustered system. The work on dynamical PGF models [129] can be seen as an elegant simplification of this pairwise approach that is valid for SIR-type infection dynamics on configuration model networks. The equations can be reformulated as S = g(θ), where g is the probability generating function for the degree distribution, p S and p I correspond to the number of contacts of a susceptible that are susceptible or infected, respectively, and θ is defined as probability that a link randomly selected from the entire network has not been associated with the transmission of infection. Here, the closure assumption is implicit in the definition of S; that is, an individual only remains susceptible if all of its links have not seen the transmission of infection, and that the probability is independent for each link, which is comparable to the assumptions underlying the formulation by Ball and Neal [69] , equation (14) . The precise link between this PGF formulation and the pairwise approach is discussed more fully in House and Keeling [134] . There are many other extensions of this general methodology that are possible. Writing ODEs for the time evolution of triples and closing at a higher order allows the consideration of the epidemiological consequences of varying motif structure [98] . Sharkey et al. [135] considered closure at triple level on directed networks, which involved a more sophisticated treatment of third-order clustering due to the larger repertoire of three-motifs in directed (as compared to undirected) networks. It is also possible to combine stochastic and network moment closure [136] . Timevarying, dynamical networks, particularly applied to sexually transmitted infections where partnerships vary over the course of an epidemic, were considered using approximate ODE-based models by Eames and Keeling [137] and Volz and Meyers [138] . Sharkey [139] considered models appropriate for local networks with large shortest path lengths, where the generic indices μ, λ in (21) stand for node numbers i, j rather than node degrees k, l. Another approach is to approximate the transmission dynamics in the standard (mean-field) differential equations models. Essentially, this is a form of moment closure at the level of pairs rather than triples. For example, in Roy and Pascual [140] the transmission rate takes the polynomial form where the exponents, p and q, are typically fitted to simulated data but are thought to capture the spatial arrangement of susceptible and infected nodes. Also, Kiss et al. [141] suggest Transmission rate to S k from I l ∝ k(l − 1)(S k )(I l ), (26) as a way of accounting for each infected "losing" an edge to its infectious parent. Finally, a very recent work [142] presents a dynamical system to capture epidemic dynamics on triangular configuration model networks; the relationship between this and other ODE approaches is likely to be an active topic for future work. This diversity of approaches leads to some important points about methods based on moment closure. These methods are extremely general and can be applied to consider almost any aspect of network structure or disease natural history; they can be applied to populations not currently amenable to direct simulation due to their size, and they do not require a complete description of the network to run-only certain statistical properties. However, there are currently no general methods for the proposal of appropriate closure regimes nor any derivation of the limits on dynamical biases introduced by closure. Therefore, closure methods sit somewhere in between exact results for highly specialised kinds of network and stochastic simulation, where intuitive understanding and general analysis are more difficult. In the papers that introduced them, the differential-equation-based approximate dynamical systems above were compared to stochastic simulations on appropriate networks. Two recent papers making a comparison of different dynamical systems with simulation are Bansal et al. [124] and Lindquist et al. [130] . There are, however, several issues with attempts to compare deterministic models with simulation and also with each other. Firstly, it is necessary to define what is meant by agreement between a smooth, deterministic epidemic curve and the rough trajectories produced by simulation. Limiting results about the exactness of different ODE models assume that both the number of individuals infectious and the network size are large, and so the early behaviour of simulations, when there are few infectious individuals, is often dominated by stochastic effects. There are different ways to address this issue, but even after this has been done, there are two sources of deviation of simulations from their deterministic limit. The first of these is the number of simulations realised. If there is a summary statistic such as the mean number of infectious individuals over time, then the confidence interval in such a statistic can be made arbitrarily small by running additional simulations, but agreement between the deterministic limit and a given realisation may still be poor. The second source of deviation is the network size. By increasing the number of nodes, the prediction interval within which the infection curve will fall can be made arbitrarily small; however, the computational resources needed to simulate extremely large networks can quickly become overwhelming. More generally, each approximate model is designed with a different application in mind. Models that perform well in one context will often perform poorly in another, and this means that "performance" of a given model in terms of agreement with simulation will primarily be determined by the discrete network system on which simulations are performed. The above considerations motivate the example comparisons with simulation that we show in Figure 3 . This collection of plots is intended to show a variety of different example networks, and the dynamical systems intended to capture their behaviour. In Figures 3(a) -3(e), continuous-time simulations have their temporal origin shifted so that they agree on the time at which a cumulative incidence of 200 is reached, and then confidence intervals in the mean prevalence of infection are achieved through bootstrapping. The 95% confidence interval is shown as a red shaded region (although typically, this is sufficiently narrow it resembles a line). Six different deterministic models are compared to simulations: HomPW is the pairwise model of Keeling [95] with zero clustering, HetPW is the heterogeneous pairwise model of Eames and Keeling [96] , ClustPW is the improved clustered pairwise closure of House and Keeling [133] , PGF is the model of Volz [129] , Pair-based is the model of Sharkey [139] , integrated using the supplementary code from Sharkey [143] , and Degree-based is the model of Pastor-Satorras and Vespignani [87] . Figure 3 (a) shows a heterogeneous network composed of two risk groups, constructed according to the configuration model [18] . In this case, models that incorporate heterogeneity like HetPW and PGF (which are numerically indistinguishable in this case and several others) are in very close agreement with simulation, while just taking the average degree as in HomPW is a poor choice. In Figure 3 (b), assortativity is added to the two group model following the approach of Newman [89] , and HetPW outperforms PGF. Figure 3 (d) the rate of making and breaking links is much faster than the epidemic process. Models like HomPW and PGF are therefore better for the former and degree-based models are better for the latter-in reality the ratio of the rate of network change to the rate of transmision may not be either large or small and so a more sophisticated method may be best [137, 138] . Figure 3 (e) shows a graph with four links per node where clustering has been introduced by the rewiring method of Bansal et al. [144] sometimes called the "big V" [133] . In this case, ClustPW performs better than HomPW and PGF, but clearly there is significant inaccuracy around the region of peak prevalence and so this model captures qualitatively the effects of clustering without appearing to be exact for this precise network. Finally, Figure 3 (f) considers the case of a one-dimensional next-nearest-neighbour lattice (so there are four links per node). This introduces long path lengths between nodes in addition to clustering, meaning that the system does not converge onto a period of asymptotic early growth and so realisations are shown as a density plot rather than a confidence interval. ClustPW accounts for clustering but not long path lengths and so is in poor agreement with simulation while the pair-based curve captures the qualitative behaviour of an epidemic on this lattice whilst being quantitatively a reasonable approximation. In order to be predictive, epidemic models rely on valid values for parameters governing outbreak dynamics, conditional on the population structure. However, obtaining these parameters is complicated by the fact that even when knowing the underlying contact network structure, infection events are censored-it is only when disease is detected either from symptoms or laboratory tests that a case becomes apparent. In attempting to surmount this difficulty, parameter estimates are often obtained by making strong assumptions as to the infectious period or through ad hoc methods with unknown certainty. Measuring the uncertainty in such estimates is as important as obtaining the estimates themselves in providing an honest risk prediction. Given these difficulties, inference for epidemic processes has perhaps received little attention in comparison to its simulation counterpart. The presence of contact network data for populations provides a unique opportunity to estimate the importance of various modes of disease transmission from disease incidence or contact tracing data. For example, given knowledge of the rate of contact between two individuals, it is possible to infer the probability that a contact results in an infection. If data on mere connectivity (i.e., a 1 if the individuals are connected and 0 otherwise) is available, then it is still possible to infer a rate of infection between connected individuals. Thus, the detail of the inference is determined to a large extent by the available detail in the network data [145] . Epidemic models are defined in terms of times of transitions between infection states, for example a progression from susceptible, to infected, to removed (i.e., recovered with lifelong immunity or dead) in the so-called "SIR" model. Statistical inference requires firstly that observations of the disease process are made: at the very least, this comprises the times of case detections, remembering that infection times are always censored (you only ever know you have a cold a few days after you caught it). In addition, covariate data on the individuals provides structure to the population and begins to enable the statistician to make statements about the importance of individuals' relationships to one another in terms of disease transmission. Therefore, any covariate data, however slight, effectively implies a network structure upon which disease transmission can be superimposed. As long as populations are relatively small (e.g., populations of farms in livestock disease analysis), it is common for models to operate at the individual level, providing detailed information on case detection times and perhaps even information on epidemiologically significant historical contact events [146] [147] [148] . In other populations, however, such detailed data may not be available due to practical and ethical reasons. Instead, data is supplied on an aggregated spatial and/or temporal basis. For the purposes of inference, therefore, this can be regarded as a household model, with areas constituting households. In a heterogeneous population, the behaviour of an epidemic within any particular locality is governed by the relationship between infected and susceptible individuals. For inference in the early stages of an epidemic, it is important to quantify the amount of uncertainty in the underlying contact networks as the early growth of the epidemic is known to be subexponential due to the depletion of the local susceptible population. This contrasts markedly to the exponential growth observed in a large homogeneously mixing population [10] . When the network is known and details of individual infections are available, contact tracing data may be used to infer the network; this data could also be used for inference on the epidemic parameters [116, 149] . Conversely, if the network is completely unknown, it would be useful if estimation of both the epidemic parameters and parameters specifying the structure of the network was possible. This is a difficult problem because the observed epidemic contains very limited information about the underlying network, as demonstrated by Britton and O'Neill [150] . However, with appropriate assumptions, some results can be obtained; the limited amount of existing work in this area is described in Section 5.3.1 below although clearly the problem is worthy of further study. For homogeneous models the basic reproduction number, or R 0 , has several equivalent definitions and can be defined in terms of the transmission rate β and removal rate γ. For nonhomogeneous models, the definitions are not equivalent; see for example [122] . Although inference for β and γ is difficult for real applications (see below), it turns out that making inference on R 0 (as a function of β and γ) is rather more straightforward. Heffernan et al. [151] summarise various methods for estimating R 0 from epidemiological data based on endemic equilibrium, average age at infection, epidemic final size, and intrinsic growth rate [114, 152, 153] . However, these methods all rely on observing a complete epidemic, and hence for real-time analysis during an epidemic, we must make strong assumptions concerning the number of currently undetected infections. An example of inference for R 0 based upon complete epidemic data is provided by Stegeman et al. [154] , where data from the 2003 outbreak of High Pathogenicity Avian Influenza H7N7 is fitted to a chain-binomial model using a generalised linear model. Obviously, complete or near-complete epidemic data is rare and hence it is desirable to perform inference based upon partial observation. This is particularly relevant for real time estimation of R 0 . For example, Cauchemez et al. [155] attempt to estimate R 0 in real-time by constructing a discrete-time statistical model that imputes the number of secondary cases generated by each primary case. This is based on the method of Wallinga and Teunis [156] who formulate a likelihood function for inferring who infected whom from dates of symptom onset where w(·) is the probability density function for the generation interval t j − t i , that is, the time between infector i's infection time and infectee j's infection time. Of course, infection times are never observed in practice so symptom onset times are used as a proxy, with the assumption that the distribution of infection time to symptom onset time is the same for every individual. Bayesian methods are used to infer "late-onset" cases from known "early-onset" cases, but large uncertainty of course remains when inferring the reproductive ratio close to the current time as there exists large uncertainty about the number of cases detected in the near future. Additionally, a model for w(·) must be chosen (see, [157] ). The tradeoff in the simplicity of estimating R 0 in these ways, however, is that although a population wide R 0 gives a measure of whether an epidemic is under control on a wide-scale, it give no indication as to regional-level, or even individual-level, risk. Moreover, the two examples quoted above do not even attempt to include population heterogeneity into their models though the requirement for its inclusion is difficult to ascertain in the absence of model diagnostics results. It is postulated, therefore, that a simple measure of R 0 , although simple to obtain, is not sufficient in order to make tactical control-policy decisions. In these situations, knowledge of both the transmission rate and removal rate are required. Inference for households models is well developed in comparison to inference for other "network" models. In essence, this is for three main reasons: firstly, it is a reasonable initial approximation to assume that infection either occurs within the home or from a random source in the population. Secondly, entire households can be serologically sampled following an epidemic, such that the distribution of cases in households of given sizes can be ascertained. Finally, it is often a reasonable approximation that following introduction of infection into the household, the within-household epidemic will go extinct before any further introductions-which dramatically simplifies the mathematics. The first methods proposed for such inference are maximum likelihood procedures based upon chain-binomial models, such as the Reed-Frost model, or the stochastic formulation of the Kermack-McKendrick model considered by Bartlett [158] . These early methods are summarised by Bailey [126] . They, and the significant majority of methods proposed for household inference to date, use finalsize data which can be readily obtained from household serology results. A simplifying assumption to facilitate inference in most methods is that the epidemics within the various households evolve independently (e.g., see the martingale method of Becker [159] , which requires the duration of a latent period to be substantial for practical implementation). Additionally, fixed probabilities p C and p H , corresponding to a susceptible individual escaping community-acquired infection during the epidemic and escaping infection when exposed to a single infected household member, respectively, were initially assumed [160] . Two important, realistic extensions to this framework are to incorporate different levels of risk factors for individuals [161] and to introduce dependence of p H on an infectious period [162] . The latter inclusion was enabled by appealing to results of Ball et al. [163] . These types of methods are largely based upon the ability to generate closed form formulae for the final size distribution of the models. The ability to relax assumptions further has been predominately due to use of Markov chain Monte Carlo (MCMC) methods as first considered by O'Neill et al. [162] for household models following earlier studies of Gibson and Renshaw [164] and O'Neill and Roberts [165] who focused on single, large outbreaks. This methodology has been used to in combination with simulation and data augmentation approaches to tailor inference methods for specific data sets of interest; for example, Neal and Roberts [166] consider a model with a spatial component of distance between households and data containing details of dates of symptoms and appearance of rash and has also resulted in a growing number of novel methods for inference, for example Clancy and O'Neill [167] consider a rejection sampling procedure and Cauchemez et al. [168] introduce a constrained simulation approach. Even greater realism can be captured within household models by considering the different compositions of households and, therefore, the weighted nature of contacts within households. For example, Cauchemez et al. [169] considered household data from the Epigrippe study of influenza in France 1999-2000 and showed that children play a key role in the transmission of influenza and the risk of bringing infection into the household. Whilst new developments are appearing at an increasing rate, the significant majority of methods are based upon final size data and are developed for SIR disease models, perhaps due in part to the simplification of arguments for deriving final size distributions. One key, but still unanswered question from these analyses of household epidemics is how the transmission rate between any two individuals in the household scales with the total number of individuals in the household (compare Longini and Koopman [160] and Cauchemez et al. [169] ). Intuition would suggest that in larger households the mixing between any two individuals is decreased, but the precise form of this scaling is still unclear, and much more data on large household sizes is required to provide a definitive answer. Perhaps the holy grail of statistical inference on epidemics is to make use of an individual-level model to describe heterogeneous populations at the limit of granularity. In this respect, Bayesian inference on stochastic mechanistic models using MCMC have perhaps shown the most promise, allowing inference to be made on both transmission parameters and using data augmentation to estimate the infectious period. An analysis of the 1861 outbreak of measles in Hagelloch by Neal and Roberts [166] demonstrates the use of a reversible jump MCMC algorithm to infer disease transmission parameters and infectious period, whilst additionally allowing formal comparisons to be made between several nested models. With the uncertainty surrounding model choice, such methodology is vital to enable accurate understanding and prediction. This approach has since been combined with the algorithm of O'Neill and Roberts [165] and used to analyse disease outbreaks such as avian influenza and foot and mouth disease in livestock populations [146, 147, 170] and MRSA outbreaks in hospital wards [171] . Whilst representing the cutting edge of inference on infectious disease processes, these approaches are currently limited by computing power, with their algorithms scaling by the number of infectives multiplied by the number of susceptibles. However, with advances in computer technology expected at an increasing rate, and small approximations made in the calculation of the statistical likelihoods needed in the MCMC algorithms, these techniques may well form the mainstay of epidemic inference in the future. Tracing. In livestock diseases, part of the standard response to a case detection is to gather contact tracing information from the farmer. The resulting data are a list of contacts that have been made in and out of the infected farm during a stipulated period prior to the notification of disease [172] . In terms of disease control on a local level, this has the aim of identifying both the source of infection and any presumed susceptibles that might have been infected as a result of the contact. It has been shown that providing the efficiency of following up any contacts to look for signs of disease is high; this is a highly effective method of slowing the spread of an epidemic and finally containing it. Much has been written on how contact tracing may be used to decrease the time between infection and detection (notification) during epidemics. However, this focuses on the theoretical aspects of how contact tracing efficiency is related to both epidemic dynamics and population structure (see, e.g., Eames and Keeling [173] Kiss et al. [174] , Klinkenberg et al. [175] ). In contrast, the use of contact tracing data in inferring epidemic dynamics does not appear to have been well exploited although it was used by the Ministry of Agriculture, Fisheries, and Food (now Defra) to directly infer a spatial risk kernel for foot and mouth disease in 2001. This assumed that the source of infection was correctly identified by the field investigators, thereby giving an empirical estimate of the probability of infection as a function of distance [148, 176, 177] . Strikingly, this shows a high degree of similarity to spatial kernel estimates based 22 Interdisciplinary Perspectives on Infectious Diseases on the statistical techniques of Diggle [178] and Kypraios [179] without using contact tracing information. However, Cauchemez et al. [155] make the point that the analysis of imperfect contact tracing data requires more complex statistical approaches, although they abandoned contact tracing information altogether in their analysis of the 2003 SARS epidemic in China. Nevertheless, recent unpublished work has shown promise in assimilating imperfect contact tracing data and case detection times to greatly improve inference, and hence the predictive capability of simulation techniques. Qualitative results from simulations indicate that epidemics on networks, for some parameter values, show features that distinguish them from homogeneous models. The principal features are a very variable length slow-growth phase, followed by a rapid increase in the infection rate and a slower decline after the peak [180] . However, in quantitative terms, there is usually very limited information about the underlying network and parameters are often not identifiable. When the details of the network are unknown, but something is known or assumed about its formation, estimation of both the epidemic parameters and parameters for the network itself are in principle possible using MCMC techniques. All the stochastic models for generating networks described in Section 2.7 above realise a distribution over all or some of the 2 N(N−1)/2 possible networks. In most cases, this distribution is not tractable; MCMC techniques are in principle still possible but in practice would be too slow without careful design of algorithms. However, with appropriate assumptions some results can be obtained, which provide some insight into what more could be achieved. When the network is taken to be an Erdös-Rényi graph with unknown parameter p and the epidemic is a Markovian SIR, Britton and O'Neill [150] showed that it is possible to estimate the parameters, although they highlight the ever-present challenge of disentangling epidemiological from network parameters. The MCMC algorithm was improved by Neal and Roberts [181] and the extension from SIR to SEIR has been developed by Groendyke et al. [182] . However, the extension to more realistic families of networks remains a challenging problem and will undoubtedly be the subject of exciting future research. The use of networks is clearly a rapidly growing field in epidemiology. By assessing (and quantifying) the potential transmission routes between individuals in a population, researchers are able to both better understand the observed distribution of infection as well as create better predictive models of future prevalence. We have shown how many of the structural features in commonly used contact networks can be quantified and how there is an increasing understanding of how such features influence the propagation of infection. However, a variety of challenges remain. Several open problems remain if networks are to continue to influence predictive epidemiology. The majority of these stem from the difficulty in obtaining realistic transmission networks for a range of pathogens. Although some work has been done to elucidate the interconnected structure of sexual encounters (and hence the sexual transmission network), these are still relatively smallscale compared to the population size and suffer from a range of potential biases. Determining comparable networks for airborne infections is a far greater challenge due to the less precise definition of a potential contact. One practical issue is therefore whether new techniques can be developed that allow contact networks to be assessed remotely. Proximity loggers, such as those used by Hamede and colleagues [12] , provide one potential avenue although it would require the technology to become sufficiently robust, portable and cheap that a very large proportion of a population could be convinced to carry one at all times. For many human populations, where the use of mobile phones (which can detect each other via Bluetooth) is sufficiently widespread, there is the potential to use them to gather network information-although the challenges of developing sufficiently generic software should not be underestimated. While these remotely sensed networks would provide unparalleled information that could be obtained with the minimum of effort, there would still be some uncertainty surrounding the nature of each contact. There is now a growing set of diary-based studies that have attempted to record the personal contacts of a large number of individuals; of these, POLYMOD is currently the most comprehensive [17] . While such egocentric data obviously provides extensive information on individual behaviour, due to the anonymity of such surveys it is not clear how the alters should be connected together. The configuration method of randomly connecting half-links provides one potential solution, but what is ideally required is a more comprehensive method that would allow clustering, spatially localised connections and assortativity between degree distributions to be included and specified. Associated with the desire to have realistic contact networks for entire populations, comes the need to characterise such networks in a relatively parsimonious manner that provides important insights into the types of epidemiological dynamics that could be realised. Such a characterisation would allow for different networks (from different times or different locations) to be compared in a manner that is epidemiologically significant and would allow artificial networks to be created that matched particular known network features. This clearly relies on both existing measures of network structure (as outlined in Section 3) together with a robust understand of how such features influence the transient epidemic dynamics (as outlined in Section 4.2). However, such a generic understanding of all network features is unlikely to arise for many years. A more immediate challenge is to understand ways in which local network structure (clustering, cliques, and spatially localised connections) influence the epidemiological dynamics. To date the vast majority of the work into disease transmission on networks has focused on static networks where all links are of equal strength and, therefore, associated with the same basic rate of transmission. However, it is clear that contact networks change over time (both on the short-time scale of who we meet each day, and on the longer time-scale of who our main work and social contacts are), and that links have different weights (such that some contacts are much more likely to lead to the transmission of infection than others). While the simulation of infection on such weighted time-varying networks is feasible, it is unclear how the existing sets of network properties or the existing literature of analytical approaches can be extended to such higher-dimensional networks. For any methodology to have any substantive use in the field, it is important both to have effective data gathering protocols in place and to have the statistical techniques in place to analyse it. Here, three issues are perhaps most critical. Firstly, data gathering resources are almost always limited. Therefore, carefully designed randomised sampling schemata should be employed to maximise the power of the statistical techniques used to analyse data, rather than having to reply on data augmentation techniques to work around the problems present in ad hoc datasets. This aspect is particularly important when working on network data derived from population samples. Secondly, any inference on both network and infectious disease models should be backed up by a careful analysis of model fit. Although recent advances in statistical epidemiology have given us an unprecedented ability to measure population/disease dynamics based on readily available field data, epidemic model diagnostics are currently in their infancy in comparison to techniques in other areas of statistics. Therefore, it is expected that with the growth in popularity of network models for analysing disease spread, much research effort will be required in designing such methodology. We have highlighted that the study of contact networks is fundamentally important to epidemiology and provides a wealth of tools for understanding and predicting the spread of a range of pathogens. As we have outlined above, many challenges still exist, but with growing interest in this highly interdisciplinary field and ever increasing sophistication in the mathematical, statistical and remote-sensing tools being used, these problems may soon be overcome. We conclude, therefore, that now is an exciting time for research into network epidemiology as many of the practical difficulties are surmounted and theoretical concepts are translated into results of applied importance in infection control and public health.

Assessing the In Vitro Fitness of an Oseltamivir-Resistant Seasonal A/H1N1 Influenza Strain Using a Mathematical Model

In 2007, the A/Brisbane/59/2007 (H1N1) seasonal influenza virus strain acquired the oseltamivir-resistance mutation H275Y in its neuraminidase (NA) gene. Although previous studies had demonstrated that this mutation impaired the replication capacity of the influenza virus in vitro and in vivo, the A/Brisbane/59/2007 H275Y oseltamivir-resistant mutant completely out-competed the wild-type (WT) strain and was, in the 2008–2009 influenza season, the primary A/H1N1 circulating strain. Using a combination of plaque and viral yield assays, and a simple mathematical model, approximate values were extracted for two basic viral kinetics parameters of the in vitro infection. In the ST6GalI-MDCK cell line, the latent infection period (i.e., the time for a newly infected cell to start releasing virions) was found to be 1–3 h for the WT strain and more than 7 h for the H275Y mutant. The infecting time (i.e., the time for a single infectious cell to cause the infection of another one) was between 30 and 80 min for the WT, and less than 5 min for the H275Y mutant. Single-cycle viral yield experiments have provided qualitative confirmation of these findings. These results, though preliminary, suggest that the increased fitness success of the A/Brisbane/59/2007 H275Y mutant may be due to increased infectivity compensating for an impaired or delayed viral release, and are consistent with recent evidence for the mechanistic origins of fitness reduction and recovery in NA expression. The method applied here can reconcile seemingly contradictory results from the plaque and yield assays as two complementary views of replication kinetics, with both required to fully capture a strain's fitness.

Influenza is the most important respiratory disease in terms of mortality and morbidity. Each year, between 3 and 5 million severe cases and 250,000 to 500,000 deaths due to seasonal influenza are reported worldwide [1, 2] . Cyclic pandemics due to antigenic shifts constitute an important threat [3] as was demonstrated by the swine-origin pandemic of 2009 [4] . Since vaccines for novel influenza virus strains require approximately 6 months to develop and produce [5] , antivirals remain the first line of defense. There are only two classes of antivirals approved for treatment of influenza [6] . The adamantanes, such as amantadine and rimantadine, are ineffective against B-type viruses [7] and have recently become ineffective against most A/H3N2 and some A/H1N1 viruses due to a mutation in the M2 gene [8] . The neuraminidase inhibitors (NAI), which include zanamivir and oseltamivir, were approved a decade ago and have shown excellent activity against all influenza A subtypes and B viruses [9] . A recent rapid increase in resistance to oseltamivir, however, has become a cause for concern. The H275Y mutation in the neuraminidase (NA) gene (H274Y in N2 numbering), first described in 2000 [10] , is the most frequent mutation associated with oseltamivir-resistance in the N1 subtype, but it had long been thought to critically reduce viral fitness [11] . With a location on the framework residue of the enzyme catalytic site [12] , the mutation has been shown to cause a reduced affinity for the substrate in enzyme activity assays [12, 13] , an impaired viral fitness in vitro [14] [15] [16] , and up to a 100-fold reduction in transmission efficiency in ferrets [14, 17] . For these reasons, strains carrying the H275Y mutation were not thought to be a great concern for public health [10, 18] . During the 2007-2008 influenza season, however, the A/ Brisbane/59/2007-like (H1N1) H275Y mutant emerged and rapidly disseminated worldwide in the apparent absence of antiviral pressure [8, [19] [20] [21] . Recently, our group performed a study on the replicative capacities of the A/Brisbane/59/2007 H275Y mutant strain where we showed that its fitness, based on in vitro and animal studies, was similar to that of its wild-type (WT), oseltamivir-susceptible, counterpart [22] . These observations, and those of others [23] , correlate with the clinical situation encountered in the 2008-2009 season where almost 100% of the A/H1N1 viruses isolated in North America and Europe were resistant to oseltamivir due to the H275Y mutation [1, 19, 23] . Recent work suggests that the origin of both the fitness reduction conferred by the H275Y mutation and the unique fitness of the A/Brisbane/59/2007 mutant strain is found in the virus NA activity and surface expression. Specifically, the reduction of NA activity conferred by the H275Y mutation has been associated with a reduced expression of surface neuraminidase, possibly due to defects in the folding of the molecule or its transport through the cellular membrane [24] . It has been shown, however, that two other mutations in the NA gene (V234M and R222Q) can provide a compensatory effect by increasing NA surface expression, and that these two substitutions indeed occurred in the evolution of the H1N1 seasonal strain between 1999 and 2007 [24] . The neuraminidase of contemporary (A/Brisbane/59/2007-like) strains susceptible to oseltamivir have shown a higher affinity for the substrate in NA activity assays than older H1N1 seasonal strains (e.g., A/New Caledonia/20/99 and A/Solomon Islands/3/06); contemporary oseltamivir-resistant strains (with the H275Y substitution) have shown a decrease in that affinity, but remain above the level of older strains [25] . Thus, it seems that these pre-existing mutations led to an over-expression of NA and provided a favorable environment for the appearance of the H275Y mutation. The eventual dominance of the H275Y mutant may be due to a better balance between the hemagglutinin (HA) and NA activity [25, 26] . It remains an open question, however, precisely how these mechanistic changes lead to viral fitness changes. The answer to this should be found in the details of the infection kinetics in the interaction of virus and cell. In this paper, we present a method to extract the values of viral kinetics parameters, specific to a particular strain, from parallel experiments of plaque and viral yield assays. Our previous study [22] assessed the in vitro replicative capacity and fitness of pairs of WT influenza virus strains and their H275Y mutant counterparts by use of viral yield assays and by qualitatively comparing plaque sizes. Here, we show that it is possible to use these experimental measures to quantitatively characterize the kinetics parameters responsible for the replicative efficiency of influenza virus strains. As a proof of concept, we apply our method for extracting the viral kinetics parameters to the oseltamivir-susceptible/-resistant pair of A/Brisbane/59/2007 influenza virus strains in order to determine how the known genotypic differences in these two strains map to quantitative changes in the viral kinetics parameters characterizing their replicative efficiency. The kinetics parameters extracted through our method suggest that the H275Y mutant has weaker NA activity compared to its WT counterpart -confirmed by NA activity assays -which manifests itself as a longer phase of latent infection before viral release -confirmed by single-cycle viral yield experiments. However, the results also indicate that this longer latent infection period for cells infected by the H275Y mutant is compensated for by a shorter infecting time required for that cell, once releasing virions, to successfully infect other cells. In order to obtain two complementary views of the infection kinetics for the A/Brisbane/59/2007 WT and H275Y mutant strains, virus growth over time was observed in two different in vitro systems: the viral plaque assay and the multiple-cycle viral yield assay. Viral plaque assays. Figure 1 shows representative plaques for the WT and H275Y mutant strains of A/Brisbane/59/2007 (H1N1) viruses at each time point. The average plaque radius of each strain over time, calculated by averaging three independent experiments of three such wells at each time point, is shown in Figure 2 . The plaque growth is characterized by an initial delay where no growth is observed, followed by a period of linear increase of the plaque radius over time. After 60 h, the rate of plaque growth declines and the linear approximation is no longer valid. The growth attenuation could be due to a number of factors including a hardening of the overlay, a depletion of nutrients required for viral production and cell maintenance, and the widespread destruction of the cell monolayer leading to holes and irregularities disrupting and limiting further growth. The plaque assay is a long-standing and standard technique in virology [27, 28] and plaque sizes have been used in many in vitro studies to qualitatively evaluate the phenotypes of various viruses [22, 29, 30] . Plaque assays are often used for strain comparison and, in that context, plaque diameters at a single time point are reported. These plaque diameters are then typically used to conclude that, for example, if the plaques observed at 48 h for strain A are larger than those for strain B, then strain A must have a higher replicative fitness than strain B. However, one should question whether such conclusions are valid and robust to experimental variability. Looking at Figure 2 , one can see that at 36 h, the plaque radius of both A/Brisbane/59/2007 WT and its H275Y mutant counterpart are comparable in size. Yet at 60 h, the WT strain has significantly larger plaques than the H275Y mutant, and the situation is reversed at 96 h. Thus, relying on single time point measurements for comparing strains can be misleading. Here, instead, we exploit the fact that the average plaque growth is approximately linear in time between 36 and 60 h. This allows us to extract a novel measure, the plaque velocity, which is the slope in the linear regression of the plaque radius to the 36, 48 and 60 h time points. The plaque velocity, unlike the plaque radius at a given time point, is a robust measure in that it takes into account plaque radius at several time points and is not affected by differences in the length of the delay period which precedes the period of linear growth. Using this method, the measured plaque growth was more rapid for the WT than the H275Y mutant, with a plaque velocity of 1:79+0:1d cell =h compared to 1:48+ 0:1d cell =h, where d cell is the diameter of one cell. Thus, using plaque velocity alone, it would appear that the WT strain has a replicative fitness advantage over the H275Y mutant. Multiple-cycle viral yield assays. In order to complement the information provided by the plaque assay, namely the plaque velocity, we also conducted multiple-cycle viral yield experiments. The results of these experiments for the WT and H275Y mutant strains are shown in Figure 3 . The kinetics of the viral yield experiments can be broken into two different phases: an exponential growth of virus concentration, characterized by the viral titer growth rate, followed by an exponential decay of virus concentration, characterized by the viral titer decay rate, after the viral titer peak. The viral titer decay rate was the same for both strains at 0:19+0:02h {1 . The viral titer growth rate of the H275Y mutant was 0:83+ 0:02h {1 , slightly greater than that of the WT which was 0:75+0:04h {1 . Thus, it would appear from the viral titer growth rate alone, that the H275Y mutant has a replicative advantage over the WT strain, in contrast with the findings using the plaque velocity extracted from the plaque assay alone. This discrepancy between the conclusions drawn from each experimental measure points to a complementarity between the two assays: they appear to emphasize different aspects of viral replication. Thus, combining the information provided by these two assays is key to obtaining a complete and consistent picture of what shapes a particular strain's replicative fitness. The plaque growth experiment yields a single experimental measure for each strain: the plaque velocity. The multiple-cycle viral yield assay provides two quantities: the viral titer growth rate, and the viral titer decay rate. It is the goal of this paper to associate these broad experimental measures to the values of fundamental infection parameters, specific to each strain, which quantitatively characterize replicative efficiency. To this aim, we have constructed mathematical models which allow us to simulate each in vitro assay in a computer experiment. The basic mathematical model used here is similar to other within-host models of viral infection [31] [32] [33] [34] [35] [36] . A cell can be in one of four states -target (uninfected), latently infected (infected but not yet releasing virus), infectious (releasing virus) and dead (no longer releasing virus) -and its passage through these states ( Figure 4 ) is determined by five infection kinetics parameters. Target cells interacting with virus become latently infected at a constant infection rate per virus, b. The average time a cell remains latently infected is called the latent infection period, t L , and the average time a cell releases virus is called the infectious lifespan, t I . Virus is produced by infectious cells at a constant viral production rate, p, and this free virus loses infectivity exponentially at a constant rate of viral infectivity loss, c (as is observed in experiments [35] ). In applying this model, it is assumed that the growth of a particular influenza virus strain in a particular cell line is determined by a single, unique set of values for these five parameters. Thus, although the mathematical structure of the models used for each experimental assay is different (see Materials and Methods for a detailed description of each), the parameter values for a particular virus strain are assumed to be constant from assay to assay. With only three experimentally measured quantities, it would be impossible to uniquely identify all five parameters for a particular virus strain. Fortunately, it is possible to reduce the number of parameters considered and obtain unique identification of a few key parameter values. One parameter can be determined immediately from the multiple-cycle viral yield results. The viral titer decay rate, characterizing the decline of the virus concentration after the peak (Figure 3 ), corresponds to the slowest of the rate of loss of virus-producing cells and the rate of viral infectivity loss [37] . Since prior in vitro experiments have shown that infectious cell death is nearly complete shortly after the viral titer peak [38, 39] , we set the rate of viral infectivity loss, c, equal to the viral titer decay rate. Because this decay rate was determined to be 0:19h {1 for both A/Brisbane/59/2007 strains, we have fixed the rate of viral infectivity loss to this value for all simulations. This corresponds to a virion half life of approximately 3.6 h, which is consistent with prior measurements for influenza virus in the experimental literature (see, e.g., [35, 40] ). Having fixed the rate of viral infectivity loss, we are left with four undetermined parameters and two experimental measures. For the experiments considered here, the infection rate per virus, b, and the viral production rate, p, can be combined into a single parameter, leaving only three parameters to be determined. The rationale for this simplification is the fact that, during an infection, the two parameters play equivalent roles: doubling the rate at which virus is produced by cells will have the same effect on new infections as doubling the rate at which virus infects cells. Therefore, the only identifiable quantity is the product of the two rates, pb. Since their product has units of inverse time squared, we have chosen to express this quantity as a new characteristic time, the infecting time, t inf~ffi ffiffiffiffiffiffiffiffiffiffiffiffiffi 2= pb ð Þ p , which corresponds to the average time it takes a single virus-producing cell to cause the latent infection of one more (see Materials and Methods). We are left then with two experimental measures -the viral titer growth rate and the plaque velocity -whose values may depend on three unknown infection kinetics parameters: the infecting time, t inf ; the latent infection period, t L ; and the infectious lifespan of a cell, t I . To determine how each of these parameters affect the infection dynamics, we varied each individually about a base value and measured the effect on the simulated experimental quantities ( Figures 5A and 5B ). One parameter, the infectious lifespan of a cell, t I , had very little effect on either the plaque velocity or the viral titer growth rate. In the latter case, this parameter was explicitly neglected in the derivation of the growth rate, because earlier viral yield experiments have shown little cell death prior to the peak of the virus concentration (see, e.g., [38, 39] ). The fact that, over a wide range of infectious lifespan values, the resulting plaque velocity remained unchanged, is perhaps more surprising. Indeed, a shorter infectious lifespan will lead to the earlier appearance of plaques, resulting in larger plaque sizes at any given time. We have shown, however, in earlier work where influenza virus plaques were observed by immunostaining [41] , that the same plaque velocity can be measured from both the progress of dead cells, as we consider here, and the progress of newly infected cells. This indicates that plaque velocity is established at the advancing edge of an infection wave, and is likely unaffected by cell death in the wake of that wave. In those experiments, the infectious lifespan of a cell appears only as a time-delay between the infected cell plaque growth and the dead cell plaque growth. This has also been observed for the plaques of other viruses [42] . Since the infectious lifespan has little effect on the experiments we consider, and is therefore not identifiable here, we have fixed its value for both strains and for all simulations to value of t I~1 2h, obtained from the literature ( Table 1) . This leaves only two parameters, the infecting time, t inf , and the latent infection period, t L , to be determined from our two experimental measures, the plaque velocity and viral titer growth rate. The full dependence of each experimental measure on the two remaining parameters are presented as contour plots in Figures 5C and 5D . Because the plaque velocity and the viral titer growth rate depend on both the infecting time, t inf , and the latent infection period, t L , the experimental measurement of either quantity alone is not sufficient to specify the values of these infection parameters for a given strain. The measurement of both, however, can provide enough information for this specification, provided that the dependence on the parameters is sufficiently different for the two quantities. To demonstrate this concept using the A/ Brisbane/59/2007 (H1N1) WT and H275Y mutant strains, we have plotted the experimentally-measured values of plaque velocity and viral titer growth rate as functions of the infecting time and latent infection period, using the model dependence determined above ( Figure 6 ). Figures 6A and 6D show the values of the kinetics parameters most consistent with the measured plaque velocities of the A/ Brisbane/59/2007 WT and H275Y mutant strains, respectively. Rather than plot a single line at the average measured value, we have accounted for the error in the measurement of the plaque velocities by plotting regions of contour denoting the one-and two-standard deviations (for a detailed description see Materials and Methods). We can see that while the plaque velocity does constrain the two parameters to a specific region, that region is too large to allow any useful comparison of the two strains. Similarly, Figures 6B and 6E show the values of the kinetics parameters most consistent with the measured viral titer growth rate. The consistency of a particular pair of parameter values with each of the two experimental measures can be combined by finding the intersection of the two parameter regions. This region of intersection corresponds to those parameter values most consistent with the parallel plaque and viral yield experimental measurements for a particular strain. The extent of these regions, shown in Figure 6C for the A/Brisbane/59/2007 WT strain and Figure 6F for the H275Y mutant strain, is summarized in Table 2 . The region of intersection suggests that the latent infection period for the H275Y mutant (w7 h) is longer than that of the WT strain (1-3 h), while the infecting time of the mutant (v 5 min) is much shorter than that of the WT (30-80 min). In order to test the predictions made in the previous section by applying the mathematical model to parallel plaque and viral yield assays, we performed two additional experimental tests which could provide some qualitative and quantitative confirmation. To independently estimate the latent infection period for the two A/Brisbane/59/2007 influenza virus strains, we performed a single-cycle viral yield experiment. Single-cycle experiments were performed at an MOI of 1 such that most cells would be infected simultaneously and pass through the phases of latency and viral release at the same time. Therefore, the observed virus production of the cell culture can be considered roughly proportional to that of an individual cell. The results of two independent experiments for each strain are shown in Figure 7A ; one experiment shows the viral titer over one full day post-infection and the other over only 14 h but with more frequent sampling. For each replicate, the viral titer of each strain was observed to grow rapidly after 4 h postinfection, with the WT viral titer reaching a plateau at approximately 8 h post-infection and that of the H275Y mutant reaching a plateau between 10 h and 14 h post-infection. Although the viral titer data in each replicate followed a relatively smooth curve, the inter-replicate variation was quite large, with peak virus titer varying from 2|10 3 PFU mL to almost 10 5 PFU mL. It is also notable that all of these peak values were well below the values seen in the multiple-cycle viral yield assay (Figure 3) , by a factor of * > 1000. Both of these features could be explained by the action of a relatively large defective interfering particle population [43, 44] within the viral stock, which is not uncommon for the influenza virus [45] . The delay in the peak of viral titer between the two strains is qualitatively consistent with the model predictions of the previous section: the H275Y mutant strain appears to have a longer latent infection period than the WT strain. To make this comparison more quantitative, we scaled each experimental data set such that the peak virus was equal to one and then performed a least-squares fit to the full set of normalized data for each virus strain ( Figure 7B ). We utilized a model similar to that used for the analysis of the multiple-cycle viral yield assay, but allowed for a normal distribution of the latent infection period among cells rather than a fixed value for all cells, as assumed previously (see Materials and Methods). The fitted value of the average latent infection period, t L , was found to be 5.6 h for the WT strain and 7.5 h for the H275Y mutant, with fitted values of the standard deviation in the normal distribution, s L , of 0.5 h and 1.2 h, respectively. These results are summarized in Table 2 , along with 95% confidence intervals determined by fitting 1000 bootstrap replicates [46] . The longer latent infection period predicted for the H275Y mutant strain, could be the result of poorer NA activity. This would also explain the shorter infecting time for the mutant strain in that its virions would more easily bind to new cells with less interference from its NA activity. To investigate whether the H275Y mutant had poorer NA activity compared to the WT, we directly measured the enzymatic activity of the NA of each virus strain using the Through modeling and simulation of two common in vitro experiments, the plaque and viral yield assays, we have extracted We have shown that seemingly contradictory results from the two experiments -plaques of the susceptible strain grow more quickly than the resistant strain, while the reverse is true of their titer growth in viral yield assays -can be considered complementary views of the infection kinetics which allow for the determination of parameter values controlling the replication of each strain. Specifically, we have found that the latent infection period of the H275Y mutant strain -equal to the time elapsed between the successful infection of a cell by a virion and the significant release of virus progeny by the newly infected cell -is much longer than that of the WT strain (by 4-10 h). The infectivity of the mutant strain, however, was found to be much higher than the WT, as quantified by the infecting time -equal to the time for a single infectious cell to cause the latent infection of one other, within a completely susceptible cell population. Independent single-cycle viral yield assay results lend support to the hypothesis of a longer latent infection period for the mutant strain than the WT, but suggest a more moderate (*2h) difference between the two. These results are consistent with the larger NA activity of the susceptible (WT) strain compared to the H275Y mutant, reported here and by others [25] , and its increased NA surface expression [24] . Since neuraminidase is the viral surface enzyme responsible for cleaving the virus from its sialic-acid receptors at the cell surface [47] , it can be expected that an increase of its expression would lead to more rapid viral release (a shorter latent infection period for the WT strain) but may also hinder the subsequent attachment of virions to other cells, leading to decreased infectivity (longer infecting times). A complete understanding of the viral kinetics requires investigation of the HA/NA balance. It has been shown that the A/Brisbane/59/2007-like strains of the 2007-2009 influenza seasons differ from earlier H1N1 seasonal strains by a few amino acid substitutions in the HA gene [25] , but none of these involve interaction with the receptor and are therefore not likely to have influenced the changes in fitness. We have recently sequenced the entire genomes of our A/Brisbane/59/2007 strains, and found three amino acid substitutions in the HA gene for the H275Y mutant compared to the WT strain. Two substitutions, G189V and L264F, do not involve interaction with the receptor, but the third, A193T, lies within the receptor-binding site. This latter substitution has been noted in earlier work in relation to oseltamivir-resistant strains of the influenza virus [48] and an investigation of its influence on viral kinetics is a necessary direction for future work. Mathematical models have been successfully applied previously to characterize the in vivo virus replication kinetics of HIV [31, 32] , hepatitis B and C [33, 49] , and influenza [34, 36, 50] , as well as in vitro viral yield experiments studying the effects of antiviral drugs [35] and the optimization of vaccine production [51, 52] . Models of viral plaques have also been considered [53] [54] [55] , although these were primarily directed at phage growth in an agar suspension of bacteria, a slightly different system than the cell monolayers considered here. The method presented here for the determination of the infection parameters differs from previous mathematical modeling approaches to viral dynamics in that we have considered the explicit dependence of two experimental quantities on the parameters, rather than fitting a full dynamical model to the time-course of an experiment. There are a number of benefits to this approach. First, we have been careful to determine that the two experimental quantities under consideration, plaque velocity and viral titer growth rate, depend on only two unknown infection parameters, the infecting time, t inf , and the latent infection period, t L . This ensures that the problem of parameter extraction is theoretically solvable, which is often not the case when fitting a multi-parameter model to experimental data (see, e.g., [56] ). Second, the experimental quantities themselves are robust and easily measurable in repeated experiments. The viral plaque is formed by the progression of an infection wave across the monolayer of cells [42] whose constant velocity is determined by the infection kinetics averaged over many thousands of cells. Therefore the measured plaque velocity depends on the average interaction of virus and cell, and is insensitive to stochastic effects on a small scale. Similarly, the viral titer growth rate is due to the collective infection of thousands of cells and is independent of the details of initial infection (i.e., the precise value of the multiplicity of infection) or the total number of cells in the system. Other quantities of in vitro experiments, such as the time and value of the viral peak in a yield experiment, are much more sensitive to experimental details. Finally, the method we have applied here is robust to changes in the construction of the mathematical model itself. We have, for example, performed the same analysis of the plaque and yield assays using a stochastic model with more general assumptions about the cell transitions from latently infected to infectious and found nearly identical results (not shown here). It is important to note that the results we present here are preliminary, a proof of concept for the method which requires further verification and refinement. In particular, it would be useful to develop an experimental assay which could measure the infecting time for a given strain, in the same way that single-cycle viral yield experiments give an approximate measure for the latent infection period. It is also of interest to design a set of experiments which may be less expensive and laborious than those presented here, perhaps using fluorescent or photographic observations of cell cultures rather than virus titrations, and which can identify a fuller set of viral kinetics parameters. We are currently designing competition experiments for the A/Brisbane/59/2007 WT and H275Y mutant strains in which the predictions which follow from the parameters extracted here can be tested directly. When verified, the basic method of analyzing parallel plaque and viral yield experiments introduced here should be useful in other contexts. For example, the investigation of other drug-resistant viruses (e.g., that of the pandemic A/H1N1), the rapid characterization of fitness for emerging strains, and assays measuring the activity of new antivirals would all be enhanced through the application of our method. The A/Brisbane/59/2007-like (H1N1) strains used were the oseltamivir-susceptible A/Québec/15230/08 (WT) and the oseltamivir-resistant A/Québec/15349/08 (NA-H275Y mutant). These clinical isolates were obtained from two distinct, untreated, immunocompetent patients during the 2007-2008 influenza season [22] . All experiments were performed on ST6GalI-expressing MDCK cells [29] which over-express the a -(2,6) sialic acid receptor predominantly found in the human upper respiratory tract. Prior to infection, cells were grown to confluence, achieving an average diameter of *20mm (used herein as a unit of length, d cell ). Plaque assays were prepared using a semi-solid overlay of 1.2% Avicel RC-581 (FMC Biopolymers, Newark, Delaware, USA) as described by Matrosovich et. al. [57] and stained with crystal violet. Six-well plates (Corning Life Sciences, Lowell, MA, USA) were infected with 25+10PFU=well, representing a multiplicity of infection (MOI) of approximately 10 {5 , and stained every 12 h for 96 h. The plates were then photographed using a DSLR camera (Fujifilm S2 with a 60 mm Nikkor macro objective) and the areas of viral plaques were measured using the Threshold and Analyze Particle features of ImageJ, an NIH opensource image analysis software [58, 59] . All plaque radii at one timepoint (three independent experiments of three wells each) were averaged and the standard error of the mean was calculated. The radial growth rate was determined by linear regression to the average radii at time points prior to 72 h. Multiple-cycle viral yield assays for the A/Brisbane/59/2007 WT and H275Y mutant were performed with MOI&10 {5 . Supernatants were harvested every 6 h for the first 36 h of infection and every 12 h subsequently, then titrated by plaque assay as previously described [22] . The geometric average and standard deviation was determined from three replicates at each time point. High MOI single-cycle yield assays were performed as described by Hurt et. al. [60] . Monolayers of ST6GalI-MDCK cells were grown to confluence in 12-well plates and infected with 10 6 PFU well (MOI = 1) in 1 mL of infection medium. Virus was adsorbed for 1 h at 37 0 C in a CO 2 incubator. The supernatant was then removed and cells were quickly washed once with acidic saline (0.9% NaCl in water, pH 2.2) and twice with PBS (pH 7.4). Fresh maintenance medium was added and plates were returned to the incubator. Supernatants were harvested every two hours for 24 h ( Figure 7A , Experiment 1) or every hour for 14 h (Experiment 2), in duplicate. Samples were frozen at {80 0 C until titrated in duplicate by plaque assay [57] . The enzyme kinetics of the neuraminidase was measured in duplicate for each strain as described in [61] , using the MUNANA reagent (4-methyl-umbelliferyl-N-acetyl neuraminic acid (Sigma-Aldrich, St-Louis, CO, #M8639)). Briefly, 10mL of live viruses diluted to 1:2|10 6 PFU mL were incubated at 37 0 C in Opaque Black Microfluor B CS50 96-well plates (VWR, Montreal, QC, #62402-983) with 30mL of MUNANA reagent ranging from 0 to 3000mM final concentration and 10mL of enzyme buffer [1:1 mix of 325mM MES (2-[N-Morpholino]ethanesulfonic acid) pH 6.5 (Sigma-Aldrich, St-Louis, CO, #M8250) and 10mM CaCl 2 (Sigma-Aldrich, St-Louis, CO)]. Fluorescence was measured in a Viktor 3 Multilabel Counter (PerkinElmer, Waltham, MA) every 90 seconds for 45 minutes. The excitation wavelength was 365nm and the emission wavelength 450nm with a 2:5nm excitation slit and a 20nm emission slit. K m and V max were calculated using a homemade Excel macro, created following [62] , and confirmed using the built-in ''Enzyme Kinetics'' features of the GraphPad Prism 5.01 software (GraphPad Software, La Jolla, CA). Plaque growth was simulated using a one-dimensional, timedelayed, partial differential equation (PDE) model: where T(r,t) and I(r,t) are the densities of target and infectious cells, respectively, and V (r,t) is the virus concentration. s (20-fold smaller than the Stokes-Einstein value for a 100 nm particle in 37 0 C water); the rate of viral infectivity loss was fixed at c~0:19h {1 based on the observed viral titer decay rate for both A/Brisbane/59/2007 strains in the multiple-cycle viral yield assays (see Results); and the infectious lifespan was held fixed at 12 h (Table 1) . Simulations were initialized with a ''top-hat'' central region of infectious cell density with radius d cell =2, and with all other cells in the target state. All fields rapidly take the form of traveling waves T(z), I(z) and V (z), where z~r{vt, with the same velocity, v. The multiple-cycle viral yield assay was modeled using a meanfield, delay-differential system of equations: where T and I are now the number of target and infectious cells, N is the total number of cells, V is the homogeneous virus concentration and all parameters have the same meaning as in the PDE model above. These equations can be derived from Equation (1) by assuming spatial homogeneity and integrating over space. An expression for the exponential growth rate, l g , of viral titer in the multiple-cycle yield assay can be derived from this system by assuming that in the early phases of an infection (well before the viral titer peak): the number of target cells is approximately constant T(t)~T 0 &N; there is an exponential growth of infectious cells I(t)~I 0 e lgt and virus V (t)~e lgt with common rate l g ; and infectious cell death can be neglected (t I ??). Substituting these expressions into (2) yields a transcendental equation for the viral titer growth rate: For any values of p, b, t L and c, this equation can be solved numerically for the viral titer growth rate, l g . The assumptions made in deriving the expression require that the viral titer growth rate be measured early in the course of infection, well before the time of peak, when the number of infected cells is small compared to the total number of cells. Both the plaque velocity and viral titer growth rate depend on the infection and production rates only through pb. Since this quantity has units of inverse time squared (units of virus cancel), it is useful to rewrite the dependence on these rates as a characteristic time, ffiffiffiffiffi ffi 2 pb s . A physical meaning can be ascribed to this quantity by considering Equation (2) in the case of a single infectious cell (I 0~1 ), within a completely susceptible cell population (T 0~N ). If loss of viral infectivity, c, is neglected, the equations can be then integrated to show that t inf~ffi ffiffiffiffi ffi 2 pb s is the time for that single infectious cell to cause the (latent) infection of one more cell. Therefore, we call this characteristic time the infecting time. The contour plots in Figure 6 were created using the functional dependence of the plaque velocity and viral titer growth rate on the infecting time, t inf , and the latent infection period, t L , as determined by model simulation, along with the experimentally measured values of these quantities and their associated measurement error, under the assumption that these errors are normallydistributed. For example, the function F v , plotted in Figure 6A and Figure 6D , takes values between zero and one, according to where v exp is the experimentally-measured plaque velocity with measurement error s v and v mod (t inf ,t L ) is the theoretical dependence of the plaque velocity determined by model simulation. Contours for the one and two-s values are drawn at F v~0 :6065 and 0:1353. A function on the parameter space, F lg , for the viral titer growth rate is constructed analogously. The product of these two functions is plotted in Figures 6C and 6F to show the likely regions of viral kinetics parameters controlling growth for each virus strain. In fitting the single-cycle viral yield data, a more biologicallyrealistic model was used which assumes that the set of latent infection periods for a collection of cells is normally-distributed about t L , rather than fixed [63] . In this model, target cell and virus dynamics are identical to that of Equation (2) where L 0 and I 0 are the number of cells latently infected and infectious, respectively, at the start of the experiment, f L (t) is the probability density function for the latent infection period and P I (t) is the probability that a cell remains infectious for at least a time t after the latent-infectious transition. If a Dirac delta function is used for f L (t) and a Heaviside step function for 1{P I (t), then the infectious cell dynamics of Equation (2) are recovered. In the fits to the single-cycle data (Figure 7 ), f L (t) was taken to be normal (truncated at t~0 and renormalized) with parameters t L and s L ; the function P I (t) was also derived from a normal distribution f I (t) with P I (t)~1{ d dt f I (t) , with fixed parameters t I~1 2h and s I~1 h.

Genomic Signatures of Strain Selection and Enhancement in Bacillus atrophaeus var. globigii, a Historical Biowarfare Simulant

BACKGROUND: Despite the decades-long use of Bacillus atrophaeus var. globigii (BG) as a simulant for biological warfare (BW) agents, knowledge of its genome composition is limited. Furthermore, the ability to differentiate signatures of deliberate adaptation and selection from natural variation is lacking for most bacterial agents. We characterized a lineage of BGwith a long history of use as a simulant for BW operations, focusing on classical bacteriological markers, metabolic profiling and whole-genome shotgun sequencing (WGS). RESULTS: Archival strains and two “present day” type strains were compared to simulant strains on different laboratory media. Several of the samples produced multiple colony morphotypes that differed from that of an archival isolate. To trace the microevolutionary history of these isolates, we obtained WGS data for several archival and present-day strains and morphotypes. Bacillus-wide phylogenetic analysis identified B. subtilis as the nearest neighbor to B. atrophaeus. The genome of B. atrophaeus is, on average, 86% identical to B. subtilis on the nucleotide level. WGS of variants revealed that several strains were mixed but highly related populations and uncovered a progressive accumulation of mutations among the “military” isolates. Metabolic profiling and microscopic examination of bacterial cultures revealed enhanced growth of “military” isolates on lactate-containing media, and showed that the “military” strains exhibited a hypersporulating phenotype. CONCLUSIONS: Our analysis revealed the genomic and phenotypic signatures of strain adaptation and deliberate selection for traits that were desirable in a simulant organism. Together, these results demonstrate the power of whole-genome and modern systems-level approaches to characterize microbial lineages to develop and validate forensic markers for strain discrimination and reveal signatures of deliberate adaptation.

Bacillus atrophaeus is a soil-dwelling, non-pathogenic, aerobic spore-forming bacillus related to B. subtilis. For more than six decades, this organism has played an integral role in the biodefense community as a simulant for biological warfare and bioterrorism events (BW) and is commonly referred to by its military two-letter designation ''BG'' [1, 2] . B. atrophaeus has served in studies of agent dispersal [3] , decontamination simulations [4, 5] and large-scale process development [6] . In addition to its historical use as a BW simulant, it is currently in widespread commercial use as a surrogate for spore-forming bacteria [5, 7] and is the basis of numerous assays for spore inactivation [8, 9] . In addition to its role as a simulant, the organism plays an important role in the biotechnology industry as a source of restriction endonucleases and of the glycosylation inhibitor nojirimycin [10] . The taxonomic placement of B. atrophaeus has changed dramatically over the years. Originally isolated as B. globigii in 1900 (Migula) as a variant of B. subtilis, it was originally distinguished from B. subtilis by the formation of a black-tinted pigment on nutrient agar and by low rates of heterologous gene transfer from B. subtilis [11] . It has been alternately known as B. subtilis var. niger, B. niger, and has been confused with B. licheniformis [12] . Other than the formation of the dark pigment, it is virtually indistinguishable from B. subtilis by conventional phenotypic analysis [13] , and the lack of distinguishing metabolic or phenotypic features has contributed to the confusionin the taxonomic placement of this organism. Low interspecies DNA transfer frequencies suggested substantial divergence [11] . Based onanalysis of comparative DNA hybridization, phenotypicand biochemical tests, Nakamura advocated that pigment-producing B. subtilis-like isolates should be classified as a distinct species termed B. atrophaeus [13] . Recently, more sensitive typing methods such as amplified fragment length polymorphism analysis showed that B. atrophaeus strains could be classified into two major biovars: var. globigii encompassing the classical, commonly used BG isolates, and var. atrophaeus encompassing other closely related yet genetically distinct strains [14] . Here we report the definitive molecular typing of several BGstrains using whole-genome sequences, and develop a plausible microevolutionary history of a commonly used lineage based on the accumulation of mutations over time and during transfer between laboratories.The selected strains span more than six decades of development, use, and transfer of BGbetween various institutions and laboratories and offer an unparalleled opportunity to investigate mutation under selection and drift over time. Phenotypic analysis revealed substantial heterogeneity both between and within strains, even in type strains, while highthroughput metabolic profiling revealed metabolic ''enhancements'' to a population that had returned to the University of Wisconsin (UW) from Camp Detrick in 1952. Whole-genome comparisons of single-nucleotide polymorphisms (SNPs), small insertion/deletion motifs (indels), and large-scale genomic architecture analysis by optical maps are combined to generate a plausible history of acquisition and use of operationally relevant strains by the American Type Culture Collection (ATCC) and by several laboratories within the biodefense community. Finally, our analysisof mutation profiles revealed potential signatures of the deliberate selection of strains with properties of enhanced growth and spore yields, properties that were deemed desirable in a simulant [6] . We also report genetic differences between strains in use in the biodefense community and the commercial sector that argue for adoption of a more uniform standard for B. atrophaeus as a simulant. Strains and growth conditions B. atrophaeus strains and their sources are indicated in Table 1 . Archival strains were maintained as spores in sterile soil at the University of Wisconsin (Figure 1 ). The 1013 lineage, originally founded from the 1942 strain, was extensively passaged by serial transfer every 12-18 months on agar slants for 30 years. Unless otherwise indicated, strains were grown using LB agar plates, LB agar brothor Tryptic Soy agar containing 5% sheep's blood (SBA, HealthLink) at 37uC. Spores were germinated by plating on LB media at 37uC. Plates were examined by stereomicroscopy using indirect lighting and imaged usinga Nikon SMZ1500 with a total magnification of 166. Colonies exhibiting distinct morphologies were repeatedly streaked to confirm stability of the phenotype. Genomic DNA was prepared from all isolates using the Blood and Cell Culture DNA Midi Kit for Bacteria (QIAGEN) from 10 ml overnight cultures in LB. BACI051-N was sequenced at the Naval Medical Research Center, while all other isolates were sequenced to .25-fold coverage at the US Army Edgewood Chemical Biological Center by massively parallel pyrosequencing on the Roche/454 GS-FLX using the Titanium reagent package. Draft genome sequences of all isolates were assembled de novo using Newbler [15] (Roche) and analyzed using both Newbler and Lasergene (DNAStar, Madison, WI). The 1942 Vogel isolate was designated as the reference strain and was brought to completion using standard finishing techniques. The draft genome of Bacillus atrophaeus var.globigii was finished at the Department of EnergyJoint Genome Institute (JGI) using a combination of Illumina [16] and 454 datasets [15] . For this genome, we constructed and sequenced an Illumina GAii shotgun library which generated 15120217 reads totaling 544 Mb, which was combined with 454 Titanium standard library which generated 387327 reads totaling 137 Mb of 454 data. All general aspects of library construction and sequencing performed at the JGI can be found at http://www.jgi.doe.gov/. The initial draft assembly contained 25contigs in 25scaffolds. The 454 Titanium standard data were assembled with Newbler, version 2.3. The Newbler consensus sequences were computationally shredded into 2 kb overlapping fake reads (shreds). Illumina sequencing data wereassembled with VEL-VET, version 0.7.63 [17] , and the consensus sequences were computationally shredded into 1.5 kb overlapping fake reads (shreds). We integrated the 454 Newbler consensus shreds, the Illumina VELVET consensus shreds and using parallel phrap, version SPS -4.24 (High Performance Software, LLC). The software Consed [18, 19, 20] was used in the following finishing process. Illumina data was used to correct potential base errors and increase consensus quality using the software Polisher developed at JGI (Alla Lapidus, unpublished). Possible mis-assemblies were corrected using gapResolution (Cliff Han, unpublished), Dupfinisher [21] , or sequencing cloned bridging PCR fragments with subcloning. Gaps between contigs were closed by editing in Consed, by PCR and by Bubble PCR (J-F Cheng, unpublished) primer walks. A total of 79additional reactions and 10shatter libraries were necessary to close gaps and to raise the quality of the finished sequence. The total size of the genome is 4 168 266 bp and the final assembly is based on 137 Mb of 454 draft data which provides an average 33.46 coverage of the genome and 544 Mb of Illumina draft data which provides an average 1336 coverage of the genome. The complete sequence and WGS were deposited at DDBJ/ EMBL/GenBank under accession numbers listed in Table 2 . The WGS versions described in this paper are the first versions, e.g. AEFM01000000. Templated assembly of the remaining strains were mapped to the 1942 finished sequence using the GSMapper tool in Newbler (Roche). High-confidence mutations were selected from Newbler ''HCDiffs'' calls (Table S1 ) by applying additional selection criteria that mandated high quality scores in both reference and templated assemblies with .80% of the sequencing reads differing from the reference, elimination of mutation calls associated with homopolymer tracts (with the exception of tracts that were formed by a deletion -see below), and a minimum coverage depth of 56 with bidirectional sequence reads. Finally, the raw 454 reads from the 1942 isolate were mapped to the finished sequence to assess error bias in the 454 process and to correct for residual sequencing errors in the finished sequence. Accession numbers of the relevant wholegenome shotgun sequences are found in Table 3 . Phylogeny was calculated using PAUP 4.0b10. Fifty-eightnucleotide positions were used with gaps being treated as a ''5th base'' and all characters assuming equal weight. One thousandbootstrap replicates were computed using a heuristic search with the optimal criterion set to ''parsimony''. The tree was created using stepwise addition. Nineteen loci in which putative mutations were identified from the 454 dataset were re-sequenced from PCR products by standard Sanger dye-terminator methods. No false-negatives or false-positives were identified among the re-sequenced loci; however resequencing of the apparent mutation at position 1486408revealed mixed genotypesin several isolates that are artifacts of a large duplication in the 1942 chromosome. Therefore, this signalcannot be considered a true SNP. Annotation, comparative genomic analysis, and multiple alignments Preliminary annotations were generated using a combination of the RAST [22] algorithm (rast.nmpdr.org). Loci containing mutations were used to query the non-redundant (nr) databases and Refseq protein databases at NCBI using directed BLASTx and BLASTp. The comparative BLAST tool from RAST was utilized for genome-wide protein sequence comparisons to B. subtilis. Results were filtered for bidirectional hits. Multiple alignments were generated by MegAlign from the LaserGene software package using the CUSTALW algorithm. Genomic DNA was prepared from live bacteria on agar slants to maximize the yield of extremely high-molecular weight DNA. Optical maps were generated by digestion with NcoI of DNA arrayed linearly on glass slides and the resulting maps were aligned and compared with the MapSolver software package (OpGen, Inc., Gaithersburg MD). Using an information-based method for genomic classification [23] , the sequence contigs from BG isolates 1942, 1013-2 and 49822 were analyzed in order to map the phylogenetic relationships of these isolates to other Bacillus species. In this method, genomic content is characterized by the frequencies of occurrence of short n-mers contained within each sequence (n typically from 3 to 16). These n-mers are then rank ordered by genome. The pair-wise comparison of the rank of n-mers within two different genomes is then used to compute an informationbased genetic distance (IBGD), where the sum of the differences in rank for all possible n-mers is weighted by an entropy factor that depends on the frequencies of occurrence of the respective n-mers in the two genomes. The pair-wise IBGD values are then used to construct a phylogenetic network [24] . Bacilli genomes were obtained from Genbank. This method for phylogenetic characterization enables computation even with the unassembled reads, and it can be applied to draft or partial genome sequence data, which was the case for the three B. atrophaeus genomes studied here. The first seven BGstrains listed in Table 1 were streaked for single colonies on BHI plates and incubated at 33uC overnight, followed by subculturing a second time under the same conditions. Subsequently, cell suspensions were prepared according to Biolog specifications, with OD readings ranging between 0.35-0.45 at 600 nm. Biolog phenotypic microarray plates PM1 through PM20, were inoculated according to the manufacturer's specifications, and incubated at 37uC for 72 hours. Readings were taken every 15 minutes, and data processed by OmniLog Phenotype Microarray File Management/Kinetic Plot and Parametric modules. Two biological replicates of the experiment were conducted for each strain. PM1-10 contain single wells for each growth condition whereas PM11-20 contain quadruplicate wells for each condition. The area under the curve (AUC) values were computed by adding all OmniLog values at all time points for each of the 1200 distinct phenotypes produced from the OmniLog software. The AUC values from the two different biological replicates for each unique phenotype were averaged. The ratio for each AUC was calculated between the 6 query strains (Detrick-1, Detrick-2, Detrick-3, 1013-1, 1013-2, and Dugway) and reference parent strain (1942) . For the purpose of visualization, 1920 phenotypes were included in the heatmap (i.e. this better represents the locations of the phenotypes which correspond to different modes of action categories). The same ratios were used for the phenotypes that have replicates. The ratio values were formatted as PM1 to PM20 for each strain across the columns and wells A i to H i , where i = 1 to 12 for the rows. The results were plotted in a heatmap using R [25] . Positive growth wells are represented by greenblocks while negative growth wells are represented by red blocks. Catalase activitywas assayed by spotting drops of hydrogen peroxide (3%) onto isolated colonies on LB agar plates. Colonies were monitored for bubble formation, signifying the release of water and oxygen. A colony was considered to be catalase positive by observation of bubbles. Streaks of Detrick 1, Detrick 2, and 1013 strains were grown for two days on TSA plates containing SBA.Bacterial cell mass was scraped using an inoculating loop (1 ml) from the streak and resuspended in PBS. Sporulation was evaluated by bright field phase-contrast microscopy. Phase-bright free sporesand phasedark vegetative cellswere counted. Five representative viewing fields were counted from each strain for each experiment. This experiment was completed in triplicate by repeating once per day over the course of three consecutive days. In order to compare the percent sporulation between Detrick 1 and Detrick 2, and Detrick 1 and 1013, a mixed analysis of variance (ANOVA) was used to complete the analysis. Strain and viewing field were evaluated as fixed factors, and replicate was included as a random factor. The natural log of the percent sporulation was taken to obtain a normal distribution of the residual error. Tukey's method was applied to compare the difference between the mean log percent sporulation. We traceda potential provenance of the commonly used BGstrains through an exhaustive search of the open literature and the archives of the University of Wisconsin,which suggested a possible lineage from which the ''military'' BGstrains were derived. The original source of the strains were the collections at the University of Wisconsin during the 1930s and 1940s, from which the strains were transferred to Camp Detrick at the initiation of the US Army's BW program at the beginning of the Second World War [6, 26] . At Camp Detrick, BG was used as a non-pathogenic surrogate in process development for sporeforming bacteria It is tempting to speculate that the University of Wisconsin supplied BG to Porton Down: A note found in the archive of Dr. Baldwin's papers, dated February 19, 1943 , contained an order from Dr. Fildes (presumably Sir Paul Fildes, a noted bacteriologist active in the British BW program at the time), for a batch of B. subtilis spores. It is not clear whether BG or B. subtilis subsp. subtilis was supplied, or whether this material was actually delivered. Unfortunately, original records describing in detail the maintenance of the strains during the period 1942-1955 were destroyed as per US Army policy at the time (Dr. Mark Wolcott, USAMRIID; personal communication), and the personnel who had first-hand knowledge of the strain passage histories and methods are deceased. Therefore, the actual source of the Camp Detrick isolates must be inferred from published work [6] , limited available documentation (e.g. ATCC 9372) and the genome sequences presented hereFrom Camp Detrick the isolates were eventually transferred to ATCC as B. subtilis var. niger ''red strain.'' The desire to maintain a phenotypically and genotypically uniform simulant throughout the biodefense communityprompted us to elucidate whether significant phenotypic and/or genomic differences had accumulated in any of the commonly used isolates during the growth and transfer of strains to different institutions and to compare the isolates in broad use today to the so-called ''Mil-Spec'' strain (ATCC 9372).In contrast, the origin of ATCC 49822 prior to acquisition F. Young's laboratory (the depositor) is unclear. We obtained isolates from archival spore suspensions in sterile soilfrom the University of Wisconsin with legible labels dating back as far as 1942 ( Figure 1 ; Table 1 ). These isolates included an archival stock dated 1942 that likely predated the transfer to Camp Detrick, as well as material that had been returned to the University of Wisconsin from Camp Detrick in 1952. A derivative of the 1942 strain that had been repeatedly passaged in vitro on agar slants over a period ofapproximately 30 years allowed us to compare the genomic signatures of deliberate selection with the effects of long-term in vitro passage. In addition, a sample of strain NRS-356 [13] , which is mentioned as a possible parent strain in correspondence between various academic laboratories and Camp Detrick, was also obtained from the same source as the 1942 ''Vogel'' strain. These isolates were subsampled, germinated on LB plates, screened for colony morphology variation (see below). Genomic DNA was prepared from these isolates for sequencing. Upon initial plating of the archival and modern-dayBG stocks, we noted distinct colony morphotypes for many of the strains, with some strains containing multiple variants ( Figure 2 , Table 2 ). Some of these morphotypes were consistent with those observed by Hayward et al. [6] whooriginally described the emergence of colony variants in ''B. globigii.'' As in the earlier report, individual morphotypes were stable and did not interconvert with high frequency (data not shown), suggesting that these morphotypes were the result of relatively rare chromosomal mutations, although 1013-1 occasionally threw off papillae in heavier streaks (not shown). Multiple morphotypes were noted for ATCC9372, ATCC 49822, Detrick, and 1013, while the archival 1942 isolate, the isolate obtained from Dugway Proving Ground (Dugway) and BACI051 appeared to be pure populations on LB. All strains tested positive for BGusing Real Time-PCR primers specific to the recF gene (Methods S1) [27] . The appearance of multiple colony morphotypes even within single ''strains'' strongly suggested an asyet undescribed level of genetic diversity within these samples that likely affected the expression of cell-surface components and/or sporulation. The intra-strain colony morphology variation was particularly dramatic in the in vitro passaged 1013 and ATCC9372 isolates, in which one variant of each lineage had lost the production of color on LB orSBAplates (Figure 2 ), suggesting more dramatic alterations to the genome. Draft genome sequences were generated from several BGstrains in our collection. A summary of the results from the sequenced isolates is indicated in Table 3 . All of the ''military'' isolates (Detrick clones1through 3, BACI051, Dugway) were extremely closely related to each other and to both ATCC9372 variants. The ATCC isolates possessed additional mutations that were absent in the ''military'' isolates. The size of the finished and closed genome of B. atrophaeus var. globigii 1942 was 4,168,266 bp, and annotation using RAST [22] revealed 4433 features, including 4343 proteincoding sequencesand 90 RNA molecules [28] . The preliminary annotations derived from RAST are available as Genbank .gbk files in the supplementary material. On average, the genome of B. atrophaeus is approximately 86% identical to B. subtilis on the nucleotide level,supporting its delineation as a distinct species and agreeing well with previous estimates [29] . Analysis of the IBGD using whole-genome sequences (N-mer length .4) supported the identification of B. subtilis 168 as the closest relative among sequenced bacterial genomes ( Figure 3 ). For this particular case, n = 5 (i.e., there were 4 5 = 1024 total 5-mers used to compute the IBGD). The IBGD values were relatively insensitive to the choice of n over the range of 4-8. Thethree BGgenomes analyzed grouped closely together, and our analysis of the Bacillus-wide phylogeny using IBGD revealed the phylogenetic distance of that B. subtilis/B. atrophaeus species from B. anthracis, supporting the inferences published elsewhere from rRNA sequence analysis ( Figure 3 ) [30] . Primary amino acid sequences of RAST-annotated proteins are on average 72% (median 83%) identical between B. atrophaeus and B. subtilis. When only the proteins that yielded bidirectional BLAST hits in RAST are examined, the predicted proteome of B. atrophaeus is, on average, 83% identical (86% median) to B. subtilis. We utilized the finished sequence of the 1942 isolate as a reference strain for templated assembly of the remaining BG draft sequences. Two additional ATCC isolates of B. atrophaeus (49337 and 6537) were distinguishable from var. globigii on the basis of very high SNP/indel counts, lower coverage ofand percentage of reads mapping to the 1942 reference, and unique genomic features which supported their proposed classification as var.atrophaeus [14] . The distinguishing genomic features of var.atrophaeus strains and the delineation of the B. atrophaeus clade from B. subtilis will be published elsewhere. Optical restriction mapping [31, 32, 33] was used to compare the overall genomic structure of selected isolates. No differencesin overall genome architecture between the ''military'' BG isolates, the archival 1942 isolate, or 1013-1 were observed (Figure 4 , data not shown), suggesting that the global architecture of these isolates is relatively stable, even over 30 years of serial in vitro passage.However, the optical maps and sequence coverage analysis of 1013-2 and 9372-1revealed substantial deletions of approximately 72,727and 23,678 bases, respectively, of genomic materialspanning from positions 3,992,613 to 4,065,341 (1013-2) or 4,022,138-4,045,817 (ATCC 9372-1) (Figure 4 ; Table S2 ). The genes within this deleted region are listed in Table S3 but notably contain genes encoding for nitrite reduction, germination (gerKABC), and biosynthesis of the lipopeptide surfactin (srfCAB) [34, 35] . A defect in surfactin production is a particularly intriguing candidate for the morphology and pigmentation variations in 1013-2 and ATCC 9372-1, since disruption of srfA has been shown to have dramatic effects on spreading motility on semisolid media, on biofilm formation [34, 36] , and low-grade hemolytic activity. Using the de novo assembled draft sequence from the 1942 isolate as a template for subsequent analysis of SNPs and small indels in the other ''military'' isolates, we generated a list of highconfidence, discriminatorymutations that differentiate the strains ( Figure 5A ). The nature and annotation of the mutations are found in Table 4 and can be assigned an approximate temporal order in which they occurred ( Figure 5B ). Based on this analysis, 1942 is the most likely parental strain for all of the isolates in this study, with the 1013 lineage diverging earliest, followed by 49822, then the ''military'' lineage prior to the transfer to Camp Detrick. This conclusion is based on the observation that 49822 shares three SNPs with Detrick-1. The latter is the most likely progenitor of the other ''military'' isolates, since it has the fewest mutations relative to strain 1942. Detrick-1 can be differentiated from other ''military'' isolates by possessing the parental allele of spo0F rather than the H101R allele (position 3231470) that is characteristic of all of the other ''military'' BGisolatesand the ATCC9372 strains. The two colony morphology variants of ATCC9372 each exhibited distinct mutation profiles indicating that the reference strain is in fact a mixed population of at least two genetically distinct substrains. The 72 kb deletion in 1013-2 included the structural genes for biosynthesis of surfactin, a cyclic lipopeptide with a mild hemolytic activity [34] . To test whether the ''military'' and in vitro passaged strains possessed low-grade hemolytic activity, we streaked these variants on rich agar media containing 5% sheep's blood and looked for hemolysis. To our surprise, all strains exhibited striking variation in their coloration (Figure 2A) , with the 1942, 9372-1 and Detrick-1 isolates considerably darker on blood agar than the other ''military'' and in vitro passaged isolates. In addition, on LB the 1013-2 and 9372-1 isolates appeared white and off-white, respectively. Pigmentation of B. subtilis colonies is associated with production of a melanin-like pigment by the CotA protein, a major component of the spore coat [37] . In addition to the variations in pigmentation, streaks of the 1942 and Detrick-1 isolates were consistently translucent under transillumination ( Figure 2B ). These zones of translucency are suggestive of weak b-hemolysis, which has previously been observed in B. subtilis strains that produce high levels of surfactin [34, 35, 38] . The other strains exhibited either weak a-hemolysis or none at all, with the exception of the strongly hemolytic 49822-1 variant. At least in the ''military'' lineage, the quasi-hemolytic phenotype and darkbrown colony pigmentation correlated with the presence of a wildtype spo0F allele, suggesting that the ability of B. atrophaeus to lyse red blood cells may be regulated in part by spo0F. However this was not universally the case; the BACI051 strain had two discernible variants on SBA (not shown), one of which appeared to have recovered partial hemolytic activity ( Figure 2B ). To gain insight into the effects of genetic divergence of the adapted isolates on their metabolic capacity, the Detrick isolates and the separate 1013 isolates were compared by multiphenotype analysis using the Omnilog system, which allows the highthroughput comparison of 96620 discrete growth conditions, including carbon, nitrogen, phosphorus, sulfate, nutrient supplements, pH, osmolytes as well as a broad class of growth inhibitors. The growth of the 1942 strain was used as a reference for determining relative growth rates of the other strains. The results of Mutations exhibiting high quality scores in both reference and query sequences and with differences from the template exhibited in .85% of the individual sequencing reads are indicated as a blackened box. In one case (position 259001 in ATCC 9372-1) an initial false-negative due to the formation of a homopolymeric tract was found by direct inspection of the assemblies. The genes whose functions are altered by the given mutation are indicated in Table 4 these experiments are summarized in Figure 6 and Table S4 . In general, growth of the 1013 isolates was significantly diminished relative to the 1942 in many different growth conditions, most notably in the ability to use amino acids and peptides as carbon and nitrogen sources, to withstand osmotic stress, and to grow under reduced pH. In addition, the strains had developed sensitivity to beta-lactams, quinolones, and membrane-disrupting activities. These results suggested broad combined effects of several mutations on the phenotype of the strains. In addition to the spo0F(A98P) allele, which is a likely candidate for highly pleiotropic effects on the decision to sporulate under many different conditions, both strains contain substitutions in the yetF and yqgE genes that may be contributing to the phenotypes observed. The more pronounced defect in 1013-2 may be attributable to defects in the gerAB and gerAC genes and the large 72 kb deletion which contains several genes involved in germination. By contrast, the Detrick isolates in general grew more robustly than the 1942 strain under multiple growth conditions. Increased relative growth rates were particularly pronounced for acidic conditions and media containing osmolytes, but particularly for wells containing sodium lactate [6] . Another isolate in the ''military'' lineage, Dugway, is clearly derived from the Detrick lineage by SNP/indel profiling yet has a metabolic profile that is much closer to the parental strain. Like the Detrick isolates, the Dugway strain grows better at low pH, but many of the other conditions do not promote elevated growth relative to 1942. Only one mutation differentiates that isolate from the Detrick-2 isolate -a 2-bp insertion in the yojO gene encoding a putative activator of nitric oxide (NO) synthesis. Again, the physiological role of this mutation is unclear, although nitric oxide synthesis plays a critical role in modulating antibiotic resistance in Bacillus spp. [39] . In addition to its role in promoting resistance to antibacterial drugs, NO is known to modulate B. subtilis genes involved in nitrate respiration when oxygen is limited [40] ; thus the lowered growth in this strain may reflect the inability to grow to higher densities and overcome the resulting lower oxygen tension. An additional isolate, BACI051 is clearly derived from Dugway, yet two variants have accumulated additional mutations in sigH (spo0H), hpr/scoC, and ebrB. Notably, the phenotype of BACI051-E on plates more closely resembles the 1942 strain ( Figure 2B ). sequencing of the ''military'' isolates revealed a frameshift mutation in the katA gene encoding the major vegetative catalase [41] . The absence of catalase activity in ''military'' isolates was confirmed by adding a solution of 3% H 2 O 2 to smears of various strains. In contrast to the 1942 strain, which exhibited immediate and robust catalase activity, the strains containing the frameshift lacked this activity. A small amount of bubbling could be seen, probably due to the presence of a second catalase normally packaged in spores [42] . To test whether the phenotype observed on blood agar was associated with differences in sporulation, selected strains were grown for two days as patches on blood agar, resuspended in PBS Annotations are a combination of RAST and directed tBLASTn and BLASTp searches vs Bacillus databases. 4 Forms part of a large polypeptide synthase containing highly homologous regions. 5 Also shared with strain ATCC 49822. 6 The conserved start codon of the radA gene (BG3899) of B. atrophaeus and B. subtilis falls within the BG3898 ORF. Therefore BG3898 as called by RAST is not deemed likely to be a protein-coding gene. 7 In putative transmembrane region. doi:10.1371/journal.pone.0017836.t004 Table 4 . Cont. and counted directly. Strain Detrick-2exhibited significantly higher percentages of phase-bright spores than the Detrick-1 strain (Figure 7 , Mean +/2 standard error of the mean). Similar results were observed for the 1942 and Dugway strains (not shown). The 1013-1 strain exhibited an even higher degree of sporulation than the Detrick-1 strain under identical conditions ( Figure 7 ). Bacillus atrophaeus has historically been grouped with B. subtilis, and is usually described as a black-pigmented variant (var. niger) because of its many phenotypic similarities to the bettercharacterized B. subtilis. Both organisms are soil-dwelling, nonpathogenic saprophytes, but have been differentiated by the ability to produce pigment on nutrient media containing an organic nitrogen source [13] . The orange pigmentation of B. atrophaeus var.globigii spores made it an attractive simulant for B. anthracis, facilitating the detection of dispersed spores in complex environmental samples. Recently, more sensitive phylogenetic approaches using AFLP have delineated B. atrophaeus as a separate species [13, 14] . The taxonomic confusion has arisen due to inadequately sensitive typing methods, and has led to misattribution of pathogenic qualities associated with some B. licheniformis strains to the B. atrophaeus strains currently in use as simulants [12] , for which no direct evidence of pathogenicity exists. This report defines the genomic composition of B. atrophaeus var.globigii and clearly separates the species by wholegenome phylogenetic analysis. In this study, we generated a high-quality, closed reference genome for the 1942 isolate using a combination of 454, Illumina, and directed Sanger sequencing. We expect the final genome to have an error rate of ,1 in 50,000 basepairs. When we mapped the 454 datasets for all of the isolates back to the finished sequence that was generated using the same DNA, we noted several putative SNPs that were common to all datasets (Table 4) . We believe these represent errors introduced during generation of the final consensus sequence, as they did not appear when the isolates were mapped against draft sequence generated exclusively using the 454 platform; these are currently being verified and the final sequence will be updated. Our sequences of multiple, closely related strains of this organism allow us to trace the derivation of the ''military'' BG isolates currently in use to a culture present at Camp Detrick during the 1940s and 1950s. The origin of ATCC 49822 is not as clear, but a publication from that era suggests a possible common origin at the University of Wisconsin [43] . While that strain is unlikely to be NRS-356 itself, given the presence of several strain-specific SNPs in our sequence, the SNPs common to both 49822 and the ''military'' lineage suggest a common ancestor that is not represented among the strains sequenced for this study. Given the lack of original records, it is unclear whether the NRS-356 variant in this study might have passed through Camp Detrick and been returned to the University of Wisconsin. However, given the date on the label and the general secrecy of operations at Camp Detrick during the Second World War [26] we consider this possibility unlikely. During development of BGas a simulant for B. anthracis, strains were selected that exhibited the most desirable characteristics, those being rapid growth, high spore yield, and experimental reproducibility. Without being aware of the nature of the genetic alterations in their ''optimized'' strains, BW workers at Camp Detrick selected a mutant that provided dramatically higher total and relative spore yields, and generated consistent experimental results [6] . These strains were adopted into the inventories of numerous biodefense laboratories and have been used for many Figure 7 . The spo0F(H101R) and spo0F(A98P) alleles are associated with hypersporulation. Phase-contrast microscopy of BG strains after two days of growth on SBA. Vegetative cells appear as phase-dark rods, while spores appear as round, phase-bright globules. The mean percentage sporulation of each strain in a representative experiment is given 6SEM. The experiment was repeated on three consecutive days; representative results of a single experiment are shown. Statistical significance was determined by mixed ANOVA (Tukey's method, p,0.05). doi:10.1371/journal.pone.0017836.g007 Figure 6 . Omnilog phenotypic arrays of B. atrophaeus subsp. globigii strains. Six strains were each inoculated into twenty 96-well Omnilog plates and grown at 37uC. Reduction of tetrazolium dye by respiring cells was measured every 15 minutes by optical density. Dye reduction relative to the 1942 strain is shown; the red ratio values indicate less respiration while the green ratio values indicate more respiration as compared to the 1942 strain. Individual arrays or strains are displayed in each of the six major columns labeled Detrick 1, Detrick 2, Detrick 3, 1013-1, 1013-2, and Dugway. A) Heat map of all conditions for each strain. Each of the twenty plates for each strain is represented by the notation PM01-PM20 (left-toright for each strain) along the x-axis. The rows represent the well position, and are denoted as A i to H i (i = 1 to 12) from the bottom to the top of the plot in each array along the y-axis. Each cell ratio value represents the average of two biological replicates for each strain. Plates PM01-PM10 contains single wells for each growth condition, while plates PM11-PM20 contain quadruplicate wells for each growth condition. Solid circle indicates wells containing sodium lactate; dotted circle indicates well containing L-serine at pH 4.5. The details of the 1920 growth conditions can be found in the first worksheet labeled ''All strain AUC data'' in Table S4 . B) Most significant phenotypes for each of the six test strains as compared to the 1942 strain. The phenotypes with statistically significant increases and/or the decreases in ratio values for each of the six strains are presented. For the 1013 isolates only the conditions giving the five largest changes are presented. The number in each color block indicates the ratio for the test strain relative to the parent strain for the phenotype specified. The details of all significant phenotypes for each test strain can be obtained in Table S4 . Bold Italic font indicates p,0.05. doi:10.1371/journal.pone.0017836.g006 decades in simulations of decontamination and dispersal [12] . By applying a combination of genomic and biochemical profiling techniques, our data demonstrate that the BG isolates were ''enhanced'' by researchers at Camp Detrick during the development of the organism as a simulant. The selection of a strain with the desired properties appears to have occurred in at least two discrete steps, as shown by the genome sequences and metabolic profiles. The initial step appears to have been the adaptation of a strain to growth in corn steep liquor, an acidic medium rich in protein and lactate [44] . The robust growth of the Detrick strains relative to 1942 in low-pH medium containing high lactate levels is likely due to mutations in mmgD (2-methylcitrate synthase, position 2029530), or a short-chain 3-oxoacyl-[acyl-carrierprotein] reductase (position 3437350), or both. The most likely candidate for a mutation in the Detrick isolates that increases growth is the frameshift in mmgDthat occurred following the divergence from the 49822 lineage and results in an altered C-terminus ( Figure S1 ). The mmgD geneencodes a 2-methylcitrate synthase that is expressed in the mother cell at the intermediate stages of sporulation [45] . A null mutation in mmgD had no perceptible effect on sporulation, although other TCA-cycle enzymes when mutated led to a loss of sporulation [45] . The effects of the frameshift mutation on sporulation and cellular physiology on the function of the enzyme are not clear at this time. We speculate that the frameshift mutation alters the substrate specificity of MmgD in favor of citrate, thus increasing the flux of lactate-derived intermediates through the tricarboxylic acid cycle. Evidence for this possibility includes the observations that 2-methylcitrate synthases can have partial citrate synthase activity [45] and that the B. subtilis mmgD gene can complement a gltA (citrate synthase) mutant of E. coli [46] . Alternatively, alteration of function of mmgD may have predisposed the lactate-adapted strain to acquisition of a hypersporulating phenotype, which is not readily isolated or stable in B. subtilis (see below); however the presence of a hypersporulating phenotype in an independently evolved lineage (1013) of BG indicates that the species may have an intrinsic predisposition to evolving such a phenotype in vitro. The ''military'' strains also grow more readily on media containing D,L-diaminopimelic acid (meso-DAP), a major component of bacterial peptidoglycan. Corn steep liquor is derived from the incubation of corn in water at 42-55uC, during which a lactic fermentation by a community of wild organisms including numerous uncharacterized Bacillus spp. occurs. Total bacterial counts at the conclusion of CSL production can be quite high [44] , thus the availability of such compounds for growth is not surprising. Another potential source of meso-DAP could be bacterial autolysis during sporulation. The relative roles of each of the alleles in growth on lactate and/or meso-DAP is the subject of current investigation in our laboratory. The second step in the development of BG as a simulant appears to have been the deliberate selection of a hypersporulating variant [6, 47] . Importantly, the selection of a strain optimized for spore yield resulted in the fixation of a new spo0F allele that has no counterpart among the available spo0F sequences (Figure 8 ). The sole Spo0Fsequence that differs at position 101 is that of B. clausii, in which tyrosine replaces histidine. Notably, the spo0F(H101R) mutation is distinct from a separate spo0F(A98P) mutation present in the in vitro passaged 1013 isolates. Given that the amino acid sequence of B. atrophaeus Spo0F is identical to that of B. subtilis but for two conservative substitutions, it is likely to have very similar if not identical biochemical properties. Detrick-1 and 1942 likely represent one of the two R colony morphotypes described by Hayward et al. [6] , whereas the hypersporulating F morphotypes likely arose due to the emergence of the spo0F(H101R) mutation. However, the possibility that Detrick-1 represents a reversion mutant at this locus from Detrick-2 cannot formally be excluded, but since it represented the dominant morphotype in the 1952 Detrick vial we believe this is unlikely. The presence of the spo0F(H101R) allele in the ATCC 9372 strains suggests that these strains were acquired by ATCC after this mutation appeared within the Detrick lineage. Experiments to verify the roles of each allele in modulating sporulation are currently in progress. Preliminary results indicate that transformation of B. subtilis Dspo0F with B. atrophaeus DNA and selection of spo+ cells dramatically alters colony morphology independently of the spo0F allele introduced; additional studies to verify the effects of each allele are currently in progress (James Hoch, personal communication). The H101R and A98P allelesare likely to alter the response to signals promoting sporulation. Aspo0F(H101A) allele results in a sporulation-proficient strain that throws off sporulation-deficient papillae [48] , and the same mutation has been shown to suppress the spo 2 phenotype of a strain containing a defective kinA allele. H101 has been proposed as a potential metal-binding site with particular affinity for Cu 2+ [49] . Binding of Cu 2+ (or another divalent metal) at this site may modulate interaction with one or more sensor kinases that promote sporulation. Substitution of positively charged arginine at this position could potentially mimic the binding of a metal cation in the loop containing H101, resulting in altered sporulation of the strains due to a change in the interaction with the kinases governing sporulation. It is unclear why, given the proposed role of divalent Cu 2+ in suppressing sporulation, H101R would result in a hypersporulation phenotype. The mechanistic relationship between spo0F(H101R) and the hypersporulation phenotype will be tested in future experiments. Both variants in the 1013 lineage possess an A98P allele in spo0F. Although the presence of several other mutations within this lineage confounds the attribution of the hypersporulating phenotype to this allele at this time, the presence of a mutation in the same gene as another hypersporulating mutant is highly suggestive. The effect of proline substitution at position 98 on Spo0F functionis not immediately obvious, but the relatively inflexible proline residue can disrupt alpha-helices in protein structures. The 1013-1 lineage exhibits a hypersporulating phenotype even more pronounced than spo0F(H101R) strains in the ''military'' lineage. The observation that hypersporulating phenotypes have emerged during cultivationof two independent B. atrophaeus lineages point to the possibility that certain in vitro selection pressures may actually favor hypersporulating variants. The selection pressures acting on the sporulation pathwayare highlighted by the sheer number of mutations discovered within the entire data set that occur in proteins known to play roles in sporulation. Nine of the 38 mutations (23%) found in all lineages were in genes that directly or indirectly regulate either entry into stationary phase or sporulation; this number exceeds the number that would be expected if mutations were to occur by chance, since less than 5% of B. subtilis genes are dedicated to regulatory processes of any kind [50, 51] . In addition to the mutations found within the ''military'' lineage, the two variants of ATCC 49822 shown in Figure 2 differ by mutations in rpoB (Table S5) which also plays a role in entry into sporulation [52] . Null mutations in spo0F resulting in asporogenous phenotypes contribute to colony morphology variation in B. anthracis, B. thuringiensis and B. subtilis [53, 54, 55] . Enhanced in vitro ''fitness'' is also a likely driver behind the recovery of asporogenic B. anthracis mutants that were discovered during the investigation into the B. anthracis attacks of 2001 [56] . Because the process of sporulation is highly energy-intensive and irreversible once commenced, mutants that delay sporulation (or fail to sporulate altogether) to take advantage of remaining nutrients would out-compete wild-type cells during repeated passage in vitro in the absence of other selection pressures, as has been demonstrated in extended in vitro evolution studies with B. subtilis under relaxed sporulation conditions [57] . This may not be universally the case, since gain-of-function mutations in sporulation such as those observed in this studymay compete favorably with wild-type cells if cannibalism of vegetative cells by sporulating bacteria is the dominant selective pressure [58] . Finally, horizontally transferred genetic elements can have dramatic effects on sporulation: for example, recent studies of phage lysogeny in B. anthracis have revealed the ability of several integrated phages to positively affect the kinetics of sporulation upon lysogeny of commonly used B. anthracis strains [59] . This study identifies the spo0F(H101R) allele as the signature of a deliberate selection during the development of B. atrophaeus as a simulant. However, without the knowledge of the history and the analysis of the phenotypes of the strains originating from ''Camp Detrick'' as published in the open literature, attribution of this genotype to a deliberate selection event would not have been definitive, since a similar phenotype is observed in the 1013 lineage which to our knowledge was not deliberately selected for any specific trait. Any study designed to determine genomic ''signatures'' of deliberate enhancement or selection is likely to require an analysis of the baseline likelihood that mutations conferring a similar phenotype would emerge and become fixed by natural processes within an evolutionary timeframe consistent with a known time interval or number of passages. Available evidence suggests that hypersporulation is not easily evolved in vitro. Maughan and coworkers attempted to evolve populations of a laboratory strain of B. subtilis with a hypersporulating phenotype by repeatedly heat-shocking cultures. While their efforts to enrich for hypersporulators failed, other studies revealed that asporogenous mutants evolved readily [60, 61] , confirming many early studies ( [62] and references therein). With the exception of the studies by Maughan et al., most ofthese investigators applied selections intended to inhibit sporulation rather than to enrich for strains with elevated sporulation rates. The 1013 lineage was never heat-shocked during its many transfers; thus the adaptations seen in this work are the result of balancing sporulation versus vegetative growth for prolonged periods on agar slants. However, because undomesticated isolates were observed to sporulate to 98-100% [60] , we cannot formally exclude the possibility that in vitro culture of the 1942 strain following its isolation for an unknown period by the University of Wisconsin might have selected for a hyposporulating variant. In this scenario, the H101Rand A98P mutations would represent suppressor mutations. We consider this possibility unlikely, given the phenotypic similarity of two environmental isolates in the UW collection (1942 and NRS-356). Furthermore, a progression toward darker pigmentation and greater hemolysisis evident in the ''military'' lineage ( Figure 2B ). These phenotypic changes are associated with the accumulation of additional mutations including a P145L substitution mutation in sigH, a positive regulator of sporulation [63, 64] and an A13P mutation in scoC, a negative regulator of sporulation [65] . Together, the strains analyzed in this study suggest strong selective pressures on the genes in the sporulation pathway, and more carefully controlled studies should be carried out to determine the dynamics of in vitro evolution and adaptation of spore-forming organisms, as has been done extensively in E. coli [66, 67, 68, 69, 70] . Unexpectedly, the ''military'' lineages were also marked by the loss of catalase activity, whose presence is an identifying feature of both B. subtilis and B. atrophaeus [13] . This activity was present in a separate lineage of in vitro passaged organisms, so it is not immediately clear why ''military'' isolates, i.e. those subjected to selection within the early days of the development of BG as a simulant organism, would have lost the catalase activity characteristic of the parental isolate. Because the KatA gene product is not found in spores [41, 71] , we consider it unlikely that the absence of this activity would impact the resistance of spores to decontamination reagents, and thus any antioxidant resistance phenotype exhibited by spores of ''military'' isolates would likely have gone unnoticed. However, direct comparisons of the ''military'' B. atrophaeus lineages to the progenitor strains have not been done, and pleiotropic effects of a spo0F mutation on spore physiology cannot currently be excluded. Whole-genome approaches are becoming critical components of microbial forensics. The SNPs and indels identified in the analysis of evidentiary materials currently become the basis for higherthroughput assays to screen large numbers of samples [56, 72] . Decreasing costs of whole-genome sequencing, and the comprehensive nature of the analysis, may make this the preferred method of forensic analysis of microbial samples in the future. With recently developed techniques of allele quantitation within populations by mass spectrometry [73] , real-time PCR [74] , and census-bysequencing [68, 75] , it may be possible to quantitate accurately rare alleles within any given microbial population. We are particularly intrigued by the possibility that, given a mixture of different variants and sufficient sequencing power, ultra-high coverage sequencing may prove to be a more quantitative means of enumerating the relative populations in a sample even before the presence of variants has been established. The results from sequencing two strains of BACI051 in this study provide evidence of such hidden diversity. The genomic basis of interlaboratory strain variation is only beginning to become evident, with recent studies tracing the histories of commonly used lab strains of B. subtilis 168, E. coli, Salmonella enterica serovar Typhimurium 14028s, Pseudomonas aerugi-nosaPA01 and Mycobacterium tuberculosis H37Rv [76, 77, 78, 79, 80, 81] . These have revealed significant divergence of putatively identical strains from one laboratory to another, largely arising from mutations that accumulate during serial passage. Like the earlier work, our study highlights the utility of approaches based on wholegenome sequencing for the discrimination of closely related strains, especially when investigating the provenance for a given isolate. Tragically, at least 13 institutions are known to have destroyed archival collections of Select Agents [82] following the implementation of mandatory monitoring and reporting requirements, representing an incalculable loss of phenotypic and genomic diversity. This report underscores the importance of maintaining the genetic heritage preserved in the culture collections of individual investigators and institutions.

Differential Induction of Functional IgG Using the Plasmodium falciparum Placental Malaria Vaccine Candidate VAR2CSA

BACKGROUND: In Plasmodium falciparum malaria endemic areas placental malaria (PM) is an important complication of malaria. The recurrence of malaria in primigravidae women irrespective of acquired protection during childhood is caused by the interaction between the parasite-expressed VAR2CSA antigen and chondroitin sulfate A (CSA) in the placental intervillous space and lack of protective antibodies. PM impairs fetal development mainly by excessive inflammation processes. After infections during pregnancy women acquire immunity to PM conferred by antibodies against VAR2CSA. Ideally, a vaccine against PM will induce antibody-mediated immune responses that block the adhesion of infected erythrocytes (IE) in the placenta. PRINCIPAL FINDINGS: We have previously shown that antibodies raised in rat against individual domains of VAR2CSA can block IE binding to CSA. In this study we have immunized mice, rats and rabbits with each individual domain and the full-length protein corresponding to the FCR3 VAR2CSA variant. We found there is an inherently higher immunogenicity of C-terminal domains compared to N-terminally located domains. This was irrespective of whether antibodies were induced against single domains or the full-length protein. Species-specific antibody responses were also found, these were mainly directed against single domains and not the full-length VAR2CSA protein. CONCLUSIONS/SIGNIFICANCE: Binding inhibitory antibodies appeared to be against conformational B-cell epitopes. Non-binding inhibitory antibodies reacted highly against the C-terminal end of the VAR2CSA molecule especially the highly polymorphic DBL6ε domain. Differential species-specific induction of antibody responses may allow for more direct analysis of functional versus non-functional B-cell epitopes.

Animal models are required for preclinical development of new generation vaccines against infectious diseases [1] [2] [3] . The ideal animal model mimics the human immunological response, the pathogen infection pathway, and allows analysis of the mechanism of the vaccine-induced protective immune response. However, despite the availability of humanised animal models such as transgenic mice [4] , immunological responses in most animal models are only indicative of what to expect in humans. The development of a recombinant vaccine is initiated with identification and selection of an antigen that induces a desired immune response. At this step, multiple antigens are tested, which requires an inexpensive and easy to handle animal model. Following selection of antigen, the optimal route of vaccine administration together with a proper adjuvant formulation has to be evaluated. The design of the individual vaccine components and their delivery thus relies on the performance in these pre-clinical animal tests, emphasizing the importance of the animal model. Vaccines are one of the future strategies to prevent and control malaria [5] , one of the most widespread, pathogenic and deadly parasitic diseases in the world. Immunity is only acquired after successive infections in areas of high and stable malaria transmission [6] . Pregnant women become susceptible to placental malaria (PM) independent of any pre-existing immunity acquired during childhood. Placental malaria can have serious consequences for both mother and child such as maternal and infant anaemia, premature labour, low birth weight, and increased neonatal mortality [7] . However, after successive pregnancies women rapidly acquire immunity to PM, indicating that a vaccine strategy against PM may be feasible [8] . PM is caused by the binding of Plasmodium falciparum infected erythrocytes (IE) to chondroitin sulfate A (CSA) present on placental syncytiotrophoblast cells located in the intervillous space [9, 10] . Placental parasites express on the IE surface VAR2CSA, which is a Plasmodium falciparum Erythrocyte Membrane Protein 1 (PfEMP1) that mediates binding of the IE in the placenta [11, 12] . VAR2CSA is a large (350 kDa) polymorphic protein with six Duffy-binding-like (DBL) domains, which complicates the development of a vaccine. A potential animal model for studying protection induced by immunizations with recombinant domains of PfEMP1 could be chimpanzees (Pan troglodytes), whose natural malaria parasite, Plasmodium reichenowi, is closely related to P. falciparum [13] . However, for ethical and economic reasons this is not a feasible model for malaria vaccinology. To date, most malaria antigens are produced by recombinant technology and tested in lower mammalian species, such as rodents and rabbits. These models have allowed analysis of antibody responses to different combinations and boundaries of the six DBL-domains of VAR2CSA. Initially, we expressed all DBL-domains fromVAR2CSA-3D7 and found that all domains except DBL4e, could induce antibodies against native VAR2CSA expressed on the surface of IE, with limited differences in immunizations of mice and rabbits [14] . However, these VAR2CSA-3D7 specific sera did not show significant inhibition of IE binding to CSA (unpublished data). Recently, in a large screening of the immunogenicity of different VAR2CSA antigens from the parasite line FCR3 using rat immunizations, we found that the DBL4e domain of VAR2CSA could elicit the desired adhesion-inhibitory antibodies [15] . However, the induction of functional inhibitory antibodies to certain DBL-domains is poorly reproducible and the immune response tends to focus on non-adhesion blocking epitopes [16, 17] . In addition higher levels of inhibitory antibodies are acquired using full-length VAR2CSA (FV2) as compared to single DBLdomains [18] . To further investigate the induction of adhesion blocking antibody responses, all single recombinant domains of VAR2CSA-FCR3 were used in immunizations of mice, rats and rabbits, and compared to responses raised against the full length recombinant protein. To examine the antigenicity of the recombinant proteins the levels of antigen-specific IgG from immunized animals were measured using ELISA. All animal species immunized with individual DBL-domains or FV2 produced an IgG response, with a typical sigmoid shaped dose response curve of the antibody titrations ( Figure 1 ). However, the levels of antibodies to each of the DBL-domains as well as to FV2 differed depending on the animal species used (for all statistical analysis of antigenic differences see Table S2 ). In mice, the DBL6e immunizations resulted in the most potent IgG response with a significantly higher titre than that of FV2 and the other DBL-domains. Following DBL6e in potency was DBL5e and thereafter a lower response induced by DBL1X, DBL2X, DBL3X, DBL4e and FV2 of similar potency. As observed in mice, immunizations of rats with DBL6e resulted in a significantly higher IgG response compared to all other immunogens. DBL5e and FV2 induced in rats a more potent response than DBL1X, DBL2X, and DBL3X ( Figure 1B ) but also DBL4e induced a high antibody response. In rabbits, the seven tested antigens produced IgG responses that can be grouped in two according to the levels of responses ( Figure 1C ). The DBL6e, DBL5e and FV2 produced a high antibody response whereas IgG responses to DBL1X, DBL2X, DBL3X and DBL4e domains were lower and essentially similar. In summary, the tested VAR2CSA derived recombinant proteins induced medium to high levels of antibodies. FV2, DBL6e and DBL5e immunizations resulted in the highest antibody titers. FV2 appeared more antigenic in rabbits and rats compared to mice, and only rats appeared to induce high levels of antibodies to DBL4e. Antibody reactivity against native VAR2CSA protein expressed on the surface of IE Epitopes exposed in the recombinant single-domain proteins may not be surface exposed in the native full-length protein, primarily due to the proposed globular folding of the VAR2CSA molecule [19] . Therefore we tested the reactivity of the induced antibodies against native VAR2CSA exposed on the surface of the IE. Erythrocytes infected with the FCR3 parasite line were used in flow-cytometry to test the IgG reactivity in animal sera against native VAR2CSA. The parasite cultures were continuously selected for binding to CSA by panning on BeWo cells, resulting in surface expression of VAR2CSA and sex specific recognition by IgG from human female serum, compared to male and Danish serum (data not shown). In line with the ELISA results measuring the reactivity against the recombinant proteins, we found that the antibody reactivity against native VAR2CSA on the surface of the IE differed both with respect to animal species, and with antigen type. For mice and rabbits (Figure 2A and Figure 2C ), the reactivity against native VAR2CSA on the IE surface was higher in sera from animals immunized with DBL5e and DBL6e, the reactivity against DBL1X, DBL2X, DBL3X and DBL4e was low. In general, it appeared that rats acquired higher levels of antibodies specific to the native protein compared to mice and rabbits, the difference being especially marked for DBL4e. However, as for mice and rabbits, the FV2, DBL5e and DBL6e immunizations resulted in high specific reactivity towards native VAR2CSA expressed on the IE ( Figure 2B ). To test whether the reactivity against the native protein on the IE surface was associated with the apparent antigenicity of the recombinant domains, as measured by ELISA, the mean fluorescence values (MFI) was correlated to the EC50 titer values for all single domains. There was a highly significant linear correlation for mice and rabbits (data not shown) (r = 0.847, P,0.0001 & r = 0.809, P = 0.0008, respectively, Pearson), whereas the linear dependency for rats was somewhat lower (data not shown) (r = 0.476, P = 0.01, Pearson). In summary, we found high levels of concordance between antibody reactivity against the recombinant domains and reactivity against the native protein in serum from rabbits and mice, but less so in serum from rats. The proposed mechanism of protection against PM is that antibodies block adhesion of IE to CSA in the placenta [8] . Therefore, the functional capacity of the induced antibodies was measured in inhibition of CSA-binding assays. IE were incubated on decorin, either with or without specific serum or purified IgG. The specific inhibition was compared to the binding of IE incubated with negative control samples. Only sera from rabbits and rats were attained in sufficient volumes allowing three independent CSA binding experiments. The inhibitory capacity of the sera was tested prior to IgG purification, since crude purification of IgG may introduce a bias due to differential efficacy of purification of different antibody isotypes. Only FV2 and DBL2X specific rabbit serum appeared to inhibit more than the negative control rabbit serum ( Figure 3A ). This was in contrast to the responses measured using FCM against the native protein in DBL5e and DBL6e specific rabbit sera ( Figure 2C ), where high reactivity was found. The inhibitory capacity of rat serum was also not determined by the surface reactivity, for instance there was prominent inhibition using DBL4e and FV2 specific serum and lack of inhibition by DBL6e specific antibodies ( Figure 3A ), although DBL6e specific antibodies also were highly reactive against native VAR2CSA ( Figure 2B ). To rule out that IgM or unspecific effects of serum did not mediate the inhibitory capacity, we purified IgG from both inhibitory and noninhibitory sera and tested them in inhibition of binding assays at 0.5 mg/ml. The level of inhibition was essentially sustained in the IgG preparations ( Figure 3B&C ). In summary, differential induction of functional antibody responses against DBL4e, but not to the FV2 protein, were demonstrated in rats compared to rabbits. To examine how the immune response against the full-length recombinant protein is focused with respect to individual DBL domains, we tested the levels of antibodies against individual DBLdomains in serum from animals immunized with FV2. In general, we found that the pattern of antigenicity when immunizing with FV2 was similar to immunizations with single domains (Figure 4 ). In rats, FV2 antibodies targeted all DBLdomains with similar reactivity, with exception of DBL5e and DBL2x, which showed highest and lowest reactivity, respectively. The levels of reactivity were more or less in line with the reactivity against the native protein of single-domain specific antibodies ( Figure 2B ). In rabbits, the reactivity against the domains appeared to be overall lower than for rats. Furthermore, the rabbit-FV2 reactivity was more focused towards DBL5e and DBL6e, as was also the case for the reactivity of single-domain specific antibodies towards the native erythrocyte surface exposed protein ( Figure 2C ). In summary, it appeared that the pattern of antigenicity found when immunizing with single domains was essentially similar to immunizations with the FV2 protein in the different animals. The DBL4e recombinant protein appeared previously to be the only single domain that induces highly inhibitory antibodies [15] . However, the inhibitory antibodies were not induced in rabbits. Therefore, we tested the specificity of the antibody response in rabbits and rats against linear epitopes using a DBL4e peptide ELISA. IgG induced by DBL4e immunizations reacted with the DBL4e peptides and no reactivity was detected with IgG from control IgG (data not shown). The overall level of reactivity was lower in rabbit serum (n = 4) compared to rat serum (n = 10) and a lower proportion of the rabbits reacted with a given peptide (data not shown). In general, the same areas of the linear sequence were targeted by IgG from the rats and rabbits, with the most distinct reactivity in both species of animals observed against peptide 2 to 5 ( Figure 5A ). However, it appeared that the rat antibody reactivity was skewed one to three peptides N-terminally compared to rabbit antibodies. Hence, the major peaks of reactivity recognized by rabbit-antibodies were peptides: 1-5, 14-19, 21-28, 31-33, 35-36 & 58-61, whereas rats recognized peptides 2-6, 14-20, 22-26, 31-38 & 60-63. Other peptides in the array were recognized by one species only, albeit some of them with quite low OD values. The major peak of reactivity (peptides 3-6) recognized by both rabbit-antibodies and rat-antibodies and one minor peak with distinct skewed reactivity (peptides [33] [34] [35] [36] [37] [38] were mapped on the structure model ( Figure 5B ). Both these areas of the model were predicted to contain unstructured loops. In addition, peptide 3-6 contained a b-sheet hairpin motif and peptide 33-38 contained several a-helixes. The other minor peak primarily targeted by rat antibodies (peptides 61-62) was also characterized by unstructured amino acid sequences (data not shown). In summary, the average reactivity to linear peptides covering the amino acid sequence of DBL4e was, apparently slightly skewed, but otherwise remarkably similar between rabbits and rats. The VAR2CSA protein is a vaccine candidate against PM due to its surface localisation, function as an adhesin for placental CSA and its role in immunity against PM [11, [20] [21] [22] [23] [24] [25] [26] [27] . Although the fulllength protein can be produced, a vaccine based on a single domain is more likely to be feasible, due to the complexity and size of the antigen. To analyze if the immune responses towards singledomains and the full-length protein were comparable, we tested all single VAR2CSA domains as well as the full-length protein in three different animal species. In general, there was a direct association between antigenicity and reactivity to the surface of the infected erythrocyte. The reactivity against native VAR2CSA on the IE surface was higher in sera from animals immunized with DBL5e, DBL6e and FV2 compared to other domains (Figure 2) , which was similar to the observed reactivity against the recombinant domains measured by ELISA (Figure 1 ). The correlations between the antigenicity of the domains and reactivity to the native protein on the surface of IE were highly significant in immunization of mice and rabbits, whereas the levels of rat antibodies reacting with the native protein in general were higher, which resulted in a less significant correlation. However, taken together the reactivity increased towards the C-terminal end of the external part of VAR2CSA, both in immunizations with single domains and FV2. This may reflect both an inherent antigenicity of single DBL domains of VAR2CSA and/or the accessibility of single domains in the quaternary structure of the native VAR2CSA molecule. This is in line with data showing that the functional CSA binding region is located N-terminally allowing for non-blocking antibodies to react with the C-terminal located domains (Dahlbä ck et al submitted) . The inhibitory capacities of the induced rat and rabbit antibodies following FV2 immunizations were very high. However, the different species-specific inhibitory FV2 antibodies do not necessarily target the same epitopes, but further analysis of the epitope targets of FV2 specific antibodies are needed to elucidate the mechanism of inhibition. For immunizations with single domains, the inhibition of IE binding to CSA varied between animal species. Only rats immunized with DBL4e induced IgG, which efficiently inhibited IE binding, whereas the rabbit sera specific against DBL4e were non-inhibitory. The inhibitory effect of the IgG was not an effect of antibody titer, since immunizations with DBL5e and DBL6e also induced high titers, but these were in general non-inhibitory. These findings are in contrast to what others have found [21, 28] . Differences in observations could be due to different domain boundaries or to the variance in posttranslational protein modifications introduced for example by the baculovirus transfected T. ni as opposed to the Pichia pastoris expression system, such as differences in glycosylation pattern. The above findings are not likely to be caused by differences in the MHC repertoire of the rat and rabbit strain used, since both strains used were outbred, reducing the likelihood that all of the individual rats and none of the rabbits induced functional antibodies. In addition, fair levels of antibodies was induced against the recombinant form of DBL4e in rabbits, indicating appropriate T-cell responses during the immunizations, yet the DBL4e specific rabbit sera was non-inhibitory even at rather high concentrations. Differential induction of binding-inhibitory antibodies is probably caused by species-specific differences in immuno-dominance of Bcell epitopes in different species of animals, similar to observations with antigens from other organisms [29, 30] . Responses to FV2 appeared more homogenous between species, at least with respect to the adhesion blocking properties, possibly due to presentation of more epitopes in a native form compared to single domains. Since the VAR2CSA molecule appears partly globular, immunodominant B-cell epitopes, not present in the native protein, may be exposed in individual DBL domains. This could also be the cause of the finding that antibodies induced against the DBL1X and DBL3X single domains, when based on heterologous sequences, can inhibit the binding of FCR3-IE [17] , as opposed to antibodies induced against DBL1X and DBL3X, based on the FCR3 sequence [15] . Attempting to identify epitopes targeted by inhibitory antibodies, we compared differences in rabbit and rat antibody reactivity to linear peptides covering the entire DBL4e domain. We previously did this with DBL4e-FCR3, DBL4e-3D7, DBL1X-3D7 and DBL3X-HB3 specific sera from rats [17] . The average reactivity to DBL4e linear epitopes was apparently slightly skewed when comparing rat and rabbit antibodies. Otherwise, the majority of responses in rats and rabbits were surprisingly similar. This similarity, combined with the lack of functional activity of the rabbit antibodies, both with regard to surface activity and inhibitory capacity, probably shows that surface-reactive and binding-inhibitory rat antibodies are not targeting linear epitopes in the native protein. The most pronounced areas of reactivity against the linear sequence were mapped to regions in the DBL4e model containing unstructured loops. This is in line with the findings that patches of high sequence diversity in VAR2CSA are concentrated in flexible loops [20, 31, 32] and that naturally acquired antibodies also are directed against these regions [33] . However, partly denatured antigen induced by Freund's adjuvant could increase the response to flexible loop regions in experimental immunizations [34] . Interestingly, a fraction of the expressed PfEMP1 during the maturation of the parasite in the IE remains intracellular [35] . Whether this intracellular pool of protein is denatured at the point of schizont rupture, promoting responses in natural infections to flexible loops more tolerant to mutations, as seen in other organisms [36] , remains to be investigated. It should be noted that DBL4e and DBL4e-ID4 specific antibodies are cross inhibitory on a panel of maternal parasite isolates [37] , showing that a conserved region of this recombinant protein can be targeted by antibodies interfering with the interaction between VAR2CSA and CSA. We have not analyzed the cross-reactivity of the FV2 induced IgG, consequently it is unknown if the inhibitory response is directed towards conserved or variable regions. It is possible that selection pressure can increase exposure of immuno-dominant variable epitopes targeted by non-inhibitory antibodies. In summary, it appears that the immunogenicity of the VAR2CSA protein is highest in the C terminal end, both immunizing with single domains and with the full-length protein. Identification of the epitopes targeted both by FV2 and DBL4e specific binding inhibitory antibodies may serve as a tool to generate second-generation antigens by removing or masking redundant/ unwanted epitopes. We do not know how the immune response will be focused in humans upon immunization with DBL4e and the results from a phase 1 clinical trial in human volunteers with DBL4e will be decisive for further clinical development. Retrospectively, we would probably not have identified the binding inhibitory capacity of DBL4e specific antibodies had we only tested the antigen in rabbits and mice. This emphasises the value of testing different animal species as vaccination models, specifically for antigens like PfEMP-1 and EBA, which have evolved to generate variations in humoral immune responses. Plasmodium falciparum culture P. falciparum, FCR3 parasite (laboratory strain) were cultured in RPMI-1640 supplemented with 25 mmol/L sodium bicarbonate (Sigma-Aldrich), 0.125 mg/ml gentamycin, 0.125 mg/ml Albumax II (Invitrogen), 2% normal human serum and with 5% hematocrit of group 0+ human blood. To select for VAR2CSA expression, IE were repeatedly panned on BeWo-cells as described [38] . All isolates were mycoplasma negative and were regularly genotyped using nested GLURP and MSP-2 primers in a single PCR step. All VAR2CSA single domains were cloned from genomic P. falciparum FCR3 parasite DNA (GenBank accession no. AY372123). Full length var2csa was based on a codon optimized synthetic DNA fragment as described [18] . Gene fragments were cloned into the Baculovirus vector, pAcGP67-A (BD Biosciences) modified to contain a V5 epitope upstream of a histidine tag in the C-terminal end of the constructs. Linearized Bakpak6 baculovirus DNA (BD Biosciences) was co-transfected with pAcGP67-A into Sf9 insect cells for generation of recombinant virus particles. Histidine-tagged recombinant protein was purified on Ni 2+ sepharose columns from the supernatant of baculovirus infected High-Five insect cells using an Ä KTA-express purification system (GE-Healthcare). The full-length protein VAR2CSA (FV2) and the six VAR2CSA DBL-domains were produced corresponding to the following amino acid number in The animals were subcutaneously immunized with either single VAR2CSA-domain constructs (DBL1x-6e) or with FV2. Control groups were immunized with the DBL3c-VAR1-FCR3 in the same concentration as the tested VAR2CSA proteins [15] . Groups of BALB/c mice (n = 8) (inbred) (Taconic, Denmark) were subcutaneously immunized with 15 mg per animal of recombinant protein in Freund's complete adjuvant, followed by three followup immunizations of 10 mg of protein in Freund's incomplete adjuvant at two-week intervals. However, due to insufficient amounts of protein only four mice received DBL2x, and six mice received DBL6x. For immunizations with FV2 (done on a later time point) the mouse strain used was CB6F1 mice (F1 hybrid between BALB/c females and C57BL/6 males) (Harlan, USA) following the same protocol as previously described for the BALBc mice. Seven groups of Wistar rats (outbred) (Taconic, Denmark) were immunized with 40 mg per animal of recombinant protein in Freund's complete adjuvant, followed by three booster injections of 20 mg of protein in Freund's incomplete adjuvant at 2K-week intervals. The rat groups consisted of three animals except the group receiving the DBL4e (n = 8) and DBL3c (n = 2). New Zealand White rabbits (outbred)(HB Lidköping Kaninfarm, Sweden) were immunized with 40 mg of recombinant protein in Freund's complete adjuvant, followed by three booster immunizations of 20 mg of protein in Freund's incomplete adjuvant at 3 week intervals. For each recombinant protein one rabbit was immunized with the exception of rabbits receiving DBL2x & FV2 (n = 2) and DBL4e & DBL5e (n = 3). For all animals sera were collected 8 days after the final immunization. Sera were stored at 220uC until use. ELISA plates (Nunc Immunoplate) were coated with protein overnight (4uC) at a concentration of 1 mg/ml in 16 PBS. The plates were blocked with ELISA dilution buffer (1% BSA and 1% Triton x-100) before the serial diluted serum was added to the plate. Secondary horseradish peroxidase-conjugated (HRP) antibodies added in the following dilutions: anti-rabbit-IgG 1:2000 (P0448, Dako), anti-mouse-IgG 1:2000 (P0260, Dako), and antirat-IgG 1:3000 (A9037, Sigma). The optical density was measured at 490 nm in an ELISA plate reader (VersaMax, Molecular Devices). Antibody titration curves were determined using GraphPad Prisma5. The FV2-specific IgG from both rats and rabbit's sera were affinity purified on HiTRap TM NHS-activated HP columns (GE Healthcare), according to the manufacturer's instructions. Briefly, each column was coated with 0.5 mg/ml of a DBL-domain resulting in six different columns for the individual six DBLdomains. The affinity purified IgG was verified using ELISA as previously described in this material and methods section. The assay was performed twice with similar results. The reactivity of IgG in animal sera to the VAR2CSA protein expressed on IE surface was measured by flow-cytometry (FCM) using a protocol modified from [39] . Briefly, the parasite culture was enriched for late trophozoite and schizont stage parasites in a strong magnetic field (MACS, Miltenyi). Aliquots of 2610 5 IE in a total volume of 100 ml were labelled by ethidium bromide and sequentially exposed to (1) 15 ml animal serum and (2) 1:100 dilutions of FITC labelled secondary antibodies specific for IgG from the individual species: Mouse (FL2000, Vector); rat (62-9511, Invitrogen) and rabbit (FL1000, Vector). As a negative control, IE were incubated both with sera from non-immunized animals and without animal serum and then exposed to secondary antibodies specific against IgG from different animals spec. Data from 5000 infected cells (ethidium bromide positive) was collected on a FC500 flowcytometer (Beckman Coulter). The median FITC fluorescence intensity was determined using Winlist Software (Verity Software House). For analyses of the capacity of serum to inhibit binding we used 2610 5 tritium-labeled late-stage IE and 15 ml serum in a total volume of 120 ml which were added in triplicates to wells coated with 2 mg/ml of the commercially available chondroitin sulfate proteoglycan Decorin (D8428; Sigma-Aldrich). Decorin was used as a source of CSA as this has a protein core resulting in more efficient coating to plastic than CSA. After incubation for 90 min at 37uC, unbound IE were washed away by resuspension performed by a pipetting robot (Beckman Coulter). The proportion of adhering IE was determined by liquid scintillation counting on a Topcount NXT (Perkin-Elmer). The inhibitory capacity of the induced VAR2CSA domainspecific IgG was measured using a standardized Petri dish method both due to the smaller volume in this assay, and because this a more commonly used assay [40] [41] [42] . IgG was purified from the animal sera using HiTrap Protein G HP kit (GE Healthcare) and dialyzed against PBS 16 buffer. Briefly, 20 spots in each Petri dish (Falcon 351005) were coated overnight with 20 ml of decorin (2 mg/ml in PBS). Spots were blocked with 3% bovine serum albumin. Parasite cultures were enriched for late trophozoite and schizont stages in a strong magnetic field (MACS, Miltenyi) and adjusted to reach a 20% parasitemia at 0.5% hematocrit. The parasite suspension was pre-incubated with both animal sera and purified IgG at different concentrations at 37uC for 30 min. Duplicate spots were incubated for 15 min at 37uC in a humid chamber. Unbound erythrocytes were removed by washing with PBS on a gyro-turntable (Stuart) until spots were clear of blood. Bound erythrocytes were glutaraldehyde-fixed (1.5%, 5 min) and Giemsa stained. Five fields were counted in each spot using a magnification of 206 using a Leica Microscope. As negative controls, anti-DBL3cVAR1 serum and IgG were used. Parasite binding to decorin was abrogated by soluble CSA (Sigma-Aldrich) and by chondroitinase (Sigma-Aldrich) treatment of decorin (data not shown). Epitope mapping was done using a peptide array covering the DBL4e domain. The peptide array consisted of 66 synthetic 16-28 amino acids long overlapping peptides, spanning amino acids 1607-1989 of the P. falciparum-FCR3 var2csa sequence (Table S1 ). The peptides were prepared by Schafer-N Copenhagen and the purity of the peptides was expected to be 70% or higher. Peptide binding was determined using ELISA. ELISA plates (Nunc Immunoplate) were coated overnight (4uC) with DBL4e peptides at a concentration of 3 mg/ml in 16PBS. Plates were blocked with ELISA dilution buffer (1% BSA and 1% Triton x-100) before animal serum samples were added. The serum samples were diluted 1:100. The following secondary horseradish peroxidaseconjugated (HRP) antibodies were used: anti-rabbit P0448 (DAKO), dilution 1:2000; and anti-rat A9037 (Invitrogen) and diluted 1:3000. The optical density was measured at 490 nm in an ELISA plate reader (VersaMax, Molecular Devices). The level of reactivity in this ELISA based peptide array was directly associated to the level of reactivity in an assay where identical peptides were synthesised directly on to a chip, confirming homogeneous coating of the peptides [17] . The structure model of the DBL4e sequence (C1576-C1910) was created using the same method as described for other VAR2CSA DBL-domains [20] , but using as a template the newly published structure of DBL3X-VAR2CSA (3BQK) [43] . Best-fit curves and mean EC50 values to determine compare antibody titres were done using Gradhpad Prism5 (Gradhpad Software Inc.). Sigma Stat 3.11 (Sysstat software Inc.) was used in the following analysis: one way ANOVA tests followed by multiple pair wise comparison procedures (Holm-Sidak method), which were used to determine differences between Log EC50 values to measure the antigenicity of single domains and FV2; Pearson correlation test, which was used to assess the correlation between MFI values and OD-EC50 values.

Severe pneumococcal pneumonia: impact of new quinolones on prognosis

BACKGROUND: Most guidelines have been proposing, for more than 15 years, a β-lactam combined with either a quinolone or a macrolide as empirical, first-line therapy of severe community acquired pneumonia (CAP) requiring ICU admission. Our goal was to evaluate the outcome of patients with severe CAP, focusing on the impact of new rather than old fluoroquinolones combined with β-lactam in the empirical antimicrobial treatments. METHODS: Retrospective study of consecutive patients admitted in a 16-bed general intensive care unit (ICU), between January 1996 and January 2009, for severe (Pneumonia Severity Index > or = 4) community-acquired pneumonia due to non penicillin-resistant Streptococcus pneumoniae and treated with a β-lactam combined with a fluoroquinolone. RESULTS: We included 70 patients of whom 38 received a β-lactam combined with ofloxacin or ciprofloxacin and 32 combined with levofloxacin. Twenty six patients (37.1%) died in the ICU. Three independent factors associated with decreased survival in ICU were identified: septic shock on ICU admission (AOR = 10.6; 95% CI 2.87-39.3; p = 0.0004), age > 70 yrs. (AOR = 4.88; 95% CI 1.41-16.9; p = 0.01) and initial treatment with a β-lactam combined with ofloxacin or ciprofloxacin (AOR = 4.1; 95% CI 1.13-15.13; p = 0.03). CONCLUSION: Our results suggest that, when combined to a β-lactam, levofloxacin is associated with lower mortality than ofloxacin or ciprofloxacin in severe pneumococcal community-acquired pneumonia.

Streptococcus pneumoniae is the leading causative agent of community-acquired pneumonia (CAP). Despite new antimicrobial agents and advances in supportive measures, attributable mortality linked to pneumococcal pneumonia remains unchanged and dramatically high when patient are admitted in intensive care units (ICU) [1] . Most guidelines have been proposing, for more than 15 years, a combination of a β-lactam with either a quinolone or a macrolide as empirical, first-line therapy of severe CAP requiring ICU admission [2] [3] [4] [5] [6] [7] [8] . Although a recent study demonstrated combination antibiotic therapy to be associated with a higher survival rate than monotherapy in patients with severe CAP and shock [9] , the rationale for this combination was not to increase efficacy but rather to routinely provide coverage of all common pathogens causing severe CAP and particularly, S. pneumoniae and Legionella species. In our ICU, we followed until 2003 the 1991 French recommendations [2] . Most patients received an empirical therapy based on a β-lactam-fluoroquinolone combination. Before 2003, fluoroquinolones used were ofloxacin and ciprofloxacin. Levofloxacin replaced these quinolones since its 2003 addition to the hospital formulary. Such a replacement was comforted by the ERS, French and IDSA guidelines published between 2005 and 2007 [6] [7] [8] . We wished to determine outcomes of patients treated with a combination of β-lactam plus fluoroquinolone for severe pneumococcal pneumonia. This homogenous modification of severe CAP antibiotic management in our ICU gives us the further opportunity to assess the influence of a fluoroquinolone with enhanced activity against S.pneumoniae. Firstly, we retrospectively collected all consecutive patients aged > 18 years who were admitted into our ICU (16-bed medical and surgical intensive care unit in a 450-bed general hospital) between January 1996 and January 2009 for severe community-acquired pneumonia (CAP) and who received a definite diagnosis of pneumococcal pneumonia. Secondly, we selected patients who received, as initial antibiotic treatment, a β-lactam plus a fluoroquinolone, used with an appropriate dosage by IV route. Thirdly, patients were divided into two groups according to the fluoroquinolone used, Group A for ofloxacin or ciprofloxacin, Group B for levofloxacin. The study protocol was submitted to the Institutional Review Board for University Hospital of Lille which gave an approval with waiver of informed consent, in agreement with French regulations concerning such retrospective studies. CAP was defined by the following criteria observed at initial presentation or occurring within 48 h following hospitalization: acute onset of signs and symptoms of lower respiratory tract infection and a new pulmonary infiltrate found on the hospital admission chest radiograph. We excluded patients coming from nursing homes or hospitalized within 90 days prior to developing pneumonia or hospitalized > 48 h in general medical wards before ICU admission, and those with radiographic abnormalities attributed solely to any other known cause (i.e., pulmonary embolus, lung carcinoma or congestive heart failure). The decision for admission to our ICU was made, in all cases, by the attending physicians. However, only patients having a Pneumonia Severity Index (PSI) score ≥ 4 were included in this study [10] . Streptococcus pneumoniae was considered as the causative agent of CAP when a S. pneumoniae strain was isolated from > 1 blood culture or when validated sputum (< 10 squamous epithelial cells and > 25 polymorphonuclear cells per low-power field) or tracheobronchial aspirates cultures grew with > 10 5 cfu/mL S. pneumoniae. Patients having CAP due to a penicillinresistant strain of S. pneumoniae (MIC > 2 mg/l) were excluded from our study. Appropriate drug dosages were defined in the French recommendations as: amoxicillin > 50 mg/kg/d, cefotaxime > 50 mg/kg/d, ceftriaxone > 20 mg/kg/d, piperacillin > 200 mg/kg/d, ofloxacin = 200 mg/12 h, ciprofloxacin = 400 mg/12 h, levofloxacin = 500 mg/12 h [2, 3, 7] . These drug dosages for β-lactams, ofloxacin and ciprofloxacin were unchanged during the study period. Thus, doses used in both groups were similar. Within 24 h of admission, all patients underwent clinical, radiological and biological tests. Briefly, we recorded age, gender, underlying clinical characteristics and initial vital signs. Chronic respiratory insufficiency was assessed combining the usual clinical and radiological criteria and the coexistence of ventilatory impairment assessed either before or after ICU stay. Immunosuppression was defined as recent use of immunosuppressant or systemic corticosteroids (i.e., prednisolone > 0.5 mg/kg/day for more than 1 month), human immunodeficiency virus infection, neutropenia (absolute neutrophil count < 1.000 cells/mm3), organ transplantation with ongoing immunosuppressant, cancer chemotherapy within the past 3 months, or asplenia. Shock was defined as a sustained (> 1 h) decrease in the systolic blood pressure of at least 40 mm Hg from baseline or a resultant systolic blood pressure < 90 mm Hg after adequate volume replacement and in the absence of any antihypertensive drug [11] . Severity of illness at admission to ICU was assessed using the Simplified Acute Physiology Score II (SAPS) II [12] , the Sepsis-related Organ Failure Assessment (SOFA) score [13] and the logistic organ dysfunction (LOD) score [14] . We also calculated the PSI at ICU admission [10] . For all patients, information on the following therapeutic topics instituted within 48 hours following ICU admission was recorded: supportive measures such as mechanical ventilation or hemodialysis, use of vasopressor drugs, hydrocortisone, drotrecogin alfa (activated), or intensive insulin therapy. The effectiveness of initial antimicrobial therapy was assessed within 72 h after treatment as follows: A lack of clinical improvement 3 days after treatment initiation (worsening or persistent fever or hypothermia, worsening of pulmonary infiltrates or of respiratory function assessed by PaO 2 /FiO 2 ) defined an ineffective treatment. On day 3, day 5 and day 7, body temperature, and SOFA score were determined. During the patient's stay in the ICU, occurrence of complications was recorded. We distinguished sepsis-related complications (secondary septic shock, acute respiratory distress syndrome or development of multiple organ failure), hospital-acquired lower respiratory tract (HA-LRT) superinfections and ICU-related complications (i.e., upper gastrointestinal bleeding, catheter-related infection, deep venous thrombosis and pulmonary embolism). Multiple organ failure (MOF), acute respiratory distress syndrome (ARDS) and HA-LRT were defined according to usual criteria [15] [16] [17] . Durations of mechanical ventilation, treatment with vasopressor drugs, and ICU length of stay were noted. Finally, patient mortality was evaluated on D-15, and at the time of ICU discharge. Descriptive analyses were performed in order to check and resume data. Characteristics of patients in each group were compared. Continuous variables were compared using the Student's t test. Categorical variables were compared using Chi-square test or Fisher's exact test when Chi-square was not appropriate. Differences between groups were considered to be significant for variables yielding a p value < 0.05. A stepwise logistic regression including variables collected within the first 48 hours of ICU stay and associated with a p value < 0.15 in bivariate analysis was performed. Adjusted odd-ratios were computed using a logistic regression analysis including the independent predictors of mortality. The Kaplan-Meier product limit method and the log-rank test were used to construct and compare survival curves for patients in each group. All statistical analyses were performed using the SAS Software, V9.1. During the study period, 378 patients with severe CAP were admitted in our unit. Among them, 83 (22%) patients exhibited a severe pneumococcal pneumonia and, finally, we identified 70 patients treated with a β-lactam combined with a fluoroquinolone, including 53 men (75.7%) and 17 women (24.3%). The mean age was 63.8 ± 16.8 years. S. pneumoniae was identified in blood cultures in 25 patients (35.7%). Infection was polymicrobial in 18 patients (25.7%). Causative pathogens associated with S. pneumoniae were Haemophilus influenzae (n = 7), methicillin susceptible Staphylococcus aureus (n = 4), enterobacteriaceae (n = 4), Streptococcus spp. (n = 2) and Moraxella catarrhalis (n = 2). All pathogens were susceptible to at least one drug (β-lactam and/or fluoroquinolone) received by the patients. Thirty-eight patients (54.3%) were classified as Group A. β-lactams used were a third generation cephalosporin (n = 20; 52.6%), amoxicillin ± clavulanic acid (n = 16; 42.1%) and piperacillin-tazobactam (n = 2; 5.3%) combined with ofloxacin (n = 33; 86.8%) or ciprofloxacin (n = 5; 13.2%). Thirty-two patients (45.7%) were classified as Group B. β-lactams used were a third generation cephalosporin (n = 26; 81.3%), amoxicillin ± clavulanic acid (n = 5; 15.6%) and piperacillin-tazobactam (n = 1; 3.1%) combined with levofloxacin. Main patients' characteristics on ICU admission are reported Table 1 . Most characteristics were similar in the two groups. However, underlying chronic respiratory insufficiency and bacteremia were more frequent in Group B patients. Main therapeutics instituted during ICU stay, evolution of severity scores, and occurrence of complications are reported Table 2 . The most significant differences between the two groups of patients were the more frequent use of drotrecogin alpha, intensive insulin therapy and hydrocortisone in Group B patients. On Day 15, 14 (20%) patients had died, 12 (31.6%) in Group A and 2 (6.3%) in Group B (p = 0.02). Overall, 26 patients died in the ICU, 17 (44.8%) in group A vs. 9 (28.1%) in group B (p = 0.15). So, difference in mortality rates was only significant during the first 15 days of ICU stay (Figure 1 ). In Group A, in-ICU mortality was 45% (9/20) when ofloxacin or ciprofloxacin were combined with a third generation cephalosporin and 44.4% (8/18) when combined with another beta-lactam, respectively (p = 0.97). In group B, it was 26.9% (7/26) when levofloxacin was combined with a third generation cephalosporin and 33.3% (2/6) when combined with another beta-lactam (p = 1). Results of ICU-discharge survival prognosis bivariate analysis, including factors present on ICU admission, are reported Table 3 . All underlying diseases (excepted chronic heart failure), mechanical ventilation, use of a third generation cephalosporin combined with a fluoroquinolone, and bacteraemia on ICU admission did not appear as significant prognostic variables in this analysis. Among the 25 bacteremic patients, mortality was higher in group A patients (66.6%) than group B patients (31.3%), but the difference was not statistically significant (6/9 vs. 5/16; p = 0.11). Among the 34 patients with septic shock on ICU admission, mortality was higher in group A patients (71%) than in Group B patients (47%), but the difference was not statistically significant (8/17 vs. 12/17; p = 0.30). Among variables collected during the ICU stay, use of hydrocortisone, intensive insulin therapy, haemodialysis and occurrence of HA-LRT superinfections did not appear as significant prognostic variables. Conversely, improvement on D3, SOFA > 8 on D3, D5, and D7, and occurrence of sepsis-related complications were significantly associated with outcome at ICU discharge (Table 4) . According to the results of the bivariate analysis, the following variables were entered in the stepwise analysis: chronic heart failure, age > 70 yrs, acute respiratory failure requiring mechanical ventilation, septic shock on ICU admission, use of hydrocortisone, haemodialysis, PSI score = 5, SAPS II > 50 on D1, LOD > 8 on D1, The main finding of this retrospective analysis is that levofloxacin plus a β-lactam appears to be associated with improved survival compared to ofloxacin or ciprofloxacin plus a β-lactam in severe pneumococcal CAP. Empirical antibiotic regimen for ICU-treated severe CAP has long been recommended to cover the 3 most common severe CAP pathogens (S. pneumoniae, S. aureus and H.influenzae), atypical pathogens and most relevant Enterobacteriaceae species. Levofloxacin is a fluoroquinolone active against most of these pathogens, especially S. pneumoniae with or without decreased penicillin susceptibility [18, 19] . Its clinical activity in CAP has been well documented in various clinical trials in Europe and the USA [20, 21] . Some studies demonstrated the efficacy of levofloxacin used as monotherapy in severe CAP, compared to ceftriaxone plus erythromycin or cefotaxime plus ofloxacin [22, 23] . Nevertheless, experts continue to propose, for ICU-treated severe CAP, an empirical antibiotic regimen based on an anti pneumococcal β-lactam combined with either a macrolide or a fluoroquinolone. Since respiratory fluoroquinolones with enhanced activity against S. pneumoniae (levofloxacin, moxifloxacin or gemifloxacin) became available, they replaced second generation fluoroquinolones (ofloxacin or ciprofloxacin) in the guidelines [6] [7] [8] . This fluoroquinolone generation shift has never been clearly justified and, to our knowledge, no clinical study has compared these different quinolones combined with a β-lactam in severe CAP. Our results suggest that, when severe CAP causative agent is S. pneumoniae, a combination levofloxacin plus β-lactam is associated with lower mortality than a combination ofloxacin or ciprofloxacin plus β-lactam. These results could be surprising as all patients received an appropriately dosed β-lactam active against S. pneumoniae and as numerous strains of S. pneumoniae remain in vitro susceptible to ofloxacin or ciprofloxacin. However, there might be bacteriological and clinical data explaining our results. A synergy between β-lactams and levofloxacin against S. pneumoniae has been reported [24] . Conversely, synergy was rarely observed between the combination of cefotaxime and ofloxacin [25] . Recent clinical studies suggest that combination therapies could improve the prognosis of pneumococcal pneumonia: Waterer et al. retrospectively studying 225 patients with severe bacteremic pneumococcal pneumonia demonstrated that a single effective therapy was an independent predictor of mortality (AOR = 6.2) [26] . Baddour et al. performed a prospective, multicenter, international study including 844 adult patients with S. pneumoniae bacteremia [27] . Although the 14-day mortality was not significantly different for all patients receiving monotherapy versus combination (11.5% vs. 10.4%), a combination of in vitro active agents was associated with a significantly lower mortality than a single active agent (19.4% vs. 60%; p = 0.0006). The present work has numerous limits. The most important is probably major treatment differences among the two groups. Patients were recruited during a long period (1996-2009), during which therapies such as hydrocortisone, drotrecogin alfa (activated), or intensive insulin therapy were introduced. Management of septic shock and ARDS has changed following results of large international studies [28, 29] . As most changes in management of patients with multiple organ failures overlap with our antibiotic policy changes, our results might be biased. Indeed, hydrocortisone use and intensive insulin therapy were more frequent in group B than in Group A. However, these factors were not significantly associated with ICU survival in bivariate analysis and hydrocortisone use, in multivariate analysis, was not an independent prognostic factor. Moreover, there is no evidence suggesting a survival benefit by most adjunctive therapies in patients with CAP [30] and the benefit of intensive insulin therapy in medical ICU and/or low-dose steroids is now highly questionable [31, 32] . Similarly, the use of cephalosporin is more frequent in group B than in group A. However, the use of a third generation cephalosporin rather than amoxicillin has no impact on prognosis. This is not surprising as, to our knowledge, no clinical study demonstrated a third generation cephalosporin to be superior to amoxicillin for non penicillin-resistant S. pneumoniae CAP as far as drug dosage is adequate. Finally, some important prognostic parameters such as the time elapsed between admission and the first dose of antibiotic were not taken into account in our study. Before 2006, we did not have computerized data charts thus, exact time of admission and antibiotics admission, particularly for patients transferred from other departments/hospitals cannot be obtained. Our study suggests that levofloxacin combined with a β-lactam is associated with improved survival in comparison with ofloxacin or ciprofloxacin combined with a β-lactam in severe pneumococcal patients admitted in the ICU. This combination, proposed by current guidelines as empirical treatment of severe CAP patients admitted in ICU could improve their prognosis. Obviously, only a prospective, randomized, double-blind trial could confirm this result. List of abbreviations AOR: adjusted odd ratio; ARDS: acute respiratory distress syndrome; CAP: community-acquired pneumonia; CI: confidence interval; HA-LRT superinfections: hospital-acquired lower respiratory tract superinfections; ICU: intensive care unit; LOD score: logistic organ dysfunction score; LOS: length of stay; MOF: multiple organ failure; MV: mechanical ventilation; PSI: pneumonia severity index; SAPS: simplified acute physiology score; SOFA: sepsis-related organ failure assessment; SD: standard deviation.

Is Generalized Maternal Optimism or Pessimism During Pregnancy Associated with Unplanned Cesarean Section Deliveries in China?

This research examines whether maternal optimism/pessimism is associated with unplanned Cesarean section deliveries in China. If so, does the association remain after controlling for clinical factors associated with C-sections? A sample of 227 mostly primiparous women in the third trimester of pregnancy was surveyed in a large tertiary care hospital in Beijing, China. Post-delivery data were collected from medical records. In bivariate analysis, both optimism and pessimism were related to unplanned c-section. However, when optimism and pessimism were entered into a regression model together, optimism was no longer statistically significant. Pessimism remained significant, even when adjusting for clinical factors such as previous abortion, previous miscarriage, pregnancy complications, infant gestational age, infant birthweight, labor duration, birth complications, and self-rated difficulty of the pregnancy. This research suggests that maternal mindset during pregnancy has a role in mode of delivery. However, more research is needed to elucidate potential causal pathways and test potential interventions.

Worldwide, Cesarean section rates are increasing [1] [2] [3] . Despite recommendations that cesarean section rates not exceed 15% [4, 5] , many countries have rates double or even triple that threshold [3] . China-home to one-fifth of the world's population and 12 percent of all births annually [6, 7] -is no different. Data from hospital-based studies in urban China showed c-section rates ranging from 26% to 63% during the late 1990s [8] , while a more recent WHO study combining urban and rural populations reported overall c-section rates of 46.2% [3] . Although cesarean section deliveries can be lifesaving for both mothers and their infants when indicated, their overuse is cause for concern due to their association with increased maternal morbidity and mortality, cost, and utilization of sometimes scarce health system resources [3] . Numerous researchers have investigated the predictors of higher than normal cesarean section rates [9] [10] [11] [12] [13] [14] [15] . Principal among these include including physician-related factors, insurancerelated factors, hospital and health-system factors, and maternal preferences. Additionally, cesarean section rates have also been found to vary by male versus female provider [15] , public versus private hospital setting [16] [17] [18] [19] , adoption and use of clinical guidelines [20] , public versus private insurance status [18] , and even day of the week and time of day [17, 19, 21] that women present for delivery. Patient race [16] , age [22] , income [22] , and preferences [23] have also been linked to increased c-section rates. Although the literature is replete with clinical factors associated with elective and emergency cesarean section, such as advanced maternal age, short maternal stature, heavier infant birthweight, fetal dystress, preeclampsia, prolonged/obstructed labor, or shoulder distocia [8, 24] , less is known about psychological factors that affect women who intend to delivery vaginally but ultimately deliver via csection. It is probable that the vast majority of unplanned cesarean sections are attributable to clinical indications. However, are there potential psychological variables at play as well? And in a country like China, with exceedingly high rates of cesarean section, might the impact of those psychological variables be observable? This exploratory study was designed to examine the psychological characteristic of dispositional optimism and pessimism in a woman's likelihood of undergoing an unplanned cesarean section delivery in urban China. Dispositional optimism is seen as a relatively stable personality characteristic (a "trait" rather than a "state") that is associated with general assumptions about positive future outcomes. Dispositional pessimism is the converse: it is a tendency to expect the worst when looking toward future outcomes. A meta-analytic review of the optimism literature from 2009 [25] that examined 83 separate studies found a persistent relationship between optimism and positive health outcomes [25] . In addition, women with higher levels of optimism during pregnancy have been found to have to lower levels of stress, anxiety, and peripartum depression than women with lower levels of optimism [26] [27] [28] [29] . Optimism has also been linked to birth outcomes, with one study finding that optimistic women gave birth to larger babies [30] , and a second study finding that when gestational age was controlled for, women who were least optimistic during pregnancy when compared to women with higher levels of optimism delivered smaller infants [31] . It may seem logical to conclude that if optimism can lead to better health outcomes, pessimism might be detrimental. However, pessimism has been shown to have prophylactic effects in certain circumstances. In particular, pessimism can insulate people from the psychological consequences of failure, including anxiety, depression, and diminished self-esteem [32] . Norem and Cantor [32, 33] found that individuals who expect the worst can sometimes use those expectations to help them better meet the demands of stressful challenges. These "defensive pessimists" engage in active and constructive coping efforts-which may mediate the relationship between pessimism and outcomes [34] . For example, Moyer et al. found that among pregnant women in Ghana, those who were the most pessimistic were more likely to get tested for HIV whereas their optimistic counterparts were less likely to get tested [35] . This research aimed to address the following research questions. (1) Is generalized maternal optimism or pessimism (assessed during pregnancy) associated with unplanned cesarean section among women giving birth in a tertiary care hospital in Beijing? (2) If optimism or pessimism is associated with unplanned cesarean section, which is more strongly associated, optimism or pessimism? And (3) if there is a significant relationship between optimism, pessimism, and unplanned cesarean section delivery, is that relationship robust enough to remain significant when clinical factors are included in the model? Site. Data were collected from pregnant women presenting for prenatal care at the obstetric outpatient clinic at the Peking University First Hospital between May and July 2006. As one of the largest and most well-known academic medical centers in Beijing, Peking University First Hospital draws both public and private patients from in and around Beijing. Clinics see an average of 600 pregnant women per week and 3000-3500 deliveries per year. All research protocols and survey instruments were reviewed and approved by the institutional review boards at the University of Michigan and Peking University. Pregnant women in their last trimester of pregnancy who were 18 years old or older attending antenatal care clinic were eligible. Women facing an imminent health crisis, those in active labor, or those being admitted to the hospital were excluded (despite the generally stable nature of optimism and pessimism, those women in active labor were excluded because of concerns about disproportionate reporting of a pessimistic attitude if it was assessed during painful, active labor when compared to assessments obtained during a routine prenatal visit). After describing the study and obtaining verbal approval to continue, research assistants talked patients through an informed consent form, answering any questions the women may have had. All participants signed a written informed consent document and were given a copy to keep. Women were then given a self-administered survey to complete before their appointment. Translators were used when necessary. Surveys were designed to be selfadministered, but women were given the option to have the survey administered verbally. Data were gathered using paper and pencil forms. Hospital registration numbers were collected from participants to allow for postdelivery followup. Hospital registration numbers were removed from the original survey and replaced with a unique ID number once the registration number was recorded in a separate location for follow-up purposes. Responses from the hard copies of the self-administered surveys were entered into an Excel spreadsheet and cleaned. The survey included administering a demographic and health questionnaire and the Life Orientation Test (LOT-R). The Demographic and Health Questionnaire measured patient characteristics including age, number of pregnancies, other medical conditions, and self-perceived health status. Women were asked to enumerate any pregnancy complications or symptoms they had during pregnancy, including such things as vaginal bleeding, headaches, swollen hands, troubled vision, preeclampsia, dizzy spells, swollen face, abdominal/belly pain, eclampsia, or other problems. For the purposes of this analysis, these were combined into a single dichotomous variable, and termed maternal complications. Women were also asked to rate their perception of the difficulty of their pregnancy on a scale of 1 to 4, with 1 being "extremely easy" and 4 being "extremely difficult." The Life Orientation Test (LOT), developed by Sheier and Carver in 1985 [36] and revised in 1994 [37] (Life Orientation Test-Revised, or LOT-R), was used to assess dispositional optimism. The LOT-R is one of the most commonly used measures of optimism/pessimism. It uses generalized outcome expectancies to measure dispositional optimism. The LOT-R has been widely validated [38] and used in China [39] [40] [41] [42] [43] [44] [45] [46] . It includes 6 scored items and 4 fillers that generate an overall score, as well as two possible subscales: an optimism subscale and a pessimism subscale. The items that make up the optimism subscale are (1) in uncertain times, I usually expect the best; (2) I'm always optimistic about my future and (3) overall, I expect more good things to happen to me than bad. The items that make up the pessimism subscale are (1) if something can go wrong for me, it will; (2) I hardly ever expect things to go the way I would like them to go; (3) I rarely count on good things happening to me. The participant answers each item based on a 5-point scale, with response options ranging from strongly disagree to strongly agree. The pessimism items are reverse scored and then added to the optimism items to create the overall score whereas the subscales are created by summing the items for pessimism and the items for optimism separately. For these analyses, the optimism and pessimism subscales were used separately. The instrument was pilot tested, and minor modifications were made to ensure comprehension. The survey was translated into Mandarin and back-translated into English by native bilingual speakers. The original and backtranslated versions were compared for consistency, and any inconsistencies were resolved by discussion and consensus among the research team. Chart Review. was used to collect data after women had delivered their babies. Mode of delivery was determined, which indicated vaginal delivery with and without forceps, vaginal delivery with and without vacuum extraction, planned cesarean section, or unplanned cesarean section. For the purposes of this analysis, a single dichotomous variable was created to reflect unplanned cesarean section yes/no. Thus women who delivered vaginally or via planned cesarean section were treated as one group, and women undergoing an unplanned or emergency cesarean section were treated as a separate group. Additional data collected from the medical record included gestational age of the infant at delivery, birthweight, labor duration, use of pain medication, 1-minute and 5-minute Apgar scores, and any of a number of delivery or birth complications, including such things as hemorrhage, preeclampsia, intrauterine infection, breech presentation, or delayed labor. For the purposes of this analysis, all of those factors were combined into a single dichotomous variable termed birth complications. Chart review Data were entered into a spreadsheet and cleaned. All data were analyzed using SPSS statistical software, Version 17.1 (SPSS Inc, Chicago, IL). Frequencies and basic descriptive statistics were calculated for all variables. Women with complete baseline and chart data (and could thus be included in the larger regression analysis) were compared against those women with incomplete baseline or chart data using Student's t-test for continuous variables and Chi Square analysis for categorical variables. To address Research Question 1, (is optimism or pessimism associated with unplanned cesarean section delivery?), bivariate statistics were calculated to determine if optimism or pessimism were independently associated with unplanned cesarean section. Additional demographic and clinical variables were examined to determine if there were factors aside from optimism and pessimism and expected clinical correlates that might be associated with unplanned cesarean section in this population. Bivariate analysis included Student's t-tests, ANOVAs, and Chi-Square analyses. To address Research Question 2, (which is more strongly associated with unplanned cesarean section delivery, optimism or pessimism?) Binary logistic regression analysis was conducted with both optimism and pessimism regressed on unplanned cesarean section (yes/no). Area under the curve analysis was conducted to judge the strength of the model. To address Research Question 3, (if there is a significant relationship between optimism, pessimism, and unplanned cesarean section delivery, is that relationship robust enough to remain significant when clinical factors are included in the model?), binary logistic regression analysis was conducted with optimism and pessimism regressed on unplanned cesarean section (yes/no) with the additional clinical factors of labor duration, birth complications, previous abortion, previous miscarriage, pregnancy complications, gestational age, infant birth weight, and self-rated difficulty of the pregnancy added into the model. Area under the curve analysis was conducted to judge the strength of the model. For all analyses a P value of .05 was taken as statistically significant. Two hundred fifty-one women were asked to participate, and 227 met our eligibility criteria and agreed to participate (90.4% response rate). Of the 227, 86 had missing items on their surveys or their birth outcomes data were not available in the hospital medical records system. Table 1 illustrates our sample demographics, comparing the 141 women who were ultimately included in our analysis with the 86 who were excluded. Overall, our sample is one of well-educated Han women in their last trimester of pregnancy who are married and working outside the home. They do not differ significantly from the 86 women excluded from the analysis 76 (91.5) P = .300 (NS) Missing = 3 * Women with incomplete baseline data were excluded from the regression analysis. Key variables for inclusion were age, education, income, number of previous deliveries, originally from Beijing (y/n), car ownership (y/n), work before pregnancy (y/n), intend to work after pregnancy (y/n), insurance status, previous abortion (y/n), previous miscarriage (y/n), and experience of this pregnancy. on any variable aside from education, with excluded women more likely to have lower levels of education (P = .012). Table 2 illustrates the health-related variables reported at enrollment. Again, there were no significant differences found between women included in our analysis and those excluded due to missing data. More than half of our sample has had at least one previous pregnancy that was either spontaneously or electively terminated, and only 2.8 percent of women report having anything other than mild complications in this current pregnancy. The vast majority of women in this study were primiparous. Table 3 reflects delivery data obtained via chart review. Mean gestational age at delivery was 39.6 weeks. Mean duration of labor-defined as the time from first documentation of regular contractions plus cervical dilation to vaginal delivery-was 9 hours, with a range of 1 to 21 hours. Slightly more than half of women delivered vaginally, with the remaining having planned, emergency, or posttrial-of-labor (PTOL) cesarean sections. Infants had a mean gestational weight of 3406 grams, and most had five-minute Apgar scores of 10. Forty-one percent of women had at least one birth complication. The most common complications were fetal distress (41%), preterm membrane rupture (26%), umbilical cord issues such as prolapsed, entanglement or nuchal cords (17%), and delayed labor (7%). With regard to Research Question 1, (is optimism or pessimism associated with unplanned cesarean section delivery?), bivariate analyses comparing optimism and pessimism against unplanned cesarean section indicated that both were significant: optimism (P = .047, 95% CI. 012, 1.81), pessimism (P = .003; 95% CI −2.42, −.529) (See Table 4 ). In addition, labor duration (P = .004, 95% CI 1. 009, 5.16) and the presence of birth complications (P = .01, Chi Square = 6.65) were also found to be significant. No other demographic or clinical factors were significantly associated with unplanned cesarean section. Table 5 Model 1, illustrates the findings with regard to Research Question 2 (which is more strongly associated with unplanned cesarean section delivery, optimism or pessimism?). In an unadjusted model in which both optimism and pessimism were regressed against unplanned cesarean section, pessimism remained statistically significant while optimism failed to meet the threshold for statistical significance. (pessimism OR = 1.28, 95% CI: 1.06, 1.56, P = .01; optimism OR = 0.88, 95% CI: 0.71, 1.08; P = . 22) . When the same model was then adjusted for a variety of clinical factors (see Table 5 Model 2) to answer Research question 3 "if there is a significant relationship between optimism, pessimism, and unplanned cesarean section delivery, is that relationship robust enough to remain significant when clinical factors are included in the model?", pessimism remained significantly associated with unplanned cesarean section (OR = 1.42; 95% CI: 1.11, 1.81; P = .004). Of note, This study showed an association between higher levels of generalized maternal pessimism during pregnancy and an increased likelihood of an unplanned c-section delivery among women presenting for prenatal care and delivering their infants at a tertiary care hospital in Beijing, China. This association was robust enough to remain, even when adjusted for clinical factors likely to be linked to a risk of unplanned cesarean section delivery. Interestingly, pessimism not optimism remained significant throughout the analysis. However, what is not clear, and what the cross-sectional study design of the study does not allow us to explore, is the mechanism of action. What is it about being pessimistic that is related to unplanned c-section delivery? It is possible that pessimists have qualitatively different or less effective coping skills than their less pessimistic counterparts [47] [48] [49] [50] . Additionally pessimists, by virtue of believing that negative outcomes are likely, may be more fearful during labor. Emotional factors such as fear of delivery or fear of pain [51] have been linked to increased risk of c-section. Pessimists may also be more likely than their optimistic counterparts to abandon a traditional vaginal delivery and opt for a csection if given the opportunity. Conceivably pessimism may serve as a proxy for another latent variable. Previous studies have linked optimism and pessimism to age, spirituality, and even SES, [52] [53] [54] but an additional, as yet undescribed and measured variable could explain the relationship between pessimism and unplanned cesarean section rates. By contract, optimists have been found to be more likely to adopt active coping strategies and reappraise a situation in a positive way if an important goal is blocked [50] . It is possible that such coping strategies may allow optimists to relax during delivery more easily than their more pessimistic peers, reducing the likelihood of "failure to progress." Our findings do not support this possibility: pessimism showed a significant association with unplanned cesarean section deliveries, while levels of optimism did not. This is not only useful in reaffirming that optimism and pessimism are two separate constructs rather than poles on a continuum [55, 56] , but is also instructive in potential interventions during pregnancy. Encouraging positive thoughts may not be nearly as helpful as discouraging negative ones. The idea that cognitive predispositions that precede delivery may be associated with type of delivery is worthy of further exploration-including whether interventions can be designed to influence women's predispositions. For example, could cognitive behavioral therapy be used to reframe pessimists' negative thoughts, and might that result in lower cesarean-section rates? Perhaps more fundamentally, can pessimism be unlearned? Despite a dearth of information on pessimism, research suggests that optimism can be learned and practiced [57] . Avoiding negative environments, seeking the company of positive individuals and reframing challenges as opportunities are some of the ways experts suggest "activating" one's optimism [57] . Yet it is unclear whether such techniques would be effective enough to impact health outcomes. Nevertheless, our findings are noteworthy for two reasons. First, they demonstrate the potential relationship/association between psychological factors assessed during pregnancy and eventual delivery outcomes, and second, they illustrate the potential strength of psychological factors such as pessimism. That birth complications were significantly associated with unplanned cesarean sections is to be expected-given that complications such as fetal distress, preeclampsia, prolonged/obstructed labor, or shoulder dystocia are primary indications for cesarean section delivery [8, 24] . It is also not surprising that duration of labor is associated with unplanned cesarean section. We also observed that women who delivered vaginally in this sample had longer labors than those who had unplanned cesarean sections (data not shown). It was interesting to note that in this study the length of time women were allowed to attempt labor before a cesarean section was chosen was much shorter than in the United States (average unplanned cesarean section labor duration in this study was 3.5 hours, compared to 16.0 hours among nulliparas and 12.4 hours among multiparas in the United States [58] ). There are several limitations to this study. First, the use of the LOT-R has not been formally validated among Chinese pregnant women. However, the instrument has been used repeatedly in China [39] [40] [41] [42] [43] [44] [45] [46] 59] and it was carefully pretested in this population prior to study implementation. Our focus groups and pilot testing did not indicate any difficulties in interpretation of these items. Nonetheless, the instrument may benefit from a more rigorous validation study in this population. Also, the use of a cross-sectional convenience sample that includes mostly primiparous women limits inference to a wider population of pregnant Chinese women. In this study, all women presenting to the clinic were asked to participate, and it is possible that the women presenting during this study period were different from the larger population of pregnant women in Beijing. Future studies would benefit from a design that includes random selection at a variety of institutions across Beijing and across China. This study also includes women in their last trimester of pregnancy. Although optimism/pessimism is considered a stable construct, it would be valuable to determine the potential impact of earlier recruitment. This study also reveals what some would call excessively high episiotomy, cesarean section, and forceps rates, limiting its generalizability to settings without similar rates. Nonetheless, we believe these findings reflect clinical practice at one large tertiary care center in China, and as such provide valuable insight. Finally, this study asked women to self-report their pregnancy complications. It was not possible to verify these self-reports against medical records data. We were able to elicit birth complications from the medical record, but this study relies upon self-reported complications during the gestation period. We do not believe this to be a significant limitation, however, given the high probability that women will know whether they are experiencing nausea, vomiting, or abdominal pain, or whether they have vaginal bleeding or swollen hands and feet. We also expect that women will remember if a doctor has told them they have high blood pressure, gestational diabetes, or other more serious pregnancy complications. Implications. This research has several important implications. First, it confirms what many women and practitioners may have believed anecdotally: that a woman's mindset during her pregnancy may have an impact on her delivery. It also raises questions about the value of positive thinking-the predominant advice given to pregnant women-versus the value of not thinking negatively. Second, it raises important questions about whether inexpensive cognitive behavioral therapy or other mindsetaltering interventions among pregnant women could be used to reduce unplanned cesarean section rates. More research is needed to elucidate the relationship between pessimism and pregnancy outcomes. Is this study replicable? Is the finding real, or is it masking some other yet to be determined variable? Is a negative outlook merely associated with a risk of unplanned cesarean section delivery, or can a causal pathway be identified? In addition, is it possible to change women's levels of pessimism? And would interventions to decrease pessimism translate to reduced rates of c-sections? These are just some of the questions in need of answers as researchers continue to explore the relationship between psychosocial variables and pregnancy outcomes. (i) Cesarean section rates are rising, due, in large part, to nonclinical factors. (ii) Physician factors, insurance status, hospital policies, and maternal preferences are all non-clinical factors that influence csection rates. (iii) Maternal cognitive predispositions during pregnancy (specifically optimism/pessimism) have not been examined in relationship to unplanned cesarean section deliveries. What This Study Adds. (i) Pessimism during pregnancy appears to be associated with an increased risk of unplanned cesarean section delivery in this population. (ii) Pessimism during pregnancy remains associated even when clinical factors are controlled. (iii) Pessimism appears to be a stronger correlate than optimism-suggesting that having positive thoughts/expectations may not be as helpful as not having negative thoughts/expectations during pregnancy. The project/study described was supported by Grant no.T37 MD001425-08, from the National Center of Minority Health and Health Disparities, National Institutes of Health. Its contents, including the design and conduct of the study, the collection, management, analysis and interpretation of the data, and the preparation, review, and approval of the paper, are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health. No author has a financial conflict of interest in this research or in its publication.

In Vitro Gene Delivery Mediated by Asialofetuin-Appended Cationic Liposomes Associated with γ-Cyclodextrin into Hepatocytes

The purpose of this study is to evaluate in vitro gene delivery mediated by asialofetuin-appended cationic liposomes (AF-liposomes) associating cyclodextrins (CyD/AF-liposomes) as a hepatocyte-selective nonviral vector. Of various CyDs, AF-liposomes associated with plasmid DNA (pDNA) and γ-cyclodextrin (γ-CyD) (pDNA/γ-CyD/AF-liposomes) showed the highest gene transfer activity in HepG2 cells without any significant cytotoxicity. In addition, γ-CyD enhanced the encapsulation ratio of pDNA with AF-liposomes, and also increased gene transfer activity as the entrapment ratio of pDNA into AF-liposomes was increased. γ-CyD stabilized the liposomal membrane of AF-liposomes and inhibited the release of calcein from AF-liposomes. The stabilizing effect of γ-CyD may be, at least in part, involved in the enhancing gene transfer activity of pDNA/γ-CyD/AF-liposomes. Therefore, these results suggest the potential use of γ-CyD for an enhancer of transfection efficiency of AF-liposomes.

The principle of somatic gene therapy is that genes can be introduced into selected cells in the body in order to treat genetic or acquired diseases. The liver may be potentially an important target for gene therapy, because crucial diseases such as amyloidosis, primary biliary cirrhosis, familial hypercholesteremia, phenyl ketonuria, and virus hepatitis occur in this organ [1] . In addition, the liver has the ability to synthesize a wide variety of proteins, to perform various posttranslational modifications, and to secrete them into the blood. Of various nonviral methods, the lipofection method, by which cationic lipids (cationic liposomes) are used for transfection and interact with plasmid DNA (pDNA) to give a lipoplex, has recently attracted attention [2] . Cationic liposomes have great advantages as gene delivery carriers such as (1) low cytotoxicity and immunogenicity [3] , (2) regulation of the pharmacokinetics through the modification of particle size or lipids components of liposomes [4] , (3) entrapment of pDNA into inner water phase of liposomes and suppression of DNA degradation by DNase [5] , and (4) delivery of gene to target cells by the addition of target ligands and/or antibody [6] . Asialofetuin (AF) is a glycoprotein that possesses three asparagine-linked triantennary complex carbohydrate chains with terminal N-acetylgalactosamine residues. The protein displays affinity to asialoglycoprotein receptor (ASGP-R) on hepatocytes and enters the cells through the receptor [7, 8] . Thus, AF has been used as a ligand to deliver drugs to hepatocytes and a competitive inhibitor to ASGP-R [9, 10] . In fact, the widespread use of AF-appended liposomes (AFliposomes) as a hepatocyte-selective gene transfer carrier has been reported [11, 12] . Cyclodextrins (CyDs) have recently been applied to gene transfer and oligonucleotide delivery [13] [14] [15] [16] . CyDs are 2 Journal of Drug Delivery cyclic (α-1,4)-linked oligosaccharides of α-D-glucopyranose containing a hydrophobic central cavity and hydrophilic outer surface, and they are known to be able to act as novel host molecules by chemical modification [17] . Davis and his colleagues reported that the ternary complex of a water-soluble β-CyD polymer with 6 A ,6 D -dideoxy-6 A ,-6 D -di-(2-aminoethanethio)-β-CyD and dimethylsuberimidate (βCDP6), galactosylated, or transferrin polyethylene glycol conjugates with adamantane, and pDNA possesses higher transfection efficiency in hepatoma or leukemia cells, respectively, through receptor-mediated endocytosis [18, 19] . Recently, we reported the potential use of PAMAM dendrimer functionalized with α-CyD (α-CDE) [20] and lactosylated α-CDE (Lac-α-CDE) as a hepatocyte specific gene delivery in vitro and in vivo [21] . Meanwhile, Lawrencia et al. reported that lipoplex transfection of pDNA with DOTAP (N-[1-(2,3-dioleoyloxy)propyl]-N,N,N-trimethylammonium methyl-sulfate) in the presence of cholesterol, which is solubilized by methyl-β-cyclodextrin (methyl-β-CyD), has significantly improved transfection efficiency in urothelial cells due to change in membrane fluidity by methyl-β-CyD [22] . In addition, we previously demonstrated that intravenous injection of the pegylated liposomes entrapping the doxorubicin (DOX) complex with γ-CyD in BALB/c mice bearing Colon-26 tumor cells showed DOX accumulation in tumor tissues and the potent antitumor effect, compared with those of DOX solution and pegylated liposomes entrapping DOX alone [23] . These lines of evidence suggest that transfection efficiency and pharmacokinetics of pDNA can be altered by the association of CyDs with AF-liposomes. Based on these backgrounds, the purpose of this study is to evaluate in vitro gene delivery of AF-liposomes associated with CyDs as a hepatocyte-selective nonviral vector in HepG2 cells. In addition, the mechanisms by which γ-CyD enhanced transfection efficiency of pDNA/AF-liposomes were investigated in the view of a receptor recognition, physicochemical properties (particle size, ζ-potential, and encapsulation ratio), membrane fluidity, cellular uptake, and cytotoxicity of AF-liposomes. 2.1. Materials. Dilauroylphosphatidylcholine (DLPC), dioleoylphosphatidylethanolamine (DOPE), dipalmitoylphosphatidylethanolamine (DPPE), and diacylphosphatidylethanolamine-N-lissamine rhodamine B sulfonyl (RH-PE) were obtained from Avanti Polar-Lipid (Alabama). N-(α-Trimethylammonioacetyl)-didodecyl-D-glutamate chloride (TMAG) was purchased from Sogo Pharmaceutical (Tokyo, Japan). Asialofetuin (AF) and 2-mercaptoethanol were obtained from Sigma Chemical (St. Louis, MO). Nhydroxysulfosuccinimide (Sulfo-NHS) was purchased from Fluka (Buchs, Switzerland). 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) was from Dojindo (Kumamoto, Japan). 2-(N-morpholino) ethanesulfonic acid (MES) and 2-[4-(2-hydroxyethyl)-1-piperazinyl] ethanesulfonic acid (HEPES) were purchased from Nacalai Tesque hydroxylamine HCl, AF-liposomes were obtained. To remove the free AF, we performed gel filtration using Sepharose CL-4B column (Amersham Pharmacia Biotech, Freiburg, Germany) and determined phospholipids and AF by the Bartlett method [24] and the Bradford method [25] , respectively. The fractions number 26-30, which eluted both phospholipids and AF, were collected as the AF-liposomes fraction (Figure 1(b) ). The ζ-potential value and particle size of AF-liposomes were 31.2 ± 0.1 mV and 230.7 ± 4.2 nm, respectively. The amount of AF modification in 1 μmol of AF-liposome lipids was 35.8 μg/μmol lipids, indicating that AF modification rate against DPPE in liposomes was 0.75%. Liposomes without AF modification (N-liposomes) was prepared by the solution without Sulfo-NHS-AF, and the other procedure was the same as that of AF-liposomes. Preparation of pDNA/CyD/AF-liposomes was performed according to the method reported by Hara et al. with some modifications [12] . Briefly, 2 μL of the solution containing pDNA (1 μg/μL) and CyDs (1 μM/μM lipids) dissolved in TE buffer were added to AF-or N-liposomes suspension. After mixing, the solution was freeze-dried. Then, the sample was rehydrated with THBS (10 mM, pH 7.5) for 30 min. After freezingthawing for three times, the vesicles were extruded through PVDF membranes (Nucleopore, Plesanton, CA) with pores of diameter 450 and 200 nm. The filtrates were used for further experiments as pDNA/CyDs/AF-liposomes or pDNA/CyDs/N-liposomes. The entrapment ratios of CyDs were evaluated by the anthrone-sulfuric acid method [26] . Briefly, 3 mL of anthrone reagent was added to 0.5 mL of the suspension containing CyDs/liposomes. The tube was covered with a glass ball and was heated for 10 min in boiling water. After quenching with cold water, absorbance of the suspension was measured by a U-2000A spectrophotometer (Hitachi, Tokyo, Japan) at 620 nm. The encapsulation ratios of pDNA were determined by a fluorescent spectrometer F-4500 (Hitachi, Tokyo, Japan). Briefly, 350 μL of the suspension containing pDNA/CyD/AF-liposomes in 10 mM THBS (pH 7.5) were mixed with 200 times diluted Picogreen dsDNA reagent (350 μL). After incubation for 30 min at 25 • C, fluorescent intensity (F po ) was determined. Next, after addition of 20% of Triton-X (20 μL) to the sample, fluorescent intensity (F pt ) was determined and the encapsulation ratio of pDNA was calculated as follows: encapsulation ratio (%) = [(F pt · r − F po )/F pt · r] · 100, where r is compensation coefficient (r = 1.03). Particle size and ζ-potential value of liposomes in 10 mM THBS (pH 7.5) were measured by a submicron particle analyzer N4 Plus (Beckman Coulter, Fullerton, CA) and ELS-8000 (Otsuka Electronics, Osaka, Japan), respectively. AF-liposomes encapsulating calcein were prepared by the freezing and thawing method after addition of 0.1 mM calcein in 10 mM THBS (pH 7.5). The vesicles were extruded through two stacked polycarbonate membranes (Nucleopore, Plesanton, CA) with pores of diameter 1 μm. The sample was subjected to 10 passes through the filter at 40 • C. The filtrates were extruded through the polycarbonate membranes (pore size 0.2 μm) as described above. Two milliliters of CyDs solution adjusted at the appropriate concentration (5-20 mM) using 10 mM phosphate buffer were added to 20 μL of the liposomal suspension, and then the resulting suspension was incubated for 30 min at 25 • C. The fluorescence intensity of calcein (F t ) was measured with a fluorophotometer (Hitachi F-4500, Tokyo, Japan) at 25 • C; excitation and emission wavelengths were 490 and 520 nm, respectively. After addition of 20 μL of cobalt chloride solution (10 mM) to the sample to quench the fluorescence of nonencapsulated calcein, the intensity of fluorescence of encapsulated calcein (F in ) was also determined. Then, the liposomes were completely disrupted by the addition of 20 μL of Triton X-100 (20%) solution, and the intensities of fluorescence after quenching by cobalt chloride (F q ) were measured. Calcein encapsulation ratio was calculated by the equation as follows: encapsulation ratio 2.5. Cell Culture. HepG2 cells, a human hepatocellular carcinoma cell line, A549 cells, a adenocarcinomic human alveolar basal epithelial cells, and NIH3T3 cells, a mouse embryonic fibroblast cell line, were obtained from Riken Bioresource Center (Tsukuba, Japan). HepG2, A549, and NIH3T3 cells were grown in DMEM, containing 1 × 10 5 U/L of penicillin, 0.1 g/L of streptomycin supplemented with 10% FCS at 37 • C in a humidified 5% CO 2 and 95% air atmosphere. Transfer. In vitro transfection of the pDNA/CyDs/AF-liposomes was performed utilizing the luciferase expression of pDNA (pRL-CMV-Luc or pGL3control vector) in HepG2, A549, and NIH3T3 cells. The cells (2 × 10 5 cells per 24 well plate) were seeded 6 h before transfection and then washed twice with 500 μL of serum-free medium. Two hundred μL of serum-free medium containing pDNA/CyDs/AF-liposomes in the absence and presence of AF as a competitor protein or BSA as a control protein were added to each dish and then incubated at 37 • C for 3 h. After washing HepG2 cells with serum-free medium twice, 500 μL of medium containing 10% FCS were added to each dish and then incubated at 37 • C for 21 h. After transfection, the gene expression was measured as follows: Renilla and firefly luciferase contents in the cell lysate were quantified by a luminometer (Lumat LB9506, EG&G Berthold Japan, Tokyo, Japan) using the Promega Renilla and firefly luciferase assay reagent (Tokyo, Japan), respectively. It was confirmed that CyDs and AF-liposomes had no influence on the luciferase assays under the present experimental conditions. Total protein content of the supernatant was determined by Bio-Rad protein assay kit (Bio-Rad Laboratories, Tokyo, Japan). EGFP-expressing cells were determined by a confocal laser scanning microscopy (CLSM, Olympus FV300-BXCarl Zeiss LSM-410, Tokyo, Japan) with an argon laser at 488 nm after fixation. Briefly, the cells (2 × 10 5 cells per 35 mm glass bottom dish) were seeded 6 h before transfection and then washed twice with 500 μL of serum-free medium. Transfection with pEGFP N1 DNA was performed using the same protocol as described above. The EGFP expression ratio was determined by the number of EGFP-expressing cells per 100 cells. To observe the cellular uptake of Rhodamine-labeled AF-liposomes (RH-AFliposomes) and Alexa-labeled pDNA (Alexa-pDNA), HepG2 cells (2 × 10 5 cells/dish) were incubated with the Alexa-pDNA/CyDs/RH-AF-liposomes for 3 h. After incubation, the cells were rinsed with PBS (pH 7.4) twice and fixed in methanol at 4 • C for 5 min prior to observation by a CLSM. CyDs have been reported to interact with cell membrane constituents such as cholesterol and phospholipids, resulting in the induction of hemolysis of human and rabbit red blood cells at high concentrations of CyDs [28] [29] [30] . In addition, CyDs are well known to disrupt liposomal membranes, depending on CyD cavity sizes and membrane components [31, 32] . Then, we evaluated the interaction of CyDs with AF-liposomes These results suggest that the interaction of γ-CyD and HP-CyDs with AF-liposomes was weaker than that of α-CyD and DM-β-CyD. CyDs. Next, we evaluated in vitro gene transfer activity of pDNA/CyDs/AF-liposomes in HepG2 cells, ASGP-R positive cells ( Figure 3 ). Here, we used pRL-CMV (CMV promoter) as pDNA and the charge ratio (AF-liposomes/pDNA) of 1.6 optimized by our previous study (data not shown). The gene transfer activity of pDNA/γ-CyD/AF-liposomes in HepG2 cells was significantly higher than that of pDNA/HP-α-, HP-β-, and HP-γ-CyDs/AF-liposomes ( Figure 3 ). Next, we measured the Renilla luciferase mRNA level after transfection of pDNA/CyDs/AF-liposomes in HepG2 cells by the RT-PCR method ( Figure 4 ). As predicted, γ-CyD significantly increased the luciferase expression, but HP-γ-CyD did not, suggesting that γ-CyD is involved in the enhancing effect on luciferase expression at or prior to a transcription process. To evaluate the enhancing effects of γ-CyD on gene transfer activity of AF-liposomes associating pDNA encoding EGFP controlled by a CMV or SV40 promoter, we examined EGFP and firefly luciferase gene expression after transfection of pDNA/γ-CyDs/AF-liposomes in HepG2 cells ( Figure 5) . The charge ratio of AF-liposome/pDNA was 1.6. γ-CyDs were added to AFliposome suspension before freeze-drying. The concentrations of γ-CyDs were 1 μM/μM lipids. The ζ-potential was measured by a light-scattering method. The particle size was determined using a photon correlation spectroscopic analyzer. Each value represents the mean ± SEM of 3 experiments. * P < .05 versus without CyD. The extent of EGFP-expressing cells in the pDNA/AFliposomes system without CyDs was found to be 8%, while that with γ-CyD and HP-γ-CyD were 19% and 10%, respectively ( Figure 5(a) ). Additionally, the enhancing effect of γ-CyD was observed in the pGL3-control vector encoding firefly luciferase and having a SV40 promoter ( Figure 5(b) ). These results suggest that the enhancing effect of γ-CyD on gene transfer activity of AF-liposomes is a geneand promoter-independent manner. To confirm whether pDNA/γ-CyD/AF-liposomes have ASGP-R-mediated gene transfer activity, we performed transfection experiments in HepG2 cells in the presence and absence of AF, as an ASGP-R competitive inhibitor. Here, we confirmed that ASGP-R are expressed in HepG2 cells by the RT-PCR method (data not shown), which is consistent with previous findings [33] [34] [35] . As shown in Figure 6 , gene transfer activity of AF-liposomes was markedly inhibited by the addition of AF, but not BSA, a control protein. These results suggest that AF-liposomes had the ASGP-R-mediated gene transfer activity. Interestingly, the similar enhancing effects of γ-CyD on Renilla luciferase protein expression after transfection of pDNA/AF-liposomes were, however, observed in A549 cells and NIH3T3 cells, ASGP-R negative cells (Figure 7) . These results suggest that γ-CyD enhances the transfection efficiency of pDNA/AFliposomes in ASGP-R-independent manner. To reveal cytotoxicity of pDNA/CyDs/AFliposomes, we examined the WST-1 method (Figure 8 ). Although the cell viability after treatment with pDNA/AFliposomes and pDNA/CyDs/AF-liposomes for 3 h slightly decreased as a charge ratio of AF-liposomes/pDNA was increased in both HepG2 and NIH3T3 cells, that is, more than 80% of cell viability after application of pDNA/CyDs/AF-liposomes was observed at the charge ratio of 1.6 used in the transfection study as described above. These results suggest that pDNA/CyDs/AF-liposomes have great advantages as a nonviral vector, that is, superior transfection efficiency and less cytotoxicity, and the enhancing effect of γ-CyD on gene transfer activity of pDNA/AFliposomes is not associated with cytotoxicity. To clarify physicochemical properties of the pDNA/AF-liposomes, we determined the particle sizes and ζ-potential values of the pDNA/γ-CyD/AFliposomes at the charge ratio of 1.6. The mean diameter of the pDNA/γ-CyD/AF-liposomes was smaller than that of pDNA/AF-liposomes or pDNA/HP-γ-CyD/AF-liposomes (Table 2) . Meanwhile, the ζ-potential values of pDNA/AFliposomes, pDNA/γ-CyD/AF-liposomes, and pDNA/HP-γ-CyD/AF-liposomes were almost comparable ( Table 2 ). These results indicate that γ-CyD reduced the particle size of AF-liposomes but did not change the ζ-potential value of pDNA/CyD/AF-liposomes. The charge ratio of AF-liposome/pDNA was 1.6. γ-CyDs were added to AFliposome suspension before freeze-drying. The concentrations of γ-CyDs were 1 μM/μM lipids. The encapsulation ratios of pDNA were determined using Picogreen assay. The encapsulation ratios of γ-CyDs were determined by an anthrone-sulfuric acid method. Each value represents the mean ± SEM of 3 experiments. * P < .05 versus without CyD. Next, we examined the effects of γ-CyDs on encapsulation ratios of pDNA into AF-liposomes ( Table 3 ). The encapsulation ratio of pDNA in pDNA/γ-CyD/AF-liposomes was significantly higher than those of pDNA/HP-γ-CyD/AFliposomes and pDNA/AF-liposomes. Meanwhile, the encapsulation ratios of γ-CyD and HP-γ-CyD into AF-liposomes were approximately 10.2% and 11.0%, respectively. These results suggest that γ-CyD improves the encapsulation of pDNA into AF-liposomes, although the extent of γ-CyD encapsulation into AF-liposomes is not high. In addition, both the encapsulation ratio of pDNA into AF-liposomes and gene transfer activity of pDNA/AF-liposomes were raised, as the number of the freeze-thaw cycle was increased, suggesting that the encapsulation of pDNA in AF-liposomes is correlated with the gene transfer activity of pDNA/AFliposomes (Table 4) . It is known that membrane fluidity of liposomes affects the release profiles as well as a retention time of drug encapsulated into liposomes. Therefore, we investigated the effects of γ-CyD on membrane fluidity of AF-liposomes using a Microcal MC2 scanning calorimeter. In the present study, we utilized N-liposomes to eliminate the effects of the heat degeneration of AF in a DSC thermograph. Figure 9 shows the effects of γ-CyD and HP-γ-CyD on DSC thermograms of N-liposomes (5 mM of total lipids). The peak derived from gel-to-fluid state transition in N-liposomes was observed at 55 • C, while new peak was appeared at 42 • C in the presence of 50 mM of γ-CyD (Figure 9(a) ). On the other hand, no significant change in DSC thermographs was observed in the presence of HP-γ-CyD (Figure 9(b) ). These results suggest that γ-CyD may affect the membrane fluidity of Nliposomes. Generally, membrane fluidity and phase transition of liposomes are determined by the intensity of lipid-lipid interactions such as hydrophobic interaction, van der Waals forces, and hydrogen bond. Therefore, to evaluate the effects of γ-CyD on lipid-lipid interactions in AF-liposomes, we performed DSC analysis of DLPC-liposomes in the presence and absence of γ-CyD. The reason why we used DLPC-liposomes is due to clear observation of lipid-lipid interaction using DLPC composed of AF-liposomes. Figure 10 The charge ratio of AF-liposomes/pDNA was 1.6. pDNA/AF-liposome was prepared using a freeze-thaw method which was repeated from 0 to 5 cycles. Cells were incubated with pDNA/γ-CyDs/AF-liposome for 3 h in FCS-free medium. After washing twice, the cells were incubated for 21 h in culture medium supplemented with 10% FCS. The luciferase activity in cell lysates was determined using a luminometer. Each value represents the mean ± SEM of 3 experiments. effects of γ-CyD and HP-γ-CyD on DSC thermograms, phase transition temperature, and enthalpy (ΔHcal) of DLPCliposomes (5 mM of total lipids). The phase transition temperature (Tc) of DLPC-liposomes in the presence of γ-CyD was shifted to high temperature as the concentration of γ-CyD was increased (Figures 10(a) and 10(c) ). The ΔHcal value of DLPC-liposomes in the γ-CyD system was drastically elevated at 20 mM of γ-CyD (Figure 10(d) ). Meanwhile, in the case of HP-γ-CyD, there was no significant change in DSC thermograms, phase transition temperature, and the ΔHcal values (Figures 10(b) , 10(c) and 10(d)). Taken together, these results strongly suggest that γ-CyD enhances the lipid-lipid interaction of DLPC-liposomes, leading to the membrane stabilization of DLPC-liposomes. 3.6. Cellular Uptake of pDNA/AF-Liposomes. Next, we examined the cellular uptake of pDNA/γ-CyD/AF-liposomes into HepG2 cells using a CLSM. Figure 11 shows the CLSM images for distribution of RH-AF-liposomes and Alexa-pDNA in HepG2 cells at 3 h after transfection. The strong fluorescence derived from RH-AF-liposomes and Alexa-pDNA in the presence of γ-CyD was mainly observed in cytoplasm of HepG2 cells. Meanwhile, the fluorescence RH-AF-liposomes and Alexa-pDNA in the absence of CyD and with HP-γ-CyD was mainly observed on cell surface. Hence, these results suggest that pDNA/γ-CyD/AF-liposomes can be internalized into HepG2 cells to a larger extent, compared to pDNA/AF-liposomes and pDNA/γ-CyD/AF-liposomes. In this study, we clarified that pDNA/γ-CyD/AF-liposomes have potent hepatocyte-selective gene transfer activity and negligible cytotoxicity, compared to pDNA/AF-liposomes and pDNA/HP-CyDs/AF-liposomes. In cationic liposome-mediated gene transfection, lipid composition and lipid type are the most important physicochemical factors, because they affect not only the interaction with pDNA but also the affinity to target cells [36, 37] . In the present study, we prepared AF-liposomes with a lipid composition of TMAG/DOPE/DLPC/DPPE (2/4/3/1, molar ratio). DOPE is known to enhance the endosomal escape of pDNA due to its structural change into hexagonal II form at pH 5-6, an endosomal pH range, resulting in destabilizing endosomal membranes [38, 39] . TMAG, a cationic lipid, makes it possible to interact with pDNA in AF-liposomes. Additionally, DPPE was used as a binding lipid with AF. Actually, cationic liposomes composed of TMAG/DOPE/DLPC (1/2/2, molar ratio) are commercially available transfection reagents as GeneTransfer, which have already been utilized in a clinical trial for a nonviral vector to deliver the interferon-β gene for the treatment of brain tumor in Japan [40] . Therefore, we used AF-liposomes composed of these lipids in the present study. The most important finding found in the present study is that γ-CyD enhances transfection efficiency of pDNA/AFliposomes in HepG2 cells (Figures 3-5 ) with negligible cytotoxicity (Figure 8 ). The enhancing mechanisms of γ-CyD presumed are discussed as follows. In the present study, we revealed that transfection efficiency of the pDNA/γ-CyD/AF-liposomes, not N-liposomes, was inhibited by the addition of AF in HepG2 cells ( Figure 6 ). Meanwhile, in NIH3T3 cells, transfection efficiency of the pDNA/AF-liposomes was not suppressed by the addition of AF. These results strongly suggest that pDNA/γ-CyD/AFliposomes can be entered HepG2 cells through ASGP-Rmediated endocytosis, consistent with Aramaki and his colleague's report [41] , and the enhancing effect of the γ-CyD may by associated with the ASGP-R-mediated endocytosis. However, γ-CyD also enhanced gene transfer activity of pDNA/AF-liposomes even in A549 cells and NIH3T3 cells, ASGP-R negative cells (Figure 7) . These results suggest that the enhancing effect of γ-CyD on transfection efficiency of pDNA/AF-liposomes is in an ASGP-R-independent manner. The charge ratio of AF-liposomes/pDNA was 1.6. γ-CyDs were added to AF-liposomes suspension before freeze-drying. The concentrations of γ-CyDs were 1 μM/μM lipids. Cells were incubated with Alexa-pDNA/γ-CyDs/RH-AF-liposomes for 3 h in FCS-free medium. After washing twice, the cells were observed using a confocal laser scanning microscopy. The particle sizes and encapsulation ratio of pDNA/γ-CyD/AF-liposomes should be involved in the enhancing effect of γ-CyD on gene transfer activity of AFliposomes. The particle size of pDNA/γ-CyD/AF-liposomes was decreased in the presence of γ-CyD, although the ζpotential value of pDNA/γ-CyD/AF-liposomes was almost equivalent to that of pDNA/AF-liposomes and pDNA/HP-γ-CyD/AF-liposomes ( Table 2 ), suggesting that γ-CyD inhibits the aggregation of pDNA/AF-liposomes, because the particle size shows more than 200 nm, despite the fact that the liposomes were extruded through a filter membrane having a pore size of 200 nm. Meanwhile, the encapsulation ratio of pDNA was significantly increased by adding γ-CyD to pDNA/AF-liposomes (Table 3) . Thus, these lines of evidence speculate that addition of γ-CyD enhances cellular uptake of pDNA/AF-liposomes. In fact, the CLSM study demonstrated that cellular uptake of pDNA/γ-CyD/AF-liposomes was higher than that of pDNA/γ-CyD/AF-liposomes ( Figure 11 ). Furthermore, we confirmed that transfection efficiency of pDNA/γ-CyD/AF-liposomes was increased, as the encapsulation ratio of pDNA into pDNA/γ-CyD/AF-liposomes was augmented (Table 4 ). In view of the findings, the particle size of pDNA/γ-CyD/AF-liposomes and encapsulation ratio of pDNA into pDNA/γ-CyD/AF-liposomes are crucial role for enhancing transfection efficiency of pDNA/AF-liposomes. The important question regarding the enhancing effect of γ-CyD on transfection efficiency of pDNA/AF-liposomes still remains, because three types of HP-CyDs did not have the enhancing effect. To address this question, the DSC analysis was performed. This study indicated that γ-CyD, but not HP-γ-CyD, changed membrane fluidity and stabilized the N-liposomal membranes (Figure 9 ). In fact, γ-CyD increased the Tc value of DLPC-liposomes, although HPγ-CyD did not increase anymore ( Figure 10 ). This increase in the Tc value induced by γ-CyD could be attributed to a compactness of lipid bilayer of DLPC liposomes. It is thereby possible that γ-CyD may increase Tc values of the other liposomes such as DMPC, DPPC, and DSPC. Here, it is well known that γ-CyD is highly hydrophilic and surface inactive [17] . Therefore, we presumed that γ-CyD encapsulates the aqueous compartment of liposomes rather than in the bilayer of liposomes. Anyhow, it is clear that the magnification of the interaction of liposomal membranes containing DLPC with HP-γ-CyD is weaker than that with γ-CyD, possibly due to the steric hindrance of the HP group in a HP-γ-CyD molecule. Taken together, it is likely that the stabilizing effects of γ-CyD on AF-liposomal membrane may lead to inhibition of pDNA leakage from AF-liposomes and increase in cellular uptake of pDNA, eventually leading to the enhancement of in vitro transfection efficiency of pDNA/γ-CyD/AF-liposomes in cells. Finally, we investigated the role of free γ-CyD on transfection efficiency of pDNA/γ-CyD/AF-liposomes in HepG2 cells. The physical mixture of pDNA/AF-liposomes and γ-CyD in culture medium had no enhancing effect on transfection efficiency of pDNA/AF-liposomes (data not shown). Therefore, encapsulation of γ-CyD into AF-liposomes may be pivotal for enhancing gene transfer activity. To reveal the detailed mechanism for the enhancing effect of γ-CyD associated in AF-liposomes on transfection efficiency of pDNA/AF-liposomes, further elaborate study should be necessary. In the present study, we demonstrated that γ-CyD enhanced gene transfer activity of pDNA/AF-liposomes in not only HepG2 cells but also A549 and NIH3T3 cells, probably due to various effects of γ-CyD on AF-liposomes such as inhibition of aggregation of the liposomes, high encapsulation of pDNA into the liposomes, and stabilization of lipid bilayer of the liposomes. Consequently, the potential use of γ-CyD could be expected as an enhancer of gene transfer activity of AFliposomes. Also, these data may be useful for design of cellspecific cationic liposomes as a nonviral vector.

Participation of the Cell Polarity Protein PALS1 to T-Cell Receptor-Mediated NF-κB Activation

BACKGROUND: Beside their established function in shaping cell architecture, some cell polarity proteins were proposed to participate to lymphocyte migration, homing, scanning, as well as activation following antigen receptor stimulation. Although PALS1 is a central component of the cell polarity network, its expression and function in lymphocytes remains unknown. Here we investigated whether PALS1 is present in T cells and whether it contributes to T Cell-Receptor (TCR)-mediated activation. METHODOLOGY/PRINCIPAL FINDINGS: By combining RT-PCR and immunoblot assays, we found that PALS1 is constitutively expressed in human T lymphocytes as well as in Jurkat T cells. siRNA-based knockdown of PALS1 hampered TCR-induced activation and optimal proliferation of lymphocyte. We further provide evidence that PALS1 depletion selectively hindered TCR-driven activation of the transcription factor NF-κB. CONCLUSIONS: The cell polarity protein PALS1 is expressed in T lymphocytes and participates to the optimal activation of NF-κB following TCR stimulation.

Establishment and maintenance of cell polarity is chiefly orchestrated by a tightly regulated interplay between three multi-protein complexes: i) Scribble (SCRIB)/Discs Large (Dlgh1)/Lethal giant larvae (Lgl) complex, ii) partitioning-defective (PAR) 3 and PAR6/ atypical protein kinase C (aPKC) complex, and iii) Crumbs (CRB)/ Protein Associated with Lin Seven 1 (PALS1)/ PALS1-associated tight junction protein (PATJ) complex [1, 2] . However, each complex is not exclusive, as PAR6 links PALS1 to PAR3/PAR6/aPKC [3] . In T lymphocytes, cell polarity proteins were shown to partition the leading edge from the uropod at the cell rear, and therefore participate to cell migration, homing, and scanning [4, 5, 6] . In addition, SCRIB and Dlgh1 are transiently recruited to the nascent immunological synapse formed with an antigen-presenting-cell (APC) [4] . Their depletion in lymphocytes has been associated with an alteration of antigen receptor-mediated activation [7, 8, 9, 10] . The adaptor PALS1 is crucial for cellular architecture as it maintains the apico-basal polarity in epithelial cells and authorizes indirect interactions between CRB and PATJ [11, 12] . Interestingly, Dlgh1 and PALS1 share a COOH-terminal part composed of a PSD-95/Dlg/ZO-1 (PDZ) domain followed by an SH3 domain adjacent to an inactive Guanylate kinase (GK) homology region [2] . This unique sequence of PDZ/SH3/GK defines the so-called membrane-associated guanylate kinase (MAGUK) proteins family, a group of molecules that serve as scaffolds to organize multi-protein signalosomes through their protein-protein interaction domains [13] . For example, the MAGUK-containing CARMA1 emerges as a central regulator of lymphocytes activation and proliferation downstream of antigen receptor stimulation [14] . Indeed, CARMA1 operates as scaffold to recruit the heterodimer BCL10/ MALT1 (CBM complex), a key step for conveying NF-kB signaling [14, 15, 16] . In addition to its established role in polarity, Dlgh1 was shown to modulate lymphocyte proliferation upon T-cell receptor ligation, possibly through p38 recruitment or via the transcription factor NF-AT [9, 10, 17, 18] . Although the MAGUK PALS1 plays a central role in the establishment of cell polarity, its contribution to lymphocyte activation remains elusive [8] . Here we show that PALS1 mRNA and protein is expressed in human lymphocytes. Furthermore, knocking down of PALS1 with small interfering RNAs (siRNAs) led to a decreased proliferation of human T lymphocytes, resulting from a reduced activation of the transcription factor NF-kB. Although several cell polarity proteins have been characterized in lymphocytes [4, 5] , PALS1 expression in T cells remains to be determined [8] . To address this question, we first performed RT-PCR analysis on resting human CD3 + T cells and Jurkat lymphocytes extracts, and detected mRNA for PALS1 ( Figure 1A ). These mRNA were efficiently translated into protein, as antibodies against PALS1 detected a band, which was absent from PALS1-siRNA transfected primary T lymphocytes lysates ( Figure 1B ). Similar results were obtained with Jurkat T cells ( Figure 1B ). Of note, PALS1 levels remained unchanged in cells stimulated with antibodies to CD3 and CD28, or with PMA and ionomycin ( Figure 1C ). We next investigated PALS1 subcellular location by confocal microscopy. In contrast to epithelial cells where it accumulate to tight junctions [12] , PALS1 did not reach membrane domains and remains essentially cytosolic with punctuate structures. Additional staining revealed that these structures coalesced with the Golgi apparatus ( Figure 1D and Figure S1 ). Accordingly, Brefeldin A-triggered disassembly of the Golgi apparatus also disrupted PALS1 punctuate structures ( Figure 1D ). This is reminiscent of PALS1 relocation to the Golgi apparatus in cells infected with SARS coronovirus [19] . Last, we observed that TCR-mediated stimulation only promoted a discrete redistribution of PALS1 within the cytosol of Jurkat cells ( Figure S1 ). Altogether, our results suggest that similarly to Dlgh1, SCRIB, CRB3, and PKCf [4, 5] , the cell polarity protein PALS1 is expressed in lymphocytes at both mRNA and protein level. Because SCRIB and Dlgh1 were proposed to modulate lymphocyte proliferation [7, 8, 9, 10] , we evaluated whether PALS1 might also participate to T cell activation. To this end, peripheral blood lymphocytes (PBL) were purified on Ficoll-isopaque gradients. Primary human T cells were nucleofected for three days with siRNA targeting PALS1, prior stimulation with anti-CD3 and anti-CD28 antibodies. PALS1 knockdown led to a significant decrease in TCR-mediated induction of the activation markers CD69 and CD25 on cell surface (Figure 2A , B). This was accompanied by a reduction in Carboxyfluorescein Succinimydyl Ester (CFSE) dilution, which reflects cell proliferation ( Figure 2C ). Collectively, these data suggest that PALS1 participates to the optimal lymphocyte activation and subsequent proliferation upon TCR stimulation. To further explore how PALS1 impacts on lymphocyte proliferation, early signaling pathways emanating from the TCR were examined in Jurkat cells transfected with PALS1 siRNA. We did not detect major alteration in the general pattern of tyrosine phosphorylation, or mitogen-activated protein kinase (MAPK) extracellular signal-regulated kinases (ERK) 1/2 phosphorylation upon TCR stimulation ( Figure 3A ). Only a slight but consistent increase in TCR-mediated phosphorylation of p38 was noted ( Figure 3A ). Moreover, CD3-induced calcium mobilization was largely normal in PALS1-knockdown Jurkat cells ( Figure 3B ). We next analyzed TCR-mediated activation of NF-AT and NF-kB transcription factors. siRNA-treated Jurkat T cells were cotransfected with firefly luciferase constructs driven by NF-AT or NF-kB binding sequences and with a renilla luciferase control. PALS1 knockdown had only a marginal effect on NF-AT activity following stimulation with PMA and ionomycin, or with antibodies to CD3 and CD28 ( Figure 3C and Figure S2 ). In sharp contrast, NF-kB activity was significantly reduced without PALS1 ( Figure 3D ). Interestingly, tumor necrosis factor-a (TNFa)induced NF-kB activation remained essentially unaffected, underscoring the selective involvement of PALS1 in the TCR-NF-kB pathway ( Figure S3 ). Altogether, our data unveiled an unexpected role for PALS1 in TCR-mediated NF-kB activation. To gain insights on how PALS1 modulate NF-kB, we first investigated the transcription factor binding ability by electrophoretic mobility shift assay (EMSA). Less NF-kB bound to its specific probe in nuclei extracts from PALS1-siRNA transfected cells following TCR stimulation ( Figure 4A ). As expected, Oct-1 binding remained unchanged without PALS1. Consistent with a diminished NF-kB activity, both the phosphorylation and subsequent proteasomal degradation of NF-kB inhibitor, IkBa, were severely decreased in the absence of PALS1 ( Figure 4B ). Because TCR-induced NF-kB activation relies on the assembly of the CBM complex [15] , BCL10 was immunoprecipitated from nonspecific (NS-) and PALS1-siRNA transfected Jurkat cells. MALT1, which forms an heterodimer with BCL10, coprecipitated with BCL10 regardless of stimulation. Although PALS1 was not found bound to BCL10, its absence diminished CARMA1 recruitment ( Figure 4C , and data not shown). Hence, our data suggest that PALS1 participates to the optimal translocation and activation of NF-kB upon TCR stimulation, possibly by favoring the CBM assembly. Since PALS1 nucleates a ternary complex containing CRB3 and PATJ, and further binds PAR6 to maintain cell polarity [3, 20, 21] , their contribution to TCR-mediated NF-kB was evaluated. Similarly to PALS1, mRNA for PATJ, CRB3, PAR6, were efficiently detected by RT-PCR ( Figure 5A ). The same hold true for the unrelated cell polarity protein SCRIB ( Figure 5A ). siRNA-based knockdown of PALS1 and CRB3 significantly decreased NF-kB activation in cells treated by antibodies against CD3 and CD28, or with a mixture of PMA and ionomycin. Although less dramatic, similar results were observed with PAR6 or PATJ knockdown. By contrast, NF-kB was normally activated in the absence of SCRIB ( Figure 5B ). In agreement, IkBa phosphorylation was diminished in lysates from CRB3-depleted cells, and to a lesser extent from PATJ-or PAR6-siRNA transfected cells, and not from SCRIB-depleted cells. Again, ERK phosphorylation occurred normally ( Figure 5C , D, E, and F). Altogether, our data suggest that PALS1 implication in the TCR-NF-kB pathway is inextricably linked to its cell polarity partners. In summary, our data show that the cell polarity protein PALS1 is expressed in lymphocytes and contributes to their optimal activation. Although Dlgh1 and SCRIB were proposed to modulate NF-AT or p38 [9, 10, 17, 18] and NF-AT [7] respectively, a distinct scenario likely occurs for PALS1. Our results support a model in which PALS1 participates to NF-kB activation, upstream of IkBa phosphorylation and degradation. However, how precisely PALS1 modulates NF-kB remains unclear. Because MAGUK function as scaffold units to organize and integrate multi-molecular signaling complexes [13] , it is tempting to speculate that PALS1 nucleates its own signalosome. For example, CARMA1 anchors a .900 kDa complex including the heterodimer BCL10/MALT1 [22] , and Dlgh1 was reported to bind to Lck, Zap70, Wasp [17] , and p38 [10] . In our hands, PALS1 did not integrate the CBM, but its absence reduced CARMA1 binding to BCL10. It will therefore be interesting to identify PALS1 partners in the context of lymphocyte activation. In line with this, CRB3, PATJ and PAR6, which all bound PALS1 to maintain cell polarity [2] , also participate to NF-kB signaling upon TCR ligation in lymphocytes, and might therefore complex with PALS1 in lymphocytes. Altogether, our results strengthen the unexpected function of cell polarity proteins in lymphocyte proliferation [7, 8, 9, 10] , and unveil an original role for PALS1 during TCR-mediated NF-kB activation. Jurkat T cells E6.1 were purchased from ATCC. CD3 + human T lymphocytes from healthy donors (Etablissement Francais du Sang) were isolated with the MidiMacs system (Miltenyi Biotec). Cells were activated with a mixture of soluble anti-CD3e (HIT3a, BD Biosciences) and anti-CD28 (BD Biosciences), or with 20-40 ng.ml 21 phorbol 12-myristate 13-acetate (PMA, Sigma) and 300 ng.ml 21 ionomycin (Calbiochem). Carboxyfluorescein Succinimydyl Ester (CFSE) and Brefeldin A were purchased from Sigma, and the calcium-sensitive dye Fluo-4 was from Invitrogen. Cells were washed twice with PBS 1X and lysed with 50 mM Tris pH 7.4, 150 mM NaCl, 1% Triton X-100, 1% Igepal, 2 mM EDTA, supplemented with complete protease inhibitors (Roche). Lysates were cleared by a centrifugation at 10,000g at 4 o C, and protein concentration determined (micro BCA kit, Pierce). Samples were resolved on 5-20% SDS-PAGE gels and transferred to nitrocellulose membranes (Amersham). For Immunoprecipitations, samples were precleared with protein G-sepharose beads (Roche) for 30 min prior to overnight incubation with antibodies and additional protein G-sepharose beads at 4uC, as previously described [23] . Antibodies to BCL10 (A-6), IkBa (C-21), MALT1 (B-12), Tubulin (TU-02), PALS1 (H-250), SCRIB (C-6), PAR6 (G-9) and p65 (C-20) were purchased from Santa Cruz. Phosphospecific antibodies against IkBa, ERK, p38, and antibodies to CARMA1 and to ERK were from Cell Signaling Technologies. Anti-phosphorylated Tyrosine (4G10, Millipore), anti-GAPDH (Sigma), and Immobilon (Millipore) chemiluminescent substrates were also used. Firefly luciferase constructs downstream of promoters for NF-kB or NF-AT were co-transfected with renilla luciferase pRL-TK (Int -) plasmid (Promega). Luciferase activities were analyzed using the Dual-Luciferase Kit (Promega), with firefly fluorescence units normalized to renilla luciferase fluorescence units (BMG microplate reader). were purified from blood on Ficoll-isopaque gradients. PBL were nucleofected with the Nucleofactor system and T cell solution (Amaxa, program U14), and left for three days in culture medium prior treatment. Nuclear protein extraction and electrophoretic mobility shift assay (EMSA) 4 mg of nuclear extracts from Jurkat cells were examined for NF-kB-and Oct1-binding activity by electromobility shift assay (Panomics kit). Samples were resolved on a 6% native polyacrylamide DNA retardation gel in 0.5X TBE buffer and analyzed using a FUJI LA4000 system. Cells were left for 10 min on poly-lysine coated slides (Thermo Scientific) prior fixation with PBS1X containing 4% paraformaldehyde. For TCR crosslinking experiments, cells were incubated with 5 mg.ml 21 anti-CD3 at 4uC for 15 min. After two washes, cells were incubated with 5 mg.ml 21 of goat antimouse (Jackson) for 20 min either at 4uC or 37uC. To disassemble Golgi apparatus, cells were treated with 10 mg.ml 21 Brefeldin A for 60 min. Samples were permeabilized with 0.05% Triton-X100 in PBS1X for 5 min, and non-specific sites blocked with 10% FCS in PBS1X. Antibodies used were: PALS1 (Millipore), 58K Golgi (Abcam), Alexa-488 conjugated goat anti-rabbit IgG or Alexa-594 conjugated goat anti-mouse IgG (Invitrogen). Samples were analyzed using a Leica confocal microscope SP6. Cells were incubated for 30 min at 4uC with FITC-and PEconjugated antibodies against CD25 and CD69 (ImmunoTools) and the respective isotype controls in PBS containing 0.5% BSA. After one wash with ice-cold PBS-BSA, cells were analyzed by flow cytometry with a FACSCalibur (BD Biosciences).

A Canadian Critical Care Trials Group project in collaboration with the international forum for acute care trialists - Collaborative H1N1 Adjuvant Treatment pilot trial (CHAT): study protocol and design of a randomized controlled trial

BACKGROUND: Swine origin influenza A/H1N1 infection (H1N1) emerged in early 2009 and rapidly spread to humans. For most infected individuals, symptoms were mild and self-limited; however, a small number developed a more severe clinical syndrome characterized by profound respiratory failure with hospital mortality ranging from 10 to 30%. While supportive care and neuraminidase inhibitors are the main treatment for influenza, data from observational and interventional studies suggest that the course of influenza can be favorably influenced by agents not classically considered as influenza treatments. Multiple observational studies have suggested that HMGCoA reductase inhibitors (statins) can exert a class effect in attenuating inflammation. The Collaborative H1N1 Adjuvant Treatment (CHAT) Pilot Trial sought to investigate the feasibility of conducting a trial during a global pandemic in critically ill patients with H1N1 with the goal of informing the design of a larger trial powered to determine impact of statins on important outcomes. METHODS/DESIGN: A multi-national, pilot randomized controlled trial (RCT) of once daily enteral rosuvastatin versus matched placebo administered for 14 days for the treatment of critically ill patients with suspected, probable or confirmed H1N1 infection. We propose to randomize 80 critically ill adults with a moderate to high index of suspicion for H1N1 infection who require mechanical ventilation and have received antiviral therapy for ≤ 72 hours. Site investigators, research coordinators and clinical pharmacists will be blinded to treatment assignment. Only research pharmacy staff will be aware of treatment assignment. We propose several approaches to informed consent including a priori consent from the substitute decision maker (SDM), waived and deferred consent. The primary outcome of the CHAT trial is the proportion of eligible patients enrolled in the study. Secondary outcomes will evaluate adherence to medication administration regimens, the proportion of primary and secondary endpoints collected, the number of patients receiving open-label statins, consent withdrawals and the effect of approved consent models on recruitment rates. DISCUSSION: Several aspects of study design including the need to include central randomization, preserve allocation concealment, ensure study blinding compare to a matched placebo and the use novel consent models pose challenges to investigators conducting pandemic research. Moreover, study implementation requires that trial design be pragmatic and initiated in a short time period amidst uncertainty regarding the scope and duration of the pandemic. TRIAL REGISTRATION NUMBER: ISRCTN45190901

Background: Swine origin influenza A/H1N1 infection (H1N1) emerged in early 2009 and rapidly spread to humans. For most infected individuals, symptoms were mild and self-limited; however, a small number developed a more severe clinical syndrome characterized by profound respiratory failure with hospital mortality ranging from 10 to 30%. While supportive care and neuraminidase inhibitors are the main treatment for influenza, data from observational and interventional studies suggest that the course of influenza can be favorably influenced by agents not classically considered as influenza treatments. Multiple observational studies have suggested that HMGCoA reductase inhibitors (statins) can exert a class effect in attenuating inflammation. The Collaborative H1N1 Adjuvant Treatment (CHAT) Pilot Trial sought to investigate the feasibility of conducting a trial during a global pandemic in critically ill patients with H1N1 with the goal of informing the design of a larger trial powered to determine impact of statins on important outcomes. Methods/Design: A multi-national, pilot randomized controlled trial (RCT) of once daily enteral rosuvastatin versus matched placebo administered for 14 days for the treatment of critically ill patients with suspected, probable or confirmed H1N1 infection. We propose to randomize 80 critically ill adults with a moderate to high index of suspicion for H1N1 infection who require mechanical ventilation and have received antiviral therapy for ≤ 72 hours. Site investigators, research coordinators and clinical pharmacists will be blinded to treatment assignment. Only research pharmacy staff will be aware of treatment assignment. We propose several approaches to informed consent including a priori consent from the substitute decision maker (SDM), waived and deferred consent. The primary outcome of the CHAT trial is the proportion of eligible patients enrolled in the study. Secondary outcomes will evaluate adherence to medication administration regimens, the proportion of primary and secondary endpoints collected, the number of patients receiving open-label statins, consent withdrawals and the effect of approved consent models on recruitment rates. Discussion: Several aspects of study design including the need to include central randomization, preserve allocation concealment, ensure study blinding compare to a matched placebo and the use novel consent models pose challenges to investigators conducting pandemic research. Moreover, study implementation requires that trial design be pragmatic and initiated in a short time period amidst uncertainty regarding the scope and duration of the pandemic. Trial Registration Number: ISRCTN45190901 Background Influenza and Swine Origin Influenza A/H1N1 Infection (H1N1) On June 11, 2009 , the World Health Organization (WHO) declared that infection with the Swine Origin Influenza A/H1N1 virus had reached pandemic proportions [1] . Cases were recorded in more than 180 countries and outbreaks that strained national resource capacities were documented in Canada, Australia, Chile, Argentina, and elsewhere. Throughout history, pandemic influenza has posed a recurrent threat to human populations. Seasonal influenza is responsible for more than 50,000 deaths per year in the United States [2] . The capacity of the influenza virus to mutate and spread from animals to humans has resulted in intermittent pandemics. The 1918 pandemic was the largest in recent history and caused between 40 and 50 million deaths worldwide [3] . Smaller pandemics in 1957 and 1968 were associated with mortality spikes but their effects were mild at the population level [4] . Experts believe further pandemics will certainly occur, but are uncertain about when. Several years ago, the avian H5N1 influenza virus threatened to be the vector of the next pandemic. While highly virulent when transmitted from infected chickens to humans, the absence of human-to-human transmission resulted in a small number of cases worldwide [5] . In early 2009, a novel strain of influenza, swine origin influenza A/H1N1 infection (H1N1) emerged in swine and rapidly spread to humans [6] . Originating in Mexico, the strain proved highly infectious and was spread by person-to-person contact with a predilection for younger hosts. While early epidemiologic data suggested that although H1N1 was highly infectious, it was less virulent [7, 8] than anticipated with a case fatality rate of approximately 0.5% of infected individuals. For the majority of infected individuals, symptoms were mild and selflimited; however, a small percentage of infected individuals developed profound respiratory failure requiring extraordinary means of oxygenation support including high frequency oscillation (HFO) ventilation and extracorporeal membrane oxygenation (ECMO) [9] . Caring for the most severely ill patients during a pandemic results in an increased need for intensive care unit (ICU) resources and strains available personnel and equipment. There is little excess capacity to care for critically ill patients in most developed countries, and minimal capacity in developing countries. Supportive care and antiviral agents, especially neuraminidase inhibitors (such as oseltamivir and zanamivir), are the mainstay of treatment for influenza. While efficacious in reducing viral load and abating symptoms in ambulatory patients, their effects on outcomes in critically ill patients have not been established. Their utility in treating severe pandemic H1N1 influenza may be compromised by widespread use and emergence of viral resistance [10, 11] , limited supplies, policies regarding treatment strategies, and cost and availability, especially in developing countries [12] . Human and animal data suggest that the course of influenza may be favorably influenced by certain agents not classically considered as treatments for influenza [13] [14] [15] [16] that are comparatively inexpensive and readily available. Such agents may provide independent benefit in treating viral infection and are attractive as adjuvant treatments. Statins lower plasma lipid levels by inhibiting HMG CoA reductase, the enzyme responsible for converting HMG CoA to mevalonate, a rate-limiting step in cholesterol biosynthesis. Multiple observational studies have suggested that statins may be of benefit in patients with a variety of severe infections [17] [18] [19] [20] [21] [22] by exerting an effect in attenuating inflammation [23, 24] . The mechanism of this activity is uncertain but may involve their ability to restrict cholesterol availability in cell membranes of the innate immune system. Cholesterol is a key component of lipid rafts, membrane-associated microdomains that support cell signaling in response to exogenous inflammatory stimuli. Raft disruption may attenuate the cellular response to inflammatory stimuli [25, 26] . Experimental studies show that pre-treatment with statins attenuates the severity of acute lung injury (ALI) following intestinal ischemia-reperfusion [27] and the inflammatory response to intravenous lipopolysaccharide challenge in human volunteers [28] . The combination of a statin and caffeine inhibits viral replication and attenuates lung injury in murine influenza models [29] . Population-based studies suggest that statins are associated with reduced inflammatory morbidity in critically ill patients receiving them [20] and reduced mortality in patients with influenza and chronic obstructive pulmonary disease (COPD) patients [30] . Reviewing 3,921 hospitalized patients with laboratory confirmed influenza, of whom 1,019 were receiving a statin at the time of admission, Vandermeer and colleagues found that statin use was independently associated with a reduced risk of death (adjusted OR = 0.34, 95% CI: 0.16-0.70) in a multivariable logistic regression model [31] . The benefits of statins appear to be best established in patients receiving them prior to the onset of infection. Cohort studies report conflicting results on the ability of statins to attenuate organ dysfunction with Schmidt and colleagues finding that statins reduce mortality in critically ill patients with multiple organ dysfunction syndrome [32] and Kor et al reporting that statin use was not beneficial in resolving organ dysfunction [33] . Statins are widely used and well-tolerated medications. Rosuvastatin differs from other HMG CoA reductase inhibitors in that only up to 10% of the parent compound is metabolized by cytochrome P450 2C9 and 2C19 and not by P450 3A4 [34] . As a result, it exhibits fewer drug-drug interactions, and serum concentrations are not affected by CYP2D6 gene polymorphism. Adverse events seen with rosuvastatin are generally mild and may include muscle symptoms, however, myopathy and rhabdomyolysis occur infrequently in patients taking 40 mg/day or less. As with other HMG-CoA reductase inhibitors, a dose-related increase in liver transaminases and creatinine kinase (CK) has been observed in small numbers of patients taking rosuvastatin. The overall occurrence of clinically significant transaminase increases is low (< 1%) and similar across rosuvastatin doses ranging from 5 mg/day to 40 mg/day [35] . Patients with severe liver disease may have increased exposure to rosuvastatin [35] . The International Forum of Acute Care Trialists (InFACT) is an informal alliance of investigator-led clinical trials networks whose remit is to improve the care of critically ill patients through the promotion of scientifically rigorous clinical research. Based on our preliminary understanding of H1N1 influenza, members of the Canadian Critical Care Trials Group (CCCTG), working in collaboration with members of InFACT, designed the Collaborative H1N1 Adjuvant Treatment (CHAT) Pilot Trial to investigate the feasibility of conducting an international therapeutic trial during a global pandemic and the potential for adjuvant rosuvastatin, in addition to standard treatment, to influence clinical outcomes in influenza A (H1N1) associated critical illness. A multi-national feasibility RCT involving adult ICUs in Canada, Saudi Arabia, Mexico, Argentina Australia/ New Zealand. An overview of the study design is provided in Figure 1 (see Figure 1 ). [1] The ability to recruit the desired patient population under pandemic conditions (i.e., the proportion of eligible patients enrolled in the CHAT Pilot Trial). [2] Adherence to the medication administration regimen as outlined in the study protocol. [3] The ability to collect the required primary and secondary endpoints for the planned full CHAT trial. [4] The number of patients who receive open-label statins. [5] The number of consent withdrawals. [6] The impact of approved consent models on recruitment rates. A dedicated research coordinator will screen patients for eligibility on a daily basis in the ICUs at participating sites. If the study inclusion criteria are fulfilled and no exclusion criteria are present, the research coordinator will identify the patient as a potential study participant. The attending physician or intensivist will confirm eligibility for participation in the CHAT Pilot RCT. Criteria to identify potential candidates for study inclusion will include: 1) Critically ill adult patients ≥ 16 years of age admitted to an adult ICU for any reason with suspected, probable or confirmed novel swine origin influenza A/H1N1 infection (see Additional File 1). 2) Requiring mechanical ventilation (invasive or noninvasive) 3) Receiving antiviral therapy (any medication at any dose and for any intended duration) for ≤ 72 hours 4) Attending physician or intensivist must have a 'moderate' to 'high' index of suspicion for H1N1. 3) Weight < 40 kg 4) Unable to receive or unlikely to absorb enteral study drug (e.g., incomplete or complete bowel obstruction, intestinal ischemia, infarction, short bowel syndrome) 5) Rosuvastatin specific exclusions: a. Already receiving a statin b. Allergy or intolerance to statins. c. Receiving niacin, fenofibrate, cyclosporine, gemfibrozil, any protease inhibitor (including but not limited to lopinavir and ritonavir) or planned use of oral contraceptives or estrogen therapy during the ICU stay. d. CK exceeds 5,000 U/L or ALT exceeds 8 times the upper limit of normal (ULN). 6) Severe chronic liver disease (Child-Pugh Score 11-15) (see Additional File 2) 7) Previous enrolment in this trial 8) Pregnancy or breast feeding 9) At the time of enrolment, receipt of > 72 hours of antiviral therapy. 10) Known or suspected clinically significant myositis or myopathy. We will record reasons why eligible patients are not randomized into the CHAT Pilot Trial under the following categories: a) Substitute Decision Maker (SDM) consent refusal b) Physician refusal of consent c) SDM not available to provide consent and waived/ deferred consent not permitted d) SDM does not exist and waived/deferred consent not permitted e) Coordinator workload f) Coordinator not available during eligible timewindow g) Enrolment in a competing trial h) Other (specification required) To enhance trial feasibility and given the need to conduct a pragmatic trial, co-enrolment in other prospective observational studies or RCTs (not investigating similar or alternative H1N1 treatments) in operation in the ICU setting will be permitted and recorded in sites where permitted by the local Research Ethics Boards (REBs) [36] . To preserve allocation concealment, participants will be randomized centrally. Randomization lists will be distributed by the study methods centre to the research pharmacies of participating centres. Stratified variable block randomization, based on centre alone, will be performed to take into consideration differences in patient characteristics at participating ICUs. Day one will be considered the day of study treatment initiation, which may or may not be the same day of randomization. A waiver of consent is the preferred option for participant enrollment given the context of a global pandemic. We have constructed a consent algorithm to direct the consent process at sites where the local REB has not approved a waiver of consent (see Additional File 3). In this trial, we will request that patients be enrolled in the study, and consent be deferred to SDMs or to the patient (whomever is able to provide consent first), when it is not possible to obtain consent within 24 hours. Consent may be obtained in person or by telephone as per local practices. In the event that patients die before providing consent, we request permission from REBs to include data collected during study participation. Using randomization lists provided by the study methods centre, research pharmacists will assign critically ill adults to once daily enteral administration of rosuvastatin or matched placebo for 14 days. Only the research pharmacy staff will be aware of the assigned treatment arm. The site investigators, research coordinator, clinical pharmacist involved in the care of the patient and all other study personnel will remain blinded to treatment assignment. An oral placebo for nasogastric administration, identical in appearance (colour and consistency matched) to crushed rosuvastatin, will be prepared by Pharmacy 1 (Toronto, Canada) and supplied to the study sites. All other aspects of patient management will be left to clinician discretion as per pragmatic trial design. The Applied Health Research Centre of the Keenan Research Centre and Li Ka Shing Knowledge Institute (St Michael's Hospital, Toronto, Ontario) will be the Study Methods Centre. Study drug will be administered once daily through an enteral feeding tube or orally if the patient is able to safely take oral medications. The type and placement of the enteral feeding tube (nasogastric, nasoenteric, percutaneous endoscopic gastrostomy, orogastric, oroenteric, etc.) will be at the discretion of the attending clinical team. The ability to safely take oral medications will be determined by the patient's primary care team. The first study drug dose (rosuvastatin or placebo) will be administered within 4 hours of randomization as a loading dose of 40 mg unless the subject is of Asian (Chinese, Korean, Japanese, Phillipino, Vietnamese, or Asian-Indian) descent, age <18 years or has serum creatinine greater than or equal to 248 umol/L (2.8 mg/dL) (see requirements for dose adjustments below). Thereafter doses of 20 mg will be administered at 22:00 hrs daily (+/-4 hours) starting on the next calendar day (study day 2) as a maintenance dose. If the patient is of Asian descent, is <18 years, or serum creatinine is greater than or equal to 248 umol/L (2.8 mg/dL) dose adjustments will be required according to the dose adjustment algorithm (see requirements for dose adjustments below). Dose adjustment is only necessary for patients with renal impairment and not receiving dialysis. Once dialysis is started and serum creatinine remains elevated, dose adjustment is not required. If for any reason a maintenance dose is not administered at the intended time, it may be administered subsequently but not more than 12 hours after the intended time of administration. If greater than 12 hours has elapsed since the last scheduled dose, the patient will receive another loading dose, and then maintenance dosing will resume on the next calendar day as outlined above. A missed dose, for reasons other than outlined under medication discontinuation will be considered a protocol violation. The loading (40 mg) and daily maintenance (20 mg) doses will be reduced by 50% for patients: a) who have at least one parent of Asian descent (loading 20 mg and daily 10 mg), b) whose age < 18 years (loading 20 mg and daily 10 mg), c) whose serum creatinine concentration is greater than or equal to 248 umol/L (2.8 mg/dL) who are not on renal replacement therapy (loading 20 mg and daily 10 mg), and by 75% for patients: d) who have at least one parent of Asian descent and/or age < 18 years and serum creatinine is greater than or equal to 248 umol/L (2.8 mg/dL) (loading 10 mg and daily 5 mg). Intermittent oral antacids should be administered no closer than 6 hours before or after administering rosuvastatin to avoid influencing study drug absorption [37] Duration of Treatment Given the substantial potential for false negative influenza results, we will continue adjuvant treatment administration, regardless of H1N1 testing results (positive or negative), for 14 days or until a criterion for cessation is met. If patients are liberated from mechanical ventilation (invasive or non-invasive) and discharged from the ICU between days 1 and 9 they will be advanced to day 10 of study drug administration. In the event that patients remain in the ICU for observation (e.g., possible reintubation or initiation of non-invasive ventilation) then study drug administration will NOT be advanced to day 10 until they are discharged. Study drug administration will be stopped when one of the following conditions is met, whichever comes first: 1. 14 days after randomization 2. Hospital discharge (including transfer to an alternate care facility) 3. Death 4. CK noted to exceed 5,000 U/L (in the absence of an alternative diagnosis) or patient is determined to have clinical myositis or myopathy (at the discretion of the primary care team, patient may be re-challenged with study drug if CK or clinical findings no longer meet this criterion). 5. Alanine aminotransferase (ALT) exceeds 8 times the ULN (in the absence of an alternative diagnosis) 6. Co-administration of any of the following: niacin, fenofibrate, cyclosporine, gemfibrozil, lopinavir, ritonavir, oral contraceptives or estrogen 7. Attending physician or intensivist or SDM request to stop treatment. We will request for data collection to continue in patients withdrawn from therapy prematurely. Clinicians will be encouraged to continue study medication despite negative H1N1 testing due to the potential for false negative results and the potential role for an antiinflammatory agent (rosuvastatin) in severe lung disease. Decisions regarding continuation or discontinuation of antiviral treatment will be left to the discretion of the attending physician or intensivist. It will not be feasible to protocolize ventilator and general clinical management under pandemic conditions. Adjunctive non-antibiotic, non-interventional management of sepsis, acute respiratory distress syndrome, and glycemic control will be at the discretion of the patient's primary clinicians. Key aspects of clinical management (such as choice of antiviral, dose of administration, duration of treatment, ventilator strategy, treatment with inhaled nitric oxide, HFO, prone positioning, ECMO, additional antiviral/anti-inflammatory treatments and treated episodes of infection) will be documented either as part of the Influenza A H1N1 (Swine Flu) ICU (Registry) Study for participating centres or on separate data forms for centres not participating in the registry. Antibiotic therapy may be prescribed for suspected or confirmed concomitant bacterial infection at the discretion of the attending physician. Local testing procedures may be used to facilitate diagnosis of H1N1. Where no local testing procedures exist, we recommend using the following initial and repeat testing procedures. For all new admissions to adult ICUs meeting study inclusion criteria and having no exclusion criteria with non-confirmed H1N1 (i.e., all suspected or probable cases) or where uncertainty exists regarding prior testing, we request the following sequence of laboratory tests: Initial Diagnostic Testing (assuming diagnosis not confirmed at ICU admission) Repeat Testing (for patients with one positive test) 1. Repeat both tests (PCR and viral culture) from the site that was positive previously (i.e., ET or BAL or NP swab). If both ET and NP swab were positive, send repeat ET aspirate specimen at day 7 and at weekly intervals thereafter. 2. After the first negative specimen(s) from a previously positive site, send a repeat specimen from the same site at 48 hours after the first specimen was collected. To verify adequate absorption of rosuvastatin, we will draw venous blood for peak and trough plasma rosuvastatin concentrations (total of 2 specimens) on day 7 (+/-1 day). Day 7 will be the preferred day for trough and peak specimen collection. A trough level specimen will be drawn prior to day 7 (+/-1 day) dose. A peak concentration specimen will be drawn 3 to 5 hours after the dose of study drug is administered. A maximum of 80 patients will have blood drawn for drug concentration analysis; however, only patients randomized to rosuvastatin will have their samples analyzed. Samples will be labeled and batched at the site, for shipment to St Michael's Hospital (Toronto, Canada) for analysis after the study is unblinded. Research staff will assess participants daily for adverse effects for the duration of treatment. CHAT study participants will have daily CK and liver function [aspartate aminotransferase (AST) and ALT] levels collected as part of the study protocol. Study drug will be discontinued if hypersensitivity is suspected (see Criterion 7 Completion of Study Drug Administration). We will record the number of patients receiving full treatment and reasons for the inability to complete the assigned treatment duration (i.e., death, transfer to an alternate care facility, study withdrawal, etc). The study Data Safety and Monitoring Board (DSMB) will review all patients withdrawn from the study, safety data, and deaths. All patients will be followed until death or hospital discharge. We will record the vital status of all patients at 90 days and hospital discharge, whichever occurs first. At day 60 patients remaining on the ventilator will be deemed ventilator dependent. Randomized patients will be considered successfully extubated when they remain off positive pressure ventilation (invasive or non-invasive ventilation) for 48 consecutive hours. If patients are re-intubated within 48 hrs following extubation, they will be followed until they achieve one of the aforementioned outcomes. Patients discharged from the ICU and requiring readmission and re-initiation of mechanical ventilation (invasive or noninvasive) will be treated according to usual practice and will not be randomized on a second occasion to this study. [1] Proportion of eligible patients enrolled in the CHAT pilot study. [2] Adherence to the medication administration regimen as outlined in the study protocol. [3] Proportion of completed primary and secondary endpoints for the planned full CHAT trial that are collected. [4] Number of patients who receive open-label statins. [5] Number of consent withdrawals. [6] Recruitment rates by approved consent model. Estimates are not available to allow precise sample size estimation of the primary outcome for the proposed CHAT pilot RCT. We propose to undertake a pilot study in a convenience sample of 80 patients with suspected, probable or confirmed H1N1 infection to assess trial feasibility. We will collect CHAT specific data starting at ICU admission using paper-based versions of the electronic data collection forms developed for the Influenza A H1N1 (Swine Flu) ICU (Registry) Study [7] . The forms will document baseline characteristics, enrolment into concurrent influenza research studies, co-morbidities, illness severity (see Additional File 4), vaccination status, co-interventions, feasibility outcomes and clinical outcomes for the planned definitive trial (primary: the proportion of patients successfully weaned from mechanical ventilation in less than 10 days; secondary: impact of rosuvastatin on ICU, 60 and 90 day, and hospital mortality and on ICU free days at day 60). In addition to the data forms developed for the Influenza A H1N1 (Swine Flu) ICU study, we developed 13 additional forms including an (i) eligibility and randomization form, (ii) severity of illness form, (iii) consent form, (iv) drug administration form, (v) H1N1 diagnostic test results form, (vi) laboratory data form, (vii) drug level (serum) specimen collection form, (viii) 60 day and 90 day outcomes form, (ix) comments and end of study investigator sign off, (x) protocol violation form: biochemistry, (xi) protocol violation form: medication administration/discontinuation, (xii) adverse event form, (xiii) serious adverse event form in randomized patients. For centres not participating in the registry, we drafted 10 additional forms to capture necessary demographic, treatment and outcomes information. In addition, we drafted a form to capture demographic data and outcomes on eligible but not randomized patients. Descriptive statistics will be used to summarize the data. For univariate analyses, we will use the Chi-square test (alternatively, Fisher's exact test when the expected cell size is ≤ 5) and Student's t-test (alternatively, the Mann-Whitney U-test, if normality assumptions are not satisfied) for binary and continuous outcomes, respectively. All analyses will be conducted on an intention-to-treat basis. Feasibility for the pilot study will be assessed by metrics that reflect our capacity to ultimately recruit a representative sample of 1,050 patients in the planned full CHAT trial. We will consider the study to be feasible if we recruit at least 30% (commonly used threshold in ICU studies) of all eligible patients in participating ICUs through careful review of site screening logs. Additionally, we expect that: (i) less than 10% of medication doses will fail to be administered in the absence of meeting one of the medication discontinuation criteria; (ii) less than 5% of data forms will be missing important primary and secondary outcomes data required for the planned full CHAT trial and (iii) no more than 10% of enrolled patients will be withdrawn prematurely due to open label use of statins or withdrawal of consent. We will describe recruitment rates based on approved consent models. Since centres in Australia and New Zealand will be permitted to use Atorvastatin and matching placebo (instead of Rosuvastatin/matching placebo), we propose to conduct the planned primary and secondary analyses (i) using the pooled data (rosuvastatin plus atorvastatin) and (ii) using rosuvastatin (as the predominantly used statin in the CHAT Trial) data alone. A DSMB will oversee the trial and will consist of 3 individuals with expertise in viral infectious diseases, statistics and clinical critical care of which one will be international. The DSMB will hold a teleconference after either 30 patients have evaluable data or approximately 8 months after study initiation. Should the trial be completed, feasibility data from the pilot study will be analyzed by the DSMB at the end of the study. This information will be conveyed to the Steering Committee. Together the DSMB and Steering Committee will formulate a decision whether to proceed with the full trial. Clinical outcomes will remain blinded by study group assignment with a view to including them in the planned larger trial. Investigators will evaluate any changes in laboratory values and physical signs and will determine if the change is clinically important and different from what is expected in the course of treatment of critically ill patients requiring mechanical ventilation for suspected, probable or confirmed influenza. If clinically important and unexpected adverse experiences occur, they will be recorded on an adverse event case report form. We will characterize adverse events (see Additional File 5) as expected, serious unexpected and study related or unanticipated. We considered other factors (see Additional File 6) including patient withdrawals, consent (including telephone consent and waivers of consent) (see Additional Files 7 and 8), eligible non-randomized patients, equitable selection of subjects, justification for including vulnerable subjects, women of childbearing age, justification for excluding pregnant women, trial oversight and the trial data safety and monitoring board (see Additional File 6) in designing the CHAT Trial protocol. The investigators plan to make changes to the larger study protocol based on their experience in implementing the pilot trial. Regardless, we will publish the findings of the CHAT Pilot Trial, either alone or pooled with another trial evaluating the role of statins in a similar population, if recruitment ensues even if the study protocol is modified in important ways following conduct of the pilot trial or the planned larger trial never comes to fruition. Pilot trial data may also be combined with data from the larger trial if the latter trial comes to fruition, study personnel (including the data analyst) remain blinded to treatment assignment and no important modifications are made to the study protocol following the pilot trial. Global concern arose from the threat of the H1N1 influenza pandemic. Despite the potential virulence of the illness, little is actually known about how severe disease develops or what treatments may confer benefit to critically ill patients. Even less is known about how to conduct clinical research in the setting of an evolving pandemic. Severe H1N1 infection primarily affects young and often previously healthy individuals. Early reports supported that aboriginal populations in Canada and Australia, obese individuals and women, especially pregnant women, appear to have a predilection for severe disease. Unlike the pandemic of 1918, the availability today of antiviral agents, antibiotics for secondary infection, and ICU supportive care interventions holds promise that the majority of patients with severe illness can be saved. The burden of severe H1N1 disease falls prominently on the ICU [8, 38] . Consequently, the opportunity to learn about treatments for severe H1N1 disease and how to conduct pandemic critical care research rests within the ICU community. Data from observational studies in humans and interventional studies in animals, suggests that the course of influenza may be favorably influenced by relatively inexpensive and readily available agents, such as rosuvastatin, that are not classically considered to be treatments for influenza. These agents are attractive as adjuvant treatments amidst emerging reports of oseltamivir resistance and threatened drug supply shortages. However, the ability to administer and test the efficacy of an adjuvant agent as a treatment for severe H1N1 infection remains to be established. The CHAT pilot trial is designed to evaluate the feasibility of implementing a randomized controlled trial of adjuvant rosuvastatin for treating severe H1N1 infection under a pandemic. We aim to evaluate whether centres can adhere to the study treatment regimens, collect the required primary and secondary endpoints for the subsequent planned full CHAT trial, and document patients who receive open-label statins and consent withdrawals. We also seek to evaluate the impact of approved consent models on recruitment rates. Several time-honored aspects of RCT design including the use of central randomization, preservation of allocation concealment, multi-level study blinding, and use of a matching placebo posed challenges to us in designing a pandemic protocol. We contemplated the necessity of including each of these study design features. After careful deliberation, we decided to include central randomization (using lists distributed by the study methods centre to participating centres), preserve multi-level blinding (by involving pharmacies at participating centres) and contract a local pharmaceutical company to prepare crushed drug and matching placebo (in the absence of industry supply of study drug and an available placebo). Recognizing that trial initiation may be delayed and recruitment curtailed if a priori in-person SDM consent was required, we considered use of alternative consent models. A priori in-person SDM consent not only hinges on the existence and availability of SDMs [39] , but also the ability of SDMs to access hospitals during a pandemic. Strengths of the proposed pilot trial design include the use of central randomization, allocation concealment, multi level blinding, standard criteria for medication discontinuation, and 90 day follow up. To ensure feasibility during a pandemic, we did not protocolize H1N1 testing, ventilator and sedation management or the clinical use of antibacterial agents. By merging study data with an Influenza Registry and capturing data on unique forms for centres not participating in the registry, we will, however, document key aspects of clinical management. The conduct of an RCT during an evolving pandemic poses unique challenges not encountered during other forms of clinical research [40] . First, it is necessary to initiate studies quickly. The normal time interval from concept to first patient enrolment for a new RCT is typically of the order of two years or more. Second, the scope and duration of the pandemic is unknown and unpredictable. Third, the mitigating effects of large-scale vaccination programs and changes in H1N1 infectivity resulting from virus mutation are unknown. Fourth, the practicalities of conducting clinical research during a pandemic are unknown. For example, research personnel may be seconded to provide clinical care, pharmacists may face challenges in dispensing drug and placebo, REBs may not permit alternative consent models and it may be difficult to obtain consent from patients who may lack decision-making capacity and families, who may be unwell themselves, unable to visit the hospital or requested to stay away during the pandemic. Finally, faced with the clinical imperative of treating gravely ill and previously well, young patients, clinicians may opt to use open label treatment rather than permit enrolment into a blinded RCT. Because we believe that it is important to develop the capacity to initiate RCTs under pandemic conditions and to test study procedures prior to implementing a large scale RCT, we propose to conduct a multi-centre pilot trial to assess the feasibility of our clinical protocol and study procedures.

Dissection of the Influenza A Virus Endocytic Routes Reveals Macropinocytosis as an Alternative Entry Pathway

Influenza A virus (IAV) enters host cells upon binding of its hemagglutinin glycoprotein to sialylated host cell receptors. Whereas dynamin-dependent, clathrin-mediated endocytosis (CME) is generally considered as the IAV infection pathway, some observations suggest the occurrence of an as yet uncharacterized alternative entry route. By manipulating entry parameters we established experimental conditions that allow the separate analysis of dynamin-dependent and -independent entry of IAV. Whereas entry of IAV in phosphate-buffered saline could be completely inhibited by dynasore, a specific inhibitor of dynamin, a dynasore-insensitive entry pathway became functional in the presence of fetal calf serum. This finding was confirmed with the use of small interfering RNAs targeting dynamin-2. In the presence of serum, both IAV entry pathways were operational. Under these conditions entry could be fully blocked by combined treatment with dynasore and the amiloride derivative EIPA, the hallmark inhibitor of macropinocytosis, whereas either drug alone had no effect. The sensitivity of the dynamin-independent entry pathway to inhibitors or dominant-negative mutants affecting actomyosin dynamics as well as to a number of specific inhibitors of growth factor receptor tyrosine kinases and downstream effectors thereof all point to the involvement of macropinocytosis in IAV entry. Consistently, IAV particles and soluble FITC-dextran were shown to co-localize in cells in the same vesicles. Thus, in addition to the classical dynamin-dependent, clathrin-mediated endocytosis pathway, IAV enters host cells by a dynamin-independent route that has all the characteristics of macropinocytosis.

Influenza A virus (IAV) is an enveloped, segmented negativestrand RNA virus infecting a wide variety of birds and mammals. As its first step in infection IAV attaches to host cells by the binding of its major surface protein, the hemagglutinin (HA), to sialic acids, which are omnipresent on the glycolipids and glycoproteins exposed on the surfaces of cells. Where the structural requirements for this interaction have been studied in great detail, much less is known about whether and how the attachment to specific sialylated receptors (e.g. to N-linked glycoproteins, Olinked glycoproteins or gangliosides or even to specific receptors within these groups) affects the subsequent endocytic steps. Obviously, knowledge about the repertoire of endocytic pathways that can successfully be used by IAV will increase our insights into cell and species tropism of IAV. In turn, this will contribute to our understanding of the requirements for the generation of novel viruses with pandemic potential that can arise by exchange of RNA segments between currently circulating human serotypes and an animal virus during occasional co-infection in a human or an animal host. Clathrin mediated endocytosis (CME) has for long been identified and studied as the major route of IAV cell entry [1, 2] and is, by far, the best characterized endocytic pathway. Evidence obtained from live cell imaging has revealed the de novo formation of clathrin-coated pits at the site of virus attachment [3] and the requirement for the adapter protein epsin 1, but not eps15, in this process [4] . Still, specific transmembrane receptors linking viral entry to epsin 1 or to other adapters have not been identified although a recent study performed in CHO cells indicated the specific requirement for N-linked glycoproteins in IAV entry [5] . Some recent papers provided indications for the utilization of alternative entry pathways by IAV. Studies in which CME was obstructed by pharmacological or genetic intervention indicated the ability of IAV to enter host cells via alternative endocytic routes [4, 6, 7] . Also live cell imaging revealed the simultaneous availability of entry routes involving non-coated as well as clathrin-coated pits [4] . However, this alternative IAV entry route has not been characterized in any detail and requirements for any specificity in receptor usage apart from the need for the proper sialic acid moiety have not been established. During the past decades quite a variety of endocytic pathways have been identified in eukaryotic cells [8, 9, 10] . Their occurrence, abundance and mechanistic details appear to vary between cell types, tissues and species and their utilization by viruses as a route of entry makes them an important factor in host and cell-type permissiveness for infection [11, 12] . Besides by CME, different viruses have been shown to enter cells via caveolae, macropinocytosis or other, less well described, routes [11, 12] . Most often, the selection of a specific endocytic route is linked to the utilization of a specific receptor that facilitates traveling via that particular route. Nevertheless, many receptors allow flexibility by their capacity to enter through multiple pathways. For IAV, an additional level of complexity to the dissection of potential entry routes is added by the apparent lack of an IAV-specific protein receptor. A full experimental characterization of the IAV entry pathways will benefit from separation of the IAV entry pathways into routes that can be studied independently. Whereas co-localization with clathrin is an established marker for endocytosis via this route, the complete lack of unique markers for macropinosomes or most other endocytic compartments [13, 14] complicates such studies. Furthermore, crucial to any study concerning endocytic pathways is the abundantly documented fact that such pathways are highly dependent on experimental cell culture conditions [15] [16] [17] [18] [19] . Pathways that are constitutive in one cell type may be absent or inducible by specific experimental conditions in other cell types. Moreover, the manipulation of specific endocytic pathways may result in up or down regulation of other specific pathways. Here we have established entry assay conditions that allow dissecting cell entry of IAV into a dynamin-dependent (DYNA-DEP) and a dynamin-independent (DYNA-IND) component. Dynamin is a large GTPase forming multimeric assemblies around the neck of newly formed endocytic vesicles. GTP hydrolysis is required for pinching off of the vesicles [20] . Whereas CME is completely dependent on dynamin, several other endocytic routes do not require dynamin [21] . We performed an extensive characterization of the dynamin-independent IAV entry route using pharmacological inhibitors as well as by expressing dominant-negative mutants and applying siRNA induced gene silencing as tools. Taken together the results identify a pathway that closely resembles macropinocytosis as a novel entry pathway for IAV. To identify and characterize potential non-CME entry routes taken by IAV, we adapted a luciferase reporter assay [22] to enable the quantitative determination of infection or entry by measuring the activity of secreted Gaussia luciferase. Twentyfour hours prior to infection HeLa cells were transfected with a plasmid (pHH-Gluc) allowing constitutive synthesis (driven by the human PolI promoter) of a negative strand viral RNA (vRNA) encoding a Gaussia luciferase under control of the untranslated regions (UTRs) of the NP segment of Influenza A/WSN/33 (H1N1) (hereafter called IAV-WSN) NP segment. Upon IAV infection, the combined expression of the viral polymerase subunits and NP will drive transcription of luciferase mRNA from the negative strand vRNA and subsequent synthesis of Gaussia luciferase. A dose-response curve demonstrating the applicability of the assay to inhibitor screening (Fig. 1A) was obtained for Bafilomycin A1 (BafA1), a known inhibitor of IAV entry [23] . BafA1 acts upon the vacuolar-type H(+)-ATPase, thus preventing endosomal acidification and thereby trapping IAV in peri-nuclear immature endosomes with a lumenal pH that does not permit viral membrane fusion. Remarkably, dynasore, a small molecule inhibitor of the GTPase dynamin 2 that is crucial for endocytic vesicle formation in clathrin-and caveolin-mediated endocytosis [8] as well as in a poorly described clathrin-and caveolin-independent endocytic pathway [8, 19] , did not give significant inhibition (Fig. 1B) . BafA1 specifically inhibits IAV during the entry phase as demonstrated in Fig. 1C . The continuous presence of 10 nM BafA1 (added to the cells 1 hr prior to infection) for 16 hrs completely prevents infection. In contrast the addition of BafA1 at 1 hr or 2 hrs post infection resulted in high levels of luciferase activity (again measured at 16 hrs p.i.) that were 63% or 90% respectively of the control to which no BafA1 was added, indicating that entry was essentially completed within 2 hrs. The last bar of Fig. 1C shows that the inhibition by BafA1 is reversible as withdrawal of the inhibitor after 2 hrs resulted in high levels of infection. The specific effect of BafA1 on IAV entry was confirmed by confocal microscopy demonstrating that BafA1, as expected, traps IAV particles in a peri-nuclear location, presumably in nonacidified endosomes (Fig. 1D) . BafA1 was subsequently exploited to establish a specific IAV entry assay (hereafter further referred to as the Gluc-entry assay). HeLa cells transfected with pHH-Gluc were inoculated with IAV at a range of MOIs and incubated for 2 hrs after which the entry medium was replaced by complete growth medium containing 10% FCS and 10 nM BafA1 to prevent any further entry of virus. Entry was indirectly quantified by determination of luciferase activity after further incubation for 14 hrs demonstrating a quantitative correlation between infection dose and luciferase activity across a wide range of MOIs (Fig. 1E) . The indirect Gluc-entry assay was next tested for its capacity to examine the effects of inhibitors on IAV entry. Dynasore or BafA1 (Fig. 1F) were included in the medium (DMEM containing 10% Attachment to and entry into a host cell are the first crucial steps in establishing a successful virus infection and critical factors in determining host cell and species tropism. Influenza A virus (IAV) attaches to host cells by binding of its major surface protein, hemagglutinin, to sialic acids that are omnipresent on the glycolipids and glycoproteins exposed on the surfaces of cells. IAV subsequently enters cells of birds and a wide variety of mammals via receptormediated endocytosis using clathrin as well as via (an) alternative uncharacterized route(s). The elucidation of the endocytic pathways taken by IAV has been hampered by their apparent redundancy in establishing a productive infection. By manipulating the entry conditions we have established experimental settings that allow the separate analysis of dynamin-dependent (including clathrin-mediated endocytosis) and independent entry of IAV. Collectively, our results indicate macropinocytosis, the main route for the non-selective uptake of extracellular fluid by cells, as an alternative IAV entry route. As the dynamindependent and -independent IAV entry routes are redundant and independent, their separate manipulation was crucial for the identification and characterization of the alternative IAV entry route. A similar strategy might be applicable to the study of endocytic pathways taken by other viruses. FCS) during entry (the first 2 h of infection) and were removed when the inoculum was replaced by growth medium containing BafA1. Concentrations up to 80 mM dynasore did not inhibit entry which is in agreement with the result shown in Fig. 1B . In contrast, 1.25 nM BafA1 already inhibited entry for more than 60% (Fig. 1F) . As a control, dynasore was also added at 2 hrs post infection to analyze whether the drug affected IAV replication during the post entry phase. As expected, 80 mM dynasore did not significantly inhibit IAV replication when present from 2 to 16 hrs p.i. (Fig. 1F ). Thus, with the Gluc-entry assay we can study the effect of specific inhibitors on IAV entry in a quantitative manner, at least as long as the inhibitors do not irreversibly affect IAV replication during the post entry phase. Furthermore, the lack of inhibition of IAV entry by dynasore demonstrates that under these experimental conditions IAV is able to enter cells via a pathway that is fully redundant to any dynamindependent (DYNA-DEP) entry route, including the classical CME pathway. Also when IAV travels via this novel dynamin-independent (DYNA-IND) route, IAV apparently enters via low pH compartments as entry is fully sensitive to BafA1. As factors present in serum are known for their potential to induce specific endocytic pathways, we further explored the conditions required for the novel DYNA-IND IAV entry pathway (using the Gluc-entry assay) by inoculating cells in PBS in the presence of increasing concentrations of fetal calf serum (FCS). Whereas dynasore completely inhibited entry in PBS, inclusion of 5% and 10% FCS resulted in increasing levels of dynasore resistant entry ( Fig. 2A) , suggesting the existence of a serum-inducible DYNA-IND IAV entry pathway. This effect was not caused by inactivation of dynasore during the experiment as vesicular stomatitis virus (VSV), which enters cells by CME [24, 25] , was still sensitive to 80 mM dynasore in the presence of 10% FCS (Fig. 2B) . In agreement herewith, the uptake of transferin, known to occur via CME, was inhibited by dynasore regardless of the HeLa cells were grown on glass cover slips and infected with IAV (strain WSN; MOI of 10) and fixated after 30 min, 3 hrs or 6 hrs (column 1, 2 or 3 respectively). Infection was performed in 0.2% DMSO (upper row panels) or in the presence of 10 nM BafA1 (lower row panels). The nucleus was visualized by DNA staining with TOPRO-3 (red). IAV infection was visualized by staining with monoclonal antiserum directed against NP (green). In the absence of inhibitor, IAV localized to the nucleus after 3 hrs, while new virus particles spread to the cytoplasm after 6 hrs. BafA1 (lower row panels) caused accumulation of incoming virus particles at a peri-nuclear location. (E) Quantitative determination of IAV entry by a single-cycle Gluc-entry assay. HeLa cells (10,000 cells/well in DMEM supplemented with 10% FCS) were transfected with pHH-Gluc 24 hrs prior to infection with a serial dilution of infectious IAV particles (plotted on the x-axis). Two hours after infection 10 nM BafA1 was added to block any further entry. Cells were incubated for a further 14 hrs to allow expression of luciferase activity (y-axis; Relative Light Units, RLU). (F) Effect of Dynasore and BafA1 on IAV entry in the Gluc-entry assay. Dynasore (DY, dark grey bars; 20, 40 or 80 mM) or BafA1 (light grey bars; 1.25, 2.5 or 5 nM) were present from 1 hr prior to infection (strain WSN; MOI 0.5) to 2 hrs p.i. after which the inhibitor-containing medium was replaced with medium containing 10 nM BafA1 to block any further entry. Cells were incubated for a further 14 hrs to allow the quantitative expression of luciferase activity (y-axes; RLU relative to the control infection without inhibitor). Whereas BafA1 displayed dose-dependent inhibition of IAV entry, dynasore did not significantly inhibit IAV entry. presence of FCS (Fig. S2, panel A) . As expected, both DYNA-DEP entry in PBS and DYNA-IND entry in the presence of 10% FCS and 80 mM dynasore required sialic acid receptors for efficient entry as pre-treatment of HeLa cells with neuraminidases almost completely abolished entry via either pathway (Fig. 2C ). The kinetics of the DYNA-DEP and DYNA-IND entry pathways were compared by performing a time-course experiment in which IAV entry was terminated by the addition of 10 nM BafA1 at different time points (Fig. 2D) . In comparison to entry via the DYNA-DEP pathway (the only pathway available in PBS) entry in the presence of FCS (when presumably both the DYNA-DEP and DYNA-IND entry pathways are available) showed similar kinetics. In contrast, entry via the DYNA-IND pathway (which is the only pathway that is active in the presence of 10% FCS and 80 mM dynasore) was slower. The difference was most prominent after 15 min, while after 4 hrs similar levels of entry were reached. To validate and extend these results we visualized the reduction of the number of infected cells by immunoperoxidase staining using an antibody against NP (Fig. 3) . A number of different cells of mammalian and avian origin were infected for 2 hours at an MOI of 1 in PBS with or without serum. After 2 hours the inoculum was replaced by growth medium containing 10% FCS and 10 nM BafA1 and the expression of NP was examined after 14 hours later. After incubation in PBS, staining was completely prohibited by the presence of 80 mM dynasore whereas in the presence of serum dynasore had no effect. A serum-inducible, DYNA-IND route of entry was thus functional in all five cell lines, including the human epithelial airway carcinoma cell line A549. To confirm our results and to obtain further proof for the utilization of DYNA-DEP and DYNA-IND entry routes by IAV, we additionally used an IAV virus-like particle (VLP) direct entry assay [26] . These VLPs contain IAV HA and NA in their envelope and harbor a beta-lactamase reporter protein fused to the influenza matrix protein-1 (BlaM1), which allows the rapid and direct detection of entry, independent of virus replication. Upon fusion of viral and endosomal membrane, BlaM1 gains access to the cytoplasmically retained fluorigenic substrate CCF-2 that, after cleavage by BlaM1, shifts to a shorter fluorescent emission wavelength that can be detected by flow cytometry. Entry into HeLa cells was performed in the absence or presence of 10% FCS using VLPs containing HA and NA either from IAV-WSN (having a strict alpha 2-3 linked sialic acid binding specificity) or from the pandemic 1918 IAV (HA from A/ NewYork/1/18, binding to alpha 2-3 and alpha 2-6 linked sialic acids; NA from A/BrevigMission/1/18). Entry of VLPs of both IAV strains was severely inhibited by dynasore when no serum was added to the inoculum (Fig. 4A, 4D) , whereas the presence of 10% FCS rendered entry completely dynasore resistant. (Fig. 4B, 4E ). Quantification of VLP entry is shown in Fig. 4C and F. Importantly, to confirm the existence of the serum-inducible entry pathway by a method that is independent of dynasore, we used siRNA induced silencing of dynamin 2. Fig. 4G shows that two different siRNAs had a significant inhibitory effect (48 hrs after siRNA transfection) on entry of the Renilla luciferaseencoding pseudovirus WSN-Ren [27] in HeLa cells in the absence of FCS, whereas the presence of 10% serum no reduction in entry levels was observed, confirming the results obtained with dynasore. Knockdown of dynamin 2 protein levels (48 hrs after siRNA transfection) was analyzed by western We conclude that a DYNA-IND entry pathway can be induced by serum in different cell types from several species. The evidence was obtained using both replication-dependent (Gluc-entry assay and immunodetection of infected cells) and replication-independent assays (entry of VLPs), the latter allowing immediate detection of the fusion-mediated delivery of viral M1 protein into the cytoplasm. Cascade Blue). In the histograms entry is displayed by a shift to higher fluorescence (the grey area represents background fluorescence of noninfected cells). (C and F) Quantification of FACS results. Background fluorescence was subtracted from each measurement (geometric mean) and data were normalized to VLP entry in Optimem without dynasore (DY) (red curve of panel A and and D). VLP entry was not inhibited by dynasore in presence of 10% FCS whereas the access of BlaM1 to its CCF2 substrate in the cytoplasm was blocked by dynasore in PBS. VLP entry was more efficient in the presence of serum. (G) Effect of downregulation of dynamin 2 by siRNA silencing. Serum-inducible DYNA-IND entry was analyzed in HeLa cells that were transfected 48 hrs prior to infection with two different siRNAs targeting dynamin-2 (dyna). siRNA treated cells were infected with the pseudovirus WSN-Ren in PBS (grey bars) or in PBS containing 10% FCS (black bars) and luciferase activity was determined after 16 hrs post infection (y-axis; RLU relative to infection of cells transfected with a scrambled siRNA). Entry of pseudovirus WSN-Ren (MOI 0.5) was reduced by 50% to 70% when entry was performed in PBS (grey bars) whereas entry was not significantly affected in the presence of 10% serum (black bars). (H) Western blot showing the knockdown of dynamin 2 (in comparison to tubulin) at 48 hrs after transfection with siRNAs. (I) Quantification of the residual levels of dynamin 2 (dyna) mRNA (determined by quantitative RT-PCR) and protein (determined by densitometric scanning of the western blot) 48 hrs after siRNA transfection. Data were normalized to 18S RNA (RT-PCR) or tubulin protein levels and calculated relative to the levels obtained after transfection with a scrambled siRNA that served as a control. doi:10.1371/journal.ppat.1001329.g004 Inhibitors of growth factor receptor tyrosine kinases and actomyosin network dynamics reduce DYNA-IND entry of IAV The DYNA-IND entry pathway was further characterized by inhibitor profiling using an 80-compound kinase inhibitor library. Serum-induced DYNA-IND entry was examined in 10% FCS using the Gluc-entry assay. 80 mM dynasore was added in order to block CME and any other potential DYNA-DEP entry pathways. This allowed the independent inhibitor profiling of the novel pathway by avoiding the potentially masking effect of the presence of redundant entry pathways. Cells were preincubated with the kinase inhibitors (10 mM) for 1 h at 37uC and then inoculated with virus (MOI 0.5) in the presence of 10% FCS and 80 mM dynasore for 2 h at 37uC (DYNA-IND entry). In parallel, inoculations were also done in PBS to compare the effects of the inhibitors on DYNA-DEP entry. After 2 hr the medium and inhibitor were replaced by full growth medium containing 10% FCS and 10 nM BafA1 to allow the subsequent expression of Gluc activity under identical conditions for the DYNA-IND and -dependent entry assay. Six kinase inhibitors appeared to act non-discriminatively, inhibiting both DYNA-DEP and DYNA-IND entry (Fig. 5A ): the protein kinase C (PKC) inhibitors Ro 31-8220, rottlerin (both displaying moderate cytotoxicity, result not shown) and hypericin, which have all three been previously identified as IAV inhibitors [28, 29] ; the highly cytotoxic pan-specific serine/threonine protease inhibitor staurosporine; the irreversible PI-3 kinase inhibitor wortmannin and the receptor tyrosine kinase inhibitor TYR9. In order to investigate whether some of these inhibitors affect IAV replication during the post-entry phase, we performed the same experiments but now adding the kinase inhibitors after viral entry. Four of the inhibitors thus appeared to induce significant inhibition of post-entry processes (Fig. 5A ). Although unlikely, we cannot formally exclude that post-entry processes specific for only one of the two entry pathways are affected. Interestingly, whereas no specific DYNA-DEP entry inhibitors were identified, 15 inhibitors (none displaying cytotoxic effects, data not shown) caused significant (p,0.05) inhibition (.5-fold) of DYNA-IND entry (Fig. 5B ). This included inhibitors of the calmodulin dependent kinases myosin light chain kinase (MLCK) and CaMKII and seven inhibitors of different growth factor receptor tyrosine kinases. In contrast to the three non-specific PKC inhibitors mentioned above, the PKC inhibitors BIM-1 and HBDDE appeared to have a specific inhibitory effect on DYNA-IND entry. The specific effect of these drugs on DYNA-IND entry is not only shown by the lack of inhibition of DYNA-DEP entry in PBS, but also by the observation that none of the fifteen compounds induced .2-fold inhibition when added post-entry (at t = 2 hr post infection). The kinase library screen was repeated on A549 human epithelial lung carcinoma cells in order to confirm the results in a potentially more natural host cell line. The inhibition profiles obtained were very similar to those found for HeLa cells with the exception of the strong effect of AG879 (99% inhibition) and moderate effects of AG825 (39% inhibition) and Tyr51 (68% inhibition) on DYNA-DEP entry. (Fig. 5C) . MLCK inhibitors ML-7 and ML-9 have been reported to be highly specific for their target kinase [30] . Phosphorylation by MLCK activates non-muscle myosin II light chain, indicating that a functional actomyosin network might be essential for DYNA-IND entry of IAV. This was further examined by testing the effect of Blebbistatin, an inhibitor of myosin II heavy chain activity, and of several inhibitors that affect actin dynamics by disrupting actin microfilaments (Cytochalasin B and D), by enhancing actin polymerization (Jasplakinolide) or by inhibiting actin polymeriza-tion (Latrunculin A). Actin inhibitors were used at the minimal concentration required to induce clearly visible changes in the actin cytoskeleton as pre-determined by staining with FITCphalloidin (results not shown). Whereas the inhibitors did not affect DYNA-DEP entry (Fig. 6A) using Gluc-entry assay, all inhibitors as well as ML-7 and ML-9 significantly inhibited DYNA-IND entry (Fig. 6B) . Next, HeLa cells were transfected with plasmids encoding dominant negative or wildtype Rab5 fused to green fluorescent protein (Rab5 DN and Rab5 wt in Fig. 6 ) 24 h prior to infection with IAV. Rab5 is a small GTPase found in association with several endosomal compartments and crucial for the function and maturation of early endosomes. It is required for the trafficking of a wide range of endocytic cargo following different routes, including DYNA-DEP as well as DYNA-IND routes [31] . Entry of IAV has been shown to require Rab5 [32] . Consistently, we found that HeLa cells expressing Rab5 DN (as identified by GFP fluorescence, Fig. 6C ) were much less susceptible to productive IAV infection (as judged by indirect immunofluorescence using Alexa-488 labeled NP antibodies) than cells transfected with Several dynamin-independent endocytic pathways have been described [8, 19] . Of these, macropinocytosis has been demonstrated to be stimulated by growth factors present in serum and to depend on actin dynamics [12] [13] [14] . Yet, studies on macropinocytosis are hampered by a lack of specific inhibitors, cargo, membrane markers and characteristic morphology. Amiloride and the more potent derivative EIPA are inhibitors of epithelial sodium channels (ENaC) as well as of several other Na+/H+ antiporters. EIPA has often been used as a hallmark inhibitor that specifically inhibits endocytosis via the macropinocytic pathway [14] . Whereas DYNA-DEP entry of IAV was not inhibited by EIPA (Fig. 7A) , DYNA-IND entry was fully blocked EIPA (Fig. 7B) . The existence of redundant entry pathways in the presence of 10% FCS is clearly demonstrated by the marginal inhibition by either EIPA or dynasore whereas the combination of EIPA and dynasore resulted in strong inhibition both in the Gluc-entry assay (Fig. 7C ) and in the direct VLP entry assay ( Fig. 7D and E) . Supplementary Fig. S1 shows that other cell lines, including the human lung epithelial cell line A549, display similar IAV inhibition patterns for EIPA and dynasore. Consistently, virus production displayed a similar inhibitor sensitivity profile (Fig.7 F and G) as virus entry indicating that the entry pathways we characterized lead to a productive infection. Clearly, VLPs and viral particles follow similar redundant entry pathways, distinguishable in a DYNA-DEP and a DYNA-IND pathway, the latter being sensitive to EIPA and dependent on actomyosin function. One characteristic of macropinocytosis is the nonselective uptake of large amounts of extracellular solutes [33] . Therefore, the uptake of soluble FITC labeled dextran (Fdx) into relatively large vesicles (0.3 to 5 mM) has often been applied as a morphological marker for macropinosomes. Using this marker we found that the addition of 10% FCS to the culture medium slightly increased the uptake of Fdx into HeLa cells (Fig. 8A) . Notably, the distribution of Fdx changes in response to serum from a random distribution into a more granular pattern. At high magnification and at color settings adjusted to higher intensity it could be seen that these Fdx granules were free of actin staining (by phalloidin) indicating that they were in the lumen of vesicles (result not shown). Interestingly, in the presence of IAV (MOI of 10) the uptake of Fdx into vesicles was clearly enhanced. At a higher magnification viral particles could be found to co-localize in Fdx loaded vesicles as well as outside these vesicles (Fig. 8B) . Phalloidin staining of actin was used to demonstrate that many virus particles localized to actin-rich protrusions at the periphery of the cell. The uptake of Fdx was studied in a quantitative manner by flow cytometry (Fig. 8 C) . A moderate, but reproducible shift to higher Fdx fluorescence was observed at 37uC when virus was added in presence of 10% FCS whereas such a shift was absent when no serum or virus was added. This result confirms the observations by confocal microscopy (Fig. 8A) which showed that the combined presence of FCS and IAV increases the uptake of Fdx as compared to FCS alone. In a control experiment the uptake of Fdx in 10% FCS in presence of IAV was shown to be specifically inhibited by EIPA, but not by dynasore (Fig. S2, panel B) . In contrast, transferrin uptake, which serves as a specific marker for CME, was affected by dynasore, but not by EIPA (Fig. S2, panel A) . In conclusion, serum induces the uptake of Fdx into large vesicles, which can be further enhanced by the addition of IAV particles that, after entry, co-localize in part with these vesicles. These results indicate the utilization of a macropinocytic pathway for entry of IAV, which is consistent with the observed sensitivity of the seruminducible DYNA-IND entry of IAV and VLPs to EIPA. Macropinocytosis has been implicated in the entry of several viruses [12, 14] . However, differences in susceptibility to inhibitors suggest that distinct forms of macropinocytosis might be used by different viruses [34, 35] . By screening specific inhibitors in the Gluc-entry assay using DYNA-IND entry conditions we evaluated the possible involvement of a few signaling cascades that have been implicated in the induction of macropinocytosis. Serum-inducible macropinocytosis has been shown to be activated via a myriad of signaling cascades initiated by growth factors binding to transmembrane tyrosine kinase receptors [14, 17, 36, 37] , consistent with the results shown in Fig. 5 . A prominent downstream effect of these signaling cascades is the activation of p21 associated kinase 1 (PAK1) which in turn can activate a number of different pathways leading to actin network rearrangements that can ultimately lead to the induction of macropinocytosis [38] . Fig. 9A -B shows that 20 mM IPA3, an inhibitor of PAK1 [39] , specifically inhibits Background fluorescence from Fdx binding to the outside of cells was determined by performing the same experiment at 4uC (at which no endocytosis takes place) and was subtracted from the mean fluorescence intensity obtained at 37uC to determine the amount of fluorescent FITC-dextran that was internalized at 37uC. Data were plotted relative to FITCdextran uptake in PBS in absence of IAV. doi:10.1371/journal.ppat.1001329.g008 DYNA-IND entry of IAV. Activation of PAK1 in response to growth factor stimulation often involves upstream signal transduction by members of the Rho sub-family of small GTPases like CDC42 and/or Rac1 [34, 40, 41] . Alternatively, activated CDC42 and Rac1 can induce actin rearrangements independently of PAK1 [34, [40] [41] [42] [43] by direct interaction with WASP or WAVE family proteins, respectively [44, 45] . However, inhibitors of CDC42 (Pirl1 [46] ), Rac1 (NSC23766 [47] ) or N-WASP (wiskostatin [48] ) did not display inhibitory effects on DYNA-IND or DYNA-DEP entry of IAV (Fig. 9C-D) . Instead, Pirl1 and wiskostatin induced a significant, concentration dependent increase of entry. This stimulatory effect was not observed for the control vaccinia virus strain WR, which enters cells via a Rac1dependent, macropinocytotic pathway [43] (Fig. 9E) , indicating that this effect is specific for IAV. The results suggest a requirement for PAK1 in DYNA-IND entry of IAV that does not require activation by either CDC42 or Rac1. Growth factor inducible activation of the tyrosine kinase src has also been linked to the induction of macropinocytosis [49] [50] [51] ; consistent with this observation the src inhibitor PP2 [52] specifically inhibited DYNA-IND entry of IAV (Fig. 9A-B) . Remarkably, 17-AAgeldanamycin, a specific inhibitor of the chaperone protein HSP90 [53] , also caused specific inhibition of DYNA-IND entry (Fig. 9A-B) . HSP90 affects the folding and activity of many proteins but the recent demonstration of direct activation of the catalytic activity of src by HSP90 [54] provides another indication of the involvement of src in DYNA-IND endocytosis of IAV. In conclusion, like for other viruses utilizing a macropinocytic entry pathway, PAK1 seems to play a crucial role in DYNA-IND entry by IAV. However, this pathway is independent of Rac1 or cdc42 but may require src, either upstream and/or downstream of PAK1. The data presented in this study demonstrate for the first time that IAV can enter cells via DYNA-IND macropinocytosis in addition to the previously described DYNA-DEP classical CME pathway [1, 2] . Several lines of evidence indicate that the DYNA-IND entry route of IAV that we identified corresponds with macropinocytosis. First of all, the entry pathway is dependent on the presence of serum, a well-known inducer of macropinocytosis. Second, IAV colocalized in vesicles with soluble FITC-dextran, a marker for macropinocytosis. Third, DYNA-IND IAV entry was sensitive to the amiloride-derivative EIPA, the hallmark inhibitor of macropinocytosis [14, [55] [56] [57] [58] . Fourth, this IAV entry pathway is sensitive to inhibitors or dominant-negative mutants affecting actomyosin dynamics. Fifth, the specific inhibition of DYNA-IND IAV entry by a number of inhibitors of growth factor receptor tyrosine kinases as well as downstream effectors thereof also points at the involvement of macropinocytosis. Finally, macropinocytosis is independent of dynamin [12, 14, 19] . Despite this extensive list of arguments, viral entry by macropinocytosis needs to be considered with caution. The characteristics of the DYNA-IND route of cell entry by IAV are similar, but not identical to the macropinocytic entry routes taken by other viruses, like two different strains of vaccinia virus and by coxsackie virus B [34, 35] . As is shown in Table 1 and discussed in more detail below, the macropinocytic pathways used by each of these viruses have a few unique characteristics. This may very well reflect the growing notion that macropinocytosis represents a number of differentially induced and regulated processes, rather than being a single endocytic pathway [13, 14] . Macropinocytosis has collectively been described as an inducible form of endocytosis by which fluid-phase cargo travels via non-coated, relatively large and heterogeneous organelles that have emanated from extensive protrusions (e.g lamellar ruffles, circular ruffles or retracting blebs) of the plasma membrane [13] . In the case of DYNA-IND IAV entry more extensive studies using electron microscopy will be required to study the morphology of membrane protrusions with which IAV may associate. In addition, live cell imaging microscopy will be required to characterize the exact itinerary that is taken by IAV virions traveling via a macropinocytic process. This is especially important as different routes of IAV entry are likely to converge at some point in the endocytic pathway. Although unlikely, co-localization of IAV particles with fluid-phase dextran as shown in Fig. 8B may thus represent a situation occurring after convergence of several different routes. The use of microscopy to study macropinocytosis is however complicated by the lack of specific membrane-associated markers for any early step of this endocytic process. A model (Fig. 10 ) based on our results explains the key steps involved in the macropinocytic entry pathway of IAV, which are described in more detail below. By manipulating the inoculation conditions we were able to experimentally dissect IAV entry into a DYNA-DEP and DYNA-IND route. The DYNA-IND route required the presence of 10% FCS in the entry assay medium. Previously, a strict dependency on a DYNA-DEP entry route for IAV was concluded from experiments with a cell line expressing an inducible dominantnegative mutant of dynamin 2 [59] . In that study, as well as in other entry studies of IAV, entry was performed in DMEM containing 2% serum or BSA. Also in our hands 2.5% serum ( Fig. 2A) or 0.2% BSA (result not shown) was not sufficient to allow DYNA-IND entry. We are currently investigating which serum component is responsible for the observed effects on IAV entry. Dialysis of FCS (MW cut off .10 kDa) did not affect its capacity to induce DYNA-IND endocytosis (result not shown), indicating that low molecular weight solutes are not responsible for the observed effect. Our evidence for a DYNA-DEP and a serum inducible DYNA-IND entry route is based on the use of pharmacological (dynasore, a highly specific inhibitor of dynamin) as well as genetic (siRNA directed against dynamin 2) tools, ruling out the possibility that the inhibitory effect of dynasore was due for instance to absorption of the inhibitor by serum components. Whereas dynasore resulted in near 100% inhibition of DYNA-DEP entry, only 65% inhibition was observed upon siRNA induced silencing of dynamin 2 indicating that the residual levels of dynamin 2 that remain after 48 hrs of silencing still support a low level of DYNA-DEP entry (Fig. 4H) . Reversible inhibitors like dynasore [60] offer a major advantage for characterization of IAV entry pathways. They can be applied for a limited period thus preventing the secondary adaptive effects of cells that may occur in response to long-term down regulation of a gene product by genetic methods like siRNA interference. Both entry routes were consistently identified by a viral entry assay quantified by virus induced expression of a luciferase reporter as well as by a VLP entry assay allowing direct analysis of the membrane fusion mediated entry step. The consistent performance of an HA with a strict preference for binding to a2-3 linked sialic acids (from IAV-WSN; our unpublished data) and an HA also binding to a2-6 linked sialic acids (from 1918 IAV [61] ) in the VLP entry assay indicates that both pathways can be utilized by HAs of different specificity and may therefore be relevant to avian as well as human IAV infections. Consistently, serum-inducible DYNA-IND entry was observed both in avian DF1 cells and in a human lung epithelial carcinoma cell line A549 (Fig. 3) . The DYNA-DEP and DYNA-IND IAV entry pathways were found by our quantitative assays to be fully redundant. In the presence of serum, the combination of dynasore (inhibiting DYNA-DEP entry) and EIPA (inhibiting DYNA-IND entry) completely abolished entry whereas either drug alone had no effect. EIPA, an inhibitor of plasma membrane Na+/H+ exchangers, has been shown to invariably inhibit macropinocytosis [14, [55] [56] [57] [58] . As other routes of endocytosis are generally not affected, EIPA is considered as a hallmark inhibitor of macropinocytosis [14] , although results obtained with EIPA should be considered with care as long as a mechanistic explanation for its effect on macropinocytosis is not yet fully clear [62] . Occasionally, a moderate two-to three-fold inhibition by dynasore alone was observed (result not shown) indicating that the capacity of the serum-inducible entry pathway is somewhat variable, possibly depending on slight variations in serum quality and factors like cell distribution in the wells that have been reported to influence viral infection [63] . A redundancy in the utilization of CME as well as a clathrin-independent route for entry of IAV has been visualized previously by quantitative live cell imaging [4] . Both routes were operative simultaneously in the same sample and the specific down-regulation of CME did not affect the total number of entry events. In response to specific extra-cellular signals (e.g. serum induction), changes in the actomyosin network occur that give rise to membrane protrusions required for macropinosome formation [13] . Compounds inhibiting actin polymerization (cytochalasin B and D), depolymerization (jasplakinolide) or sequestering soluble actin (latrunculin A) all specifically inhibited DYNA-IND IAV entry. In addition, the requirement for myosinII activity was established by a specific inhibitor (Blebbistatin) of myosin II ATPase activity and by the expression of a dominant negative mutant of myosinIIA heavy chain. Also, the regulation of myosinII activity by phosphorylation of myosin light chain through the action of MLCK is suggested by the inhibitory effect of MLCK inhibitors ML-7 and ML-9 as well as by the similar effect of an expressed MLCK dominant negative mutant. Recently, a function for the actin cytoskeleton in IAV entry was reported to be required for the entry into polarized epithelial cells but not for entry into non-polarized cells [64] . When using the low-serum conditions used in that paper (2% FCS), we only observed DYNA-DEP entry that was not affected by actin dynamics inhibitors. Perhaps, the polarized cells permit DYNA-IND entry at lower serum concentrations. The changes in actin network dynamics that can lead to the formation of macropinosomes can be triggered by a number of signaling cascades. Actin dynamics are induced by the activation of growth factor receptor tyrosine kinases by their respective Figure 10 . A model for IAV entry by macropinocytosis. The model summarizes the inhibitory (red boxes) or stimulatory (blue boxes) effects of compounds on dynamin-independent IAV entry. The effect of over-expression of dominant-negative mutants is indicated by red-lined boxes. The pathway requires the presence of serum factors in the entry medium and results in the enhanced uptake of dextran and its co-localization with IAV in large vesicles (green boxes). We hypothesize that the interaction of serum factors and/or IAV with receptor tyrosine kinases (RTKs) is the primary signal for the induction of macropinocytosis. A number of RTKs have been shown to be involved in this process in different cell lines. Remarkably, a recently published genome-wide siRNA screen of IAV infection identified the FGF receptor as a host factor required for influenza virus replication [22] . Activation of Rho family GTPases CDC42 and/or Rac1 has been shown to be essential for signal transduction leading to macropinocytosis in many cases [13, 14] but inhibitors are without effect or are stimulatory in the case of IAV entry. Downstream effectors of Rho family GTPases include scaffold proteins like N-WASP and WAVE and protein kinases like PAK1. Macropinocytic entry of IAV however seems to require a Rho family GTPaseindependent PAK1 activation mechanism. In addition, src family kinases, which can be directly activated by RTKs, play a role. PAK1 and src have previously been linked to the activation of macropinocytosis via their effect on changes in actomyosin dynamics, a process which is crucial to any form of macropinosome formation [13, 14] . Apart from N-WASP-or WAVE-containing macromolecular assemblies other actin binding proteins can induce such changes (e.g. cortactin, which can be activated by src [81] ) and thereby induce the formation of one of the different plasma membrane protrusions that can result in the formation of macropinosomes. In addition to an effect on the formation of plasma membrane protrusions and subsequent macropinosome formation, inhibitors can also affect downstream trafficking and maturation of macropinosomes which might be actindependent, but this is not depicted in the scheme. doi:10.1371/journal.ppat.1001329.g010 growth factor ligands that are normally present in serum [12] [13] [14] 17, 36, 37] The signal transduction cascades that link activation of growth factor receptor tyrosine kinases to actin remodeling and macropinocytosis are only beginning to be revealed. The specific inhibition of DYNA-IND entry of IAV by IPA3, an inhibitor of PAK1, provides proof for the involvement of these cascades. PAK1 is a key serine/threonine kinase regulating actin network dynamics but its crucial function in several pathways of endocytosis as well as numerous other cellular processes does not make it a very specific marker [65] . Even so, macropinocytosis has consistently been demonstrated to require PAK1 activation, both in the induction of the process and/or in further downstream trafficking events of macropinosomes [13, 14] . Growth factor dependent activation of PAK1 has most often been demonstrated to depend on upstream activation of small GTPases Rac1 or cdc42 [34, 40, 41] . Different strains of vaccinia virus were recently shown to induce their uptake by macropinocytosis via activation of either Rac1 or cdc42 [34] . Activation of Rac1 has been linked to the induction of macropinocytosis via actin network-mediated formation of lamellipodia and/or circular ruffles whereas cdc42 has most often been implied in the formation of filopodia [44] . An inhibitory effect of the Rac1 inhibitor NSC23766 or the cdc42 inhibitor pirl1 on IAV entry, however, could not be demonstrated. Remarkably, cdc42 inhibitor pirl1 enhanced IAV entry and a similar effect was observed by wiskostatin, an inhibitor of N-WASP which functions directly downstream of cdc42 as a scaffolding complex required for the activation of actin polymerization leading to filopodia formation. Similarly, the macropinocytosis-like entry pathway taken by Coxsackie B virus was also shown to require PAK1 activity that was independent of Rac1 activation [35] . Direct examination of the magnitude and timing of the activation of PAK1 will be required to obtain more insight in the involvement of this complex pathway. The induction of macropinocytosis by a PAK1dependent mechanism has been associated with ruffling at the cell membrane [12, 14, 15, 37] . The identification of sub-membranous regions with increased actin staining by phalloidin has been interpreted as evidence for ruffling. This was not unambiguously identified by confocal microscopy in the experiments presented in Fig. 8 and Fig. S2 and needs to be investigated in depth by life cell imaging techniques. In agreement with our observation that the DYNA-IND entry of IAV was inhibited by PP2, an inhibitor of src family kinases, the non-receptor tyrosine kinase c-src has been shown to function as a key signaling intermediate in the induction of macropinocytosis via a mechanism independent of Rac1 or cdc42 [49] [50] [51] . Downstream effects of c-src on actin networks proceed, amongst others, via phosphorylation of cortactin by c-src resulting in accelerated macropinosome formation [50] . C-src has been shown to associate with macropinosomes [49, 51] , both during their formation and their trafficking, while c-src kinase activity is required for macropinocytosis following EGF stimulation of HeLa cells [49] . Interaction of HSP90 with c-src was recently shown to induce c-src kinase activity [54] . Also HSP90 has been demonstrated to associate with macropinosomes, while its specific inhibitor geldanamycin reduced the membrane ruffling that preceded macropinocytosis [66] . Thus, the inhibition of IAV entry via macropinocytosis by AA-geldanamcyin may very well involve the effects of HSP90 on c-src. As detailed above, the DYNA-IND entry pathway of IAV shares many characteristics with the endocytic pathway macropinocytosis. This is corroborated by the observation that IAV particles and dextran colocalize in large vesicles in the presence of FCS. Several viruses have recently been reported to enter cells via macropinocytosis [12, 14] . Apart from common factors like the requirement for PAK1 activation, actin dynamics and independence of dynamin, virus specific details have been described [34, 35] (Table 1 ). In part these might be contributed to differences in experimental conditions (e.g. cell types tested) but diversity in the molecular mechanisms by which macropinocytosis can be induced and executed is likely to exist and to be exploited by viruses. Whereas vaccinia virus is able to trigger its own macropinocytic uptake [34, 43] , we have described a macropinocytosis pathway that is operational under conditions that are activated by components in serum. Still, this does not exclude signaling induced by virus-host cell interactions, which are for instance suggested by the significant increase of FITC-dextran uptake in the presence of IAV. The possible requirement for costimulatory signals from serum components and virus imposes an additional layer of complexity on the analysis of IAV entry via DYNA-IND pathways. Influenza viruses cause respiratory infections by targeting the epithelial cells lining the respiratory tract. These surfaces are covered by a mucous layer composed of a variety of small solutes and glycoproteins derived among others from goblet cells [67] . This semi-fluid layer in turn conditions the underlying cells and determines their physiological state, including the activities of their uptake and secretion pathways. It will be important to determine to what extent the DYNA-DEP and DYNA-IND IAV entry pathways are operational under the conditions prevailing along the respiratory tract. Current knowledge on the protein composition of the fluids covering the respiratory epithelium is rapidly expanding by the application of proteomic methods to determine the protein composition of bronchial alveolar lavage fluids (BALF). These studies have extended the previous notion that BALF is highly similar in composition to serum. For example, just as for the serum proteome more than 85% of the total protein mass of the BALF proteome is accounted for by albumin, immunoglobulins, transferring, a1-antitrypsin and haptoglobin. In addition, many other proteins have been identified both in serum and in BALF including growth factors that can bind to growth factor receptor tyrosine kinases [68] [69] [70] . Thus, BALF is likely to harbor, just as serum, the protein factors that can activate signaling pathways that are crucial for the induction of DYNA-IND entry of IAV. In agreement herewith, macropinocytosis has been described as a functional entry pathway of Haemophilus influenzae into primary human bronchial epithelial cells [71] although the factors involved in signaling the process have not been identified yet. In addition to infecting the respiratory tract, IAV has been shown to be able to cause systemic infections involving multiple organs. This has mainly been studied in avian infections [72, 73] or by infection of mice with human-derived H1N1 or H3N2 IAVs [74] but is poorly documented for human infections and may have been underestimated thus far. Obviously, during potential systemic spreading of IAV, the serum-rich conditions that we have demonstrated here to enable the use of alternative entry pathways will be encountered and may contribute to such spreading. MDCK, A549, DF-1 and HeLa cells were maintained in complete Dulbecco's Modified Eagle's Medium (DMEM) (Lonza, Biowittaker) containing 10% (v/v) fetal calf serum (FCS; Bodinco B.V.), 100 U/ml Penicillin, and 100 mg/ml Streptomycin. Chinese Hamster-E36 cells were maintained at 37uC in a-Minimal Essential Medium (Gibco) supplemented with 10% (v/v) FCS, 100 U/ml Penicillin, and 100 mg/ml Streptomycin. Cells were passaged twice weekly. Influenza A/WSN/33 (H1N1) (IAV-WSN) was grown in MDCK cells. Briefly, ,70% confluent MDCK cells were infected with IAV-WSN at a MOI of 0.02. Supernatant was harvested after 48 hr of incubation at 37uC and cell debris was removed by centrifigutation (10 min at 2000 rpm). Virus was stored at 280uC and virus titers were determined by measuring the TCID 50 on HeLa cells. The IAV-WSN luciferase pseudovirus (WSN-Ren) system has previously been described [27] . Briefly, WSN-Ren pseudovirus harbors a HA segment in which the HA coding region is replaced by Renilla luciferase. The pseudovirus is produced in a MDCK cell line that stably expresses the HA of IAV-WSN. WR-LUC, a firefly luciferase encoding vaccinia virus (strain WR) was previously described [75] . VSV-FL, a firefly luciferase encoding VSV virus was also previously described [76] . Stocks of bafilomycin A1 (BafA1), dynasore, cytochalasin D, cytochalasin B, Blebbistatin, 17-AA-geldanamycin, ML-7, ML-9, PP-2, 5-(N-ethyl-N-isopropyl)amiloride (EIPA), IPA-3 (all obtained from Sigma-Aldrich), Latrunculin A (Enzo), jasplakinolide, wiskostatin, NSC23766 (all obtained from Calbiochem) and pirl1 (Chembridge) were prepared in dimethylsulfoxide (DMSO). All stocks were stored at 220uC. A kinase inhibitor library composed of 80 kinase inhibitors was obtained from Biomol (2832A[V2.2]). HeLa cells (10,000 cells/well in 96-well plates) were treated with 2 mUnits of Vibrio cholerae neuraminidase (Roche) in 50 ml phosphate-buffered saline (PBS) for 2 hr. After washing with PBS cells were infected with IAV as described. Virus-like particles (VLPs) were produced as described [26] . Briefly, 293T cells were transfected using Lipofectamine 2000 (Invitrogen) with pCAGGS-BlaM1 (encoding a beta-lactamase reporter protein fused to the influenza matrix protein-1), pCAGGS-HA (encoding HA derived from either A/NewYork/ 1/1918 or IAV-WSN) and pCAGGS-NA (encoding IAV neuraminidase [NA] derived from either A/BrevigMission/1/18 or IAV-WSN) and maintained in OptiMEM. Supernatants were harvested 72 h after transfection and centrifuged to remove debris. VLPs were used for inoculation of cells without further concentration. VLPs were incubated for 30 min at 37uC with trypsin/TPCK for activation of HA. MDCK or HeLa cells grown to near confluency in 24-well plates were inoculated with 250 ul of VLPs after pre-treatment of the cells with inhibitors as indicated. Infection was synchronized by centrifugation at 1500 rpm for 90 min at 4uC and was performed by further incubation at 37uC for 2 h in the absence or presence of 10% FCS and inhibitors as indicated. Detection of beta-lactamase activity was performed as described [25] by loading cells with CCF2-AM substrate (InVitrogen) and subsequent analysis by flow cytometry on a LSRII flow cytometer (Becton Dickinson). Typically 10,000 events were collected and analyzed using FlowJo 8.5.2 software. The reporter construct pHH-Gluc was derived from plasmid pHH-Fluc [22] by replacing the firefly luciferase coding region with the Gaussia luciferase coding region of pGluc-basic (New England Biolabs). Unique SpeI and XbaI restriction sites were introduced into pHH-Fluc using the Quikchange XL Site-directed mutagenesis kit (Stratagene) and oligonucleotides Spe4262 (5-9GCCTTTCTTTATGTTTTTGGCACTAGTCATTTTACCG-ATGTCACTCAG), Spe4263 (59-CTGAGTGACATCGGTAA-AATGACTAGTGCCAAAAACATAAAGAAAGGC), Xba4260 (59-GTATTTTTCTTTACAATCTAGACTTTCCGCCCTTC-TTGG) and Xba4261 (CCAAGAAGGGCGGAAAGTCTAG-ATTGTAAAGAAAAATAC). A SpeI site was introduced by sitedirected mutagenesis in pGluc-basic directly following the start codon of the Gaussia luciferase coding sequence. The unique SpeI -XbaI fragment of pGluc-basic was subsequently cloned into the SpeI-XbaI site of pHH-Fluc resulting in plasmid pHH-Gluc. Cells were seeded in 96-well plates at a density of 10,000 cells/ well and transfected the next day with 10 ng pHH-Gluc using Lipofectamine 2000 (InVitrogen) according to the manufacturer's protocol. After 24 hrs the transfected cells were treated with inhibitors and infected as indicated. At 16 hr p.i. samples from the supernatant were assayed for luciferase activity using the Renilla Luciferase Assay system (Promega) according to the manufacturer's instructions, and the relative light units (RLU) were determined with a Berthold Centro LB 960 plate luminometer. WR-LUC and VSV-FL were used to inoculate HeLa cells (10,000 cells/well) at an MOI of 2, in complete Dulbecco's Modified Eagle's Medium (DMEM) (Lonza, Biowittaker). After 7 hr the luciferase activity was detected using the SteadyGlo assay kit (Promega). The addition of 10% (v/v) FCS did not change infection levels for both viruses. Cells were fixed with 3.7% paraformaldehyde (PFA) in PBS and subsequently permeabilized with 0.1% Triton-X-100 in PBS. After blocking with normal goat serum IAV-infected cells were incubated for 1 h with a monoclonal antibody directed against the nucleoprotein (NP) (HB-65; kindly provided by Dr. Ben Peeters). After washing, the cells were incubated with a 1:400 dilution of Alexa Fluor 488-or 568-labeled goat anti-mouse IgG (Molecular Probes) secondary antibody for 1 h. Nuclei were subsequently stained with TOPRO-3 and after three washing steps, the coverslips were mounted in FluorSave (Calbiochem). Actin was stained using phalloidin labeled with Alexa Fluor 633. The immunofluorescence staining was analyzed using a confocal laser-scanning microscope (Leica TCS SP2). FITC, GFP or Alexa Fluor 488 were excited at 488 nm, Alexa Fluor 568 at 568 nm, and TOPRO-3 at 633 nm. HeLa cells were grown in 24-well plates on glass coverslips (50,000 cells/well). Prior to FITC-dextran uptake cells were serum-starved for 2 hr in PBS. FITC-dextran (MW70,000, Sigma-Aldrich) was incubated with HeLa cells (final concentration of 0.5 mg/ml) in 500 ml PBS or in PBS containing 10% FCS in the absence or presence of IAV (strain WSN; MOI 10; concentrated and purified by centrifugation through a 15 to 30% sucrose gradient with a 50% sucrose cushion at the bottom) at 37uC. After 15 min cells were washed 4 times with PBS at 4uC, fixed with 3.7% PFA in PBS and subsequently permeabilized with 0.1% Triton-X-100 in PBS. Slides were stained for examination by confocal microscopy as described above. For quantification of FITC-dextran uptake 1.5610 5 HeLa cells were infected with IAV-WSN (MOI 10) in suspension in a volume of 1 ml in the presence of FITC-Dextran (1 mg/ml). Infections were performed for 15 min in PBS (containing 2% BSA to reduce unspecific binding of FITC-Dextran) or in PBS containing 10% FCS at 37uC or at 4uC (control for binding of FITC-Dextran to cells in the absence of endocytosis). Mock-infected samples were analysed in parallel. Infection was terminated by addition of 3 ml ice-cold PBS followed by three washes with cold PBS and fixation with 3.7% PFA. 20,000 cells were analyzed by FACS and results were represented as the mean fluorescence which was plotted relative to the uptake in the mock-infection in PBS (after subtraction of background fluorescence obtained at 4uC). The effect of dynasore and EIPA on dextran and transferrin uptake HeLa cells (grown on glass cover slips) were incubated at 4uC for 1 hr with 50 mg/ml Alexa633-labeled Transferin (InVitrogen) in PBS. After 1 hr the medium was replaced by PBS or PBS supplemented with 10% FCS containing IAV (strain WSN; MOI 10) and 0.5 mg/ml FITC-Dextran (Sigma; 70 kDa) and cells were transferred to 37uC for 15 min. After 15 min cells were fixed and stained as described above and examined by confocal microscopy. Cells were fixed with 3.7% PFA in PBS and subsequently permeabilized with 0.1% Triton-X-100 in PBS. Peroxidase was visualized using an AEC substrate kit from Vector Laboratories. IAV-positive cells were detected using bright-field light microscopy. Two siRNA duplexes targeting different sites within the coding sequences of dynamin 2 were obtained from Ambion Inc (15581 (Dynamin 2 siRNA 1) and 146559 (dynamin 2 siRNA2)). A scrambled siRNA (Ambion Inc.) was taken along as a control for non-specific effects of the transfection procedure and was used for normalization. One day after seeding in 96-well plates (6,000 cells/well), the HeLa cells were transfected with a final concentration of 10 nM siRNA using oligofectamine (Invitrogen). 48 h after transfection, the cells were inoculated with the WSN-Ren pseudovirus (MOI 0.5) in PBS or in PBS containing 10% FCS. After 2 h of infection the entry medium was replaced by complete growth medium containing 10 nM BafA1 to prevent further entry. At 16 h post infection intracellular Renilla luciferase expression was determined as described above. Each siRNA experiment was performed in triplicate. Cell viability was not affected as determined by performing a Wst-1 cell-viability assay (Roche). Functional knockdown of dynamin 2 mRNA levels was performed by quantitative RT-PCR. using a TaqMan Gene Expression Assay for DNM2 (Hs00191900_m1, Ambion) and using 18S RNA (Hs03928985_g1, Ambion) as a control for normalization. The comparative Ct-method was used for quantification of the results [77] . Reduction of dynamin 2 protein levels was determined by western blotting using polyclonal goat-anti-dynamin 2 C18 (Santa-Cruz SC-6400). A monoclonal against alpha-tubulin (DM1A, Sigma T9026) was used to detect tubulin for normalization. Results were quantified by Densitometric scanning of the dynamin 2 and tubulin signals displayed in Fig. 4H . HeLa cells were grown in 24-well plates on glass coverslips (50,000 cells/well) for 24 hrs. Cells were then transfected (1 mg of DNA with lipofectamine 2000 as described above) with plasmids encoding wild-type or dominant-negative (DN) human MLCK fused to GFP [78] , wild-type or DN Rab5 fused to GFP [79] , or MyoII-tail or MyoII-head domain fused to GFP [80] . 24 hr after transfection cells were inoculated with IAV-WSN (MOI 1) in PBS or in PBS containing 10% FCS and 80 mM dynasore. 4 hr after infection cells were fixed and stained for examination by confocal microscopy as described above. An unpaired Student's t-test was used for detemination of statistically significant differences. The use of the term significant in text refers to a comparison of values for which p,0.05.

Alpha-COPI Coatomer Protein Is Required for Rough Endoplasmic Reticulum Whorl Formation in Mosquito Midgut Epithelial Cells

BACKGROUND: One of the early events in midgut epithelial cells of Aedes aegypti mosquitoes is the dynamic reorganization of rough endoplasmic reticulum (RER) whorl structures coincident with the onset of blood meal digestion. Based on our previous studies showing that feeding on an amino acid meal induces TOR signaling in Ae. aegypti, we used proteomics and RNAi to functionally identify midgut epithelial cell proteins that contribute to RER whorl formation. METHODOLOGY/PRINCIPAL FINDINGS: Adult female Ae. aegypti mosquitoes were maintained on sugar alone (unfed), or fed an amino acid meal, and then midgut epithelial cells were analyzed by electron microscopy and protein biochemistry. The size and number of RER whorls in midgut epithelial cells were found to decrease significantly after feeding, and several KDEL-containing proteins were shown to have altered expression levels. LC-MS/MS mass spectrometry was used to analyze midgut microsomal proteins isolated from unfed and amino acid fed mosquitoes, and of the 127 proteins identified, 8 were chosen as candidate whorl forming proteins. Three candidate proteins were COPI coatomer subunits (alpha, beta, beta'), all of which appeared to be present at higher levels in microsomal fractions from unfed mosquitoes. Using RNAi to knockdown alpha-COPI expression, electron microscopy revealed that both the size and number of RER whorls were dramatically reduced in unfed mosquitoes, and moreover, that extended regions of swollen RER were prevalent in fed mosquitoes. Lastly, while a deficiency in alpha-COPI had no effect on early trypsin protein synthesis or secretion 3 hr post blood meal (PBM), expression of late phase proteases at 24 hr PBM was completely blocked. CONCLUSIONS: alpha-COPI was found to be required for the formation of RER whorls in midgut epithelial cells of unfed Aa. aegypti mosquitoes, as well as for the expression of late phase midgut proteases.

Blood meal digestion takes place in the midgut of hematophagous arthropods and provides the source of nutrients for vitellogenesis and oogenesis, as well as the entry point for pathogen transmission [1] . Mosquito midgut epithelial cell ultrastructures have previously been characterized by electron microscopy in several species, including Aedes aegypti [2, 3, 4, 5] , Anopheles gambiae [6] , An. darlingi [7] , Culex quinquefasciatus [8] , and Cx. tarsalis [9] . One of the most striking findings from these studies was that the rough endoplasmic reticulum (RER) in midgut epithelial cells of adult female Aedes and Anopheles mosquitoes is organized into large whorl-like structures prior to blood feeding. With the onset of feeding, the RER whorls undergo a dramatic reorganization, presumably to facilitate the synthesis and secretion of digestive proteases and components of the peritrophic matrix [2, 3, 5, 7, 10, 11] . Interestingly, male mosquitoes, which do not blood feed, lack RER whorls in their midgut epithelial cells [12] . The identity and functional contribution of protein components localized to mosquito midgut RER whorls have not been investigated. ER whorls have been shown to form in vertebrate cells exposed to inducers of the ER stress response [10, 13] , and as a result of pathological conditions [14] . Whorl-like smooth ER structures have also been shown to form in mammalian tissue culture cells that over express transfected ER resident proteins or fluorescent protein gene fusions containing heterologous transmembrane domains [15, 16, 17] . Immunofluorescent staining has been used in these transfection experiments to show that weak proteinprotein interactions between the cytoplasmic domains of highly abundant ER imbedded proteins are sufficient to induce membrane stacking and whorl formation. The absence of some ER proteins can also induce whorl formation, suggesting that mechanisms are in place to prevent ER whorls from forming in most cell types [18] . Plasma free amino acids and protein-bound amino acids in vertebrate blood provide the raw materials for energy conversion and reproduction [19, 20, 21] , and also function to initiate signaling through the target of rapamycin (TOR) signal transduction pathway [22, 23] . Moreover, an artificial amino acid meal is sufficient to induce translation of early trypsin mRNA using the TOR signaling pathway in midgut epithelial cells, however amino acids alone do not induce the transcription and translation of the late phase trypsin protein [24] . It is possible that amino acid signaling through the TOR pathway also stimulates RER reorganization in mosquito midgut epithelial cells. In order to investigate mechanisms controlling RER whorl unwinding and the initiation of blood meal digestion in female Ae. aegypti mosquitoes, we performed proteomic studies using mass spectrometry and identified midgut proteins associated with microsomal fractions in unfed and amino acid fed mosquitoes. The 139 kDa alpha subunit of the COPI coatomer complex was selected as one of eight candidate RER whorl-forming proteins based on bioinformatic analyses. We used RNAi-mediated knockdown to determine if loss of alpha-COPI expression had any effect on RER whorl formation in unfed mosquitoes or blood meal digestion in fed mosquitoes. Our data show that alpha-COPI functions are indeed required for RER whorl formation in unfed mosquitoes, and moreover, that a deficiency in alpha-COPI blocked the expression of three late phase midgut proteases in Figure 1 . Amino acid feeding is sufficient to induce RER whorl unwinding. Three-day-old Ae. aegypti females were kept unfed (sugar fed), or fed an amino acid meal for 30 min., and then sacrificed and dissected 30 min. or 120 min. later. The dissected midguts were fixed for electron microscopy preparation as described in Materials and Methods and representative electron micrographs of midgut sections are shown here. A) Unfed mosquito (25,0006magnification) . B) Unfed mosquito (8,8006magnification) . C) Amino acid fed mosquito dissected at 30 min. post-feeding (8,8006 magnification). D) Amino acid fed mosquito dissected at 120 min. post-feeding (8,8006 magnification). doi:10.1371/journal.pone.0018150.g001 blood fed mosquitoes (AaSPVI, AaSPVII, AaLT). Interestingly, the early phase of blood meal digestion, which is characterized by the synthesis and secretion of the early trypsin protease (AaET), was not dependent on alpha-COPI expression. To determine if amino acids in the midgut lumen are sufficient to trigger RER whorl unwinding in midgut epithelial cells of Ae. aegypti mosquitoes, females were fed an artificial amino acid meal and dissected midguts were characterized for ultrastructural changes at the subcellular level using electron microscopy (EM). As seen in representative electron micrographs (Figure 1 ), large whorls of stacked RER membranes were observed in midgut epithelial cells from unfed (sugar only) mosquitoes, with some RER whorls containing .50 stacked membranes. The whorls appeared to consist entirely of tightly packed RER membranes, with no evidence of a core lipid droplet as seen in other organisms [25] . Quantitative measurements of whorl size in 20 randomly chosen EM fields revealed that the RER whorls ranged in size from ,5-35 mm 2 in unfed mosquitoes ( Figure 2 ). However, following a 30 minute amino acid feeding period, and another 30 minutes of recovery, midgut epithelial cells were found to contain significantly fewer RER whorls of .5 membrane stacks (p,0.001) ( Figure 2 ). Moreover, most of the RER contained in the midgut epithelial cells of amino acid fed mosquitoes was arranged in short linear stacks ( Figure 1C ). Quantitative measurements of RER whorl size and number in midgut epithelial cells following a recovery period of 120 minutes after amino acid feeding gave similar results, suggesting that RER whorl unwinding is rapid and complete by 30 minutes post feeding ( Figures 1D and 2 ). Reorganization of the ER in midgut epithelial cells of amino acid fed mosquitoes may be associated with differential abundance of KDEL-containing ER resident proteins, which could provide a clue as to possible whorl-enriched proteins. To test this possibility, we prepared total protein extracts from midguts of unfed and amino acid fed (30 min and 120 min post feeding) mosquitoes, and analyzed the extracts by Western blotting using a KDELspecific antibody as shown in Figure 3 . Four distinct protein bands were observed, two of which appeared to decrease in abundance after feeding (KDEL-2 and KDEL-4), relative to the internal control protein (GAPDH). The KDEL-1 protein was only present in the 30 min sample, whereas the KDEL-3 protein was present in all three protein samples. Since Ae. aegypti protein disulfide isomerase (PDI) has a predicted molecular mass of ,56 kDa, and is a KDEL-containing ER protein known to be expressed in Ae. aegypti midgut cells [26] , we analyzed these same protein extracts with a PDI antibody that recognizes Drosophila melanogaster PDI. Results from these experiments were negative (data not shown), however we could not rule out that antigenic differences between Drosophila and Aedes PDI proteins contributed to the negative result. To more directly identify midgut proteins that might be contributing to whorl formation in unfed adult female mosquitoes, we isolated microsomal proteins from midgut tissues of unfed and amino acid fed mosquitoes (30 min post-feeding) using a modified procedure that minimized protease activity in the sample and enriched for RER associated proteins (see Materials and Methods). The protein samples were separated by molecular weight using SDS PAGE and four gel slices from each lane were processed for LC-MS/MS analysis and protein annotation ( Figure 4 ). The SDS PAGE analysis revealed that intact proteins of up to ,130 kDa were present in both samples, suggesting that protease activity was minimal in the midgut protein sample from fed mosquitoes. Moreover, no major differences were observed in the distribution of proteins isolated from the unfed or amino acid fed mosquitoes, indicating that the 30 min time point preceded feeding-induced protein synthesis. Following in-gel trypsin digestion, each of the eight protein samples was analyzed by LC-MS/MS. Bioinformatic analysis of the mass spectrometry data led to the identification of 127 proteins using the Ae. aegypti genome database in a Sequest-based search (Table S1 ). Gene Ontology (GO) analysis revealed that ,30% of Figure 2 . Quantitation of whorl size and number in unfed and amino acid fed mosquitoes. Whorls with a minimum of five stacked ER membranes were counted and the total area of each whorl was determined using NIH Image software. Data are shown for whorl size and number in 20 random EM fields covering 3335 mm 2 of mosquito midgut epithelial cell area. A) Size and number of RER whorls in 20 random fields from unfed and amino acid fed mosquitoes after 30 min. and 120 min. B) Total whorl in the same sample as shown in A. Statistical analysis using Pearson Chisquare test revealed that the total whorl area in amino acid fed mosquitoes at both 30 min and 120 min post-feeding was significantly less than the total whorl area in unfed mosquitoes (bars with different letters signify significance at p,0.001 comparing 30 min and 120 min to unfed mosquitoes). doi:10.1371/journal.pone.0018150.g002 the identified proteins could be assigned to microsomes (including ribosomal proteins and protein synthesis related proteins), ,19% were mitochondrial proteins, ,37% were cytoplasmic proteins, and ,14% were from other cellular fractions, including the nucleus. Eight of the microsomal proteins that were present in the protein sample from unfed mosquitoes, were chosen as candidate whorl-associated proteins based on their known role in ER or Golgi processes (Table 1 ). Based on peptide frequency, the most often represented protein was SND1 (Staphylococcal nuclease domain-containing protein 1), which is also known as Tudor domain-containing protein 11 [27] . Although SND1 is assigned to the Golgi apparatus cellular component by GO analysis, its primary function is gene regulation and RNA processing [28] . The peptide frequency of SND1 was similar in protein samples from unfed and fed mosquitoes, even though EM analysis revealed that the majority of RER whorl unwinding had already occurred by 30 min post amino acid feeding ( Figure 1C ). This result suggests that SND1 is most likely not associated with whorl maintenance since its presence in microsomal fractions is independent of whorl formation. Besides PDI, we also identified three of the seven known coatomer subunits of the COPI vesicle transport system. As shown in Table 1 , more peptides corresponding to the alpha-COPI, beta-COPI, beta'-COPI, subunits were identified in the microsomal fractions isolated from unfed mosquitoes than fed mosquitoes ( Table 1 ), indicating that endosomal membranes in these fractions were depleted of COPI subunits. Since COPI subunits are soluble proteins that transiently cycle between vesicle membrane bound and unbound forms as a function of interactions with Arf proteins [29, 30] , the observed differential abundance of COPI subunits in the two midgut microsomal fractions suggests that they could be associated with RER whorls. We tested this idea by knocking down alpha-COPI expression using RNAi to determine if this COPI coatomer subunit is required for RER whorl structures in mosquito midgut epithelial cells. In order to test the role of alpha-COPI in RER whorl formation, we used an efficient dsRNA based RNAi protocol we previously developed for knocking down expression of abundant midgut proteases in Ae. aegypti [31] . As shown in Figure 5 , the mean level of alpha-COPI transcripts in midguts of individual dsRNA-injected mosquitoes was .90% lower than the mean level in both uninjected mosquitoes and mosquitoes that had been injected with a control dsRNA from the firefly luciferase gene (Fluc). Since 100% of the alpha-COPI dsRNA injected mosquitoes we analyzed showed a similar decrease in alpha-COPI transcript levels ( Figure 5 ), we used this dsRNA injection protocol to knock down alpha-COPI expression in mosquitoes that were subsequently maintained on sugar alone (unfed), or fed an amino acid meal and sacrificed 120 min post-feeding. As shown in Figure 6 , EM analysis of representative midguts from alpha-COPI and Fluc dsRNA injected mosquitoes showed that while large characteristic RER whorls were present in the midguts of unfed Fluc dsRNA injected mosquitoes ( Figures 6A, 6B) , the RER present in unfed alpha-COPI dsRNA injected mosquitoes was disorganized and RER whorl structures were absent ( Figures 6C, 6D) . RER reorganization at 120 min post-feeding in Fluc dsRNA . Western blot analysis using a KDEL-specific antibody shows differential expression of four ER resident midgut proteins in response to amino acid feeding in Ae. aegypti mosquitoes. Total midgut protein extracts were prepared from three day old female mosquitoes that were unfed, or amino acid fed and dissected at 30 min. or 120 min. post-feeding. Proteins were resolved by 12% SDS-PAGE gel and Western blotted as described in Materials and Methods. The four unique protein bands were labeled KDEL-1 through KDEL-4 to denote their relative molecular weight, with KDEL-1 being the largest protein. Western blotting with an antibody that recognizes the ubiquitously expressed glyceraldehyde-3P dehydrogenase protein (GAPDH) was performed to control for equal protein loading. doi:10.1371/journal.pone.0018150.g003 injected mosquitoes looked similar to that of uninjected mosquitoes at the same time point, in that few if any whorls were present (compare Figures 7B and 1D ). As shown in Figure 8 , by quantitating the number and size of whorls in midgut epithelial cells from alpha-COPI and Fluc dsRNA injected mosquitoes, it is clear that loss of alpha-COPI expression in unfed mosquitoes was associated with a significant decrease in RER whorl formation. Moreover, a similar quantitative analysis of midgut epithelial cells from amino acid fed alpha-COPI dsRNA injected mosquitoes, showed highly disorganized endosomal structures and extended regions of swollen RER ( Figures 7C and 7D ). Taken together, these data indicate that functional alpha-COPI protein is required for normal whorl formation in the midgut epithelial cells of unfed mosquitoes. Blood meal digestion in Ae. aegpti mosquitoes requires the synthesis and secretion of numerous proteases, the best characterized of which are serine proteases. The early trypsin protease (AeET) is synthesized from pre-existing mRNA and secreted into the lumen within the first 6 hr PBM, whereas the late phase serine proteases AaSPVI, AaSPVII, and AaLT, are transcribed, translated, and secreted between 18-30 hr PBM [31] . To determine if alpha-COPI functions are required for the synthesis and secretion of early and late phase serine proteases, we injected mosquitoes with Fluc and alpha-COPI dsRNA 3 days prior to blood feeding and used Western blotting to detect protease protein expression at 3 hr PBM (AeET) and 24 hr PBM (AaSPVI, AaSPVII, AaLT). As shown in Figure 9A , the pattern of AeET protein expression is similar in Fluc and alpha-COPI dsRNA injected mosquitoes, suggesting that alpha-COPI functions are not required for early phase blood meal digestion. However, as shown in Figure 9B , 100-400 ng of alpha-COPI dsRNA blocks expression of all three abundant late phase proteases at 24 hr PBM in a dose-dependent manner. Lack of late phase serine protease expression in alpha-COPI deficient mosquitoes was associated with ejection of the undigested blood meal between 12-30 hr PBM (data not shown). Blood feeding is required by Ae. aegypti mosquitoes to obtain the necessary protein-derived nutrients for completion of the gonotrophic cycle [32] . Understanding the biochemical and cellular regulation of blood meal metabolism will provide insights into this critical process in Ae. aegypti, as well as other blood-feeding arthropods that function as vectors of blood borne pathogens. Our objective in the studies reported here was to first determine if amino acids were sufficient to induce RER whorl unwinding, and if they were, to use a combination of proteomic and RNAi approaches to identify proteins that may be required for maintenance of RER whorls in midgut epithelial cells. It is possible that RER whorls function by providing a mechanism to rapidly activate midgut secretion pathways upon feeding, rather than expend energy maintaining them in the absence of feeding. Biochemical studies have shown that in addition to globin, albumin, and immunoglobulin proteins, which make up 80% of the soluble proteins in human blood, free amino acids are also present in blood [21] . Based on physiological studies showing that artificial amino acids meals are sufficient to induce early events in the blood digestion process [33] , it has been proposed that amino acids could play a signaling role by inducing protease expression [24] . Consistent with this idea, we have recently shown that amino acid feeding stimulates translation of pre-existing early trypsin mRNA through activation of the TOR signaling pathway [23] . Since large ER whorl structures have been shown to exist in the midgut of Ae. aegypti and other mosquito species [2,3,4,5,6,8,9], we used EM ultrastructural analysis to test if amino acid feeding induces ER whorl unwinding. As shown in Figures 1 and 2 , our data clearly demonstrate that amino acid feeding does induce ER whorl unwinding, and moreover, that the most abundant whorls consist of rough ER membranes based on the high density of ribosomal particles. Recently, several studies have reported the presence of ER whorl-like structures in cultured cells following different treatment conditions. For example, treatment of mouse Leydig cells with the piperazine derivative diethylcarbamazine citrate (DEC), led to the appearance of large lipid droplets, degenerative mitochondria, and giant smooth ER whorls in some cells [25] . Since DEC has been shown to disrupt ER and golgi vesicle transport processes, it could explain the formation of ER whorls. A similar loss of function study leading to the formation of ER whorls was seen in human HeLa cells in which expression of the ER protein Yip1A was knocked down by RNAi [18] . Therefore one of the functions of Yip1A could be to maintain proper ER/golgi membrane networks under normal conditions, but when it is absent, ER membranes collapse into whorl structures. Another explanation for whorl formation is the presence of one or more abundant ER-associated proteins that stabilize stacked ER membranes in the absence of ongoing protein synthesis. Support for this model comes from studies in cultured cells showing that protein overexpression can induce ER whorl formation. Snapp et al. (2003) [15] reported that overexpression of cytochrome b5 induced the formation of smooth ER whorls, they called organized smooth ER (OSER), in transfected CV-1 cells. These same OSER structures could be observed when b5-GFP fusion proteins were over expressed, as long as they retained dimerization function in either the b5 or GFP protein domains. The authors proposed that weak noncovalent interactions between ER resident proteins on apposing stacked membranes are sufficient to maintain smooth ER whorl structures. Similar OSER structures were shown to form in HeLa cells that were infected with an adenovirus vector expressing the ER resident protein LAT linked to a heterologous protein dimerization domain [16] , or overexpression of a viral protein that associates with ER membranes in infected cells [17] . In order to identify candidate RER whorl forming proteins in Ae aegypti midgut epithelial cells, we took advantage of the fact that an amino acid meal is sufficient to induce whorl unwinding, which simplifies both EM analysis and protein identification by mass spectrometry because it eliminates interference from abundant blood meal proteins. Moreover, since we found that RER whorl unwinding occurs rapidly once feeding begins (Figure 1 ), we were able to use a short recovery time of only 30 min to enrich for microsomal proteins that are present in the mosquito midgut prior to feeding. The LC-MS/MS analysis identified 127 proteins using microsomal midgut protein samples from unfed and amino acid fed mosquitoes. Eight proteins were considered candidate whorl associated proteins based on their known location and function in ER and Golgi membrane compartments (see Table 1 ). Since three of the candidate proteins encoded COPI coatomer subunits, and the peptide frequency was lower for each of the proteins in samples from fed mosquitoes compared to unfed mosquitoes, we chose alpha-COPI as a representative coatomer subunit and performed RNAi knockdown experiments. Data presented in Figures 6, 7 , and 8 show that loss of alpha-COPI expression is associated with the absence of RER whorls in unfed mosquitoes, as well as membrane disorganization and ER swelling in midgut epithelial cells of amino acid fed mosquitoes. These data suggest that alpha-COPI, and likely two or more of the other seven COPI coatomer subunits, are involved in RER whorl formation and maintenance in midgut epithelial cells of unfed female mosquitoes. Based on the association of RER whorls with untranslated AeET mRNA transcripts in unfed mosquitoes, we predicted that RER whorls inhibited the translation of AaET transcripts. However the data in Figure 9A clearly show that this is not the case since sugar fed alpha-COPI dsRNA injected mosquitoes lacked both RER whorls and AaET protein expression. Surprisingly, while synthesis and secretion of AeET was not altered in blood fed alpha-COPI deficient mosquitoes, expression of three abundant late phase proteases (AaSPVI, AaSPVII, AaLT) was inhibited ( Figure 9B ), indicating that COPI vesicle transport is required for later events in the blood digestion process. The COPI vesicle transport system has been shown to function in most cells as an retrograde transport mechanism that returns golgi-modified proteins back to the ER where they function in cell signaling and metabolism [29, 30] . However, the COPI system has also been shown to function in anterograde transport in some secretory cells where it was found to be required for exocytosis of specific proteins [34] . The COPI vesicle transport system consists of seven coatomer subunits (alpha, beta, beta', gamma, delta, epsilon, zeta), which function as structural components that promote vesicle formation, a G protein that facilitates coatomer assembly and membrane budding (Arf), guanine nucleotide exchange factors (GEFs) that activate Arf proteins and thereby initiate coatomer assembly, and GTPase activating proteins (ArfGAPs) that stimulate GTP hydrolysis in Arf proteins and induce coatomer disassembly, which is required for vesicle membrane fusion. Based on our finding that loss of alpha-COPI expression in midgut epithelial cells of unfed mosquitoes disrupts RER whorl formation, without decreasing the total amount of RER membrane in cells ( Figure 6C and 6D) , we propose that COPI coatomer proteins directly contribute to whorl formation through subunit assembly. Further experiments are needed to directly test this idea, for example, by using immunogold EM analysis to determine if COPI coatomer proteins are tightly associated with RER whorls, and if so, which coatomer subunits are colocalized. Aedes aegypti (L.) (Rockefeller strain) mosquitoes were used for all studies. Larvae were maintained on a diet consisting of equal proportions of rat chow (Sunburst Pet Foods, Phoenix, AZ), lactalbumin hydrolysate (USB, Cleveland, OH), and yeast hydrolysate (USB, Cleveland, OH). Female pupae were separated Figure 8 . Quantitation of whorl size and number in dsRNA injected unfed and amino acid fed mosquitoes. Whorls with a minimum of five stacked ER membranes were counted and the total area of each whorl was determined as described in figure 2 legend. A) Size and number of RER whorls in 20 random fields from unfed and amino acid fed mosquitoes after 120 min. that were injected with Fluc or alpha-COPI dsRNA. B) Total whorl in the same sample as shown in A. Statistical analysis using Pearson Chi-square test revealed that the total whorl area in alpha-COPI dsRNA injected unfed and amino acid fed mosquitoes was significantly less than the total whorl area in Fluc dsRNA injected unfed mosquitoes (bars with different letters signify significance at p,0.001). In addition, the total whorl area in Fluc dsRNA injected fed mosquitoes was found to be significantly less than in Fluc dsRNA injected unfed mosquitoes using the Pearson Chi-square test (bars with different letters signify significance at p,0.001). doi:10.1371/journal.pone.0018150.g008 from males using a mosquito separator. Adult mosquitoes were routinely maintained at 28uC, 70-80% relative humidity and a photoperiod of 16:8 h (L:D), on 10% sucrose ad libitum. Preparation of amino acids meal or feeding buffer for mosquito feeding Amino acid meals was prepared according to Noriega et al. [33] with modifications. Briefly, the amino acids meal consisted of 40 ml of amino acid-deficient M199 media (dM199) (Invitrogen Corporation, Carlsbad, CA), 1.6 ml of 1006 MEM nonessential amino acid solution (Mediatech, Inc., Herndon, VA), 3.2 ml of 506 MEM amino acid solution (Mediatech, Inc.), and 10 mg of HEPES (Sigma, St. Louis, MO) (pH 7.2), with a final concentration of 2.3 mg/ml total amino. Just before feeding mosquitoes, ATP (Sigma) was added to the meal to a final concentration of 5 mM. The feeding buffer was composed of 100 mM NaHCO 3 and 150 mM NaCl, pH 7.2, as described by Kogan [35] . Dissected midgets were fixed in 2.5% glutaraldehyde+2% formaldehyde in 0.1 M PIPES buffer (pH 7.4) for 1 hr at room temperature, washed in buffer, post-fixed in 1% osmium tetroxide in buffer for 1 hr, washed in deionized water, and stained with 2% aqueous uranyl acetate for 30 mins. Specimens were dehydrated through an ethyl alcohol series, infiltrated with Spurr's resin and flat embedded at 60uC. Longitudinal sections (50 nm) were cut on a Leica UC2T ultramicrotome onto uncoated 150 mesh copper grids, counter-stained with lead citrate, and viewed in an FEI CM12S electron microscope operated at 80 kV. TIFF mages (8 bit) were collected via an AMT 4 M pixel camera and used for image analysis. Twenty sequential and adjacent visual fields from each EM slide were imaged at a magnification of 8800. Total area of the ER whorls in 20 visual fields of electron microscope was measured with ImageJ (NIH), which covered a total area of 3335 mm 2 . Only ER membrane structures containing more than four membrane stacks were considered to be whorls and used for area measurements. The data were statistically analyzed as pixels per 20 visual fields using Pearson Chi-square test with SPSS for Windows (v11.5). Western blots of early phase (AeET) and late phase (AaSPVI, AaSPVII, AaLT) protease expression in unfed and blood fed dsRNA injected mosquitoes were performed as previously described [31] . Analysis of KDEL containing proteins by Western blotting was done by dissecting fifty midguts from amino acid fed or unfed (sugar fed) female mosquitoes in pre-cold PBS buffer, dipped into 76Protease Inhibitor Cocktail in 100 mM phosphate buffer (pH 7.0) (Roche Applied Science, Germany) with a forceps, and transferred into an ice-cold 1.5-ml Eppendorf tube containing 60 ml of PBS/TDS buffer (1% Triton X-100, 12 mM Na deoxycholate, 0.2% SDS in PBS, and Protease Inhibitor Cocktail). The dissected midguts were homogenized using a blue Kontes pestle. The homogenate was incubated on ice for 10 min and then spun at 10000 rpm for 10 min at 4uC. The supernatant was transferred to a pre-chilled 1.5-ml tube and 56SDS sample loading buffer was added. The mixture was boiled for 4 min, chilled on ice for 2 min, and then spun at 13000 rpm for 5 min at room temperature. The yielded supernatant was stored at 220uC for SDS-PAGE. Protein samples normalized to equal midgut equivalents were separated on 12% SDS-PAGE using standard procedures. PageRuler TM Prestained Protein Ladder (Fermentas, USA) was used as a protein standard. The proteins were transferred onto Odyssey Nitrocellulose Membranes (LI-COR Inc. Lincoln, Nebraska, USA). The membranes were dried in the air for 1 h and blocked at room temperature with 4% nonfat dry milk in 25 mM Tris-HCl, pH 7.6, 150 mM NaCl, and 10% Tween-20 (TBST), and then incubated with mouse monoclonal antibody against the peptide sequence SEKDEL conjugated to KLH (Abcam Inc. Cambridge, MA, USA). Loading controls were performed using rabbit polyclonal antibody against full length native GAPDH protein from human erythrocytes (Abcam, USA) at 1:1000 dilution in 4% nonfat milk TBST solution at 4uC overnight. After washing with TBS containing 0.1% Tween-20 (TBST), the membranes were incubated with goat anti-rabbit IRDyeH 800CW or goat anti-mouse IRDyeH 800CW secondary antibody (LI-COR Biosciences, USA) at a dilution of 1:10000 in TBST/4% nonfat dry milk for 1 h at room temperature. Labeled proteins were visualized using Odyssey Infrared Imaging System (LI-COR Biosciences, USA). About 40 midguts from the fed or unfed female mosquitoes were dissected with pre-cold PBS buffer, dipped into 76Protease Inhibitor Cocktail in 100 mM phosphate buffer (pH 7.0) (Roche Applied Science, Germany) with a forceps, and transferred into an Figure 9 . A deficiency in alpha-COPI inhibits feeding-induced expression of late phase midgut proteases. A) Representative Western blot of AeET protein expression in the midguts of mosquitoes injected with 400 ng of Fluc or alpha-COPI dsRNA and maintained on sugar for 3 days and then dissected (unfed), or blood fed after 3 days of sugar feeding and dissected at 3 hr PBM (fed). The GAPDH antibody was used as a protein loading control. Each lane contains the same midgut equivalents obtained from pooled mosquitoes. B) Representative Western blots of protein extracts prepared from pooled mosquito midguts dissected at 24 hr PBM and analyzed with antibodies against late phase serine proteases (AaSPVI, AaSPVII, AaLT) or GAPDH. Mosquitoes were injected with the indicated amount of dsRNA 3 days prior to blood feeding. doi:10.1371/journal.pone.0018150.g009 ice-cold 1.5-ml Eppendorf tube containing 100 ml of 16Isotonic Extraction Buffer (5 mM HEPES;pH 7.8, 0.25 M sucrose, 1 mM EGTA, 25 mM KCl), and an appropriate amount of Protease Inhibitor Cocktail. Mosquito midgut ER protein extraction was performed according using the Endoplasmic Reticulum Isolation kit (ER0100; Sigma, St. Louis, USA) with modification. Briefly, the dissected midguts were homogenized with a blue Kontes pestle and spun at 1000g for 10 min at 4uC. The post nuclear supernatant was transferred into a clean 1.5-ml tube and spun at 12000g for 15 min at 4uC. The post mitochondrial supernatant was transferred into a clean 1.5-ml tube and mixed with 7.5 volumes of 8 mM CaCl 2 by vortexing. The mixture was incubated on ice for 10 min and then spun at 8,000g for 10 minutes at 4uC. The supernatant was discarded, and the pellet containing the microsomal fraction, was resuspended in 20 ml of 16 Isotonic Extraction Buffer followed by a 10-min of incubation on ice. A 5 ml portion of the extracted ER protein fraction was used for measuring protein concentration with the BCA TM Protein Assay Kit (Pierce, USA). The remaining ER proteins were mixed with an appropriate amount of 56SDS sample loading buffer, boiled for 4 min, and spun at 130006 rpm for 5 min. The supernatant was stored at 220C for SDS-PAGE. Equal amounts of ER proteins isolated from midguts of unfed and fed mosquitoes were separated by 12% SDS-PAGE and visualized using the GelCodeH Blue Staining Kit (Thermo Scientific, Rockford, IL, USA). Each gel lane was cut into 4 equal slices using a scalpel, and labeled as A, B, C, and D from high to low molecular weight regions. The gel slices were stored individually stored in a clean Eppendorf tube for in-gel digestion. Each gel slice was washed first with ddH 2 O for 15 min, then twice with 50% acetonitrile (ACN), and finally with 100 mM ammonium bicarbonate (Ambic, pH 8.0) containing 50% ACN for 15 min. After drying, the gel pieces were subjected to the standard in-gel digestion protocol. Briefly, proteins were reduced by 10 mM DTT/100 mM Ambic at 56uC for 45 min., and alkylated by 55 mM iodoacetamide (IAA)/ 100 mM Ambic at room temperature in the dark for 30 min. Gel pieces were washed with 100 mM Ambic dehydrated with ACN and dried. Proteolytic digestion was performed with 12.5 ng/ml trypsin dissolved in 100 mM Ambic and incubated on ice for 45 min. The digested mixture was acidified with 2% trifluoroacetic acid (TFA) in water for 1-2 minutes and then the supernatant was collected in a clean 1.5-ml tube. The peptides were extracted from the gel slice using 0.1% TFA in water, 0.1% TFA in 30% ACN, and 0.1% TFA in 60% ACN, respectively, with ultrasonication. The peptides extracted in the four steps were combined together, concentrated by a SpeedVac to a desired volume and subjected to LC-MS/MS analysis. Mass spectrometry analysis was performed by the Chemistry & Biochemistry department proteomics core facility using in-house protocols. Briefly, trypsin digested protein samples were acidified with TFA and diluted to 20 ml prior to separation by C18 column (75 um61 mm, LC Packings, Amsterdam, Netherlands) at a flow rate of 30-500 nl/min and introduced into Finnigan LTQ (Thermo electron corporation) through nano spray (2.8 kV). Liquid chromatography mobile phases consisted of Solution A (90% water, 10%methonal, 0.5% formic acid, 0.01% TFA) and Solution B (98% methanol, 2% water, 0.5% formic acid, 0.01% TFA). A 120min linear gradient from 0 to 90% B was typically used. Samples were subjected to nanoelectrospray mass spectrometry using standard procedures and data were analyzed using Sequest (ThermoFinnigan, San Jose, CA; version27, rev. 12) and X! Tandem (www.thegpm.org; version 2007.01.01.1) using the Aedes aegypti database. Scaffold (Proteome Software Inc., Portland, OR) was used to validate MS/ MS based peptide and protein identifications. Sequest identifications required at least DCn scores of greater than 0.08 and XCorr scores of greater than 1.8, 2.5, 3.5 for singly, doubly, triply charged peptides. X! Tandem identifications required at least 2Log(Expect Scores) scores of greater than 3.0. ER proteins listed in Table 1 and Table S1were RNAi-mediated knockdown of alpha-COPI expression and validation by QRTPCR alpha-COPI gene was subjected to a RNAi according to our previous study [31] . Briefly, a DNA fragment of alpha-COPI gene was amplified by PCR (forward primer containing the T7 promoter sequence: 59 TAATACGACTCACTATAGGGAGA TGCTGACAAATTTCGAAACCAA 39; and reverse primers containing the T7 promoter sequence: 59 TAATACGACTCAC TATAGGGAGATCCGTCGCCGTAGGATTCTT 39) and cloned into the pGEM-T easy vector (Promega). Subsequently, a double-strand RNA (dsRNA) was synthesized in vitro transcription using the MEGAscript RNAi Kit (Ambion). Two-day old female mosquitoes were injected with 400 ng of dsRNA using a Nanoject II microinjector (Drummond Scientific). The knockdown efficiency of mRNA encoding alpha-COPI was determined using mosquitoes injected with Fluc (firefly luciferase) dsRNA as a control. QRT-PCR was performed using PerfeCTa SYBR Green FastMix (Quanta BioSciences) by Real-Time PCR (7300 Real-Time PCR System, Applied Biosystems) with alpha-COPI forward primer: 59 GTGTCCGCATCGTTGGATCA 39, alpha-COPI reverse primer: 59 ACAACAGCATCAGCTTGCCCAA 39. Ribosomal protein S7 mRNA levels were used as an internal control for normalization. Statistical analysis of the gene expression was done by unpaired student t test using GraphPad Prism software (GraphPad Software, Inc.). Table S1 Proteomic analysis of Ae. aegypti midgut microsomal proteins isolated from unfed (sugar fed) and amino acid fed (30 min. post-feeding). (DOC) paper: GZ JI RLM. Edited the manuscript: GZ JI WAD RLM. Contributed equally in all aspects: GZ JI.

Interaction of a Specific Population of Human Embryonic Stem Cell–Derived Progenitor Cells with CD11b+ Cells Ameliorates Sepsis-Induced Lung Inflammatory Injury

Human embryonic stem cells differentiated under mesoderm-inducing conditions have important therapeutic properties in sepsis-induced lung injury in mice. Single cell suspensions obtained from day 7 human embryoid bodies (d7EBs) injected i.v. 1 hour after cecal ligation and puncture significantly reduced lung inflammation and edema as well as production of tumor necrosis factor-α and interferon-γ in lungs compared with controls, whereas interleukin-10 production remained elevated. d7EB cell transplantation also reduced mortality to 50% from 90% in the control group. The protection was ascribed to d7EB cell interaction with lung resident CD11b+ cells, and was correlated with the ability of d7EB cells to reduce it also reduced production of proinflammatory cytokines by CD11+ cells, and to endothelial NO synthase–derived NO by d7EB cells, leading to inhibition of inducible macrophage-type NO synthase activation in CD11b+ cells. The protective progenitor cells were positive for the endothelial and hematopoietic lineage marker angiotensin converting enzyme (ACE). Only the ACE+ fraction modulated the proinflammatory profile of CD11b+ cells and reduced mortality in septic mice. In contrast to the nonprotective ACE-cell fraction, the ACE+ cell fraction also produced NO. These findings suggest that an ACE+ subset of human embryonic stem cell–derived progenitor cells has a highly specialized anti-inflammatory function that ameliorates sepsis-induced lung inflammation and reduces mortality.

Lung inflammatory injury from septic shock is the leading cause of death in patients in the intensive care unit, 1 with mortality remaining at ϳ40%. 2 The disease is characterized by progressive respiratory failure with bilateral alveolar infiltrates and lung edema. 3 Transplantation of adult bone marrow-derived mesenchymal stromal cells, endothelial progenitor cells, and bone marrow-derived progenitor cells has been studied in models of sepsis 4 -11 ; however, the results have varied, and specific cell populations responsible for the protection have not been characterized. Although in some cases transplanted cells differentiated into specialized parenchymal cells, 7,10 the lung repair observed may also be secondary to immunomodulatory effects of the transplanted cells. 4, 6, 8 Previous studies have not addressed the effects of a well-defined progenitor population derived from embryonic stem cells (ESCs) in resolution of sepsis-induced lung injury. Because ESCs are pluripotent, it was surmised that specific progenitors derived from ESCs could effectively mitigate sepsis-induced lung inflammation and injury. Using blast progenitor cells from human ESCs (hESCs) cultured in conditions favoring development of mesoderm, 12 the present study addressed the role of a purified population of progenitor cells in the lung response to polymicrobial sepsis induced by cecal ligation and puncture (CLP). It was observed that transplantation of hESC-derived progenitor cells after induction of sepsis reduced lung inflammation and edema formation, and it also reduced production of proinflammatory cytokines tumor necrosis factor-␣ (TNF-␣) and interferon-␥ (IFN-␥) without affecting production of the anti-inflammatory cytokine interleukin (IL)-10. Recipient mice also demon-strated marked reduction in mortality. Dampening of lung inflammation was the result of progenitor cells enriched with the endothelial and hematopoietic progenitor cell marker angiotensin-converting enzyme (ACE) and was largely ascribed to the interaction of these cells with CD11bϩ cells in lungs. This interaction in turn mediated reduction in production of proinflammatory cytokines and high-output NO production by CD11bϩ cells. hESCs (H1, XY, WiCell, and National Institutes of Healthapproved WA01) were maintained on mitomycin-blocked mouse embryonic fibroblast feeders in hESC growth medium (Dulbecco's modified Eagle's medium and Ham nutrient mixture F-12) supplemented with 15% knockout serum replacement enriched with 4 ng/ml of human basic fibroblast growth factor-2, 1ϫ nonessential amino acid, 1ϫ glutamax-I, and 1ϫ ␤-mercaptoethanol (all from Invitrogen Corp., Carlsbad, CA). Half of the medium was changed every 48 hours until the colonies were close to confluence. For differentiation induction, 2 to 2.5 ϫ 10 6 hESCs were resuspended in 3 ml of stem cell medium (HEScGro; Millipore Corp., Billerica, MA) supplemented with 50 ng/ml of vascular endothelial growth factor and 50 ng/ml of bone morphogenetic protein-4, plated in one well of a six-well plate (Ultra-Low; Corning Inc., Corning, NY), and incubated at 37°C with 5% CO 2 . After 24 hours, 40 ng/ml of stem cell factor, 40 ng/ml of thrombopoietin, and 40 ng/ml of Fms-related tyrosine kinase-3 (Flt3) ligand (R&D Systems, Inc., Minneapolis, MN) were added to the cultures, followed by 25 ng/ml each of granulocyte colony-stimulating factor, granulocyte-macrophage colony-stimulating factor, IL-6, and IL-3, and 3 U/ml of human erythropoietin at day 3½ of differentiation culture. d7EB cells were fractionated using fluorescein-activated cell sorting (FACS) for ACE and kinase insert domain receptor (KDR) expression. The isolated fractions were subcultured on fibronectin-coated plates in the presence of endothelial cell basal medium and 20 ng/ml of stem cell factor, 20 ng/ml of thrombopoietin, 20 ng/ml of Fmsrelated tyrosine kinase-3 ligand, 25 ng/ml each of granulocyte colony-stimulating factor, granulocyte-macrophage colony-stimulating factor, IL-6, and IL-3, and 1 U/ml of human erythropoietin. Studies were performed using 6-to 8-week-old male CD1 mice (Jackson Laboratory, Bar Harbor, ME), which were housed in pathogen-free conditions at the University of Illinois animal care facility. Experimental sepsis was induced via CLP performed as previously described. 13 In brief, after proper sterilization, the cecum was exposed via a midline abdominal incision and ligated at 75% of the distance between the distant cecal pole and the base of the cecum, followed by a 21-gauge needle puncture in a mesenteric toward antimesenteric direction. Wound closure was achieved using separate sutures (6-0 nylon) to the abdominal musculature and the skin. Immediately after the procedure, 500 L of warmed normal saline solution was administered subcutaneously. For survival studies, the mice were monitored every 6 hours for 24 hours, and thereafter every 12 hours for 5 days. All cell injections were administered 1 hour after CLP via i.v. administration through the facial vein. Studies were performed in accordance with institutional guidelines, and approval was obtained from the institutional review board. In some xenotransplantation studies, cyclosporine A (CsA) was administered as an oral solution diluted with corn oil at a concentration of 10 mg/ml. The CsA was administered p.o. via gavage at a dose of 75 mg/kg/d, beginning 1 day before CLP and continuing for another 24 hours after the procedure. 14 Pilot experiments demonstrated efficiency of this dosage in inhibiting production of IL-2 and IFN-␥ by T cells isolated from mouse spleens. Lung tissue neutrophil (polymorphonuclear leukocyte) uptake was assessed via determination of myeloperoxidase activity. Lungs were dried and homogenized in 1.0 ml of PBS (50 mmol/L, pH 6.0) with 5% hexadecyl-trimethylammonium bromide and 5 mmol/L of EDTA. Homogenates were sonicated, centrifuged at 4 ϫ 10 4 g for 20 minutes and were frozen and thawed twice, followed by homogenization and centrifugation. The supernatant was mixed 1:30 (v/v) with assay buffer (0.2 mg/ml of o-dianisidine hydrochloride and 0.0005% H 2 O 2 ), and absorbance change was measured at 460 nm for 3 minutes. Myeloperoxidase activity based on dry lung weight was calculated as the change in absorbance over time. 15 Lung capillary filtration coefficient was measured to quantify lung vascular permeability, as described previously. 16 Venous outflow pressure was elevated using a computer-controlled two-way electronic pinch valve (P/N 98301-22; Cole-Parmer Instrument Co., Vernon Hills, IL) that channeled venous fluid into silicon tubing (1/16-inch i.d.). The weight change resulting from the venous pressure increase of 6 cm of water was recorded for 5 minutes. The weight recording has two exponential components that reflect the rapid expansion of vascular volume and the slower phase of transvascular fluid filtration. The amount of fluid filtered in 5 minutes was determined via logarithmic extrapolation of the slower component to time 0. The lung capillary filtration coefficient, a measure of lung vascular liquid permeability, was computed as milliliters per minute per centimeter of water per gram dry weight by normalizing the estimate of filtered fluid by time, venous pressure change, and lung dry weight. At 6 and 12 hours after CLP, lungs were removed from euthanized mice, minced into small pieces, and incubated in RPMI-1640 medium with 1% penicillin-streptomycin and 1% glutamine for 30 minutes at 37°C and at 5% CO 2 in the presence of 1 mg/ml of collagenase type 1 and 50 U/ml of deoxyribonuclease I. After incubation, cells were filtered through a 70-mm cell strainer and washed with RPMI medium. The cells were incubated for 15 minutes at 4°C with CD11b magnetic beads (Miltenyi Biotek, Inc., Auburn, CA) and subsequently applied to MS columns (Miltenyi Biotek, Inc.) for their positive selection. Negatively selected cells were also collected. For CD11bϩ depletion studies, 24 hours before experimental procedures, mice were injected i.v. with gadolinium chloride, 10 g/g body weight; anti-Gr-1, 50 g per mouse; and anti-NK1.1, 50 g per mouse. Depletion of CD11bϩ cells was verified using flow cytometry. Mouse lungs were inflated with 10% formalin and embedded in paraffin. Formalin-fixed paraffin-embedded tissue samples were cut into 5-m sections, mounted on slides (Starfrost Plus; Maenzel Glaeser, Braunschweig, Germany), and hydrated using an alcohol gradient. Slides were rinsed in distilled water, followed by antigen unmasking using a 10ϫ concentrated retrieval solution (antigen decloaker solution; Biocare Medical Inc., Concord, CA) according to the manufacturer's instructions, and rinsed in PBS for 5 minutes. For detection of human laminin A1, tissue sections were blocked with H 2 O 2 blocking reagent for 10 minutes at room temperature. Slides were treated with a protein-blocking solution for 10 minutes at room temperature, rinsed, and incubated with human anti-laminin monoclonal antibody (L8271, clone LAM 89; Sigma Aldrich Corp., St Louis, MO) at a titer of 1:1000 for 30 minutes at room temperature, followed by anti-mouse horseradish peroxidase (EnVision FLEXϩ autostainer kit; Daco A/S, Glostrup, Denmark). Slides were rinsed and treated using the mouse-on-mouse polymer kit (MM510; Biocare Medical Inc.) for 30 minutes at room temperature. Laminin staining was detected using a betazoid diaminobenzidine kit (Biocare Medical Inc.). Human tonsil tissue stained with anti-laminin monoclonal antibody was used as positive control, and the same tissue stained only with the secondary anti-mouse horseradish peroxidase served as the negative control. Immunocytochemistry d7EB cells were fractionated using FACS according to their cell surface expression of ACE and KDR and were subcultured on 35-mm fibronectin-coated dishes on top of No. 1.5 coverslips (0.16Ϫ0.19 mm). After 2 weeks, the cells were washed twice with PBS, fixed with paraformaldehyde 4% in PBS for 30 minutes, washed with 100 mmol/L of glycine in PBS, and incubated for 1 hour with blocking buffer containing 2% to 5% normal donkey serum, 0.2% bovine serum albumin, and 0.1% Triton X-100. Subsequently, they were incubated for 2 hours at room temperature with primary antibodies against human vascular endothelial cadherin, von Willebrand factor, and CD43 at 1:200 dilution in 2% normal donkey serum, 1% bovine serum albumin, and 0.1% Triton-X in PBS. Control cells were incubated with goat IgG, rabbit IgG, and mouse IgG 1 , respectively, under the same conditions. After washing with PBS and repeat blocking for 1 hour, the cells were incubated with secondary antibodies Alexa 488 anti-goat IgG, Alexa 568 anti-rabbit IgG, or Alexa 568 anti-mouse IgG1 at 1:700 dilution for 1 hour at room temperature. Single cell suspensions were created from d7EB cells via brief trypsinization. The cells were then fluorescently labeled via incubation with 10 mol/L of carboxy-fluorescein diacetate (Green Tracker, C2925; Invitrogen Corp.) in serum-free medium for 30 minutes at 37°C. Labeling was confirmed at fluorescence microscopy, and cells were kept on ice until use. Carboxy-fluorescein diacetate-labeled green fluorescent cells, 5 ϫ 10 5 , were injected i.v. through the facial vein. For cell tracking studies, mice were sacrificed at 6, 18, and 24 hours after injection, and 5-m snap-frozen lung sections were cut. Tissues were counterstained with DAPI (4=,6-diamino-2phenylindole) and examined for native green fluorescence at confocal microscopy (Zeiss LSM 510 META confocal microscope; Carl Zeiss AG, Oberkochen, Germany). Sections were graded using a semiquantitative scale established by counting the number of fluorescent cells per 100 nuclei per power field examined at 40ϫ magnification (0 ϭ no florescence, 1 ϭ Յ1 fluorescent cell, 2 ϭ 1 to 5 fluorescent cells, 3 ϭ Ͼ5 fluorescent cells). Ten power fields per slide were examined by two independent reviewers blinded to the animal group. Extravascular lung water content was measured via determination of lung wet-to dry-weight ratios in which intravascular lung water was corrected using a hemoglobin assay of lung homogenates and peripheral blood. 4 The following antibodies were used for flow cytometry studies and FACS: mouse anti-CD11b Alexa Fluor 488 -conjugated (clone M1/70; isotype, rat IgG 2b -488); mouse anti-F4/80 (clone 6F12; isotype, rat IgG 2a ); mouse anti-NK1.1 PE-conjugated (clone PK136; isotype, mouse IgG 2a -PE); mouse anti-Gr1 PE-conjugated (clone RB6-8C5; isotype, mouse IgG 2a -PE); rat anti-mouse IFN-␥ (clone XMG 1.2; isotype, rat IgG 1 ); rat anti-mouse IL-10 (clone JES5-16E3; isotype, rat IgG 2b ); biotin mouse anti-human TLR4 (clone HTA 125; isotype, mouse IgG 2a -biotin); and biotin anti-Annexin V (all from BD Biosciences Pharmingen, San Di-ego, CA); mouse anti-KDR PE-conjugated (isotype, mouse IgG 1 -PE) and mouse anti-ACE fluorescein isothiocyanateconjugated (isotype, goat IgG-fluorescein isothiocyanate) (both from R&D Systems, Inc.); rat anti-mouse TNF-␣ (clone MP6-XT22; isotype, rat IgG 1 ; Invitrogen Corp.); rabbit polyclonal anti-NOS 2 , (clone N-20; isotype, goat IgG; Santa Cruz Biotechnology, Inc., Santa Cruz, CA); and NO-Cu-Fl, final concentration 10 mol/L (Intracellular Nitric Oxide Sensor Kit, catalog No. 96 -0293; Strem Chemicals, Inc., Newburyport, MA). For intracellular staining, mouse lung cells were first fixed via incubation using 100 L of fixation buffer (eBioscience, Inc., San Diego, CA) for 20 minutes at room temperature, and washed twice with 1 ml of permeabilization buffer (eBioscience, Inc.). Consequently, 1 to 5 g/10 6 cells of antibodies against TNF-␣, IL-10, IFN-␥, or inducible NO synthase (iNOS) were then added in 100-L volume of permeabilization buffer, and cells were incubated for 30 minutes at 4°C. After an additional washing with 1 ml of permeabilization buffer, the cells were stained with secondary anti-rat-fluorescein isothiocyanate or anti-rabbit-PE (iNOS) conjugated antibody for 20 minutes at 4°C. The cells were resuspended in flow cytometry buffer and analyzed immediately. Intracellular NO staining with fluorescentbound copper was performed per the manufacturer's instructions and in accordance with published protocols. 17, 18 Surface and intracellular marker expression were analyzed using software (LSR and CellQuest Pro; Becton Dickinson & Co., San Jose, CA). Enzyme-linked immunosorbent assays were performed on supernatants of lung homogenates and cell culture supernatants using kits for mouse-specific TNF-␣ (Mouse TNF-␣/TNFSF1A Quantikine ELISA Kit, MTA00); IFN-␥ (Mouse IFN-␥ Quantikine ELISA Kit, MIF00); and IL-10 (Mouse IL-10 Quantikine ELISA Kit, M1000) (all from R&D Systems, Inc.) according to the manufacturer's instructions. Single-cell suspensions of mouse lung CD11bϩ and CD11bϪ cells were prepared and plated in 24-well plates at a concentration of 10 6 cells in 1 ml of complete RPMI-1640 for 1 hour. The supernatants were removed, and 2 ϫ 10 5 d7EB cells in fresh medium were added either directly with the mouse cells or on the insert membrane of the Transwell system (HTS Transwell 0.4-m pore size polycarbonate membrane; Corning, Inc.). As a control, CD11bϩ cells were plated without d7EB cells. The co-cultured cells were or were not stimulated with 1 g/ml of lipopolysaccharide (LPS) for 6 hours. At the end of the stimulation period, the supernatants were collected, and mouse-specific ELISAs (R&D Systems, Inc.) were performed on samples to detect the released mouse TNF-␣, IFN-␥, and IL-10. This protocol was repeated for the experiments with ACE and KDR sorted d7EB cells. Concentrations of NO and nitrite were determined using an NO analyzer (NOATM280; Sievers Instruments, Inc., Boulder, CO). NO concentrations were determined via analysis of nitrite accumulation in the culture supernatants of co-cultured d7EB cells and CD11bϩ cells or CD11bϩ cells alone. Media aliquots were collected at the indicated times, and were injected directly into the analyzing cell containing the reducing solution (saturated sodium iodide in 50% acetic acid). The analysis was conducted at room temperature. Authentic NaNO 2 solutions of known concentrations were used as standards. For determination of NO production from human d7EB cells, cultures were washed twice with fresh serum-free media. Medium containing 1 g/ml of LPS was replenished, and cells were undisturbed for the duration of the experiment. Media samples were collected at selected times, and NO production was assessed as nitrite accumulated in the media. Single-cell suspensions of CD11bϩ cells were prepared from septic mice as described above. The cells were plated in 24-well plates at a concentration of 10 6 cells in 1 ml of complete RPMI-1640 medium for 1 hour. The supernatants were removed, and fresh medium was added in the presence or absence of LPS, 100 ng/ml, plus a series of concentrations of the NO donor diethylenetriamine (DETA) NONOate for 12 hours. The range of concentrations was chosen to mimic approximately the amount of NO produced by d7EB cells as determined by direct measurements as described above. At the end of the experiment, supernatant free of cells was collected for cytokine measurements via ELISA. Human d7EB cells were plated in 24-well plates at concentration of 2 ϫ 10 6 cells in 1 ml of growth medium. Cells were cultured overnight in the presence or absence of LPS, 1 g/ml. The supernatant was collected, spun down for 10 minutes at 500g to remove possible cell contamination, and injected i.v. into septic mice 1 hour after CLP at a volume of 200 L per mouse. CD11bϩ cells were either co-cultured with d7EB cells as described above, exposed to DETA NO, or cultured alone. After stimulation with LPS, 1 g/ml, for 12 hours, the cells were lifted by gentle scraping. Human cells, which do not express CD11b, were removed via positive selection using CD11b magnetic beads (Miltenyi Biotek, Inc.). Live CD11bϩ cells were lysed in 1ϫ radioimmunoprecipitation assay lysis buffer containing protease inhibitor cocktail, 60 L/10 ml PBS (Sigma-Aldrich Corp.). Lysates were centrifuged at 14,000 rpm for 10 minutes at 4°C. Supernatants were collected, and the protein concentration of each sample was measured using a bicinchoninic acid assay kit with bovine serum albumin as the standard (Pierce Chemical Co., Rockford, IL). For each sample, 50 g of protein was loaded onto lanes of Nu-PAGE 4% to 12% Bis-Tris gel (Invitrogen Corp.). Proteins were transferred to nitrocellulose membranes (Millipore Corp.). After incubation in blocking solution (5% dry milk in Tris-buffered saline solution with Tween 20) at room temperature for 1 hour, membranes were immunoblotted (24 hours at 4°C) with anti-iNOS rabbit polyclonal antibody (ab3523) and anti-eNOS mouse monoclonal antibody (ab76199) (both from Abcam Inc., Cambridge, MA), followed by secondary horseradish peroxidaseconjugated goat anti-rabbit or mouse (1:1000). Peroxidase labeling was detected using the ECL Western Blotting Detection System (GE Healthcare, Piscataway, NJ). Five NOD/SCID mice were injected subcutaneously with 2 ϫ 10 6 d7EB cells in the dorsal region, and the animals were observed for 6 months for development of teratomas. At the end of the observation period, the injection areas were dissected and stained with H&E. No evidence of teratomas was observed in any of the animals. Data were analyzed and figures generated using commercially available software (Prism version 4.03; Graph-Pad Software, Inc., San Diego, CA). Quantitative data are given as mean (SEM). Comparison between transplant and control groups was made using the nonparametric Mann-Whitney test. P Ͻ 0.05 was considered significant. Survival between intervention and control groups was compared using the log-rank test. First, the effects of d7EB cells on sepsis-induced lung inflammatory injury and death induced via CLP were determined. Because the studies involved transplantation of human progenitor cells in mice, the question of xenotransplantation was considered. In mice treated with CsA, transplantation of human d7EB cells (500,000 cells in 200 L of PBS) reduced mortality after lethal CLP, from 10% in control mice receiving mitomycin-blocked mouse embryonic fibroblasts (MEFs) to approximately 40% in the transplant group ( Figure 1A ). Although the CsA-immunosuppressed septic model proved useful in establishing a therapeutic role for human cells, there remained the important bias of pharmacologic immunosuppression in the sepsis model. Based on findings that ESCs and adult bone marrow stromal cells 19 -22 might not be strongly immunogenic because they generally lack major histocompatibility complex class II antigens, 23 the effectiveness of these cells in immunocompetent septic mice without the complicating effects of previous immunosuppression was evaluated. Human d7EB cells in these mice were as protective against CLP-induced death as demonstrated in CsA-treated mice; that is, survival improved from less than 10% at 48 hours after CLP in control mice to 50% in the transplant group ( Figure 1B) . Also investigated were alterations in lung vascular permeability and edema formation at 24 hours after CLP. d7EB cell transplantation significantly reduced neutrophilic lung inflammation and lung edema and prevented lung endothelial barrier dysfunction in treated mice compared with the control group (Figure 2, A-D) . The improvement in lung injury and survival was associated with decreased production of TNF-␣ and IFN-␥ in septic lungs, whereas there was no change in IL-10 production ( Figure 2E ). Despite the absence of immunosuppression, d7EB cells were not immediately rejected, and their presence in recipient mouse lungs was verified up to 24 hours after cell transplantation. Increased numbers of cells were observed in the recipient lungs at 1 and 3 hours after transplantation, as demonstrated by human-specific laminin staining ( Figure 3A ). The number of cells in the lungs gradually decreased, but cells were still detectable during the first 24 hours ( Figure 3B) ; however, no cells were visible at 48 hours after administration (data not shown). To determine whether the protection induced by d7EB cells could be ascribed to secreted factors, cultured d7EB cells were stimulated overnight with LPS, 1 g/ml, or medium alone. Addition of LPS-conditioned medium obtained from d7EB cells did not alter mortality in mice that underwent CLP compared with controls injected with nonconditioned medium or PBS ( Figure 3C ), indicating that d7EB cells were essential for protection. Because the primary source of TNF-␣ and IFN-␥ in inflamed lungs is CD11bϩ cells, that is, resident macrophages, blood monocytes, granulocytes, and natural killer cells, 24 the possibility that d7EB cell interaction with host CD11bϩ cells might induce an anti-inflammatory cytokine profile was investigated. CD11bϩ cells were isolated from lungs of septic mice treated with either d7EB cells or MEFs. These cells were then stimulated with LPS in culture in the presence of the intracellular protein transport inhibitor Golgi stop. Cytokine staining showed that lung CD11bϩ mouse cells obtained after d7EB cell transplantation produced significantly less TNF-␣ and IFN-␥ ( Figure 4A ) compared with lung CD11bϩ cells from mice receiving control MEF. These responses were not observed in CD11bϪ cells ( Figure 4A ). Whether d7EB transplantation alone influenced the CD11bϩ cells by affecting the composition of the CD11bϩ cell population isolated from lungs of septic mice undergoing d7EB transplantation was addressed. No significant difference was observed in the number of monocytes (F4/80-positive), polymorphonuclear leukocytes (Gr-1-positive), and natural killer cells (NK1.1-positive) in the CD11bϩ populations of the d7EB recipient and control lungs ( Figure 4B) . To examine whether cell-cell interaction mediated the lung protection induced by d7EB cell transplantation, studies were performed in mice depleted of macrophage and monocyte populations using GdCl 3 in combination with anti-Gr1 and anti-NK1.1 antibodies injected i.v. 24 hours before CLP challenge. Mortality was higher in these mice after CLP, which could not be prevented by d7EB cell transplantation ( Figure 4C) . Thus, d7EB cell interaction with macrophage or monocytic cells is important in the mechanism of protection. To address whether direct interaction of mouse CD11bϩ cells with d7EB cells was required for the shift to anti-inflammatory lung cytokine profile observed in mice in the cell transplantation group, lung CD11bϩ cells were isolated from septic mice 12 hours after CLP and cocultured with d7EB cells either directly or indirectly via separation using a 0.4-L Transwell microporous filter. Both direct and indirect co-cultures were stimulated with LPS for 6 hours, and the supernatants were analyzed for TNF-␣ and IFN-␥ production using a mouse-specific Tissues were snap-frozen, cut into 5-m sections, and examined for native fluorescence at confocal microscopy. Nuclei have been counterstained with DAPI. Magnification ϫ40. Image is representative of at least 3 experiments. C: d7EB-conditioned medium has no effect on sepsis-induced mortality. In contrast to d7EB transplanted cells, conditioned medium from d7EB cells injected 1 hour after CLP did not prevent death. In this experiment, 2.5 ϫ 10 6 d7EB cells in 1 ml of medium were stimulated or not stimulated in vitro with LPS, 1 g/ml, for 12 hours. The supernatant was collected, centrifuged to remove any cells, and injected i.v. in mice 1 hour after CLP at a volume of 200 L per mouse. The two control groups received an equal volume of either DMEM/F-12 growth medium not exposed to d7EB cells or sterile PBS. N ϭ 25 in each group. ELISA. Only direct co-culture reduced TNF-␣ and IFN-␥ production by the CD11bϩ cells ( Figure 4D ). The observation that direct interaction of d7EB and CD11bϩ cells was required for the anti-inflammatory phenotype of CD11bϩ cells led to consideration of the involvement of NO, an important inflammatory mediator of sepsis, 25 in the response. A time course study of NO production in cultured d7EBs over 6 hours demonstrated that d7EB cells constitutively produced NO and that they also strongly expressed eNOS ( Figure 5, A and B) . Evidence suggests that NO supplementation in the form of NO donors exerts an anti-inflammatory effect in animal models of endotoxemia through inhibition of nu-clear factor B activation and iNOS expression, and decreased production of proinflammatory cytokines. 26, 27 Therefore, whether the NO-producing d7EB cells exerted a similar anti-inflammatory effect on CD11bϩ cells was determined. d7EBs were co-cultured with CD11bϩ cells, and nitrite concentration was measured in the supernatant after LPS stimulation. Supernatant NO concentrations at 6 and 12 hours after LPS stimulation of cocultured d7EB and CD11bϩ cells were significantly lower than the supernatant NO concentration from CD11bϩ cells alone ( Figure 5C ). This effect could not be attributed to increased apoptosis because apoptosis was not significantly different in the co-cultured CD11bϩ cells compared with the CD11bϩ cells cultured alone (see Supplemental Figure S1 at http://ajpamjpathol.org). Analysis of iNOS expression in co-cultured CD11bϩ cells showed decreased protein expression in LPS-stimulated CD11bϩ cells interacting with d7EB cells (Figure 5D ). This finding was supported by measurement of iNOS expression Co-culture of d7EB cells with CD11bϩ lung cells reduces production of inflammatory cytokines. Lung CD11bϩ cells from septic mice were co-cultured with d7EB cells (ratio, 5:1) either directly or separated by a 0.4-m pore size Transwell filter. Non-co-cultured CD11bϩ mouse cells from septic mice were used as controls. The cells were stimulated with LPS, 1 mg/ml, for 6 hours, and supernatant was collected for cytokine measurement using an ELISA specific for mouse cytokines. Directly co-cultured cells demonstrated a decrease in TNF-␣ and IFN-␥ production, whereas production of these cytokines was not significantly different from that in controls when the cells were separated using Transwell filters. N ϭ 3 experiments. *P Ͻ 0.05; error bars represent SEM. AJP January 2011, Vol. 178, No. 1 in CD11bϩ cells isolated from the lungs of septic mice 12 hours after CLP, where decreased expression of the enzyme was again observed ( Figure 5E ). To further test the hypothesis that NO production by d7EB cells contributes to the anti-inflammatory phenotype of the mouse CD11bϩ cells, an NO donor (DETA NONOate) was added directly to LPS-stimulated CD11bϩ cells, this time without d7EB cells, at concentrations ranging from 0.5 to 2 mol/L, and cytokine concentrations in the supernatant were measured. Reduction in TNF-␣ and IFN-␥ production by the LPS-stimulated CD11bϩ cells was observed ( Figure 5F ). In addition, iNOS expression in these cells after stimulation with LPS was reduced to levels comparable to those observed after co-stimulation with human d7EB cells ( Figure 5G ). ACE and KDR expression in d7EB cells was monitored because these markers are associated with differentiation of hematopoietic and endothelial progenitor cells. 12, 28 Neither ACE nor KDR were expressed in undifferentiated stem cells; however, increased cell surface expression of these markers was observed after onset of mesodermal differentiation, starting at day 3 and reaching maximum at day 7 to 8 when 23% of d7EB cells were ACEϩKDRϩ and 23% were ACEϩKDRϪ ( Figure 6A ). To determine whether these two markers identified distinct ) in supernatants of directly co-cultured human d7EB and mouse CD11bϩ cells (ratio, 1:5) isolated from septic mice. The co-cultured cells were stimulated with LPS, 1 g/ml, for either 6 or 12 hours. At both times, NO production was significantly lower in the co-cultured wells compared with CD11bϩ cells alone. N ϭ 3 for each experiment. *P Ͻ 0.05; error bars represent SEM. D: Decreased iNOS expression in CD11bϩ cells co-cultured with human d7EB cells. After 12 hours of co-culture, CD11bϩ cells were selected using CD11b magnetic beads, and lysed. Western blot for iNOS protein was performed. Blot is representative of 3 separate experiments. E: Decreased intracellular iNOS expression in CD11bϩ cells isolated from lungs of mice transplanted or not with d7EB cells 12 hours after CLP. CD11bϩ cells were selected using magnetic beads, and permeabilized for intracellular staining for iNOS. Plot representative of 3 separate experiments. Open lines represent isotype-matched control. F: Exposure of CD11bϩ cells to NO alters subsequent inflammatory profile. CD11bϩ cells isolated from septic mice were stimulated with LPS, 1 mg/ml, and the indicated amounts of the NO donor diethylenetriamine NONOate (DETA NO) for 12 hours. Supernatants were collected, and TNF-␣ and IFN-␥ concentrations were measured using an ELISA. lumen-containing tubes. The ACEϩKDRϪ cells gave rise to colonies of hematopoietic precursors that stained positive for hematopoietic markers such as CD43 ( Figure 6B ). The endothelial and hematopoietic colony-forming capacity of the FACS-sorted cells was quantified by counting the number and types of colonies formed per 10,000 ACEϩ/KDRϩ, ACEϩ/KDRϪ, ACEϪ/KDRϪ, and ACEϪ/KDRϩ cells. Endothelial colonies were formed almost exclusively by the ACEϩ/KDRϩ fraction, with a small contribution from the ACEϪ/KDRϩ cells, whereas the ACEϩ/KDRϪ fraction showed great efficiency in production of hematopoietic colonies ( Figure 6C ). To determine the importance of ACE and KDR markers in the observed protective phenotype against sepsis, fractionated d7EB cells were co-cultured according to their expression of ACE and KDR with the CD11bϩ cells from septic mice as described above. The co-cultured cell fractions were stimulated in vitro with LPS for 12 hours, and mouse TNF-␣ and IFN-␥ production was monitored. Both the ACEϩ/KDRϪ fraction and the ACEϩ/KDRϩ fraction significantly reduced TNF-␣ and IFN-␥ production ( Figure 6D ), indicating the critical role of the progenitor cells expressing only ACE in reducing TNF-␣ and IFN-␥ production by CD11bϩ cells. In addition, a strong cell surface co-expression of ACE and toll-like receptor 4 (TLR4) was observed in d7EB cells, suggesting an association between ACE expression and the capability to respond to LPS ( Figure 6E ). To further validate these observations, the ACEϩ cells were fractionated and used for survival studies comparing them with the ACEϪ fraction. Only ACEϩ d7EB cells reproduced the protective effect of the mixed d7EB population in survival against sepsis, whereas ACEϪ cells were not protective ( Figure 6F ). To examine whether the reduced production of cytokines was associated with differential NO production by the ACEϩ progenitor cells, flow cytometry was used to determine co-expression of ACE and KDR with NO production. NO was produced only by the ACE-expressing cells, both ACEϩ/KDRϩ and ACEϩ/KDRϪ, whereas ACEϪ cells did not produce NO ( Figure 6G ). hESCs, derived from the inner cell mass of the pre-implantation blastocyst, are defined by their ability to selfrenew and to differentiate into all types of mature cells. 29, 30 Although these cells and their derivatives in different stages of differentiation have been used for cell therapy applications in animal models of cardiovascular disease, 31-33 peripheral vascular disease, 12, 34 and central nervous system disorders, 21, 22, 35, 36 their potential in preventing sepsis-induced lung inflammatory injury characteristic of adult respiratory distress syndrome has not been addressed. The present study demonstrates for the first time the role of a population of progenitor cells derived from hESCs in preventing lung inflammatory injury induced by sepsis and improving survival in mice. The results show that the protection is the result of the subset of cells that are ACEϩ to respond to LPS by producing eNOS-derived NO. These cells functioned by moderating the pro-inflammatory cytokine production of the host immune CD11bϩ cells and reducing the high output of iNOS-derived NO cells and thereby mitigating lung inflammatory injury. The protective effects of hESC-derived progenitor cells were associated with decreased production of the proinflammatory cytokines TNF-␣ and IFN-␥ and maintenance of the production of the major anti-inflammatory cytokine IL-10. It was demonstrated that protection by hESC-derived progenitor cells was the result of the interaction of these cells with resident lung CD11bϩ cells. Direct interaction of hESC-derived progenitor cells with CD11bϩ cells was required to create the anti-inflammatory environment in lungs because neither injection of septic mice with d7EB-conditioned medium nor indirect co-cultures reproduced the salutary changes in the cytokine profile. Because of the need for cell-cell interaction, the possibility was considered that paracrine factors with sufficient diffusing capacity and instability in solution such as NO 37 might be involved. d7EB progenitor cells constitutively produced eNOS-derived NO in amounts comparable to those of immune cells, and strongly expressed eNOS. Co-culturing CD11bϩ cells with d7EB cells significantly reduced NO production in the culture supernatant after stimulation with LPS. This reduction in NO production was coupled with inhibition of iNOS expression in the co-cultured CD11bϩ cells and in CD11bϩ cells isolated from lungs of septic mice that received transplanted d7EB cells. eNOS-derived NO is beneficial in maintaining vascular endothelial integrity, 38 and eNOS-derived NO production suppresses nuclear factor B activity, decreases the transcription of iNOS and intercellular adhesion molecule-1, and prevents lung injury and death due to endotoxin. 39 Thus, NO produced by d7EB cells may protect against iNOS activation and resultant high NO production, which has known deleterious effects on the host. 40 To test this hypothesis, CD11bϩ cells were exposed to an NO donor at a concentration range that induces NO release equivalent to the amount of NO produced by d7EB cells. This experiment reproduced the decreased generation of pro-inflammatory cytokines and reduced the iNOS expression observed in the CD11bϩ cells cocultured with human d7EB cells. Thus, the results suggest a critical role of eNOS-derived NO by d7EB cells in down-regulating iNOS activation in CD11bϩ cells and, thereby, dampening the production of pro-inflammatory cytokines. These studies provide a mechanistic basis by which d7EB cells prevent sepsis-induced lung inflammatory injury. hESC-derived d7EB cells responsible for the protective phenotype were identified and found to be enriched in two cell surface markers, ACE and KDR. It was also demonstrated that the population consisting of ACEϩKDRϩ and ACEϩKDRϪ progenitor cells gave rise in culture to endothelial and hematopoietic colonies, respectively. In addition, the ACE-expressing cells expressed TLR4, the receptor sensing LPS, and produced NO. In contrast, ACEϪ cells did not exhibit these char-acteristics. The functional importance of these observations was reinforced by the ability of ACEϩ fractions of d7EB cells to modulate the inflammatory cytokine production profile of CD11bϩ host cells through direct cellcell interaction. ACE is constitutively expressed on the surface of endothelial and hematopoietic precursors, 41, 42 and ACE has been identified as a marker of hematopoietic stem cells present at all stages in the ontogeny of the human hematopoietic system 43 and as a novel marker of hemangioblasts differentiating from hESCs. 28 Evidence also points to an important role of ACE in sepsis. ACE knockout mice exhibited improved lung injury scores after acid aspiration, endotoxin challenge, and peritoneal sepsis. 44 Cohort studies in humans have shown a correlation between ACE polymorphisms and susceptibility to and death from adult respiratory distress syndrome, 45 and decreased plasma concentrations of ACE have been observed in patients with adult respiratory distress syndrome and sepsis. 46 A close homologue of ACE, ACE 2, has been identified as a key factor in protection from adult respiratory distress syndrome, and it also functions as a critical in vivo receptor for severe acute respiratory syndrome. 47, 48 A limitation of the present study is that the immunologic properties of human progenitor cells were studied in a xenograft model, introducing biases related to cross-species barriers. However, there is no reason to believe that species incompatibility affects the validity of the proposed mechanism, which involves interaction with host CD11bϩ cells and modulation of NO production. Human cells were not acutely eliminated by the immune-competent mice because cells were detected in lungs for 24 hours after injection. These data are in accord with observations in syngeneic mouse mesenchymal stem cells 6 and may be related to the immunosuppressive nature of the human cells. 49 It is likely, however, that some degree of immunosuppression will be required for longer term xenotransplantation studies. 50 Although, to our knowledge, the present study is the first to describe the function of mesodermally differentiated hESCs in a model of lethal sepsis and lung injury, other studies have used bone marrow mesenchymal stromal cells in similar models. 4,6 -11 These studies demonstrated an anti-inflammatory benefit of these cells but showed variable effects on the production of anti-inflammatory and pro-inflammatory cytokines. In a polymicrobial sepsis model, the beneficial effect of bone marrow stem cells was attributed to stem cell-induced production of IL-10 by host macrophages, 6 sphingosine-1-phosphate production by stem cells, 11 and homing of stem cells to lungs through integrin expression. 7,11 hESC-derived progenitor cells represent a much earlier developmental stage than bone marrow mesenchymal stem cells, and they have demonstrated ability to differentiate into niche-specific mature cells, a characteristic that distinguishes them from adult mesenchymal stromal cells. Although the importance of the interaction of hESC-derived cells with host CD11bϩ cells and the association of ACE expression with the protective phenotype were observed, factors such as homing to a niche and the engraftment potential of hESC-derived cells that may also be impor-tant in the observed protection cannot be ruled out. Findings of the present study demonstrate that d7EB cells, a population of hESC-derived progenitor cells, constitute a novel immunomodulatory cell population with therapeutic potential in sepsis that may have clinical application in cell-based therapy.

Wide Prevalence of Heterosubtypic Broadly Neutralizing Human Anti-Influenza A Antibodies

(See the editorial commentary by Donis and Cox, on pages 1010–1012.) Background. Lack of life-long immunity against influenza viruses represents a major global health care problem with profound medical and economic consequences. A greater understanding of the broad-spectrum “heterosubtypic” neutralizing human antibody (BnAb) response to influenza should bring us closer toward a universal influenza vaccine. Methods. Serum samples obtained from 77 volunteers in an H5N1 vaccine study were analyzed for cross-reactive antibodies (Abs) against both subtype hemagglutinins (HAs) and a highly conserved pocket on the HA stem of Group 1 viruses. Cross-reactive Abs in commercial intravenous immunoglobulin were affinity purified using H5-coupled beads followed by step-wise monoclonal antibody competition or acid elution. Enzyme-linked immunosorbent assays were used to quantify cross-binding, and neutralization activity was determined with HA-pseudotyped viruses. Results. Prevaccination serum samples have detectable levels of heterosubtypic HA binding activity to both Group 1 and 2 influenza A viruses, including subtypes H5 and H7, respectively, to which study subjects had not been vaccinated. Two different populations of Broadly neutralizing Abs (BnAbs) were purified from intravenous immunoglobulin by H5 beads: ∼0.01% of total immunoglobulin G can bind to HAs from both Group 1 and 2 and neutralize H1N1 and H5N1 viruses; ∼0.001% is F10-like Abs directed against the HA stem pocket on Group 1 viruses. Conclusion. These data—to our knowledge, for the first time—quantitatively show the presence, albeit at low levels, of two populations of heterosubtypic BnAbs against influenza A in human serum. These observations warrant further investigation to determine their origin, host polymorphism(s) that may affect their expression levels and how to boost these BnAb responses by vaccination to reach sustainable protective levels.

Influenza remains a major medical problem and is a constant threat to human health. Of the 3 influenza virus genera (A-C), influenza A is generally associated with more severe disease and is further subtyped by 2 surface proteins, hemagglutinin (HA) and neuraminidase (NA). Various combinations of the 16 HA (2 phylogenetic groups) and 9 NA subtypes define all subtypes of influenza A viruses. The Group 1 HA subtypes are H1, H2, H5, H6, H8, H9, H11, H12, H13, and H16; the Group 2 HA subtypes are H3, H4, H7, H10, H14, and H15. Seasonal viruses, such as influenza A H1N1 (Group 1) and H3N2 (Group 2), and influenza B viruses cause infection in 5%-15% of the population worldwide and 250,000-500,000 deaths annually. In addition to frequent annual epidemics, influenza viruses periodically cause pandemics, the most recent example being the 2009 pandemic caused by the swine-origin H1N1/2009 influenza virus. Vaccination is the principle means of preventing seasonal and pandemic influenza and it's complications. A universal vaccine that induces broad immunity against multiple subtypes of influenza A viruses is a long-sought goal in medical research. Recently, we and others identified a family of human broadly neutralizing ''heterosubtypic'' antibodies (BnAbs) that bind to a highly conserved pocket on the stem of HA present in all Group 1 influenza viruses [1, 2] and block virus-host cell membrane fusion. These BnAbs were identified by recombinant H5 panning from antibody (Ab)-phage libraries that are constructed from nonimmune B cells [1] , immunoglobulin (Ig) M memory B cells of seasonal vaccinees [3] , or bone marrow of H5N1-infected ''bird-flu'' survivors [4] . BnAbs with similar properties have also been recovered from immortalized IgGexpressing memory B cells of seasonal vaccinees [5] . An unexpected finding from these studies is the frequent contribution of VH1-69 heavy chain genes to these BnAbs, suggesting that a large fraction (up to 10%) of the human naive B cell repertoire has the capability of responding to this conserved epitope [5, 6] . These observations raise additional questions as to whether such BnAbs are present in human serum and at ''protective'' levels, whether they exist as ''natural'' Abs, and/or whether are they generated during the immune response to influenza virus infection or vaccination [5, 7] . To explore these questions, we analyzed serum samples from H5N1 vaccinees and commercial intravenous immunoglobulin (IVIG) samples for their crossreactive binding and neutralization activity against different influenza A subtypes. We also utilized a representative BnAb, F10, for which the precise epitope was determined crystallographically [1] , to probe for heterosubtypic BnAbs directed against the highly conserved pocket on the HA stem. Serum samples from 77 healthy volunteers, matched before and after vaccination (1-4 months), had been collected and stored at a single center, from a dose-escalating clinical trial (Clin-icalTrials.gov identifier: NCT00383071) on an inactivated H5N1 vaccine, conducted at the National Institute of Allergy and Infectious Diseases, National Institutes of Health [8] . The vaccine (manufactured by Sanofi Pasteur) used in the trial was a monovalent, inactivated subvirion H5N1 vaccine (rgA/ Vietnam/1203/04 X A/PR/8/34). The study was conducted in accordance with institutional review board-approved protocol. Enzyme-linked immunosorbent assays (ELISAs) were performed to test the cross-reactivity of the serum samples against multiple influenza A subtypes. Recombinant HA proteins, including H1 (A/New York/18/2009(H1N1), H1-NY18), H3 (A/ Aichi2/68 (H3N2), H3-A2/68), H5 (A/Vietnam/1203/04 (H5N1), H5-VN04), and H7 (A/Netherlands 219/03 (H7N7), H7-NL03), were expressed in insect cells as trimers of HA ectodomains [1] . HA proteins (0.2 lg) were coated onto 96-well Maxisorb ELISA plate (Nunc) at 2 lg/mL in phosphate-buffered saline (PBS) at 4°C overnight. The plate was washed with PBS for 3 times to remove uncoated proteins. Serially diluted serum samples were applied to the HA-coated plates, followed by Horseradish Peroxidase (HRP)-anti-human IgG or IgM (Pierce Biotechnology), to detect the IgGs or IgMs against various HA subtypes in the serum samples. The optical density at 450 nm was measured after incubation of the peroxidase tetramethylbenzidine substrate system. A competition ELISA assay was conducted to determine the level of F10-like Ab in the serum samples. F10 Ab (human IgG1) was biotinylated with Sulfo-NHS-SS-Biotin (Pierce Biotechnology) in accordance with the manufacturer's instructions. Biotinylated F10 (Bio-F10; 3 ng/mL) was mixed at 1:1 (vol) ratio with serum samples at various dilutions and added to ELISA plates coated with H5-VN04 trimer. The competition of serum samples for the binding of Bio-F10 to H5 was determined by measuring the remaining binding of Bio-F10 using HRP-Streptavidin (BD Bioscience). To isolate H5-bound Abs and F10-like Abs from intravenous immunoglobulin (IVIG; 100 mg/mL, Gamunex IVIG; Talecris Biotherapeutic), we first immobilized H5-VN04 proteins on magnetic beads (Dynabeads M-270 Epoxy; Invitrogen) in accordance with the manufacturer's manual, and then used step-wise Bio-F10 elution and acid elution to separate the F10-like Abs and H5-bound Abs present in the IVIG. Specifically, 5 3 10 8 of H5-beads were incubated with 1.5 mL of IVIG (100 mg/mL) overnight at 4°C, washed extensively with 0.1% BSA/PBS, followed by Bio-F10 elution (100 lg/ml, 0.5ml, total 50 lg) overnight at 4°C. The remaining H5-bound Abs on the H5-beads were then isolated with acid elution (pH 5 2.8; buffer, 500 lL/10 9 H5beads). Next, the Bio-F10 eluents and the acid eluents were incubated with Strep-T1 beads (250 lL of 4 3 10 8 beads; Invitrogen) at 4°C for 4-8 h each time for 2 and 3 times, respectively. ELISA was performed to verify the Bio-F10 in both samples was completely removed after Strep-T1 beads absorption. The purified H5-bound Ab and F10-like Abs, as well as unpurified IVIG, were quantified by ELISA using known concentrations of F10 monoclonal Ab (mAb) as standards. The aforementioned purification procedure was scaled up proportionally to obtain enough H5bound Abs and F10-like Abs to test their cross-reactivity against different influenza A subtypes by ELISA and pseudotyped virus neutralization assay, as described elsewhere [1] . Statistical analyses were performed using the paired t test for the comparisons of prevaccination and postvaccination Ab-binding levels and Bio-F10 Ab inhibition activities. The neutralization activity was compared using log 2 microneutralization assay (MN) [8] titers. The proportions of .2-fold increase in Ab level were compared by McNemar's test for paired binary outcomes. All P values were 2-sided, and P values ,.05 were considered to be statistically significant. The correlation between F10-like IgG Abs and MN titer against H5N1 in postvaccination serum samples (n 5 77) was analyzed by Spearman rank correlation coefficient analysis. In the H5N1 vaccine group, pre-vaccine IgMs against recombinant H5-VN04 protein were detectable in the majority of study subjects. The IgM binding level increased significantly after vaccination (P 5 .006); however, only 8 (10.4%) of 77 subjects demonstrated a .2-fold increase (as shown by a .2-fold increase in optical density at 450 nm by ELISA) ( Figure 1A ). All of the study subjects had pre-immune IgG Abs that bound to H5-VN04 ( Figure 1B ). As expected, the binding to H5-VN04 (P , .001) ( Figure 1B ) and neutralization activity (MN titer) against H5N1 (the vaccine strain; P , .001) ( Figure 1E ) significantly increased after vaccination. IgGs against recombinant H3-A2/68 protein ( Figure 1C ) and H7-NL03 protein ( Figure 1D ) were also detected in the pre-immune serum samples. In contrast to H5-binding, the postvaccination IgG binding to H3 did not increase significantly (P 5 .11); however, it increased to reach statistical significance for H7 (P , .001), but the proportion of specimens with a .2-fold increase in H7 binding was significantly lower than that for H5 binding (1.3% vs 76.6%; P , .001). To investigate further the presence of heterosubtypic BnAbs in serum, we probed serum samples for the presence of F10-like IgG Abs, as determined by inhibition of the binding of biotinylated F10 (Bio-F10) to H5-VN04 in a competition ELISA assay. We found that F10-like Abs that could compete for Bio-F10 binding were also present in prevaccination serum specimens, with 23 (29.9%) of 77 samples demonstrating .30% inhibition of Bio-F10 binding at 1:90 dilution ( Figure 1F) . Furthermore, the F10-like IgG titers increased significantly after vaccination (P , .001), and the majority of vaccinees (54 of 77) had a 1.5-fold increase. In addition, there is a weak but significant correlation between the F10-like IgGs and the MN titer against H5N1 in post-immune serum samples, as determined by Spearman rank correlation analysis (co-efficient 5 .44); however, these stem pocket-directed BnAbs do not reach high enough levels to render a strongly significant correlation. To determine whether these BnAbs were unique to the subject population tested or are present more broadly, we quantified the anti-H5 and F10-like Abs in a commercial IVIG and determined their breadth of heterosubtypic binding and neutralization activity. The IVIG contained pooled IgGs from thousands of donors and is representative of the pre-immune IgG Ab composition in the general population. Although it cannot be formally ruled out that IVIG is truly H5 naive, this was the best representative sample available for this study. H5-immobilized magnetic beads were used to affinity-purify F10-like and anti-H5 Abs from the IVIG. F10-like Abs were isolated by Bio-F10 competition elution from the total H5-beads bound Abs, and the remaining H5-bound Abs on the H5-beads were released with acid elution. By quantitative ELISA, we found that, from 100 mg of IVIG, 10 lg of Abs bound to the H5-coated plate (0.01% of the total). Of these, 10% could be purified using affinity purification with H5-beads by acid elution. The final yields were 1-1.4 lg acid-eluted anti-H5. The Bio-F10 competitive elution yielded 0.1 lg of F10-like Abs per 100 mg of IVIG (0.001% of the total Ig level) and with similar or higher efficiency of recovery as compared with acid elution. These anti-H5 and F10-like Abs were further tested for their heterosubtypic HA binding and neutralization activity against Group 1 and 2 viruses. The acid-eluted anti-H5 Abs are not H5-subtype specific: they showed cross-binding activity to H1-NY18, to H3-A2/68, and weakly to H7-NL219 ( In this study, we show that prevaccination serum samples have baseline heterosubtypic HA Ab binding activity to both Group 1 and 2 HA subtypes, including H5 and H7, to which these subjects are most likely unexposed because of their US geographic location. The IgM and IgG Abs to H5 increased significantly after the H5N1 vaccination, whereas this did not happen for IgG Abs to H3. F10-like IgG Abs are also detected in pre-immune serum samples and increased significantly in H5N1 vaccinees. A low level of serum anti-HA Abs that bind and neutralize H5N1 viruses has been reported to be age and influenza exposure dependent [9, 10] . Other investigators have reported enhanced levels of HA-directed anti-H5N1 neutralizing Abs in healthy donors after boosting with unrelated human influenza H1N1/ H3N2 seasonal vaccines [5, 11, 12] . These serum Abs may be directed to the Group 1 stem pocket on HA. Indeed, Corti et al [5] showed that a majority of 2007 and 2008 seasonal vaccinees had preexisting neutralizing Ab titers against pseudotyped H5N1 viruses that markedly increased after seasonal vaccination and that the HA stem-pocket directed Abs were present at relatively low levels, compared with Abs that bind to the globular head of HA. We also observed an increase of H7 (Group 2) reactivity after H5N1 (Group 1) vaccination ( Figure 1D ). The precise location of the cross-binding epitope(s) is currently not known. Observations that support the possible presence of heterosubtypic Abs of the broader type that bind Group 1 and 2 viruses have also been seen in children after primary influenza infection [13] and mucosal vaccination of experimental animals [14, 15] . In one study, a cross-neutralizing murine mAb with The binding levels are shown as optical density at 450 nm (OD 450). E, microneutralization assay (MN) titer against H5N1 virus; y-axis shows the Log 2 MN titer. F, Competition ELISA. Pre-and post-immune serum samples from H5N1 vaccinees were tested for their competition activity against a Group 1specific BnAb, F10, binding to H5-VN04. Serially diluted serum samples were mixed with 3 ng/mL Bio-F10 and applied to H5-coated ELISA plates. The serum competition for binding of Bio-F10 to H5 was determined by measuring the remaining binding of Bio-F10 using HRP-Streptavidin. The serum competition activity is shown as percentage of inhibition. For all panels except panel E, data at 1 representative serum dilution are shown, as follows: panel A, 1:270; panel B, 1:5120; panels C and D, 1:2430; and panel F, 1:90. For all panels, data are shown in a box and whiskers graph. The box extends from 25th percentile to the 75th percentile, with a line at the median. The whiskers above and below the box indicate the 95th and 5th percentiles, respectively. The dots above and below the whiskers are data points beyond the 95th and 5th percentiles. hemagglutination inhibition activity against Group 1 (H1, H2, H5, H9, and H13) and Group 2 (H3) was recovered from an intranasally vaccinated mouse. In contrast to the F10-like Abs, this mAb was directed to the globular head, and escape mutants were readily obtained [16] . Other animal vaccine studies have shown that antibodies to the fusion peptide and an HA2 linear peptide can have broad reactivity to Group 1 and 2 HAs [17] [18] [19] and neutralizing activity [20, 21] , respectively. Whether different quantitative amounts of Group 1-specific BnAbs and/or Group 1 and 2-directed BnAbs are elicited by vaccination or natural infection remains an important but unanswered question of our study. To quantify the baseline levels of anti-influenza heterosubtypic Abs in human serum, we used IVIG as representative of pre-immune IgG Ab composition in the general population, which includes individuals who are likely exposed to seasonal Figure 2 . Heterosubtypic antibodies against influenza A viruses in intravenous immunoglobulin (IVIG). H5-VN04 was immobilized on magnetic beads (H5-beads) and the beads were used to affinity purify antibodies (Abs) from the IVIG sample. F10-like Abs and the remaining H5-bound Abs were purified separately. F10-like Abs were purified by Bio-F10 competition elution followed by multiple steps of streptavidin-beads absorption to eliminate Bio-F10 completely. The remaining H5-bound Abs on the H5 beads were eluted with standard acid elution method followed by a complete absorption of Bio-F10 using Streptavidin-beads as well.An enzyme-linked immunosorbent assay (ELISA) detecting Bio-F10 was used to confirm that no residual Bio-F10 remained in either Bio-F10-eluted or the acid-eluted samples. The binding activity of these purified Abs to H1-NY18 (A), H5-VN04 (B), H3-A2/68 (C), and H7-NL03 (D) was measured by ELISA at different serially diluted concentration. The neutralization activity of these samples was measured using neutralization assay with pseudotyped viruses of H1-1918 (E) and H5-TH04 (F); 80R was used as a negative control mAb that is specifically against the spike protein of severe acute respiratory syndrome coronavirus [24] . OD 450, optical density at 450 nm. influenza A virus infection and/or vaccinations (H1N1 and H3N2). Other investigators have shown that a low level of heterotypic anti-HA Abs that bind and neutralize H5N1 viruses is present in IVIG from diverse geographic locations [9, 10] . These investigations further showed that these heterosubtypic anti-H5N1 Abs cross-react with H3N2 and H1N1; however, efforts to purify and characterize these Abs (beyond neutralization titers) have not been reported. Our data show that there are 2 populations of heterosubtypic Abs with different HA binding ability in IVIG: one can bind to HAs from both Group 1 and Group 2 viruses; the other is specifically directed against Group 1 stem pocket. Approximately 0.01% of IVIG (most certainly derived from H5-and H7-naive donors), which is purified by acid elution from H5-beads, has heterosubtypic binding activity to both Group 1 and Group 2 HAs. This fraction of IVIG Abs also demonstrated neutralizing activity against the H1N1 and H5N1 pseudotyped viruses. The F10-like stem pocket-directed Abs were also detected in IVIG and were recovered at 10% of the levels of H5 binding Abs. As expected, these F10-like Abs displayed similar binding and neutralization profiles among the tested HAs and viruses as the Group 1 specific mAbs that are directed to the stem pocket of HA [1, [3] [4] [5] . For both Ab fractions, a broader range of other subtypes were not tested due to limited amount of the Abs that we could purify from IVIG (1 lg and 0.1 lg/100 mg IVIG, respectively), larger-scale purification of H5 protein and other materials will be required to characterize these heterosubtypic BnAbs in more detail. Although we quantitatively show that BnAbs that bind to Group 1 and 2 HAs are present at very low levels, as well as that stem pocket-directed F10-like BnAbs exist at even lower levels, the question of whether serum or IVIG has protective levels of either type of heterosubtypic BnAbs is not answered in our study. However, our quantitative data support the notion that the levels of these BnAbs are borderline or below titers that would traditionally be considered protective. For example, there is up to 1 lg of F10-like Abs/100 mg IVIG; assuming 10 mg/mL IgG in normal human serum, then concentrations up to 0.1 lg/ ml of F10-like Ab could be present. Likewise, the fraction of acid eluted BnAbs with activities against Groups 1 and 2 could also be in this range. Furthermore, we did observe variability in these levels in our study patients (Figure 1) , and it remains possible that host factors, including VH polymorphism, may impact the baseline and inducible BnAb levels. In addition, the origins of these 2 populations of heterosubtypic anti-HA Abs are unknown. The possibility that they may be a component of ''natural'' polyreactive Abs cannot be excluded [22] . However, it is most likely that both our H5N1 vaccine study subjects and the IVIG donors had prior exposure to other Group 1 (H1N1) and Group 2 (H3N2) influenza A viruses, either through seasonal vaccinations and/or natural infection, and this may have given rise to heterosubtypic H5 and H7 binding Abs, respectively. In summary, our findings show that the human immune system is capable of making BnAbs-not only to the conserved pocket on the HA stem of Group 1 viruses, but also to another unknown epitope(s) that are shared by Group 1 and 2 influenza A viruses. These observations provide the basis for further investigations aimed at obtaining a better understanding of these BnAbs, their origins, and the host genetic factors that restrict or enable their induction [23] . These additional studies should bring us closer to developing a universal influenza vaccine that provides durable protection beyond seasonal vaccines and mitigates that ability of the viruses to undergo neutralization escape. Indeed, a recently reported vaccine regimen-which induced protective level of the Group 1 stem-pocket directed BnAbs in animals-provides experimental evidence that the same may be possible in man [23] .

Peptide model helices in lipid membranes: insertion, positioning, and lipid response on aggregation studied by X-ray scattering

Studying membrane active peptides or protein fragments within the lipid bilayer environment is particularly challenging in the case of synthetically modified, labeled, artificial, or recently discovered native structures. For such samples the localization and orientation of the molecular species or probe within the lipid bilayer environment is the focus of research prior to an evaluation of their dynamic or mechanistic behavior. X-ray scattering is a powerful method to study peptide/lipid interactions in the fluid, fully hydrated state of a lipid bilayer. For one, the lipid response can be revealed by observing membrane thickening and thinning as well as packing in the membrane plane; at the same time, the distinct positions of peptide moieties within lipid membranes can be elucidated at resolutions of up to several angstroms by applying heavy-atom labeling techniques. In this study, we describe a generally applicable X-ray scattering approach that provides robust and quantitative information about peptide insertion and localization as well as peptide/lipid interaction within highly oriented, hydrated multilamellar membrane stacks. To this end, we have studied an artificial, designed β-helical peptide motif in its homodimeric and hairpin variants adopting different states of oligomerization. These peptide lipid complexes were analyzed by grazing incidence diffraction (GID) to monitor changes in the lateral lipid packing and ordering. In addition, we have applied anomalous reflectivity using synchrotron radiation as well as in-house X-ray reflectivity in combination with iodine-labeling in order to determine the electron density distribution ρ(z) along the membrane normal (z axis), and thereby reveal the hydrophobic mismatch situation as well as the position of certain amino acid side chains within the lipid bilayer. In the case of multiple labeling, the latter technique is not only applicable to demonstrate the peptide’s reconstitution but also to generate evidence about the relative peptide orientation with respect to the lipid bilayer. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s00249-010-0645-4) contains supplementary material, which is available to authorized users.

Abstract Studying membrane active peptides or protein fragments within the lipid bilayer environment is particularly challenging in the case of synthetically modified, labeled, artificial, or recently discovered native structures. For such samples the localization and orientation of the molecular species or probe within the lipid bilayer environment is the focus of research prior to an evaluation of their dynamic or mechanistic behavior. X-ray scattering is a powerful method to study peptide/lipid interactions in the fluid, fully hydrated state of a lipid bilayer. For one, the lipid response can be revealed by observing membrane thickening and thinning as well as packing in the membrane plane; at the same time, the distinct positions of peptide moieties within lipid membranes can be elucidated at resolutions of up to several angstroms by applying heavy-atom labeling techniques. In this study, we describe a generally applicable X-ray scattering approach that provides robust and quantitative information about peptide insertion and localization as well as peptide/lipid interaction within highly oriented, hydrated multilamellar membrane stacks. To this end, we have studied an artificial, designed b-helical peptide motif in its homodimeric and hairpin variants adopting different states of oligomerization. These peptide lipid complexes were analyzed by grazing incidence diffraction (GID) to monitor changes in the lateral lipid packing and ordering. In addition, we have applied anomalous reflectivity using synchrotron radiation as well as in-house X-ray reflectivity in combination with iodine-labeling in order to determine the electron density distribution q(z) along the membrane normal (z axis), and thereby reveal the hydrophobic mismatch situation as well as the position of certain amino acid side chains within the lipid bilayer. In the case of multiple labeling, the latter technique is not only applicable to demonstrate the peptide's reconstitution but also to generate evidence about the relative peptide orientation with respect to the lipid bilayer. Introduction A quantitative, dynamic, and functional understanding of novel natural or synthetically modified peptide species in the native membrane environment requires information about the reconstitutability, positioning, and orientation with respect to the lipid bilayer matrix. Compared to protein complexes, membrane active peptides present relative structural simplicity and are frequently used to validate experimental approaches, techniques, and applications (Bechinger 2001; He et al. 1993) . With the presented study, we are interested in evaluating the biophysical properties of a recently published motif of straightforward designed, aromatic, b-type polypeptides. This includes analysis of peptide positioning, orientation, and changes in bilayer thickness as well as the effect of peptide reconstitution on lateral lipid packing and ordering. Furthermore, we want to introduce a general and applicable X-ray-based approach that provides combined access to significant and quantitative parameters of peptide/lipid interaction on a molecular scale within the physiological membrane state. In this way, structural information, both in the bilayer plane and along the normal direction, can be determined. While spectroscopic methods in their ''standard'' application, such as fluorescence emission (Bong et al. 2000; Krishnakumar and London 2007; Wimley and White 2000) , circular dichroism (CD) (Chen and Wallace 1997; Koeppe and Andersen 1996) , and transmission infrared (IR) spectroscopy (Küsel et al. 2007) , produce qualitative information on the structure and localization of the peptide species, more advanced applications, such as oriented CD (Wu et al. 1990 ), attenuated total reflection (ATR) IR spectroscopy (Tamm and Tatulian 1997) , Förster resonance energy transfer (FRET) experiments (parallax method) (Chattopadhyay and London 1987; Chung et al. 1992) , solid state nuclear magnetic resonance (NMR) (Andronesi et al. 2004; Strandberg and Ulrich 2004) , and electron paramagnetic resonance (EPR) techniques (Dzikovski et al. 2004; Marsh 1996) , usually account for quantitative values in one dimension of the lipid bilayer without the need for complementary simulations or geometrical models (Naik and Krimm 1986a, b) . Some of these techniques require membrane states or parameters that are far from physiologically or biologically relevant. In addition, sample preparation for some applications, e.g., frozen state samples for EPR spectroscopy, are often prone to uncertainties and artifacts. Therefore, the sample preparation method itself has to be carefully evaluated, usually with significant effort. It is obvious that no single technique can give complete structural information for each component of a fluid membrane system. However, in recent years the potential of X-ray scattering methods in probing peptide/lipid complexes, i.e., the membrane as well as the peptide structure, has been shown to be almost solely limited by the nonuniformity or patchiness of the peptide/lipid system itself (Qian et al. 2008; Salditt et al. 2006) . Therefore, if samples of suitable quality can be obtained, the potential of a simultaneous analysis at the molecular scale of the incorporated peptide species, the surrounding lipid matrix, and the peptide/lipid interactions underlines the power of this analytical method. In addition, X-ray scattering is compatible with the fluid membrane state and capable of providing information for both dimensions of the lipid bilayer space. The alignment of multilamellar membranes is quantified by the distribution function of the membrane normal vectors, which are directed along the z axis perpendicular to the solid substrate on which the membranes are deposited. The central width of this distribution is called mosaicity. A narrow distribution or equivalently small mosaicity allows for a precise distinction between the scattering vector components, vertical (q z ) and parallel (q k ) to the lipid bilayer. Thereby, the multilamellar state of the samples, which can be almost fully hydrated by controlled relative humidity (Aeffner et al. 2009 ), provides a tremendous amplification of the scattering signal compared to single bilayer preparations. However, compared to single bilayer reflectivity, the multilamellar stacks usually impede the determination of the electron density q(z) on an absolute scale, unless a full q-range fit of the electron density is performed (Constantin et al. 2003) . In principle, an ideal small-angle scattering experiment with respect to distribution, concentration, and structural integrity of the peptide species can yield multiple structural parameters (Fig. 1) . The reflection geometry for aligned lipid films (Fig. 1i ) is suited to determine the vertical density profile q(z) of the bilayers, averaged in the xy plane, e.g., by least-square fitting of the specular reflectivity curve (Constantin et al. 2003) . The specular peaks are accompanied by diffuse scattering in the form of so-called Bragg sheets extending along q k . This scattering contribution can be assigned to lateral inhomogeneities at mesoscopic length scales (nanometers to a few micrometers) due to membrane fluctuations-always present in the fluid/ hydrated state-and peptide insertions causing local orientational disorders (Fig. 1ii) , which break the lateral translation invariance. While the vertical electron density distributions may also be obtained by fitting the diffuse Bragg sheet intensities of the reciprocal space for a single angle of incidence, we have used full scans with background correction (Fig. 1iii ) in reflectivity mode (Fig. 1iv) . The lateral membrane structure at molecular length scales is accessible via the grazing incidence diffraction (GID) scattering geometry, at which the 2D detector (CCD camera) is moved out of the plane of incidence at grazing incidence and exit angles. The lipid chain correlation peak (Fig. 1v) is usually stretched around a circle with the radius q = (q k 2 ? q z 2 ) 1/2 % 1.4 Å -1 reflecting the distribution of acyl chain tilt angles with respect to the bilayer normal. The radial profile of the peak has a Lorentzian lineshape with a peak position corresponding to the inverse nearestneighbor distance of chains and its width reflecting the lateral correlation length (see below). This peak is often broadened upon peptide insertion. A lateral correlation of membrane-inserted structures, e.g., a regular oligomerization as known from pore arrangements, can be detected and characterized from ''superstructure peaks'' (Constantin et al. 2007 ) in the small-angle region at low q z (Fig. 1vi ). In addition, the conformation of the reconstituted peptide's structure can in some cases be deduced from the low q k signal of molecular form factors, the so-called helix maximum (Fig. 1vii) , or from helix peaks revealing intrahelical distances at high q k (not shown) as it has been revealed for the regularly organized and conformationally constrained alamethicin single helix (Spaar and Salditt 2003) . As we are interested in a general approach including a side-directed iodine-labeling for the reflectivity studies (Arbely et al. 2004) , the present work focuses on data recording and analysis, as can be applied for any peptide/ lipid sample. Fig. 1 Schematic representation of the reciprocal space for multilamellar membrane stacks (i) as a function of the parallel (q k ) and normal (q z ) components of the momentum transfer (q). The scheme shows the reflection geometry known as grazing incidence X-ray scattering (GISAXS) at small q k or grazing incidence diffraction (GID) at high q k . The recorded patterns include the diffuse Bragg sheets (ii) in the vicinity of the (specular) q z axis. In this study the q z components (iii) were measured by line scanning under specular conditions (iv). At low q z , the lipid acyl chain correlation maximum (v) is observed. Lorentzian fits yield the lateral lipid chain distance. The width of the acyl chain peak along q k gives information about the lipid ordering (correlation length); the angular width of the peak corresponds to the acyl chain tilt. In addition, superstructures (vi) and peptide geometries (helix maximum, vii) can be observed in some cases Eur Biophys J (2011) 40:417-436 419 Design of the b-type peptide helix motif In contrast to well established, commercially available peptide species, we concentrate our studies on a recently reported peptide model that was designed for spanning the membrane hydrophobic core and anchoring at the membrane/water interface. Therefore, this motif holds a tunable, conformationally stable peptide structure that is open for multiple functionalization ). The D,L alternation of the peptide backbone concordant with a double-helical b-type shape is related to natural peptide antibiotics such as gramicidin A (Kelkar and Chattopadhyay 2007) or feglymycin (Bunkóczi et al. 2005) . In addition, the studied sequences are unique on their own due to their almost purely aromatic side-chain composition (Alexopoulos et al. 2004; Küsel et al. 2007 ). On the basis of the crystallographically elucidated watersoluble nonameric sequence H-(Tyr-Tyr) 4 -Lys-OH (note: underlined amino acids indicate the D-conformer), the homodimeric dodecamer H-(Phe-Tyr) 5 -Trp-Trp-OH (1) (Fig. 2a, b ) and the covalently linked hairpin species H-(Phe-Tyr) 5 -Trp-Trp-Gly-Lys-Pro-Gly-(Phe-Tyr) 5 Trp-Trp-OH (2) and H-Lys(NBD)-Tyr-(Phe-Tyr) 4 -Trp-Trp-Gly-Lys-([C 2 H 4 O] 2 -CH 2 -CO-gcgtgg-Lys-Lys)-Pro-Gly-(Phe-Tyr) 5 -Trp-Trp-OH (3) (Fig. 2a) (note: gcgtgg represents the nucleobase sequence of aminoethylglycine peptide nucleic acid monomers) were derived via synthetic modification applying solid-phase peptide synthesis (SPPS) and loop design . The transmembrane helices of the latter structures (1-3) are well adapted to the membrane environment as optimized by sequence variation in length and composition (Küsel et al. 2007 ). Through the execution of design studies of the peptide substrates, the preservation of the b 5.6 helical structure ( Fig. 2b ; the superscript denotes the periodicity of the helix), as found for the homodimeric species 1, was maintained for the final hairpin prototype 2 via constant comparison of the respective circular dichroism data resulting from several tested constructs . As there has been strong evidence for a conformational and monomer-dimer equilibrium of the homodimeric structure 1 involving antiparallel oriented b 5.6 double and b 6.3 single helices (Fig. 2c , i-iii) (Küsel et al. 2007 ), a covalent linkage of both strands seemed to be mandatory in order to circumvent the described equilibria. This goal was fulfilled through the synthesis of the peptide hairpin 2 (Fig. 2c , iv) allowing for further studies that address peptide oligomerization within lipid model membrane complexes. The helix assembly is, thereby, driven by molecular recognition of the peptide nucleic acid (PNA) moieties at the membrane's exterior (Fig. 2c, v-vi; . The attachment of fluorescence probes and peptide nucleic acid (PNA) recognition moieties led to functional constructs, such as compound 3, that have already been tested for their dynamic dimerization within lipid bilayer structures via Förster resonance energy transfer (FRET) assays ). a b c Fig. 2 Molecular representation of the different D,L-alternating peptide types (a) ranging from the homodimeric species 1 to the designed hairpin 2 to the functionalized construct 3. Furthermore, the molecular structure of the homodimer 1 (b) (Küsel et al. 2007 ) as well as proposed equilibria (c) are sketched, either for homodimeric structures (i)-(iii) or the PNA-equipped transmembrane domains of 3 (v)-(vi) that undergo oligomerization driven by molecular recognition at the membrane's exterior Here, we probe the membrane reconstitution of the three presented kinds of b-helical species via GID and reciprocal space mapping (RSM) and evaluate their influence on the lateral lipid packing by addressing the lipid chain correlation peak. In the respective studies, parameters such as the sample composition, the relative humidity (RH), and the peptide-to-lipid (P/L) ratio are varied. The obtained results can be explained by a qualitative model that predominantly takes into account a hydrophobic mismatch situation as well as a lateral packing and ordering of the annular lipid shell that is correlated to the extent of peptide-lipid contact surface. Furthermore, we analyze the differentially (double) iodine-labeled hairpin species of type 2 via anomalous and in-house reflectivity elucidating their positioning with respect to the membrane normal (z axis) and confirming their transmembrane orientation. These experiments created the basis for a functional FRET assay that was reported elsewhere . Solid supported stacks of typically 1,000 aligned DLPC bilayers with a cholesterol (Chol) content of 5% and differing amounts of the respective peptide species were prepared from stock solutions following procedures described in the literature (Seul and Sammon 1990) . Polished and cleaned (sonication in MeOH and ultrapure water, 15 min each) Si wafers with h100i orientation and a thickness of 625 lm (Silchem, Freiberg, Germany) were used as substrates. Stock solutions of DLPC and Chol in chloroform were prepared at concentrations of 40 and 3 mg/ml, respectively. Peptide stocks were composed of MeOH/DCM/EtOH 4/3/3 (v/v/v) at 6 mg/ml concentration. For P/L ratios between 1/10 and 1/50 in a definite volume of 80 or 150 ll, mixtures of stock solutions were spread onto cleaned, horizontally mounted silicon substrates with dimensions of 10 9 15 mm (80 ll, beamline experiments) or 15 9 25 mm (150 ll, in-house experiments), respectively. The coated Si wafers were covered with a watch glass, and the solvent was carefully evaporated in a flowbox overnight to prevent film rupture and fast dewetting. Subsequently, reduced pressure was applied for an additional 12 h. The resulting film-covered substrates were stored at 4°C until use. For measurements, the prewarmed (40°C, 1 h) and rehydrated (saturated water vapor atmosphere) samples were placed in home-built sample cells with Teflon sealing and either PE foil or kapton windows. A setup was applied at which the RH could be adjusted by PID control, remaining stable within 0.1% (Aeffner et al. 2009 ). The temperature (T) of the sample chamber, the water reservoir, and the pipings was controlled as well. The RH/T sensors and mass flow controllers were interfaced with the diffractometer controls (in-house and beamline), enabling the usage of long and fully automated scan macros including RH and T as parameters. Unless otherwise stated, GID experiments were carried out at RH = 94%, and the reflectivity experiments were performed at RH = 90%, both at 20°C. The chambers were mounted to the respective goniometers with the sample oriented either horizontally (beamline) or vertically (in-house) depending on the diffractometer setup. The X-ray beam enters and exits the chamber through kapton (in-house) or PE foil (beamline) windows. GID and anomalous reflectivity experiments were performed with the insertion device 01 (ID01) undulator beamline at the European Synchrotron Radiation Facility (ESRF, Grenoble, France), while in-house reflectivity measurements were carried out with a home-built diffractometer. In GID mode the sample was tilted at a fixed angle of incidence close to the critical angle of total external reflection a c in order to optimize the lateral scattering intensity and to minimize background scattering caused by the substrate. The vertical scattering depth along z is, thereby, tuned by the angle of beam incidence a i and the angle of the scattered beam a f (compare to Fig. 3) . GID experiments were performed at 17 keV radiation using a Peltier cooled 2D CCD detector (Princeton Scientific Instruments, Princeton, NJ, USA) providing a resolution of 1,340 9 1,300 pixels (pixel size: 48 9 49 lm 2 ; active area: 64.3 9 63.7 mm 2 ) that was mounted at approximately 19.5 cm from the sample. The beam was cut by the entrance slits to 0.7 9 0.7 mm 2 in front of the sample; no beam restrictions were set on the detector side. Three different types of measurements were performed: 1. After placing the sample roughly horizontally and adjusting the specular axis to the center of the CCD, four images were taken along the q z direction under stepwise increase of the angle of beam incidence (Da i = 0.45°) enabling the calculation of the actual a i as well as the exact sample detector distance and allowing for scanning the low q k space for peaks resulting from superstructures (Fig. 3 ). 2. Adjustment of the specular axis at one side of the CCD and coverage by a lead stripe facilitated scanning the Eur Biophys J (2011) 40:417-436 421 reciprocal space far from the specular axis yielding combinations of exposures that provide a partial overlap. With this data collection an overview of the reciprocal space was achieved in which peaks representing intrahelical distances might occur (data not shown). 3. Quantitative images including the nonattenuated specular axis (q k = 0) on one side of the CCD plate up to q k = 2.5 Å -1 were taken with accumulated exposure series of 20-600 images applying single exposure times of 0.3-5.0 s. In analysis of the chain correlation peak using the same geometry, images were taken with exposure times of 20-30 s under attenuation of the primary beam and the specular axis. For the anomalous scattering experiments (ID01 beamline, ESRF) the polychromatic X-ray beam was monochromatized downstream from the undulator by a double crystal Si(111) monochromator with an energy resolution of DE = 1 eV yielding Dk/k B 10 -4 . The beam was cut to a size comparable to the GID experiments by the entrance slits directly in front of the sample. The reflected beam was recorded by two wide-opened slits of 4.0 9 4.0 mm 2 and 3.0 9 3.0 mm 2 . For line scanning under specular conditions, an avalanche photodiode (APD) was used as a point detector and placed directly behind the last slits reducing parasitic air scattering. The sample-to-detector distance was 28 cm. In order to reduce sample degradation (beam damage) during motor movement and in between scans, especially when applying low X-ray energies, a ''fast shutter'' was implemented in front of the entrance slits/absorber box and synchronized with the detector (Giewekemeyer and Salditt 2007) . Therefore, no automated absorbers could be used during reflectivity or offset scans; absorbers had to be calibrated in advance of data collection. A monitor, mounted directly in front of the sample, was used to control the beam intensity. To further reduce dose and avoid beam damage, the reflectivity curves and offset scans were recorded only in the vicinity of the Bragg reflections at calculated positions. In resonant reflectivity experiments, the scattering power of iodine labels was used to retrieve selective structural information of the iodine positioning with respect to the membrane normal by varying the incidence of the X-ray radiation close to the iodine L III absorption edge around E = 4.5575 keV (Gullikson 1995 (Gullikson -2008 . The scattering length becomes explicitly dependent on the X-ray energy E near the absorption edge, according to where f 0 is the number of electrons of the ion (nonresonant term) and f 0 and if 00 are the real and imaginary corrections of the scattering factor, respectively (Evans and Pettifer 2001 ). Both f 0 and f 00 , connected by the Kramers-Kronig relations, strongly vary in the vicinity of an absorption edge and can be measured (f 00 ) or obtained from databases. At an absorption edge the absolute values of f 0 and f 00 are the highest with respect to nonresonant energy regions, where f & f 0 . Thus, in a resonant diffraction experiment it is very important to determine the energy value of the absorption edge accurately to a resolution of about 1 eV. It is known that the scattering factors f 0 and f 00 of a chemical species within a sample, here iodine, depend very sensitively on its environment. Since there are many influences on these values, their accurate calculation by theoretical consideration is not feasible. Therefore, it is the most reliable strategy to measure f 00 and calculate f 0 during the experimental setup (Als-Nielsen and McMorow 2001). Experimentally, it is possible to access f 00 by absorption or fluorescence measurements. In order to Fig. 3 GID scattering geometry (left) and exemplary results from low q k scanning at different angles of a i in search of ''superstructure peaks'' (right) also required for determination of parameters concerning the scattering geometry. As the angle of incidence increases, the specular beam reflex (sb) moves along the q z axis. The primary beam (pb) is located at the origin of the q z axis, here shadowed by the sample horizon mimic the hydrophobic environment within a lipid membrane where peptides should be placed, absorption measurements were performed using quartz capillaries (700 lm diameter) filled with 0.5 M iodoform dissolved in chloroform. Instead of measuring f 00 and calculating f 0 the distinct iodine L III edge was found by varying the beam energy about ±0.025 keV around the tabulated L III edge by means of modulating the undulator gap and detecting the absorption. The iodine L III edge was found at E = 4.5578 keV. For the contrast variation in the anomalous scattering experiments, the reflectivity measurements were performed at energies of 4.5578 and 5.8000 keV. In-house X-ray reflectivity (nonisomorphic samples) For comparison and because of higher resolution (less absorption at higher energies) and enhanced contrast variation [see tabulated atomic scattering factor f 0 of iodine (Gullikson 1995 (Gullikson -2008 ], in-house reflectivity experiments have been performed with the same sample composition as applied in synchrotron experiments. As the beam energy cannot be varied at in-house experimental stations, nonisomorphic samples, with respect to the iodine labels, were used for the home-built diffractometer. The in-house experimental station is a stationary H/2H diffractometer with a Seifert long fine focus X-ray tube holding a Cu anode (U = 35 kV, I = 40 mA) and a Cyberstar point detector. The X-ray beam is monochromatized and parallelized by a Göbel mirror selecting the Cu-K a line (E = 8.048 keV, k = 1.541 Å ). The primary beam intensity is on the order of 10 9 counts per second (cps). Automatized absorbers are used to avoid detector saturation for 2H close to zero and at first Bragg reflexes. The sample is mounted to a Huber goniometer for which three linear stages for x, y, and z translation are used to place the sample in the center of rotation. The sample-todetector distance is 400 mm. The lateral dimensions of the primary beam (1 9 5 mm 2 ) are defined by two slits in the horizontal and the vertical direction (S1). A vertical slit behind the sample (S2) screens scattering that does not stem from the sample; additional slits in front of the detector (S3) define the resolution of the instrument. All scans are performed at slit widths of 2 mm (S2) and 0.5 mm (S3). For the given sample-detector distance, the latter yields a resolution of 0.07°or 0.01 Å -1 . With these settings, the profile of the primary beam typically has a full width at half maximum (FWHM) of 0.2°. The diffractometer motor and the detector controls as well as the monitor counter readout are accomplished by the SPEC software (Certified Scientific Software, Cambridge, MA, USA). Two-dimensional resolved maps of the q space were derived from the CCD images via data treatment that includes (1) positioning of the beam center, (2) polarization correction, (3) elimination of dead pixels, and (4) scaling. Data treatment was performed by applying a self-written MATLAB (The MathWorks, Natick, MA, USA) tool (Weinhausen 2010) . This tool further allows for sectioning through the reciprocal space at a definite angle (/) between q z and q k leading to extraction of the intensity courses along these sections via averaging over an angular ROI five pixels wide. The respective intensity profiles were linearized. Lorentzians with linear backgrounds were fitted to the extracted functions: where x is the half width at half maximum, q 0 is the peak center, I 0 is the maximum of the Lorentzian without the linear offset, m is the slope of the linear background, and b is the constant offset of the Lorentzian. The correlation length n = 1/x and the average chain distance in real space were calculated: For analysis of the resulting reflectivities, a self-written MATLAB software tool was applied ). The reflectivity data were plotted as a function of the vertical momentum transfer q z after subtraction of the diffuse scattering (offset scan) and illumination correction. The electron density profiles were calculated by applying an empirical Fourier Synthesis (FS) scheme, exploiting the area under Bragg peak intensities I n , as it is used for such multilamellar lipid membranes (Münster et al. 2000; Spaar et al. 2004; Wu et al. 1995) . In simple terms the onedimensional electron density profile q(z) was obtained via Fourier synthesis method from the integrated peak intensities via Gaussian fitting, applying the Lorentz correction factor 1/q z 2 and phases -, -, ?, -, -, -, -in accordance with the number of observed Bragg reflexes. The phases t n are reduced to positive/negative signs due to the point symmetry and were reconstructed in accordance with the 1D swelling method approach using Eq. 3 to obtain the continuous form factor F(q z ) with its relative amplitude |F(q z,n )| and phase t n : Eur Biophys J (2011) 40:417-436 423 The scattering vectors are represented by q z and q z,n , and d is the membrane periodicity (Aeffner et al. 2009 ). The phases have been extracted by the 1D swelling method of pure DLPC-lipid samples (Schneggenburger et al. 2009 ). The electron density profile q(z) normal to the interface is computed by N 0 Fourier coefficients f n = I n q z (Eq. 4): The factor n in front of the Bragg peak intensity I n follows from an empirical correction factor to calculate the nth Fourier coefficient. The curves have been normalized by scaling higher-order Bragg peaks to the area under the first Bragg peak. General aspects of reflectivity experiments are discussed elsewhere (Salditt et al. 2002) . Insertion and lateral lipid response by reciprocal space mapping The homodimeric representative of the applied peptide motif H-(Tyr-Tyr) 4 -Lys-OH has already been shown to enable intermolecular interaction via its phenolic sidechain pattern within the aqueous phase (Alexopoulos et al. 2004 ). The homodimer species 2 adopts a membrane spanning orientation in DLPC bilayers (Küsel et al. 2007) and likewise tends to aggregation (Schneggenburger et al. 2009 ). Therefore, the GID approach was undertaken to evaluate if the latter observation of a membrane insertion and transmembrane alignment is likewise true for the designed hairpin species (2) and the recognition system (3) (Fig. 2) . Furthermore, it was our aim to screen the samples for any evidence of lateral peptide interaction, i.e., the formation of higher-order structures (low q k , q z ), as sometimes indicated by superstructure peaks (Constantin et al. 2007 ) as well as for regular intrahelical distances that can be revealed from the occurence of so-called helix peaks (Spaar et al. 2004 ). Highly aligned membrane stacks of either a pure lipid matrix (DLPC/Chol) or differentially composed peptide/ lipid compositions (Fig. 2) were analyzed. The lipid phase of each sample contained 5% cholesterol to enhance the surface mosaicity and reduce bilayer fluctuations allowing for pronounced Bragg reflections and correspondingly higher resolution in the reflectivity experiments (Chen and Rand 1997; Mouritsen and Zuckermann 2004) . The particular sample composition, the variation of parameters, and the obtained data are shown in Table 1 . The direct comparison between undisturbed DLPC/Chol membranes and membranes with either homodimeric or hairpin structures was carried out via analysis of samples 1-3 (Table 1 ). The homodimer 4 [=H-(Phe(4I)-Tyr-(Phe-Tyr) 4 -Trp-Trp-OH] was embedded at a P/L ratio of 1/10 while the hairpin system 5 [=H-(Phe-Tyr) 3 -Phe(4I)-Tyr-Phe-Tyr-Trp-Trp-Gly-Lys-Pro-Gly-(Phe-Tyr) 2 -Phe(4I)-(Tyr-Phe) 2 -Tyr-Trp-Trp-OH] was embedded at P/L = 1/20, both leading to a peptide-to-helix ratio of 1/20. The equimolar mixture of species 3 and 6 [= H-Lys(TAMRA)-Tyr-(Phe-Tyr) 4 -Trp-Trp-Gly-Lys-([C 2 H 4 O] 2 -CH 2 -CO-gcgtgg-Lys-Lys)-Pro-Gly-(Phe-Tyr) 5 -Trp-Trp-OH] representing the recognition system could at least be incorporated into multilamellar DLPC stacks at P/L = 1/40 avoiding precipitation. A second hairpin sample (sample 4) including compound 5 was prepared at a P/L ratio of 1/40. Therefore, samples 3 and 4 were analyzed to account for the concentration dependency of the observed lipid response. In order to evaluate the applied methodology and due to the fact that the recognition system (3/6) should be most amenable to changes in RH because of its polar PNA moieties facing the water layers, a series of experiments was performed at different RHs ranging from 94% to 25%. All other samples were analyzed at RH = 94%. In searching for potential superstructure peaks in the low q z range and the helix maximum at higher q z , first, scans of the small q k space along the q z axis were performed in GID mode under variation of the angle of beam incidence (Fig. 3) . These scans were followed by addressing the far q k space under attenuation of the specular axis, yielding a combination of exposures that provide a partial overlap (data not shown). As neither reflections resulting from superstructures nor from helix peaks could be observed for the reconstituted peptide species 3-6, quantitative images (accumulations) of the q space were taken, focusing on the lipid chain correlation peak (cc peak). In analyzing the lateral lipid bilayer response of different samples upon peptide insertion, the cc peak gives indirect information on the peptide reconstitution and peptide/lipid interactions. Attenuation of the specular axis at low angles of incidence, e.g., a i & 0.4°, avoiding overexposure of the detector provided sufficient access to the cc peak at low q z (Fig. 4a) . Taking images of the reciprocal space at higher a i (e.g., a i & 2.5°) without attenuation led to unacceptably high sample horizons (shadowed low q z ) that were not applicable for analysis of the cc peak. Corrections of the scattering distribution (dark image, polarization), normalization (monitor signal, counting time), and coordinate transformation to (q k , q z ) coordinates based on the calculated sample-to-detector distance led to the 2D intensity distribution in the reciprocal space (Fig. 4a) . Intensity profiles (Fig. 4b) of the cc peak were extracted by interpolation along radial sections through the reciprocal space (black lines in Fig. 4a ) for different angles / corresponding to the tilting of lipid acyl chains in real space (Fig. 4a, inset) . The root mean square deviation of the intensity was estimated via extraction of five neighboring profiles within a distance of D/ = 0.2°( 1-2 pixels). A Lorentzian function (Eq. 1) was fitted to the intensity traces via a nonlinear least squares method using a trust region. The Lorentzian function fits the intensity profiles [I(/)] very well for all /. For calculation of the average lipid acyl chain distance a (Eq. 2) and the correlation length n only intensity traces at small / were used in order to minimize the q z component. Depending on the angle of beam incidence the sample horizon was partially shadowing the intensity profiles at / \ 8°for some of the samples. Therefore, the Lorentzian fits for / = 8° (Fig. 5a) were used for the calculations of a and n (Fig. 5b) . In addition to the acyl chain distance a and the correlation length n, the extension of the cc peak in D/ gives information about the tilt homogeneity of the acyl chains (Fig. 4a, inset) . Quantitative fits of the profiles along this dimension, as performed for pure lipids (Weinhausen 2010), were not applicable for the present data for reasons of the signal-tonoise ratio. First, it is important to keep in mind that the next neighbor distance a between lipid acyl chains as well as the In addition to the peptide component, the P/L ratio and the relative humidity ( Table 1 and visualized in Fig. 5b . As we were interested in the changes of the lipid bilayer structure upon peptide insertion we had to exclude influences on the ordering and packing of lipids exerted by other parameters. These are, on the one hand, the constitution of the lipid species in terms of chain length and saturation (Seelig and Seelig 1977) , chain branching (Perly et al. 1985) , and headgroup structure (Cullis et al. 1986) , and on the other hand, the lipid phase state (Lafleur et al. 1990b; Perly et al. 1985; Sternin et al. 1988 ) and the cholesterol content (Dufourc et al. 1980; Oldfield et al. 1978; Stockton and Smith 1976) . The use of only one lipid component with a fixed cholesterol content eliminates the mentioned impacts. The uniformity of physical parameters, namely the temperature (Davis et al. 1980 ) and the relative humidity, was assured by the controlled sample environment. Therefore, the experiments carried out with the isomorphic sample 3/6 (equimolar)/DLPC/Chol (1/38/2) at different values for RH (samples 5-9) served as a control for the suitability of the experimental protocol as well as the scattering setup and the data treatment. Furthermore, it should be tested whether a variation of RH is detectable by changes in a and n and if these changes are within an appropriate regime and follow expected trends. In general, lowering the RH promotes the gel phase character of the lipid membranes causing a reduced fluidity and increased inter-bilayer potentials (Ho et al. 1995; Long and Hruska 1970) . Upon reducing the water content of the system, the fraction of gauche bonds within the lipid acyl chains is diminished and leads to lipid chain stretching. As a consequence, the lipid chain alignment and ordering (Cevc and Marsh 1985) as well as a lateral compression or packing (Bryant et al. 2001; Chen and Hung 1996) are enhanced. Such behavior was confirmed by the results obtained from measurements for samples 5-9. In line with the literature, a lowering of the acyl chain distance of about Da max = 0.097 ± 0.006 Å and an increase in the correlation length of Dn max = 0.8 ± 0.2 Å were observed when comparing the results for RH = 94% to RH = 25% and to RH = 60%, respectively. The revealed effect that at low water content with RHs between 40 and 25%, the correlation length n seems to reach saturation and even slightly decreases for sample 9, can be associated with the steric demands exerted by the peptide and cholesterol inclusions (Fig. 5b) . Steric demands of the relatively rigid peptide helices become more critical in the more condensed phases than in the fluid phase (Marčelja 1974) . This condition of so-called intermediate fluidity (Oldfield and Chapman 1972) describes the fact that the interaction between inclusions such as cholesterol or peptides cannot be as strong as interactions among ordered all-trans lipid chains (Marčelja 1974 ). We did not observe the phenomenon of an interchain hydration at high RH that is concomitant with an increased lipid chain-chain distance and results from defect structures caused by inserted peptides (Ho et al. 1995) . As a general trend upon reconstitution of the different peptide species, we observed a significant decrease in the acyl chain distance a while the values for the lipid chain correlation length n increased (Fig. 5b) . One may initially expect the opposite effect, i.e., an increase in a and a decrease in n. This could appear conclusive, since the occupation of the lipid bilayer area by peptide inclusions would lead to additional and longer trans-helix chain-chain distances for so-called annular lipids, which are in direct contact with the peptide helix. Furthermore, the absent chain-chain correlation between those lipids could reduce n. From the experimental results it is obvious that this is not the case, or at least other phenomena are outweighing such assumed effects leading to an increase in lipid packing (density) and chain ordering. Changes in the lateral bilayer structure upon peptide insertion are mostly discussed in theoretical studies and appear contradictory with respect to their conclusions. The intensity profiles (Fig. 4b) at a certain constant / were fitted to Lorentzian lineshapes (Eq. 2). The resulting fits at / = 8°(a) were applied to calculate the mean lipid acyl chain distance a and the acyl chain correlation length n (b) Already in early work, motivated by fluorescence data, a distinction is made between lipids not in contact with protein or peptide surface and lipids 'coupled' to the reconstituted protein, which are, thereby, disturbed in their interaction (Träuble and Overath 1973) . These so-called annular lipids were shown not to undergo lipid phase transitions. The postulated protein-derived disturbance was calculated to extend up to the third lipid neighbor of a protein inclusion (Marčelja 1976) . Addressing the structure of annular lipids in greater detail by theoretical molecular field approximations revealed an increased ordering of the lipid chains in proximity to membrane-incorporated structures within the fluid lipid phase (Marčelja 1974) . With increasing fraction of ''foreign molecules'' on the phospholipid bilayer, the membrane order parameter was reported to increase and the dependency of the order parameter on temperature to become less pronounced. We will denote this scenario as the (annular) lipid ordering scenario. In the lipid condensed phase, this effect was described to be the opposite, which we denote as the lipid disordering scenario. Other authors claim that rigid inclusions such as cholesterol might cause straightening and ordering of lipid chains, while the fluid-like surface of embedded proteins does not influence the motional freedom of the lipid chains (Lafleur et al. 1990a) . Molecular dynamics simulations of three different peptide a helices, published at the same time, likewise revealed an ordering of the lipid chains close to the peptide helices for all peptide species (Edholm and Johansson 1987) . Even if the overall effect was assessed as ''not drastic,'' the annular lipid ordering could be estimated as ''much less pronounced'' for lipids with bulky side chains. More recent theoretical considerations applying statistical mechanical integral equation theories (Lagüe et al. 1998 ) predict an expansion of the area per lipid for the annular shell upon peptide insertion into different phosphatidylcholine (PC) bilayers. In these studies, peptides are treated as soft cylinders (Lagüe et al. 2001 ). The authors postulate effective long range repulsion between lipids and peptides due to the formation of a lipid depletion layer around a protein leading to an increased cross-sectional area per lipid molecule. According to this view, which also supports the lipid disordering scenario for annular lipids, a lipid molecule in close contact with the peptide must significantly reduce its order. Effective lipid-protein repulsion would arise due to its entropic disadvantage. These theoretical results were further supported by MD simulations of a hydrated 1,2-diphytanoyl-sn-glycero-3-phosphocholine (DPhyPC) bilayer containing an a-helical peptide bundle of four transmembrane domains (Husslein et al. 1998 ). Comparable to the results of Lagüe et al., Husslein and coworkers revealed an increase in the area per lipid from 7.46 nm 2 for the unperturbed bilayer to 8.50 nm 2 in the vicinity of the peptide species. Finally, for the sake of completeness, we note that a pronounced annular lipid effect is not reported by all studies. Using spin labeling techniques, Marsh and Horváth found order parameters of protein-associated lipid chains to be very similar to those in the bulk liquid-crystalline phase regions (Marsh and Horváth 1998) . Considering these opposing scenarios, it is interesting that we observed a very clear indication of the lipid ordering scenario. The decrease in the average lipid chain distance a as well as an increase in the lipid chain correlation length n for the peptides studied here can be ascribed to a reduced number of lipid gauche rotamers (Fig. 6) . For different molecular systems imposing different molecular boundary conditions, the second scenario could be valid instead. It should be taken into consideration that for the sake of simplicity, the theories supporting either one of the scenarios neglect the individual chemical properties of the incorporated molecular species that are the bases for interaction at the peptide/lipid interface. However, the particular molecular properties of different species may be the decisive parameters that determine whether a peptide/lipid interaction is attractive or repulsive. For the presented peptide species 1-6 these parameters are (1) the unique b-helical secondary structure caused by the D,L-alternation and bulkiness of the peptide side chains, (2) the exclusiveness of the side-chain composition with almost entirely aromatic amino acid residues, and (3) the presence of flanking interfacial anchoring moieties. Our findings are also in line with literature studies that report higher order parameters for hydrocarbon chains of lipids adjacent to peptide channels composed of aromatic amino Fig. 6 Schematic representation of the conclusions drawn from GID results, suggesting a positive hydrophobic mismatch and, most likely, an enhanced highly ordered packing of the annular lipid shell. These effects may be attributed to the specific properties of the peptide species, i.e., their sequences, side-chain orientations, and secondary structures Eur Biophys J (2011) 40:417-436 427 acids (Chiu et al. 1999 ). This should be particularly pronounced in the case of tryptophan-containing peptides, such as species 1-9, as tryptophan is known as an amino acid with high preference for lateral interaction (Adamian and Liang 2001; Ridder et al. 2005) . The planarity and hydrophobicity/amphipathicity of the aromatic side chains have already been shown to provide potential for a perfect alignment and interaction with the lipid bilayer core via intercalation between the lipid chains (Hite et al. 2008; Palsdottir and Hunte 2004) . Hydrophobic interaction might further direct lipid chains to approach closer contact to incorporated peptide species via lowering entropic costs for acyl chain deformation. This lipid/peptide interaction might play a predominant role for b-helical peptide species due to the fact that all peptide side chains are pinpointing radiantly outside the helix . Proper vertical alignment for the reconstituted peptide in the bilayer was concluded already in a previous reflectivity study exhibiting up to six lamellar orders nearly independent of the applied peptide concentration (Küsel et al. 2007 ). Owing to the nature of bilayer elasticity, the effects of lateral packing cannot be discussed without addressing the vertical structure and density profile, which are closely interrelated. Namely, the effects of hydrophobic matching and interfacial anchoring (Killian and Nyholm 2006) interfere with the lateral forces. Hydrophobic matching and interfacial anchoring are especially important for the presented peptide species, as all of them hold two flanking tryptophan residues. The magnitude of hydrophobic mismatch is considered to be a function of the distance between interfacial anchoring residues, here tryptophans, and not the overall peptide length (De Planque and Killian 2003) . In this respect, tryptophan residues are known to provide the highest tendency for localization at the interface between the lipid bilayer core and the polar lipid headgroup. Hydrogen bonding and electrostatic interactions are discussed to account for the interfacial anchoring and z positioning of membrane-incorporated peptides (De Planque and Killian 2003; Doux and Killian 2010) . A stretching of the lipid bilayer in the z direction caused by hydrophobic mismatch (Killian and Nyholm 2006) may alone result in pronounced lateral effects on annular lipids. Huang and coworkers observed a stretching of a DLPC lipid bilayer upon incorporation of gramicidin A that likewise adopts b-helical structures, similar to the D,L-alternating peptides applied in the presented approach (Harroun et al. 1999) . In previous studies of the homodimeric species 1, comparable effects were indicated by a thickening of the lipid bilayers of 1.4 Å (DMPC = 1,2-dimyristoyl-sn-glycero-3-phosphocholine) and 4.0 Å (DLPC), respectively (Küsel et al. 2007 ). Finally, the reflectivity experiments of the designed peptide hairpin structures 2, 5, 7, 8, and 9 unambiguously prove a positive mismatch situation, also for the hairpin structures, as presented in Table 2 . Such mismatching promotes stretching of the lipid chains and reduction in gauche conformations in the lipid acyl chains. The mismatch situation as well as the described lateral effects may be directly coupled or, alternatively, may independently contribute to the observed changes in a and n upon peptide insertion, as addressed in the following section. Differences in peptide species: the cross-sectional area and the characteristics of peptide/lipid contacts If the ordering of the lipid chains is solely a function of hydrophobic mismatch, any of the studied peptide species should yield the same values for a and n due to sequence similarity. Obviously, this is not the case (Fig. 5b) . For the sample solely composed of DLPC and cholesterol (sample 1) the acyl chain distance is a = 4.94 ± 0.01 Å . The shortening of the acyl chain distance a appears more pronounced for the homodimeric species 4 (sample 2) at a helix-to-lipid ratio of 1/20 (P/L = 1/10) but is even stronger for the hairpin species 5 at the same helix-to-lipid ratio (P/L = 1/20). Considering the recognition system 3/6 (equimolar) at RH = 94% (sample 5), the shortening of a is less distinct but can also be assigned to a lower helix-to-lipid ratio of 1/40 (P/L = 1/20). The concentration effect can be factored out by comparing different samples (3 and 4) of the hairpin species at P/L = 1/20 and P/L = 1/40, respectively, showing that higher peptide concentration yields shorter lipid chain distances. It can further be concluded that the recognition system 3/6 (equimolar ratio) at RH = 94% (sample 5) has less impact on a than the respective hairpin peptide 5 (sample 4) at identical P/L = 1/40. The values for the lipid chain correlation length n follow an almost opposing trend as that observed for a. Peptide incorporation generally leads to an increase in n. In principle, the homodimer (4), hairpin (5), and recognition system (3/6) structures are designed to adopt an identical secondary structure with respect to their TMDs. In the case of the hairpin structure 2, due to a decline in helix propensity compared to the homodimeric species 1 as revealed by CD spectroscopy , it was assumed that the hairpin structure exhibits steric clashes in the reverse-turn region leading to a less perfect alignment of the double strands and a broadening of the helix diameter. Therefore, the self-associated homodimer 4 should provide a closer alignment enabled by hydrogen bonding of the two antiparallel-oriented strands. The PNA recognition system 3/6 leading to an aggregation of two of the double helices in a close contact state would yield more than a doubled lateral occupied area within the lipid bilayer (Fig. 2c) . Under these circumstances, the lateral dimensions of the peptide species are supposed to increase in the order: homodimer (4) ? hairpin (5) ? recognition system (3/6). As the differences in the values for a and n when comparing the different peptide species cannot be solely attributed to a hydrophobic mismatch situation (see above), lateral peptide parameters such as the peptide cross-sectional area or the peptide/lipid contact surface are assumed to be the primary cause for the observed differences. In line with the lipid disordering scenario discussed above, larger peptide cross-sectional areas and peptide lipid contact surfaces would probably cause higher values for the lipid acyl chain distance a and smaller values for the lipid correlation length n. This is true if peptide/lipid interactions are neglected or a repulsive potential between lipids and peptides is assumed (Lagüe et al. 2001) . For the present sample sequence: homodimer (4) ? hairpin (5) ? recognition system (3/6), this would imply an increase in a and a decrease in n. The fact that the observed results point to the opposite may be taken as evidence for attractive peptide lipid interactions. Due to the sequence identity of the antiparallel TMD strands of all peptide species, and therefore, the constitutional similarity of the postulated attractive peptide/lipid interactions, only the peptide secondary structure, i.e., the lateral sizes of the peptides, and the size of the lipid/peptide interfacial plane should affect a and n. Therefore, a monotonous decrease in the lipid chain distance would be expected for samples 1-3, which is in line with the obtained results (Fig. 5b) . On the contrary, a monotonous increase in the chain correlation length was not observed. The value for n of the homodimeric species 4 (sample 2) appears higher than that for the hairpin 5 (sample 3). This might be explained by the fact that oligomer 4 is known to experience a monomer-dimer equilibrium (Küsel et al. 2007 ) including a structural change from a membrane-spanning b 5.6 helix to two reasonable broader and shorter b 6.3 single helices (Fig. 2) . This would actually change the peptide dimensions in the z direction as well as the molecular constitution of the helix surface and therefore cause different hydrophobic mismatch situations as well as changed lateral peptide/lipid interactions. In the case of the recognition system 3/6, a close contact state along the outer helix shells results from PNA pairing and TMD assembly . The adopted peptide complexes can be considered as one reconstituted peptide species surrounded by annular lipids (Fig. 2c) . The recognition process, therefore, diminishes the peptide/lipid contact area by shielding the inner helix flanks towards the lipid matrix. In this respect, the recognition system species 3/6 (sample 5) provides lower values for a and higher values for n than the hairpin species 5 at the same P/L ratio. Peptide positioning and orientation from reflectivity using iodine labels The membrane response in terms of changes in the lipid correlation length and lipid chain distance addressed above can neither reveal the insertion depth of a particular peptide side chain nor does it allow distinguishing between a membrane-spanning state or insertion perpendicular to the membrane normal. FRET experiments (parallax analysis) are well suited to studying membrane proteins in vivo but are often limited in resolution caused by large Förster radii of the fluorescence probes. In contrast, X-ray reflectivity in combination with contrast variation by heavy atom labeling can provide structural constraints down to subnanometer resolution for fluid or condensed peptide/lipid samples. For a fast in-house approach, the analysis of the electron density difference is a unique method to obtain highly resolved information about the localization of a certain structural element within a lipid bilayer (Fig. 7) . As shown in previous studies for the double helical homodimer 1, this method is even capable of producing evidence of a transmembrane peptide orientation within the lipid bilayer via iodine labeling at several positions (Küsel et al. 2007 ). With the iodine-labeling approach, the alignment of the membrane-spanning helical hairpin formed by the severe acute respiratory syndrome (SARS) coronavirus E protein could be elucidated by localizing the Phe23 adjacent to the lipid headgroup region (Arbely et al. 2004; Khattari et al. 2006a, b) . Instead of determining the label position by a comparison of the deduced electron densities q(z) resulting from unlabeled and labeled samples, this can also be accomplished using an isomorphic sample and applying different photon energies (Khattari et al. 2005) . In this work, we used both approaches, the label replacement by a sample series for in-house reflectivity and the contrast variation via anomalous reflectivity for the synchrotron studies. In both cases, the commercially available building block Fmoc-Phe(4-I)-OH served as iodine label, substituting the native phenylalanine. Within the centrosymmetric electron density profiles deduced from reflectivity scans (Fig. 7b) , maxima correspond to the lipid headgroup regions, while the global minimum reflects the acyl-acyl contacts of opposing hydrocarbon chains. The adjacent water layers are represented by the side minima. For anomalous scattering, the electron density profiles obtained at the L III absorption edge (E = 4.5578 keV) were subtracted from the electron density profiles at E = 5.8000 keV to estimate the position of the iodine label with respect to the z direction, indicated by rises in the electron density difference curve (Fig. 7b , blue curve). In in-house experiments such curves result from subtraction of the electron density q(z) of the samples containing iodinated peptides from the sample with unlabeled peptide species. The detection of single iodine labels via difference analysis usually requires high peptide-to-lipid ratios in the range of P/L & 1/10 ( Arbely et al. 2004; Küsel et al. 2007) . With the synthesis of the novel double iodinated amino acid building block Fmoc-5,5-diiodoalyllglycine-OH, the use of an enhanced in-house reflectivity set up including substrate sizes of 15 9 25 mm and the addition of 5 mol% cholesterol to the lipid phase enabled recording up to seven lamellar orders from samples with a P/L = 1/50. The analysis of the electron density difference profiles then allowed for the determination of the iodine positions with high accuracy (Schneggenburger et al. 2009 ). Even for the Fmoc-Phe(4-I)-OH single label, P/L ratios of 1/50 turned out to be sufficient. Therefore, in-house reflectivity studies and anomalous reflectivity using synchrotron radiation were both performed at P/L = 1/50. Three differentially labeled peptide species, 5, 7, and 8 (Fig. 8) , were used to test for a transmembrane orientation. To avoid beam damage during anomalous reflectivity experiments, the respective curves were solely scanned around the Bragg reflections. Four to five lamellar orders (Fig. 9a) were recorded by specular and offset (longitudinal diffuse) scans. Unfortunately, the specular condition (angular alignment) was lost in some of the scans, so that the diffuse offset scans, which by nature are less sensitive to angular misalignment and which show strong lamellar reflections, were used for the Fourier synthesis analysis. Due to the difficulties associated with low energies around the I-L III edge and the intrinsically small anomalous difference signal, all experiments were repeated on an in-house diffractometer without any risk of radiation damage during long scanning and with most careful alignment procedures. The one-dimensional swelling method was used to yield the phases t n . In in-house experiments, six intensive equidistant Bragg peaks were obtained for the labeled (7) as well as the unlabeled (2) hairpin species (Fig. 9b ) due to the higher radiation energies (Cu-K a = 8.048 keV) and improved alignment. For peptides labeled with the Fmoc-Phe(4-I)-OH building block, species 2 served as reference structure. In case of hairpin 8, labeled with Fmoc-5,5-diiodoallylglycine-OH, analog 9 with the sequence H-(FY) 5 WW-Gallylglycine-PG-(FY) 5 WW-OH was applied as unlabeled species. As the electron density profiles of the lipid bilayer structures obtained from in-house experiments were more resolved, these data were used to estimate the respective changes in hydrophobic thickness (see below). The X-ray reflectivity curves (Fig. 9 ) and the corresponding electron density profiles (Fig. 10a, b) are exemplarily shown for compound 7 and for its unlabeled analog 2 (in-house), respectively. More data are shown in the Electronic Supplementary Material. The derived electron density difference curves for all studied constructs 5, 7, and 8 and unlabeled compounds 2 and 9 are depicted in Fig. 10b, c. a b water-layer water-layer water-layer Si-substrate Fig. 7 Schematic representation of the reflectivity scattering geometry for multilamellar P/L complexes (a) with highlighted iodine positions (red arrows) and deduced electron density profiles (right, b) . The comparison of electron density profiles originating from samples containing labeled and unlabeled peptide species or from scattering at different photon energies yields the difference curve (blue) The reflectivity curves and deduced electron density profiles reveal a stretching of the lipid bilayers upon peptide insertion. This was found for all examined peptide species (2, 5, 7, 8, 9) and is in line with the obtained GID results confirming a positive hydrophobic mismatch. The repeat spacing and the peak-to-peak distance (d pp , compare to Fig. 7) values (Table 2 ) evidence a small thickening of the lipid bilayer up to 1.4 Å , which is more pronounced for the mono-iodinated species 5 and 7. In case of the doublelabeled species 8 and its reference analog 9, this effect is inverted. In general, the electron contrast variation resulted in changes in the headgroup densities, which were too pronounced to be explained by the contrast variation itself and therefore indicate systematic errors. These probably stem from misalignment or drift when changing the photon energies. Thus, the two approaches of contrast variation (in-house and synchrotron) are not in quantitative agreement. For the in-house measurement, the effect of a decrease in electron density contrast of the side minima compared to the global minimum is observed in most peptide/lipid systems. The decrease can be explained by a beginning lipid disorder that evens the density profile by increased density in the water layer and/or an enlarged area per headgroup ; this effect may be more pronounced for a peptide carrying a bulky iodine label. However, we also cannot rule out systematic errors for the in-house measurements, and beyond the thickening effect, which appears to be robust, the small differences between curves with and without labels may impede an unambiguous localization of the iodine position. Finally, we must be cautious since iodine-labeling itself may influence peptide/lipid interactions. However, there is no obvious reason why attachment of the iodine labels at different positions should cause different peptide positioning, especially as the iodine labels have been shown to have only minor effects on the peptide secondary structure (Schneggenburger et al. 2009 ). Notwithstanding the problems mentioned above, we will now address the difference curves in view of a tentative interpretation of label position and peptide orientation in the next subsection, step by step for each peptide and labeled construct. Species 5 The center maximum of the density difference curve for the middle-labeled oligomer 5 (both intertwined strands) indicates that a significant fraction of the iodine labels are buried deeply in the hydrophobic core of the lipid bilayer (Fig. 10b, d) . In both data sets, in-house and synchrotron, the membrane thickness (peak-to-peak distance) is approximately 32 Å (see the Electronic Supplementary Material). The distribution of the central maximum of the difference curve appears relatively broad, especially for the in-house experiment (Fig. 10c) . The synchrotron data do not only illustrate a rise in intensity at the bilayer center but also at approximately z = ± 11.1 Å , which does not occur in the higher resolution in-house study (Fig. 10b, d) . These peak positions may still be commensurate with more than one orientation, for example either a membrane-spanning orientation of the b-hairpin (Fig. 11i) or an orientation of the peptide species parallel to the membrane surface (Fig. 11ii) at the bilayer center. The latter is, however, not very likely for energetic reasons. (1/47.5/2.5) from diffuse anomalous scattering (a) and inhouse reflectivity (c). In the inhouse experiments the samples containing the iodine-labeled compound 7 were compared to the samples lacking the iodine probes (compound 2, black curve). The electron density difference curves (blue, shifted for clarity) are depicted for all peptide constructs 5, 7, and 8 either studied by synchrotron radiation (b) or by in-house reflectivity (d). The curves derived from in-house measurements are multiplied by a factor of 3 for clarity; the headgroup region is highlighted (green area) Species 7 Anomalous reflectivity of terminally labeled peptide species 7 in peptide/lipid complexes (Fig. 10a) showed relatively weak minima in the density corresponding to the water layer. If this is not an artifact of the rather problematic synchrotron experiment, it could be possibly attributed to increased lipid fluctuations or beginning lipid disorder upon localization of the rather bulky iodine labels at the lipid headgroup region. The electron density difference curve shows only two symmetrically distributed maxima at approximately z = ± 15.0 Å indicating a localization of the iodine labels at the lipid headgroup region tending to the membrane interior. This observation would perfectly fit with a transmembrane orientation of the b-hairpin (Fig. 11i) , although an orientation parallel to the membrane surface at the headgroup region (Fig. 11iii ) cannot be ruled out. Taking into account the observations made for the peptide/lipid complex 7/DLPC/Chol (1/47.5/ 2.5), this appears rather unlikely, since the depth of the insertion should not depend on the position of the label along the helix. The distance between the maxima of the electron density difference curve, associated with the distance of the two iodine labels is about 32 Å (see the Electronic Supplementary Material). This is in good agreement with the expectation based on a peptide length of 26 Å and the additional spacing of the Phe(4-I) side chains attached at both ends of the double helix and pointing to opposite directions (Schneggenburger et al. 2009 ). The electron density profiles derived from the in-house measurements of compounds 7 and 2 (Fig. 10c) show reasonable lower contrast between labeled and unlabeled samples, which is much closer to the expectation. Four small bumps in the electron density difference curve appear at z = ±17.0 and ±7.0 Å (Fig. 10d ), corresponding to possible inter-iodine distances of approximately Dz = 34, 14, 10, or 24 Å . Compared to b-structures of gramicidin A that are likewise assumed for peptide 7 or the unlabeled analog 2 with a periodicity of 5.6 or 6.3, the investigated b-hairpin with a strand length of 12 residues would give a peptide length of 28.8 or 20.8 Å , respectively (Fahsel et al. 2002) . This would not directly match a transmembrane orientation without assuming additional effects from distortions, tilt (Fig. 11iv) , or a coexisting population (Fig. 11v) . The rather flat difference curve of the in-house data for species 7 appears less conclusive than difference curves with more clearly identifiable maxima. Species 8 In case of the molecular species 8, holding a 5,5-diiodo-allyglycine label at the loop position, five lamellar orders could be obtained in the synchrotron experiment (see the Electronic Supplementary Material). The peak-to-peak distance of the electron density curves reveals a headgroup spacing of approximately 33 Å (4.55 keV) and 32 Å (5.80 keV), respectively. The corresponding electron density difference curve is pronounced due to the double signal intensity resulting from the double iodinated label, but differences in resolution at the different X-ray energies may also play a role. The profile shows strong maxima at ±15.4 Å and additional inflection points around ±7.1 Å (Fig. 10d) . The maxima at ±15.4 Å may indicate a transmembrane orientation of the b-hairpin structure 8 nearly spanning the same distance as the terminally labeled sample 7/DLPC/Chol. The electron density difference curve of in-house reference measurements does not show significant rises in the headgroup region, but only at the membranes outside (±21.4 Å ) in the membrane water layer (see above). A rather diffuse and less intense maximum appears spread out over the bilayer hydrophobic core. Summarizing the results of the iodine-labeling experiments, the electron density difference data from synchrotron experiments and in-house data are not in agreement. While the synchrotron experiment faced technical challenges related to the relative low photon energy (L III edge) and sample realignment after beam drift when changing the energy, as well as intrinsically lower contrast, the comparison of two different, not necessarily isomorphic, samples in the in-house study may also pose a challenge. In other words, the differences from one sample to the next may be on the same order as the label effect itself. Note that one needs a much higher reproducibility to interpret the difference curve than the rather gross features of the profiles before subtraction. Therefore, we consider the results regarding the membrane repeat spacing to be more reliable and consistent than conclusions drawn from the peptide labels. Fortunately, however, the reflectivity results for the different peptide species 5, 7, and 8 need not be judged alone to make conclusions about the peptide orientation Fig. 11 Exemplary schematic representation of orientations that could be adopted by labeled peptide species in lipid multilayers and fit the reflectivity data of one or the other peptide species (Fig. 11) . The observed changes of the inner hydrophobic core upon peptide insertion, as revealed by GID/reciprocal space mapping, together with the sum of the reflectivity results may not completely rule out a simple surface anchoring or surface adhesion of the peptide species 2-9, but they make an inserted orientation parallel to the membrane normal, possibly with some tilt distribution, much more likely. This would be further supported by band shifts of the fluorescence emission spectra , indicating a localization of flanking tryptophan residues in the lipid headgroup regions. The lipid response upon the insertion of peptide constructs systematically varied in the oligomerization state and secondary structure was addressed by an X-ray scattering study. The approach included grazing incidence diffraction and X-ray reflectivity and is, in principle, applicable to any multilamellar peptide/lipid complex. The preparation of multilamellar lipid samples is pretty much straightforward and only requires minor adjustments due to solvent compositions at high P/L ratios. In contrast, the requirements for the sample environment are rather high in terms of a stable and reproducible RH. Nevertheless, an advanced setup is not mandatory, since very distinct and stable RHs can also be generated by applying saturated salt solution reservoirs in a sealed sample chamber. (Greenspan 1977; Winston and Bates 1960) . It has been shown that reflectivity experiments can be carried out with an in-house diffractometer with high performance and advanced resolution. This is especially true for an analysis of the hydrophobic membrane thickness variation upon peptide insertion. For a labeling approach, presented herein, in-house reflectivity holds the advantage of a higher contrast variation compared to energy variation around the I-L III absorption edge but includes the application of nonisomorphic samples. However, in-house reflectivity in combination with heavy-atom labeling is generally suited to generate reliable results for peptide side-chain positioning within the lipid bilayer (Küsel et al. 2007; . GID experiments are normally not only carried out at beamline experimental stations; a more extended sample screening appears only reasonable for taking advantage of the relatively short exposure times resulting from usage of a high-brilliant synchrotron X-ray beam. The methods presented herein are amenable to users that are not closely related to the X-ray field. This is notably true, since macro scanning and automated data evaluation can be applied that is even recommended if it is dealt with an increased quantity of samples. In this study GID yielded quantitative values for the lipid chain correlation length n and the average lipid chain spacing a via analysis of the lipid chain correlation peak. Reflectivity studies provided additional information about the variation in the lipid spacing d and bilayer thickness d PP (Fig. 7b) . This rather general approach of monitoring the collective lipid response to peptide interaction was combined with multiple site-directed iodine labeling, extending previous studies (Küsel et al. 2007; ) through a larger variety of constructs, labels, and label positions. While a good approach in principle, this part of the presented work showed results that were less consistent and thus less conclusive due to experimental difficulties. Future improvements could include the use of the I-K a edge at 33.17 keV rather than the low energy L edges or the application of a more dedicated setup for this spectral range. Despite some of these problems, the data seem to point to a membranespanning orientation of the designed b-helical peptide hairpins. Without relying on the contrast variation, a thickening of the bilayer by stretching of lipids in the z direction could be clearly and consistently evidenced. Thus, membrane insertion of b-helical structures is characterized by a positive mismatch situation. This hydrophobic stretching of the membrane and a good accommodation of the peptide species, which might be based on their extraordinary sidechain composition, also result in an increased membrane packing and lipid ordering of annular lipids. These findings are in contrast to theoretical studies that describe peptide helices as soft cylinders (Lagüe et al. 1998 (Lagüe et al. , 2001 . Two effects could possibly contribute to the enhanced lateral packing, the hydrophobic mismatch and/or hydrophobic, attractive lipid/peptide interactions resulting from sidechain intercalation into the lipid matrix. Future experiments can be designed to evaluate the separate effects of lateral packing and ordering, systematically varying the mismatch by using longer chained lipids such as DMPC or reducing the aromatic content of peptide side chains via sequence mutation. Open Access This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

Vaccine Potential of Nipah Virus-Like Particles

Nipah virus (NiV) was first recognized in 1998 in a zoonotic disease outbreak associated with highly lethal febrile encephalitis in humans and a predominantly respiratory disease in pigs. Periodic deadly outbreaks, documentation of person-to-person transmission, and the potential of this virus as an agent of agroterror reinforce the need for effective means of therapy and prevention. In this report, we describe the vaccine potential of NiV virus-like particles (NiV VLPs) composed of three NiV proteins G, F and M. Co-expression of these proteins under optimized conditions resulted in quantifiable amounts of VLPs with many virus-like/vaccine desirable properties including some not previously described for VLPs of any paramyxovirus: The particles were fusogenic, inducing syncytia formation; PCR array analysis showed NiV VLP-induced activation of innate immune defense pathways; the surface structure of NiV VLPs imaged by cryoelectron microscopy was dense, ordered, and repetitive, and consistent with similarly derived structure of paramyxovirus measles virus. The VLPs were composed of all the three viral proteins as designed, and their intracellular processing also appeared similar to NiV virions. The size, morphology and surface composition of the VLPs were consistent with the parental virus, and importantly, they retained their antigenic potential. Finally, these particles, formulated without adjuvant, were able to induce neutralizing antibody response in Balb/c mice. These findings indicate vaccine potential of these particles and will be the basis for undertaking future protective efficacy studies in animal models of NiV disease.

Since it was first recognized in 1998, Nipah virus (NiV) has caused several outbreaks in humans of encephalitic disease associated with high lethality. In the first outbreak, which was in Malaysia and Singapore, 265 humans became sick and some ,40% of them died. Epidemiological links pointed to human contact with sick pigs in commercial piggeries, and the outbreak was brought under control through culling of approximately ,1.1 million pigs [1, 2, 3, 4] . Since then, the virus has re-emerged in Bangladesh and neighboring India, starting in 2001, and between then and now, has caused several smaller but even deadlier disease outbreaks with case fatality rates ranging between 60 and 90% [5, 6, 7, 8] . Unlike the Malaysian outbreak, the route of transmission in these outbreaks was considered to be bat-to-human via food contaminated with bat saliva [9] . In some cases, nosocomial transmissibility and person-to-person spread was also noted [5, 10, 11, 12] . An additional concern is that NiV is also potentially an agent of agro-terror since the rate of transmission of this virus in the pig population is close to 100% [13] . Effective vaccine and therapies are needed to combat the threats posed by NiV. NiV is a member of the genus Henipavirus in the subfamily Paramyxovirinae, family Paramyxoviridae. It has several distinctive genetic and biologic features [14, 15, 16, 17, 18] although its morphology and genome organization is similar to that of other members of the subfamily. NiV has six genes arranged in tandem, 39-N, P, M, F, G and L-59 [15, 16] . The N, P and L are required for reconstituting viral RNA polymerase activity, the matrix protein M is required for particle formation and budding, and the two surface glycoproteins G and F are required for attachment and entry into the host cell [19, 20] . EphrinB2 and B3 have been identified as the NiV entry receptors [21, 22, 23] . After fusion of the virus and the cell membrane, the viral ribonucleoprotein is released in to the cell cytoplasm. Following transcription and replication, the viral components migrate to the plasma membrane for assembly and budding of progeny particles [24, 25] . Two vaccination strategies for NiV disease prevention have already been explored experimentally: A canarypox virus-based vaccine vector approach was effective as veterinary vaccine [26] , and it is in the process of further development. The same approach for human use vaccines is undergoing extensive evaluation, largely for HIV and AIDS [27] . A soluble NiV G protein approach has also shown promise [28, 29] . However, subunit approaches are in general less effective than particulate immunogens, and can suffer from suboptimal presentation to the immune system [30, 31, 32] . Immunogenicity in mice to NiV glycoproteins has been reported recently using two vectored approaches for gene delivery; one using Venezuelan equine encephalitis virus replicons [33] and the other involving inoculation of a mix of two complementing defective vesicular stomatitis virus (VSVDG) vectors, one for expressing each of the two NiV glycoproteins [34] . The latter approach is new and seems promising but its regulatory approval as human vaccine might be problematic [34, 35] . In this study we have explored the potential of NiV virus-like particles (VLPs) as a vaccine. Plasmid-mediated expression of selected viral proteins results in the spontaneous assembly and release of VLPs. These particles make highly effective immunogens because they possess several features of the authentic virus such as their surface structure and dimensions [31, 36] . They are also safe because they do not contain any viral genetic material. VLPs, where one or more of the constituent proteins serve as immunogens (native VLPs), are particularly effective as vaccines for infectious disease. The fact that two such vaccines [Gardasil (Merck & Co) for human papillomavirus (HPV), and Sci-B-Vac (SciGen) and Bio-HepB (GlaxoSimthKline) for Hepatitis B virus (HBV)] have already been approved for human use, and many, for non-enveloped and enveloped viruses [31, 32, 37, 38, 39, 40, 41, 42] are at various stages of development, attests to the desirability of this approach for vaccine development. The budding capacity of virus proteins as VLPs, the proteinprotein interactions that facilitate this process, and the central role of M protein in VLP assembly and release has been described for several paramyxoviruses such as Sendai virus (SeV), Newcastle disease virus (NDV), respiratory syncytial virus (RSV), paramyxovirus simian virus 5 (PIV-5) and human parainfluenza virus type 1 (hPIV1) [20, 43, 44, 45, 46, 47] : The efficiency of VLP formation in virtually all these studies was based on M protein release in the supernatant. NiV virus-like particles have also been described [48] ; the results of this study showed 1) that NiV G and F proteins individually retained some budding capacity although it was far less efficient than that of the M protein and 2) NiV N, M, F and Gcontaining VLPs resembled the virus in some respects but differed significantly from it with respect to ratio of VLP-incorporated F protein; most of it was present in precursor F 0 form. Recently, the vaccine potential of native VLPs of NDV [49] has been described: these particles, composed of HN, F, M and NP proteins, had several virus-like properties. However, since the F protein in this formulation was modified by design to ablate the cleavage site, it remained in its precursor form; consequently, the NDV VLPs were non-fusogenic, and therefore incapable of inducing syncytia formation. Here we describe NiV VLPs composed of the two surface glycoproteins G, and F, and the matrix protein M. The G and F proteins were included because they mediate attachment and entry into the host cell [50, 51, 52] , both are major targets of neutralizing antibodies, and both are major players in vaccine induced protection [52, 53, 54] . NiV G and F together are also the most effective as immunogens; this was elucidated in a canary pox virus vector-based experimental protective efficacy study [26] . The M protein was included in our formulation because it is required for particle formation and release [20, 25, 45] . Under optimized conditions, we were able to make substantial, quantifiable amounts of NiV VLPs composed of these three NiV proteins. This has allowed us to characterize their properties in detail to show that they possessed many virus-like/vaccine desirable properties in vitro. It has also allowed us to test for immunogenicity in vivo in Balb/c mice; note that although NiV does not cause disease in these animals, NiV proteins injected in them are known to induce robust neutralizing antibody response [29, 33, 34] . Importantly, NiVspecific mouse monoclonal antibodies are protective in the hamster model of NiV disease [55] . In this study, careful assessment of immunogenicity has shown for the first time, that these NiV VLPs are able to induce neutralizing antibody response. We have also provided a detailed methodology to optimize production of the VLPs for research purposes. Beyond this, we have provided the first CryoEM study of NiV VLPs and thus provide a careful assessment of their morphology. We further demonstrate that NiV VLPs can trigger ''fusion from without'' upon addition to cells. To our knowledge this is a first for an enveloped VLP. Finally, we have shown that NiV VLPs activate innate immune signaling in ''infected'' cells and provide a transcriptional profile of this response. Based on all these attributes, NiV M, F and G-protein-containing VLPs show promise as vaccine and will be the basis for undertaking future protective efficacy studies in animal models of NiV disease. NiV expression plasmids pCAGGS-G, F, and M are all under the control of chicken beta actin promoter [56] , and they were constructed in the laboratory of one of the co-authors of this study (CB) as described previously [20] . Human embryonic kidney 293 cells (ATCC, CRL-1573) and 293T cells (ATCC, CRL-11268) were grown in Dulbecco's minimum essential medium supplemented with10% fetal bovine serum (FBS) and penicillin and streptomycin, and maintained in the same medium containing 2% FBS. The minigenome that was used for optimizing VLP formation has been described previously [57] . All the initial minigenome-based optimization steps were done in BHK-T7 cells (a gift from Dr. N. Ito). The same conditions were applicable to produce VLPs in 293T cells and they were used throughout to generate the VLPs used for the work described in this study. 293T cells were grown in Dulbecco's complete medium to achieve semi-confluent (80-90% density) cell monolayers. The cells were transiently transfected with the plasmids constructs using the lipid reagent Lipofectamine 2000 according to the general guidelines provided by the manufacturers' instructions (Invitrogen Inc). At 48 hrs post-transfection, the VLP-containing cell supernatants (SUP) were harvested for concentration and purification of the VLPs. Because of the fusogenic property of our VLPs, there was widespread syncytia formation at this time point although the cells were still adherent. VLPs released in the transfected-cell SUP were harvested and clarified by centrifugation at 3,500 rpm for 30 minutes at 4uC and concentrated by sucrose density gradient centrifugation based on previous descriptions [44, 45, 58] . Briefly, the clarified SUPs were concentrated by ultracentrifugation through 20% sucrose cushion in TN buffer (0.1 M NaCl; 0.05 M Tris-HCL, pH 7.4) at 200,0006 g for 8 hours at 4uC. The resulting VLP pellet in ,0.5 ml volume was purified on a discontinuous sucrose gradient formed by layering 80%, 65%, 50% and 10% sucrose in TN buffer. After centrifugation at 186,0006 g for 8 hours, the top ,1.5 ml of the gradient (which included the VLP-containing band at the interface between the 10% and 50% sucrose layers) was resuspended in 20% sucrose buffer and centrifuged once more at 160,0006 g for one hour. The resulting pellet was resuspended in 20% sucrose solution in endotoxin-free TN bufffer and stored at 4uC for subsequent analysis. Supernatant of 293T cells transfected with empty pCAGGS plasmid and processed similarly (referred to as ''mock'' particles) served as negative control when needed. Since the ratio of the protein expression plasmids used at transfection and the time of harvest may have a bearing on the level of VLP formation, a minigenome-based VLP infectivity assay, similar to those described previously [59, 60] was used to determine the relative concentrations of the constituent plasmids, and to determine the kinetics of VLP formation for optimal production. This assay provides only a comparative assessment of VLP formation since it only accounts for VLPs that are able to incorporate and passage minigenomes. However, based on the assumption that the ratio of empty and minigenome-containing VLPs will be equivalent in each reaction, the method provides an indirect means to determine the optimal set of conditions for VLP production as determined by VLP-incorporated minigenomeencoded CAT enzyme activity. Briefly, the steps involved in the VLP infectivity assay were 1) transfection of NiV minigenome construct and co-transfection with full complement of the NiV protein expression plasmids, N, P, L, M, F and G, using Lipofectamine 2000. 2) following replication (48 hours posttransfection), passage of equal volume of VLP-containing transfected cell SUP on to fresh cells previously transfected with N, P and L plasmids and 3), determination of CAT activity in the VLP infected cells 48 hours later. Replication of the VLPincorporated incoming mingenomes based on reporter gene activity indicates the level of particle formation and release, VLP infectivity, and successful minigenome packaging. FAST CAT Assay kit (Molecular Probes) was used according the manufacturer's instructions and allowed accurate quantification of CAT enzyme levels over a wide linear range. VLPs were purified as described. The particles were adsorbed on Formvar carbon coated copper grid by floating it on a drop of VLP suspension for 15 minutes, the grids were blotted, and then negatively stained with 2% aqueous uranyl acetate for viewing by transmission electron microscopy. The VLPs were vitrified as reported previously [61] on holey carbon film grids (C-flat TM , Protochips, Raleigh, North Carolina). VLPs were imaged at 40,000x indicated magnification using a 4k64k slow-scan CCD camera (UltraScan 895, GATAN, Inc., Pleasanton, CA) using a low-dose imaging procedure. Unfixed VLPs were used for immunogold labeling to limit antibody reactivity to the cell surface proteins. The particles were adsorbed on formvar coated nickel grids, stained with NiV specific primary antibody (hyper immune mouse ascites fluid, HMAF, obtained from Dr. P. Rollin, CDC) diluted in buffer (1% BSA in 0.05 M tris buffer) rinsed in wash buffer (0.1% BSA in 0.05 M tris buffer), stained with colloidal gold labeled goat anti-mouse secondary antibody (Jackson ImmunoResearch Laboratories), washed, and then negatively stained with 2% uranyl acetate for viewing by EM. The total protein concentration of the purified VLP preparations was measured by the BCA (Bicinchoninic Acid) method (Thermo Scientific Laboratories). VLP composition was determined by western blot analysis. Briefly, purified VLPs resuspended in endotoxin free PBS were lysed by resuspending them in equal amount of 2x SDS proteinloading buffer and loaded into a 12% SDS-polyacrylamide gel with a 4% stacking gel. 293T cell lysates processed similarly were run in parallel as negative cell control. Following electrophoresis to resolve the protein bands, and transfer to membrane, the blot was incubated with NiV-specific HMAF primary antibody at a dilution of 1:1000 dilution, overnight at 4uC, and HRP-conjugated antimouse secondary antibody (from GE Healthcare) at a 1: 20,000 dilution for one hour at room temperature. The proteins were revealed using western blot detection reagents according to instructions provided by the manufacturer (GE Healthcare). These studies were undertaken with the approval of the Institutional Biosafety Committee (Protocol# #01/08-2010-1) and the Institutional Animal Care and Use (IACUC) Committee (Protocol # 0904028). Five to six week old female Balb/c mice (Harlan Laboratories) were housed in microisolater cage for 4 days in the Animal Resource Center at the University of Texas Medical Branch before beginning the immunization protocol. Mice in groups of five were immunized by subcutaneous inoculation of four different concentrations of VLPs (1.75, 3.5, 7 or 14 mg/ mouse, referred to subsequently as treatment groups A through D respectively) prepared just prior to use in sterile endotoxin free PBS. No adjuvant was used. A group of five mice inoculated with sterile endotoxin free PBS served as negative control group. Mice in the four treatment groups (A through D) were boosted (6 mg/ mouse) on days 15 and 29; the negative control group received PBS. Blood was collected from the submandibular vein of the animals on days 21, 14, 21, 28 and 35; they were euthanized on day 35. Plaque Reduction neutralization test (PRNT). Two-fold dilutions of test sera were made in 50 ml cell culture medium. Under biohazard level 4 conditions, each of the diluted sera were mixed with 50 ml of NiV diluted to generate ,30 plaque forming units and incubated for 30 min at 37uC. The pre-incubated virusantibody mix was added to Vero cell monolayers grown in 96 well plates and incubated for 30 min at 37uC when the inoculum was removed and replaced 150 ml of cell media. After incubation at 37uC for 24 h, the cells fixed in 100% ice-cold methanol and staining by indirect immunofluorescence assay as follows: The wells in the plate were blocked with BSA/PBS and stained with rabbit sera raised against the G protein of HeV, and goat antirabbit Alexa Fluor 488 conjugate (Invitrogen) diluted 1:1000 in blocking buffer. Viral plaques were visualized and counted, and neutralizing antibody titers were reported based on reduction in plaque count by 50% relative to the untreated control (PRNT 50 ). Antibody levels measured by Immunofluorescence assay (IFA). For IFA, NiV-specific total antibody levels were measured by using NiV G, F and M expressing 293T cells as target antigen. Thirty six hours post-transfection, the cells were harvested, fixed in paraformaldehyde, cytospun (Cytocentrifuge, Thermoscientific) on glass slides to obtain monolayered preparations and then stored at 4uC, and used as antigen within three weeks of preparation. On the day of use, the slides were washed in PBS, permeabilized with Triton-X-100 and blocked with BSA/PBS. After incubation with two fold dilutions of the test sera, the cell monolayers were washed and stained with Alexa fluor 488-conjugated goat anti-mouse antibody according the manufacturer's (Molecular Probes) instructions. Negative and positive controls were run in parallel with each batch. Gene expression profile by Real-time PCR VLP-mediated transcriptional activation was tested for eighty four genes involved in Toll-like receptor (TLR)-mediated signal transduction using RT 2 Profiler PCR array (SABiosciences). The 96 well array format included mediators of TLR signaling including adaptors and proteins that interact with the TLRs, and members of NFKB, JNKp38, NF/IL6 and IRF signaling pathways downstream of TLR signaling. Briefly, 293 cells grown overnight in 60 mm dishes were exposed to 10 mg of purified VLPs suspended in 1 ml of OPTI-MEM (Invitrogen). ''Mock particles'' (see Methods, VLP harvest and purification) resuspended similarly and exposed to 293 cells served as negative control. The inoculums were adsorbed on the cell monolayers for 3 hours at 37uC when additional 1.5 ml of OPTI-MEM was added and the dishes further incubated. Twenty fours post VLP exposure, total cell RNA was extracted according to the manufacturer's (SABiosciences) instructions. The integrity of the RNA was verified by agarose gel electrophoresis and the same concentration of total cell RNA from the VLP-stimulated and ''mock'' stimulated cells were used for gene expression profiling by Real-time PCR using Eppendorf Mastercycler unit. The array plate included positive and negative controls for quality assurance, and three sets of housekeeping genes for normalization for data analysis. The fold-change in gene expression in the VLP stimulated 293 cells relative to the ''mock'' stimulated 293 cells was calculated by the DDCt method according to the manufacturer's instructions. In preliminary studies it was found that co-expression of NiV G, F and M proteins in 293T cells resulted in the formation of VLPs that bud out into the transfected cell SUP and that they can be harvested, concentrated and purified as described under Methods. However, the VLP yield was low. To improve the efficiency of VLP formation we proceeded to optimize the ratio of the three expression plasmids used at transfection. We speculated that this would be important based on the fact that a), during replication, paramyxoviruses form a transcription gradient where the 39 proximal genes are transcribed more abundantly than the successive downstream genes [52] and b), the stoichiometry of interaction of the viral proteins has proved to be critical in plasmid-driven minigenome and full-length rescue systems [62] . The importance of protein ratios for VLP formation was alluded to in a previous NDV study where the expression plasmids were co-transfected at ''pre-determined concentrations'' to produce VLP-incorporated protein ratios analogous to those in virus infected cells [49] . In a study by Patch el al [48] , equivalent amounts of NiV N, M, F and G were initially used to produce the VLPs. In that study, VLPs were subsequently also made by adjusting NiV expression plasmid concentrations by experimental variations similarly to that in the NDV study [49] . The efficiency of particle formation and budding in both these, and many other paramyxovirus VLP formation systems was based on M protein release [43, 44, 45, 46, 48] . We have chosen a minigenome-based functional assay, the VLP infectivity assay (described under Materials and Methods), to determine optimal expression plasmid ratios for efficient VLP formation based on reporter gene readouts. Briefly, 293T cells were transfected with plasmids as shown in Figure 1 . For titrating NiV G, F and M plasmids, increasing concentrations of either G, or F or M expression plasmids were, in turn, co-transfected with fixed concentrations of the other two plasmids. The minigenome and N, P and L plasmids were transfected using a predetermined ratio [57] . The VLP-containing cell SUP was harvested 48 hours post-transfection, clarified by centrifugation, and equal volume from each was passaged onto fresh cell monolayers (VLP-infected cells) previously transfected with the core proteins required to support the incoming packaged minigenomes. The VLP infected cells were harvested 48 hours later and tested for optimal particle production based on incoming minigenome-encoded CAT activity. This time point was chosen because maximal VLP formation was also found to be time dependent and optimal at 48 hours post-passage (data not shown). The reproducibility of the results was verified in an independent repeat experiment. Results presented in Figure 1 show that within the given range, and based on the levels of minigenome-encoded CAT activity, varying the concentrations of G, F and M plasmids had a bearing on VLP formation. CAT activity in the VLP infected cells appeared optimal in the boxed lanes 7 and 8 but further analysis to ensure reporter activity in the linear range (data not shown) indicated that the largest amount of minigenome-containing NiV VLPs were produced when the cells were transfected with the NiV M, F and G plasmid ratios of 3:1:1 as in lane 7. This ratio was used for making all our VLP preparations. The optimized conditions were applied to transfect G, F and M expression plasmids in 293T cells grown in 10 cm dishes. The VLP-containing culture SUPs were harvested 48 hours later, and concentrated and purified as described. Briefly, the clarified SUPs were concentrated by ultracentrifugation through 20% sucrose cushion, and then purified on a discontinuous sucrose gradient. The VLP pellet was resuspended in TN buffer and viewed by EM after negative staining. The result presented in Figure 2A shows a VLP-containing band in the sucrose gradient. Viewing of the negatively stained purified particles by transmission electron microscopy ( Figure 2B) showed numerous virus-like particles. The size variation of these VLPs was consistent with the parental virus: NiV is a pleomorphic virus ranging in size from 40-1900 nm [63, 64] ; the sizes of the VLPs ranged from ,40-500 nm. The particles also resembled authentic NiV morphologically, and this is seen more clearly in the magnified images presented in Figure 2C ; here, the fringe of the glycoproteins is clearly visible on the VLP surface. An occasional VLP had what appeared to be a double fringe (shown with an arrow), a feature more frequently associated with Hendra rather than NiV virus particles [64] . The image in Figure 2D is a cryoelectron micrograph of one of our VLPs; the overall surface appearance is virus-like, which is described as dense, ordered and repetitive [31] , and it shows the surface glycoproteins and their spatial arrangement even more definitively. To verify whether the NiV proteins were incorporated into the VLPs as designed, purified particles were analyzed by western blotting using NiV-specific mouse antibody, and HRP-conjugated anti-mouse secondary antibody as described under Methods. The right hand panel in the Figure 3 shows VLP-incorporated proteins in two different preparations of NiV VLPs. The protein bands are consistent in size to NiV proteins G, F 0 , F 1 and M proteins [17, 48] . The relative amounts of the VLP-incorporated G and M proteins appeared to be similar to that reported in NiV virions also [17] . This was in spite of the fact that the viral proteins in that study were revealed using rabbit sera raised against bacterially expressed Hendra virus proteins. However, the ratio of the VLPincorporated F 1 to F 0 was different from that in the virions. This difference is more likely to be a reflection of timing and protein turnover rather than the reagents used to reveal them since in a previous study, pulse chase experiments have shown that similarly to the VLP-incorporated F 1 to F 0 , intracellular cleavage of the precursor NiV fusion protein by cathepsin L results in near equal mix of mature fusogenic, and the precursor forms [65] . Absence of the NiV-specific bands in two different 293T cell lysate preparations processed similarly (and shown in the left hand panel in Figure 3 ) confirms specificity of the VLP-incorporated proteins. The immunoreactivity of the VLP surface glycoproteins was verified by staining purified unfixed VLPs by the immunogold labeling technique using NiV-specific mouse antiserum and 6 nm colloidal gold particle-conjugated goat anti-mouse secondary antibody (Jackson ImmunoResearch Laboratories Inc). The particles were viewed by EM after negative staining. The use of unfixed particles assured that only the surface-exposed antigens would be reactive. Numerous VLPs with the gold particles decorating their surface were seen; Figure 2E shows two such VLPs. Syncytium formation is a classical feature of NiV and other paramyxovirus-induced cytopathology that can be blocked by virus-specific neutralizing antibody. A similar observation was made when 293 cells were ''infected'' with the NiV VLPs. Briefly, the VLPs were pre-incubated with NiV-specific antibody, Junin virus (JV)-specific antibody, and with OPTI-MEM I (Invitrogen Inc) medium only for one hour at 37uC before inoculating onto near confluent 293 cell monolayers grown overnight in 60 mm dishes. The inoculum was removed after incubation for 3 hours at 37uC, replaced with OPTI-MEM I, and the plates were further incubated overnight at 37uC overnight. The monolayers were then viewed for the formation of syncytia after staining with crystal violet. The results in Figure 4 show that 293 cells exposed to NiV VLPs induced syncytium formation and that this process was neutralized by NIV-specific antibodies; prior incubation with the unrelated JV antibodies failed to block this process. Mice in groups of five were inoculated subcutaneously with four different concentrations of purified VLPs and boosted as described under Methods. The negative control group of five mice were inoculated with sterile endotoxin free PBS for each inoculation. The mice were bled from the submandibular vein on the day before primary inoculation, and then on days 14, 21, 28 and 35. For initial evaluation, sera from each treatment group were pooled, and the IFA method used to determine levels of NiV-specific antibodies. The results in Figure 5A show that titers (reciprocal of the highest serum dilution showing reactivity) increased with time post primary inoculation, i.e., the highest titers (1:2560) were seen on day 35. Titers also increased with VLP dosage although by day 35, the three higher treatment groups seemed to produce similar titers. As expected, the mice in the negative control group remained nonresponsive. All sera were tested individually by plaque reduction neutralization method by doubling dilution of each sample (1:5 to 1:80) as Figure 5B ) showed distinct association between VLP dosage and the ability to mount a neutralizing antibody response. Mice inoculated with the two highest VLP doses (treatment groups C and D) were each able to induce neutralizing antibodies by day 35. When samples from mice receiving the two lower concentrations of VLPs (3.5 ug/dose and 1.75 ug/dose, corresponding to treatment groups B and A respectively) were similarly tested, 3 of 5 and 1 of 5 mice respectively induced neutralizing antibody response; the titers ranged from 1:5 to .1:80. As expected, the control mice did not induce neutralizing response. NiV VLP-induced activation of genes involved in signaling innate immune response A PCR array format (SABiosciences) was used to investigate modulation in transcription profile of 84 genes involved in innate immune responses to include TLR signaling family and members of the downstream signaling pathways, NFKB, NF/IL6, IRF and JNKp38. These genes represent key sensors of non-self that signal, and ultimately shape the nature of innate immune response that modulates the type and duration of adaptive immune responses [66, 67] . The differential expression of genes in VLP-exposed 293 cells relative to the ''mock'' infected 293 cells was measured by real-time PCR. Same concentration of total cell RNA from the VLP-stimulated and the ''mock'' stimulated control cells were used for first strand synthesis and Sybr green PCR amplification of the relevant genes as described under Methods. The integrity of RNA in each sample was confirmed by gel electrophoresis (Figure 6A ). Data representing the differential transcription profile of VLP exposed vs. ''mock'' stimulated cells is shown as a heat map ( Figure 6B) . A 4-fold cutoff threshold was used to determine modulation in gene expression. We noted significant VLP-stimulated up-regulation (89 fold and 7 fold) in the expression of NFKB2 and TBK1 genes respectively. Close to four fold (3.9 fold) up-regulation was noted also in IL-8 and MAPK8 genes. NFKB2 and IL-8 are target genes in the downstream NFKB pathway, and TBK1 which are in the IRF and JNK/p38 pathways respectively. Using a minigenome-based functional assay, we have established conditions (described under Results and shown in Figure 1 ) that have allowed us to produce substantial quantities of NiV VLPs to be able to undertake the studies described in this manuscript. We have shown that these particles are functionally assembled, biologically active and are able to induce innate immune responses, and a neutralizing antibody response. Native VLPs have been used to study various aspects of the virus lifecycle, as carriers to deliver heterologous proteins for vaccination, and to deliver small molecules for gene therapy purposes. Particularly importantly, they have been used highly effectively as vaccines in their native form [31, 32, 39, 40, 41] . No vaccine for NiV disease has been developed so far that would be both safe and protective for humans. The two vaccination strategies that have already been explored are the canary pox-based vector approach [26] and soluble subunit approach [28, 29] . NiV vaccine by the former method is undergoing development as a veterinary vaccine [26] . The same approach is being evaluated for human use vaccines, mainly for the prevention of HIV and AIDS [27] . The subunit approach has limitations as already mentioned above [28, 29, 30, 31, 32] . One particular challenge revealed by studies that tested a soluble NiV G protein-based subunit vaccine formulated with adjuvant is the potential difficulty of eradicating infection in the central nervous system. In that study [28] , live virus was present in the brain of one cat, and viral RNA was present throughout the 21 day postchallenge period in the brains of the remaining challenged animals. A recently reported vaccination strategy [34] requires simultaneous inoculation of two VSVDG vectors, one expressing NiV G, and the other expressing NiV F proteins. It was of interest to note that supernatants of cells co-infected with these two defective viruses were infectious and could be passaged indefinitely in the absence of VSV G trans-complementation. This vaccination approach seems promising since self-propagated stock of these two viruses induced robust neutralizing antibody response in mice. However, potential pathogenicity of VSV-based vaccine vectors remains a concern [34, 35] . The potential of a recombination event resulting in a single VSV vector virus expressing both these NiV proteins is unlikely, but it may still be problematic for a human use vaccine. Native VLPs like the ones we have produced allow the viral proteins to be presented to the immune system in the same conformation as in the virion for effective B and T cell response [31] . VLPs are particularly effective in producing a protective antibody response because of their virus-like size range, their particulate nature, and their virus-like dense, repetitive and ordered surface structure [31, 36] . The spacing of the antigenic epitopes on the VLP is also optimal for B cell activation [31] : EM analysis showed that our particles resembled the real virus in terms of size and surface structure [63, 64] . The image in Figure 2D is the first elucidation of VLP structure of any paramyxovirus imaged by CryoEM, and it provides a careful assessment of their morphology; it alludes to a surface similar to that revealed for measles virus by the same imaging technique (Dr. Elizabeth Wright, Emory University). The proteins on the VLP surface are clearly visible here; the average distance between the spikes was 9.13 nm and standard deviation was 1.72 nm. This is of interest given that epitopes spaced between 5 and 10 nm are known to be sufficient to drive optimal B cell activation [36] . NiV M, F, G and N protein-containing VLPs consistent in size and morphology to the parental virus have also been reported in a previous study which evaluated protein-protein interaction that facilitate VLP formation [48] . However, in that study, most of the particle-incorporated NiV F protein was predominantly in the uncleaved precursor form. This finding is clearly distinct from ours since our VLPs contained substantial amounts of cleaved F protein, and this may have been related to ratios of the interacting proteins expressed in 293T transfected cells. In a recent study of NDV VLPs [49] , the particle-incorporated proteins were reported to have virus-like protein ratios, but the F protein remained in its precursor form because the cleavage site required to produce the fusion competent form was mutated by design. What effect a VLP-incorporated non-fusogenic F protein may have, relative to the fusogenic form, on the level and quality of VLP-induced immune response is not clear at present since difference in immunogenicity between fusion-competent and fusion-defective VLPs has not been experimentally evaluated so far. However, a recent report suggests that viral fusogenic membrane glycoproteins may enhance vaccine potency [68] . Immunogold labeling of our unfixed NiV VLPs confirmed that the surface proteins in our VLPs were functionally assembled and they were biologically active ( Figure 2E ). We could deduce the presence of biologically active G and F proteins on the VLP surface by the fact that they were able to induce the formation of syncytia in 293 cells ( Figure 4 ); this is a process that requires the interaction of both the surface glycoproteins, the attachment protein G, and the fusion competent F protein, when they come in contact with the cognate receptor-bearing cells. Formation of syncytia or multinucleated cells in replication competent enveloped viruses, especially paramyxoviruses, is induced by a process that is described as ''fusion from within'', and it can be blocked or neutralized by prior treatment of the virus with specific antisera. In contrast, ''fusion from without'' is induced by non-replicating viruses at high multiplicities of infection, and it too can be blocked by pretreatment with virus-specific antibodies ( [69] , and references therein; [70] ). Our non-replicating particles likewise induced syncytia formation in 293 cells that could be neutralized with NiVspecific antibodies (Figure 4 ). To our knowledge, this is the first study describing fusion from without induced by VLPs of any paramyxovirus, or any other enveloped viruses, although it has been described for the VLPs of the non-enveloped rotavirus [71] . The mechanism(s) of fusion from without is not clear but two models have been proposed [72, 73] ; one proposes that particles connecting adjacent cells effectively promote fusion between them, and the other is that when particles decorated with the surface glycoproteins fuse with the target cell membrane, the glycoprotein complexes diffuse freely in the lipid bilayer, and mimic fusion from within. The type of VLP-induced syncytia formation and eventual cell death is also not known. We are in the process of investigating it. Neutralizing antibody response is the critical correlate of protection mediated by prophylactic vaccines [53, 54] and native VLPs promise to be highly effective prophylactic vaccines for paramyxoviruses like NiV, and others like NDV and measles where neutralizing immune response is known to play a pivotal role in protection against disease [53, 54, 55, 74] . Our VLPs were highly effective immunogens, and all, especially in the three higher treatment groups produced high levels of response by day 35 ( Figures 5A) . Importantly, NiV VLPs were able to induce neutralizing antibodies. This response was clearly dose-dependent ( Figure 5B ). All ten mice receiving a primary inoculation of 7 or 14 mg VLPs (subgroup C and D) were able to produce such response; but even of those animals that received a first dose of only 3.5 or 1.75 ug/mouse (treatment group B and A respectively), 3 of 5 and 1 of 5 produced neutralizing antibodies. Neutralization antibody response was first seen on day 28, and increasing titers were seen in some animals within a week of it; we believe that this response, induced by our non-replicating and potentially safe particles, formulated without adjuvant, compares favorably with the levels of such response induced at an equivalent time point by some replication competent pseudotype viruses [34] . Immunogenicity to native VLPs has been reported previously for one other paramyxovirus namely NDV [49] . In that report, immune response to NDV VLPs was evaluated by primary inoculation of mice intraperitoneally with VLP concentrations ranging between 10 and 40 mg, and a booster dose of 10 mg, without adjuvant. NDV-specific titers by ELISA were high in each mouse in each treatment group. Neutralizing antibody response to 20 and 40 mg of these particles was also detected. The nature of innate immune response dictates the type and duration of adaptive immune response [67, 75] . The mechanism by which NiV VLPs are recognized by host cells and trigger the induction of innate immune response, and how this translates into effective adaptive immunity is not known. Here we have taken the first step ( Figure 6 ) towards understanding this process. With the experimental conditions as described, we observed VLP-induced activation of some of the genes that are known to be involved in the induction of an effective innate immune response [75] . Results presented in the heat map in Figure 6 show that relative to the ''mock'' treated cells, NFKB2 gene (in the NFKB pathway) was up-regulated 89 fold as a result of VLP exposure, and TBK1 (in the IRF pathway) was 7 fold higher. In the light of these findings, we are testing PCR array expression profiles of the same set of 84 genes in 293 and other cells at earlier and later time points to identify their upstream effectors, and NiV VLP-responsive signaling networks. In this respect, the murine system, with the many available immunological reagents and knockout strains may provide the best system to identify these host sensors. Currently there is minimal information on live NiV infection-responsive cellsignaling changes [76] and there is none on array-based transcriptional alterations for comparative analysis. Likewise, it has not been possible to compare the NiV VLP-induced transcription modulation with those induced by other paramyxovirus VLPs since to our knowledge, such studies have not been undertaken so far. Lastly, a growing number of reports point to viral surface glycolproteins as relevant in host cell signaling and triggering of innate immune response. We believe that particles like NiV VLPs, with many virus-like properties (including their surface glycoproteins organized to resemble the parental virus, Figure 2D ) would induce an effective innate immune response for the promotion of the desired adaptive immunity [67, 76] . Finally, as described above, our VLPs were highly effective as immunogens, able to induce neutralizing antibody response in all animals with primary inoculation of as little as 7 mg VLP protein each. Fusogenic property of our VLPs may be critically relevant in this regard in the light of recent findings, and would need to be experimentally verified by comparing the potency of fusioncompetent and fusion-defective VLPs as vaccine [68] . In conclusion, we have been successful in producing substantial quantities of NiV VLPs needed to characterize NiV VLPs, we have demonstrated their many virus-like properties, and their effectiveness as immunogens in Balb/c mice. These findings are the basis on which we will be undertaking future challenge studies in the hamster model of NiV disease [77] .