Datasets:

keerthanapacific

/

sample3

like 0

493271

Smoking cigarettes of low nicotine yield does not reduce nicotine intake as expected: a study of nicotine dependency in Japanese males

Background Many Japanese believe that low-yield cigarettes are less hazardous than regular cigarettes, and many smokers consume low-yield cigarettes to reduce their risks from smoking. We evaluate the association between actual nicotine intake and brand nicotine yield, and the influence of nicotine dependence on this association. Methods The study subjects included 458 Japanese male smokers, aged 51.2 ± 9.9 years, who participated in health check-ups in a hospital in 1998 and 2000. Each subject filled out a self-administered smoking questionnaire and the score of each on the Fagerström Test for Nicotine Dependence was calculated. Urinary cotinine concentration was measured at the time of participation. Results The geometric mean of urinary cotinine concentration was 535 ng/mgCr for those who smoked brands with the lowest nicotine (0.1 mg on the package), compared with 1010 ng/mgCr for those who smoked brands with the highest (0.9–2.4 mg, weighted mean of 1.1 mg). Thus, despite the 11-fold ratio of nicotine yield on the packages, the ratio of urinary cotinine level was less than twofold. Both nicotine yield on the package and nicotine dependence significantly increased urinary cotinine concentration, and the negative interaction between them almost attained statistical significance. Cotinine concentration in heavily dependent smokers was consistently high regardless of the nicotine yield of brands. Conclusions The nicotine yield of cigarettes measured by machine-smoking does not reliably predict the exposure of smokers. Smokers consuming low-yield nicotine cigarettes did not reduce actual intake of nicotine to the level that might be expected, especially for those heavily dependent on nicotine. Current labeling practices are misleading for the two-third of smokers who are moderately or highly dependent on nicotine.

Background 'Low-yield nicotine' cigarettes, which have brand names that include 'light,' 'mild,' or similar words, and which have nicotine yields on their packages of 0.8 mg or less, are widely consumed inside and outside Japan, and their market share is increasing. The Tobacco Institute of Japan reported that, in 2001, of the 20 top brands that share 84% of the total cigarettes consumed in Japan, 6.9% had a nicotine yield reported on their package of 0.1 mg, 73.4% had a yield of 0.2–0.8 mg, and 19.7% had a yield of 0.9 mg or higher [ 1 ]. The average nicotine yield of these 20 top brands was 0.8 mg, weighted by number of cigarettes consumed [ 1 ]. Many smokers would like to avoid the health risks associated with smoking, but not want to quit. These individuals would like to use less hazardous cigarettes or cigarettes that cause less irritation to their throats [ 2 ]. In response, the tobacco industry has developed low-yield nicotine brands [ 3 - 5 ]. There have been many studies examining whether low-yield cigarettes are less hazardous than regular brands. For example, in the 1980s and 90s, the actual intake of nicotine [ 6 - 9 ], as well as tar and carbon monoxide [ 7 , 9 ], from smoking low-yield brand cigarettes was similar to that from high-yield brands. More recent studies, which have included ultra-low yield cigarettes (0.1 mg nicotine yield on the package), have shown similar results [ 10 , 11 ]. In addition mortality from lung cancer in the United States has not decreased over the past 30 years, although low-yield brand cigarette increased in market share during that time [ 12 ]. Thus consumers of low-yield cigarettes are at a higher health risk than they expected. It has been assumed that high nicotine levels in the blood of smokers of low-yield cigarettes are caused by compensatory behavior due to nicotine dependence. Most of these comparisons have been determined in Western countries. Fewer comparisons have been reported in Japan, and nicotine dependence was partially taken into account during analysis [ 11 ]. In Japan, the rate of smoking is still high, being >50% among males[ 13 ]. These smokers are likely to consume low-yield cigarettes and to decrease the number of cigarettes consumed in order to reduce the health risks of smoking. For example, a study of smokers in a medical school showed that about 70% of males and 100% of females consumed low-yield nicotine brands [ 14 ]. In 1999, some physicians recommended that smokers change to low-yield nicotine cigarettes as the first step toward quitting [ 15 ]. In 2000, however, a TV program on scientific issues in Japan reported that low-yield nicotine cigarettes did not reduce the health hazards of smoking [ 16 ]. In 2002, the Ministry of Health, Labour and Welfare of Japan published in major Japanese newspapers the findings of a study showing that the nicotine and tar yield of 7 popular cigarette brands in Japan measured by simulating the manner of actual human smoking was larger than that obtained by machines using the Federal Trade Commission (FTC) method [ 17 , 18 ]. Two days later, Japan Tobacco Inc. advertised in these newspapers that the FTC method was authorized throughout the world and that the nicotine yield on the cigarette package was valid for consumers [ 19 ]. Thus, issues regarding nicotine yield are less known in Japan than in Western countries. Moreover tobacco companies seem to target young people, especially young women, by intensive advertisement of low-yield cigarette brands [ 20 ]. It is important, therefore, to emphasize to smokers the health hazards of low-yield nicotine cigarettes, but evidence in Japanese smokers is still scarce [ 11 ]. In the present study, we examined the relationship between nicotine yield and nicotine metabolites excreted in the urine, and the influence of nicotine dependence on this relationship among the Japanese male smokers. Methods The subjects of this study were male smokers who participated in a health check-up at the Kyoto First Red Cross Hospital from July to December in 1998, or from January to February in 2000. The latter subjects were supplementary to the main group, but the two groups exhibited similar demographics. Smokers were recruited using a routine health check-up questionnaire and were defined in this study as those who smoked at least one cigarette per day. Of the 1,579 male participants in the health check-up during the study period, 513 were identified as smokers, and 479 agreed to participate in this study. Each participant filled out a self-administered questionnaire, which was checked during an interview with a physician. This questionnaire included questions determining score on the Fagerström Test of Nicotine Dependence (FTND) [ 21 ]. These included questions on the number of cigarettes smoked per day, time from awaking to the first cigarette, difficulty in refraining from smoking in places where smoking is forbidden, the number of cigarettes smoked during the morning compared with the number smoked during the rest of the day, cigarettes that could not be give up, and smoking for most of the day while ill in bed. Also, included were questions about the brand(s) of cigarettes smoked, inhalation pattern (deep inhalation, some deep inhalations, or no inhalation), attempt to quit smoking, and stage of behavioral change in the quitting process [ 22 ]. Smokers who would continue to smoke during their lifetimes were defined as being on precontemplation-1, and smokers who would continue to smoke for at least one year further but who would quit smoking some day were defined as being on precontemplation-2. About three quarter of plasma nicotine is converted to cotinine, which is excreted in the urine. The half-life of nicotine is 30 minutes [ 23 ] and that of cotinine is about 20 hours [ 24 ]. Measurement of cotinine in the plasma, urine is widely used to assess the level of nicotine intake [ 25 ]. Therefore, each subject's urinary cotinine concentration was measured. Actual nicotine intake was evaluated from urinary cotinine concentration adjusted for urinary creatinine concentration. Although collection of urine over 24 hours may represent nicotine intake more accurately than a spot urine test, for practical reasons we measured cotinine concentration in the first urine in the morning, as this can reflect smoking from the previous day. Each participant was asked to fast from 21:00 the night before until urine was collected around 9:00 the following morning. The urine samples were frozen at -80°C with in the same day and transported to SRL Laboratory, Hachioji, Tokyo, at which cotinine was measured by gas chromatography [ 26 , 27 ]. For machinery nicotine yield by the FTC method, we used the value indicated on the cigarettes packages. Statistical analysis was performed using data from 458 male smokers who completed the FTND question and whose urinary cotinine levels were measured. We compared the characteristics of subjects among three groups categorized by machine-measured nicotine yield (0.1 mg, 0.2–0.8 mg, 0.9+ mg). For each group, we calculated mean machine-measured nicotine yield weighted by the number of subjects. Log-transformed data were used for the urinary cotinine concentrations because it was distributed log-normally. Means were compared using Student's t-test or analysis of variance, and proportions were determined using the chi-square test. The effects on urinary cotinine concentration of machine-measured nicotine yield, number of cigarettes consumed per day, and nicotine dependence were analyzed using a regression model, in which urinary cotinine concentration was the dependent variable, and two of the other parameters were independent variables. Main effect and interaction were evaluated by regression coefficients and partial correlation coefficients. The effect of different cigarette brands was also examined. All statistical procedures were performed by SPSS [ 28 ]. A P value <0.05 was considered statistically significant. Results Of the 458 subjects, 87 (19.0%) smoked cigarette brands yielding 0.1 mg nicotine, 223 (48.7%) smoked cigarettes of 0.2–0.8 mg, and 148 (32.3%) smoked cigarettes of 0.9+ mg (Table 1 ). The highest machine-measured nicotine yield for cigarettes consumed by the subjects was 2.4 mg. The weighted mean of brands yielding 0.2–0.8 mg was 0.5 mg, whereas the weighted mean of brands yielding 0.9–2.4 mg nicotine was 1.1 mg. The subjects ranged in age from 23 to 83 years, and the number of cigarettes consumed per day was 1 to 60. Smokers of brands yielding nicotine of 0.1 mg were slightly older than those smoking brands yielding 0.2–0.8 mg and of 0.9–2.4 mg nicotine (p = 0.08). These two groups did not differ with respect to the numbers of cigarettes consumed per day and the FTND score (p = 0.93 and p = 0.20, respectively). Table 1 Characteristics of the subjects by machine-measured nicotine yield of cigarette Characteristics Machine-measured nicotine yield (mg/cigarette) Total 0.1 0.2–0.8 0.9–2.4 p Number of subjects 87 223 148 458 Mean and SD of mahchine-measured nicotine yield (mg/cigarette) 0.1 0.5 ± 0.2 1.1 ± 0.4 0.6 ± 0.4 Mean and SD of age 53.6 ± 10.4 50.6 ± 9.34 50.8 ± 10.4 51.2 ± 9.9 0.07 Mean and SD of number of cigarettes per day 23.4 ± 12.2 24.5 ± 10.7 24.4 ± 9.5 24.4 ± 10.6 0.93 Mean and SD of FTND 5.1 ± 2.5 5.4 ± 2.3 5.6 ± 2.0 5.4 ± 2.2 0.21 Number of smokers at each category of FTND FTND score 0–3 19 (21.8%) 47 (21.1%) 21 (14.2%) 87 (19.0%) FTND score 4–6 40 (46.0%) 100 (44.8%) 78 (52.7%) 218 (47.6%) 0.41 FTND score 7–10 28 (32.2%) 76 (34.1%) 49 (33.1%) 153 (33.4%) Number of smokers having attempted to quit 50 (61.0%) 132 (59.5%) 84 (56.8%) 269 (58.9%) 0.73 Number of smokers at each stage of behavioral change in quitting process Precontemplation-1 16 (18.6%) 47 (21.3%) 58 (39.7%) 121 (26.7%) 0.001 Precontemplation-2 48 (55.8%) 131 (59.3%) 62 (42.5%) 241 (53.2%) Contemplation 20 (23.3%) 33 (14.9%) 23 (15.6%) 76 (16.8%) Preparation 2 (2.3%) 10 (4.5%) 3 (2.1%) 15 (3.3%) Mean and SD of urinary cotinine concentration (mean-SD, mean+SD*) 535 (1782,160) 770 (1981,299) 1010 (2071,492) 784 (484, 1264) <0.001 FTND:Fagerström Test for Nicotine Dependence P for difference is examined by analysis of variance or chi-square test * Back-transformation of log-transformed data Of all subjects, 153 (33.4%) were heavily dependent on nicotine (FTND score>= 7), whereas 87 (19.0%) had low dependence (FTND score< = 3). About 60% of all subjects had attempted to quit, with the proportion similar in the high- and low-nicotine dependent groups (p = 0.73). The stage of behavioral change was different (p = 0.001), however, with smokers of cigarettes yielding 0.1 mg of nicotine being at more advanced stages. The geometric mean of urinary cotinine concentration in all subjects was 784 ng/mg creatinine (Cr), with a distribution of 484 ng/mg Cr (mean-SD) to 1264 ng/mg Cr (mean+SD), as determined by back-transformation of log-transformed data (range; 10–4770 ng/mgCr). The levels differed significantly between the machine-measured nicotine yield groups (p < 0.001). Urinary cotinine levels did not differ among smokers of individual brands of yielding 0.1 mg of nicotine (p = 0.51 by analysis of variance to adjust for number of cigarettes consumed per day). After integration of similar brands, geometric cotinine concentration means were 686 ng/mgCr for those who smoked American brands and 460 ng/mgCr for those who smoked Japanese brands (p = 0.19). Urinary cotinine concentrations also did not differ among smokers of individual cigarettes brands yielding 0.2–0.8 mg nicotine (p = 0.71, adjusted for number of cigarettes and nicotine yield). The geometric means were 823 ng/mgCr for smokers of Japanese brands and 724 ng/mgCr for smokers of American brands. Mentholated cigarettes were consumed by only 5 subjects and were therefore not examined. When we assayed the relationship between urinary cotinine concentrations and number of cigarettes consumed per day by machine-measured nicotine yield of cigarettes, we found that cotinine concentrations were related to number of cigarettes consumed per day (Figure 1 ). The correlations were different between machine-measured nicotine yield groups, in that there was a stronger correlation for the low nicotine-yield group. There was some negative interaction between the number of cigarettes smoked and machine-measured nicotine yield (Table 2 upper). When the data restricted with <30 cigarettes consumed per day in which the relationship was assumed to be linear (n = 394), the regression coefficient for machine nicotine yield was 0.834 (p = 0.006), 0.074 (p < 0.001) for number of cigarettes consumed per day, and -0.017 (p = 0.20) for interaction term. Thus, among smokers who consumed a small number of cigarettes, cotinine level of those who smoked high nicotine cigarettes was considerably higher than the level of those who smoked low nicotine cigarettes. In contrast, cotinine level differed little among smokers who consumed 40–60 cigarettes per day, regardless of machine-measured nicotine yield of cigarettes. Figure 1 Number of cigarettes per day and urinary cotinine concentration by machine yield of nicotine Table 2 Regression coefficients and partial correlation coefficients for urinary cotinine concentration in a multiple regression model Variable Regression coefficients Partial correlation coefficients B p r p Figure 1 Intercept 5.270 <0.001 Nicotine yield by machine 0.762 0.001 0.23 <0.001 Number of cigarettes consumed 0.045 <0.001 0.43 <0.001 Interaction -0.012 0.16 Figure 2 Intercept 4.994 <0.001 Nicotine yield by machine 0.793 0.001 0.18 <0.001 FTND score 0.268 <0.001 0.52 <0.001 Interaction -0.079 0.057 FTND: Fagerström Test for Nicotine Dependence When we assayed the relationship between urinary cotinine concentrations and machine-measured nicotine yield of cigarettes by FTND score (Figure 2 ), we found little difference between machine-measured nicotine yield groups among heavily nicotine dependent smokers, although there was a correlation between urinary cotinine concentration and nicotine yield among smokers with low dependence. According to the regression model, there was an almost significant negative interaction between FTND score and machine-measured nicotine yield (Table 2 , lower). Figure 2 Geometric means of urinary cotinine concentration by nicotine yield and nicotine dependence The ratio of mean nicotine yield was 0.45 for cigarettes yielding 0.2–0.8 mg nicotine, and 0.09 for cigarettes yielding 0.1 mg nicotine, compared with the brands yielding 0.9–2.4 mg (Table 3 ). In contrast, the ratio of mean cotinine concentration was 0.76 for those who smoked cigarettes yielding 0.2–0.8 mg nicotine, and 0.53 for those who smoked cigarettes yielding 0.1 mg nicotine, compared with those who smoked cigarettes yielding 0.9–2.4 mg nicotine. Among heavily dependent smokers the ratios of urinary cotinine concentration were much nearer to 1 (0.92 and 0.85, respectively) than among smokers with low dependence (0.58 and 0.32, respectively). Table 3 Ratios of mean urinary cotinine concentration for nicotine yield by nicotine dependence Machine measured nicotine yield Ratio of mean urinary cotinine concentration FTND score Category (mean) (mg) Ratio of mean nicotine yield Total 0–3 4–6 7–10 0.9–2.4 (1.1) 1 1 (560) 1 (993) 1 (1333) 1 (1010) 0.2–0.8 (0.5) 0.45 0.58 (327) 0.81 (808) 0.92 (1226) 0.76 (770) 0.1 (0.1) 0.09 0.32 (179) 0.53 (529) 0.85 (1138) 0.53 (535) Numbers in parentheses are geometric means of urinary cotinine concentration (ng/mgCr) FTND:Fagerström Test for Nicotine Dependence Self-reported inhalation patterns did not influence the average urinary cotinine concentration (p = 0.54) when the variable of inhalation pattern was added to the above model with nicotine yield and FTND score. Discussion In Japan, low nicotine-yield cigarettes seem to be recognized as less hazardous, and smokers likely think that the hazards of smoking are directly proportional to nicotine or tar yield shown on the cigarette packages. This is supported by the results of the present study, which indicate that smokers of low-yield nicotine cigarettes were more advanced behaviorally in wishing to quit. This is additionally supported by circumstantial evidence [ 1 - 5 , 14 - 18 ] and by our experience in a check-up clinic, despite the paucity of formal studies of this issue in Japan. We have shown here that smokers of low nicotine cigarettes did not reduce their actual intake of nicotine to the degree that would be expected from the nicotine yield on the packages. Although smokers of cigarettes yielding 0.1 mg nicotine would be expected to ingest one-eleventh of the nicotine ingested by smokers of cigarettes yielding 0.9–2.4 mg nicotine (average of 1.1 mg), the average urinary cotinine concentration of the former group was more than half that of the latter (535 ng/mg Cr vs. 1010 ng/mg Cr). Moreover, smokers of cigarettes yielding 0.2–0.8 mg nicotine (average of 0.5 mg) had about a 25% decrease in urinary cotinine concentration (770 ng/mg Cr) compared with smokers of cigarettes yielding 0.9–2.4 mg nicotine, despite the two-fold reduction expected from the nicotine yield on the packages. These differences were even smaller in smokers who consumed large numbers of cigarettes per day as well as in smokers with heavy nicotine dependence. The number of cigarettes consumed per day is regarded as a component of nicotine dependence and is included in FTND. There were negative interactions between machine-measured nicotine yield and number of cigarettes consumed per day, and between machine-measured nicotine yield and nicotine dependence. In particular, smokers with heavy nicotine dependence tended to have a high urinary cotinine concentration (about 1200 ng/mgCr) despite differences in machine-measured nicotine yield of cigarettes, which may explain this negative interaction. In contrast, the actual nicotine intake of smokers who consumed small numbers of cigarettes and smokers with a low level of dependency was more strongly correlated with the machine-measured nicotine yield of the cigarettes they consumed. That is, those who smoked light cigarettes absorbed a smaller amount of nicotine, but, again, the amount absorbed was not equal to the difference in nicotine yields on the packages. These associations are evident in the ratios of means shown in Table 3 . Significantly high values of the intercept in the regression models in Table 2 also provide an explanation for the insufficient decrease in urinary cotinine compared with the decrease in nicotine yield on the packages. We determined the full FTND score in each of our subjects, although some components of FTND, including the number of cigarettes consumed and the time from awakening until the first cigarette, were measured in the previous study of smoking in Japan [ 11 ]. It has not been previously reported that smokers with a strong dependency on nicotine showed constantly high levels of urinary cotinine regardless of nicotine yield of the cigarette brands they consumed. Moreover, we recruited a larger number of smokers of cigarettes yielding 0.1 mg nicotine than the previous report [ 11 ]. We were thus clearly able to show associations among nicotine dependency, the machine yield of nicotine, and urinary cotinine concentration. Smokers heavily dependent on nicotine obtained no advantage by smoking low-yield cigarettes. Moreover, they may actually increase their risk due to compensatory behavior, for example, by inhaling more carbon monoxide or other harmful substances contained in cigarette smoke. Our results suggest that tobacco industry advertising may have led these smokers, especially those heavily dependent on nicotine, to underestimate the health risks posed by low-yield cigarettes. This is similar to the results of other studies, which suggested that 'Light' or 'Ultra Light' cigarettes could deliver as much tar and nicotine as 'Regular' cigarettes [ 6 - 11 ]. The compensation mechanisms that may keep blood nicotine at a high level include more puffs per cigarette, greater volume per puff, and greater depth of inhalation, all of which may be conscious or unconscious on the part of smokers. In addition, the filters of low-yield cigarettes are sometimes treated with ammonium to increase absorption of nicotine, thus eliminating the need for deep inhalation [ 5 ]. These filters may also be processed to reduce throat irritation, for example in mentholated cigarettes, so that smokers do not realize that are inhaling deeply or frequently [ 5 , 29 ]. This mechanism may increase the inhalation volume, and consequently increase the absorption of carbon monoxide or other harmful substances. In our results, the urinary cotinine concentration did not differ according to self-reported inhalation pattern, suggesting that smokers regulate nicotine intake without being aware of their inhalation patterns. In this study, mentholated cigarettes were not consumed by a sufficient number of subjects for examination. Cigarette brand, however, showed no apparent difference in urinary cotinine level. Another important mechanism by which smokers of low-yield cigarettes increase their nicotine intake is by blocking the ventilation holes on the filter with their fingers or lips while holding the cigarette and smoking [ 29 ]. These holes are made to inspire fresh air and dilute the smoke. During measurement of nicotine yield by the FTC method, however, these holes are not blocked [ 17 ]. The tobacco companies have admitted that they have known of the relationship between the nicotine yield reported on the packages and the actual quantities of tobacco smoke components inhaled [ 30 ]. Low-yield brands appear to mislead smokers who want to avoid health risks without quitting smoking. Smokers who had not previously considered quitting smoking, however, were found to begin to consider quitting after learning that low-yield cigarettes are processed to make them less irritating and that they did not reduce health risks [ 31 , 32 ]. Most smokers of low-yield brand should be informed of these findings. The subjects of this study were participants in a so-called 'human dry dock,' a detailed health check-up system for middle-aged and elderly people, which started in Japan in the 1950's and currently enroll about 10 million [ 33 ]. Most of them are socioeconomically well off, because each subject pays seven or eight thousand yen ($70–80) on average (up to about forty thousand yen) out-of-pocket for this check-up, and most subjects get annual check-ups. Most participants in this system are health-conscious and have some knowledge about health hazards of smoking. It is likely, therefore, that the subjects of our study are similar with respect to those characteristics, although their occupation, education level, and socioeconomic state were not surveyed. Moreover, in the health check-up associated with this study, many smokers were advised to quit smoking [ 34 ], which may explain the relatively low smoking rate of male candidates for this study (32%), compared with an average of 50% or higher in the general community. Female participants were not examined in this study, mostly because smoking rate of females in our health check-up clinic is less than 10% and some participants refused to answer questions about smoking. However, the advertisement of 'light' cigarettes seems to target young people, especially young women [ 20 ], and the smoking rate among young women is increasing in Japan [ 13 ]. Thus further studies focused on young people, particularly young females, would provide important information. In conclusion, we have shown here that the difference in intake of nicotine into a smoker's body was smaller than the difference in machine-measured nicotine yield among cigarette brands. Smokers consuming cigarettes with a low nicotine yield did not reduce actual intake of nicotine to the level that they expected. This was especially true for smokers with heavy nicotine dependence. This result should be emphasized in public health messages to smokers as well as to young people likely to start smoking. Abbreviations FTND: Fagerström Test of Nicotine Dependence, FTC: Federal Trade Commission Competing interests None declared. Authors' contributions AN participated in the design of the study, collected the data, performed the statistical analysis, drafted the manuscript, and was the principal investigator. MNS conceived of the study, participated in its design, and collected the data. KO supervised the data analysis and the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC493271.xml

526390

A further critique of the analytic strategy of adjusting for covariates to identify biologic mediation

Background Epidemiologic research is often devoted to etiologic investigation, and so techniques that may facilitate mechanistic inferences are attractive. Some of these techniques rely on rigid and/or unrealistic assumptions, making the biologic inferences tenuous. The methodology investigated here is effect decomposition : the contrast between effect measures estimated with and without adjustment for one or more variables hypothesized to lie on the pathway through which the exposure exerts its effect. This contrast is typically used to distinguish the exposure's indirect effect, through the specified intermediate variables, from its direct effect, transmitted via pathways that do not involve the specified intermediates. Methods We apply a causal framework based on latent potential response types to describe the limitations inherent in effect decomposition analysis. For simplicity, we assume three measured binary variables with monotonic effects and randomized exposure, and use difference contrasts as measures of causal effect. Previous authors showed that confounding between intermediate and the outcome threatens the validity of the decomposition strategy, even if exposure is randomized. We define exchangeability conditions for absence of confounding of causal effects of exposure and intermediate, and generate two example populations in which the no-confounding conditions are satisfied. In one population we impose an additional prohibition against unit-level interaction (synergism). We evaluate the performance of the decomposition strategy against true values of the causal effects, as defined by the proportions of latent potential response types in the two populations. Results We demonstrate that even when there is no confounding, partition of the total effect into direct and indirect effects is not reliably valid. Decomposition is valid only with the additional restriction that the population contain no units in which exposure and intermediate interact to cause the outcome. This restriction implies homogeneity of causal effects across strata of the intermediate. Conclusions Reliable effect decomposition requires not only absence of confounding, but also absence of unit-level interaction and use of linear contrasts as measures of causal effect. Epidemiologists should be wary of etiologic inference based on adjusting for intermediates, especially when using ratio effect measures or when absence of interacting potential response types cannot be confidently asserted.

1. Introduction A large portion of epidemiologic research is devoted to etiologic investigation, and so techniques that may facilitate mechanistic inferences are sought by researchers and are applied frequently in their work. Unfortunately, some of these techniques have been found to provide far more ambiguous evidence on which to base mechanistic conclusions than was first believed. For example, analysis of patterns of joint effects has been proposed as a means of identifying causal structure [ 1 ], but simple counterexamples show that in general the underlying etiologic model cannot be readily identified[ 2 ]. Typically, some method is proposed under a sound theoretical argument in a specific analytic setting, but this method is subsequently applied in a more general context in which those specific theoretical conditions no longer hold. For example, Greenland and Poole[ 3 ] provide a rational justification for deviation from additive joint effects as the benchmark for identifying mechanistic interaction between two factors [[ 4 ], pp. 332–339]. But this argument is not generally valid as is often assumed; it doesn't hold for all causal structures and target populations[ 5 ]. The list of such untenable overgeneralizations in epidemiologic practice is surely large and varied, and has led to any number of false conclusions and misunderstandings. We describe here one particular epidemiologic technique that is applied frequently in practice, and yet is invalid in all but a surprisingly narrow range of circumstances. It is a remarkable example in that the analytic strategy is exceedingly common, and yet is described infrequently in epidemiologic texts or methodologic articles. The few textbook citations that do exist provide no formal justification, and therefore there is little guidance available from within the sources in our field to guide users and warn them of important limitations of this approach. This situation motivates the present article, in which we will show that although widely applied, this analytic approach is almost never justifiable on the basis of reasonable assumptions about the data. The methodologic approach of interest in this article is the decomposition of effects purportedly accomplished by contrasting two adjusted effect estimates for the exposure of interest: an estimate adjusted for potential confounders, and an estimate adjusted for the same set of potential confounders plus one or more additional variables hypothesized to be causal intermediates, i.e., to lie on pathway(s) through which the exposure exerts its effect. This contrast is then typically used to distinguish the exposure's indirect effect, through the specified intermediate variables, from its direct effect, transmitted via pathways that do not involve the specified intermediate variables. If control of hypothetical causal intermediates greatly attenuates an exposure's estimated effect, it is generally inferred that the exposure's effect is mediated primarily through pathways involving these quantities; a small degree of attenuation is interpreted as evidence that other pathways predominate. These mechanistic inferences then inform policy recommendations concerning the utility of potential interventions. Although this effect decomposition approach is quite common in the epidemiologic literature, its general validity has not been adequately investigated. This analytic strategy for effect decomposition in epidemiologic research is recommended by Susser [[ 6 ], pp. 121–124], and more recently by Szklo & Nieto [[ 7 ], pp. 184–187]. The latter authors quantify the degree of mediation as follows: " The degree to which a given mechanism...explains the relationship of interest is given by the comparison of adjusted (A) and unadjusted (U) measures of association (e.g., a relative risk, RR). This comparison can be made using the ratio of the unadjusted RRs, RR U /RR A , or the percent excess risk explained by the variables adjusted for: ". Calculations similar to this "% Excess Risk Explained" are the most common framework for describing the effect decomposition analysis in epidemiologic research. For example, a study by Heck and Pamuk investigated the relation between education and postmenopausal breast cancer incidence using data from the National Health and Nutrition Examination Survey I Epidemiologic Follow-up Study[ 8 ]. Proportional hazards modeling was used to estimate the relation between breast cancer incidence and education level. The authors then reported that reproductive factors including nulliparity were found to mediate this relation. This assertion was based on the observation that adjustment for these factors reduced the magnitude of the positive relation between education level and risk of postmenopausal breast cancer. Furthermore, because the association between exposure and outcome was no longer statistically significant after adjustment for the putative mediators, the authors concluded that "the association between higher education and increased risk of breast cancer appears to be largely explained by differences in the known risk factors for breast cancer" [[ 8 ], p. 366]. This methodology is commonly applied, and therefore there are many similar examples in the published literature. On the basis of this approach numerous authors have made many similar mechanistic claims about mediation, for example, that blood pressure mediates the causal relation between homocysteine and cardiovascular risk[ 9 ], that behavioral risk factors mediate the causal relation between hostility and incident myocardial infarction[ 10 ], and that the protective effect of gene CCR5 heterozygosity on clinical AIDS occurrence is completely mediated through an effect on CD4 cell count[ 11 ]. The results of decomposition analyses are also frequently used to anticipate the impact of a potential intervention or policy related to the intermediate variable(s). For example, Lantz and colleagues adjusted for several measured behavioral intermediates in assessing the relation between income and mortality[ 12 ]. They noted that even after adjustment for these measured intermediates "the risk of dying was still significantly elevated for the lowest-income group (hazard rate ratio, 2.77; 95% CI, 1.74–4.42)..." and on this basis they offered the conclusion that "socioeconomic differences in mortality ...would persist even with improved health behaviors among the disadvantaged." [[ 12 ], p. 1703] In a seminal article on the topic, Robins & Greenland[ 13 ] employed a causal framework based on latent potential response types in order to describe the limitations inherent in the effect decomposition analysis. Subsequent authors have for the most part focused on the Robins & Greenland finding that direct effect estimates, decomposed from the total effect by adjusting for an intermediate, may be biased if there is unmeasured confounding between the intermediate and the outcome [ 14 - 16 ]. As Robins & Greenland showed and these later authors reiterated, the decomposition strategy may fail even when the total effect is unconfounded. While this consideration is important, this is not our concern in the present discussion. Rather, we will show that even when there is no confounding of any relevant causal effect, the decomposition strategy will still generally fail, in the sense that a contrast such as that described above as the "% Excess Risk Explained" will fail to provide an unbiased estimate of the proportion of the causal effect that is relayed through the intermediate. Our critique would appear to contradict standard practice in the social sciences, in which decomposition analysis is also commonly applied[ 17 , 18 ]. We suggest two explanations for this state of affairs. The first is that the development of the decomposition methodology by Wright[ 19 ] and other pioneering social science statisticians did not make use of an explicitly casual framework, but rather was derived algebraically from linear regression theory. One consequence is that the causal assumptions necessary for the model to be substantively meaningful were not readily apparent until the advent of a notational system for potential outcomes[ 20 ]. Secondly, we suggest that this is another example in which unwitting users have extrapolated a technique beyond the strictly defined original set of assumptions without assessing the impact of this extrapolation on the validity of the estimation. In this case, assumptions imposed in the original development of the decomposition methodology involved additivity of effects and linear contrast measures, neither of which are typical of the analysis of discrete events, such as occurrence of disease. Epidemiologists as well as others have generally been remiss in failing to attend to these crucial assumptions when applying these techniques more broadly. However, as we will describe, the causal assumptions required for the validity of the decomposition method are not verifiable from observed data, and furthermore are unrelated to any typical substantive knowledge. It may therefore be essentially impossible to apply this methodology with any confidence in a real-world analysis of data. 2. Framework, Notation and Causal Structure For clarity, we limit our exposition to the simplest possible decomposition problem, which is the structure that includes three measured binary variables and sample size sufficiently large to justify the assumption of zero sampling error (see endnote 1). The three variables are designated as X, Y and Z. The causal relationships between these nodes are described by the directed acyclic graph (DAG)[ 21 ] in Figure 1 . X is a randomly assigned (i.e., exogenous) treatment and therefore there are no arrowheads terminating at this node in the graph. X takes the value of 1 if treated, 0 otherwise. Y equals 1 if the outcome occurs, and 0 otherwise. Z takes the value of 1 if the intermediate occurs, and 0 otherwise, and like X is manipulable (i.e., may be fixed through external intervention to take either level). The framework adopted here is a deterministic counterfactual model in which each individual unit in the population is assumed to have a fixed potential response to each possible input pattern at each endogenous node of the DAG. As such, the observed data reveal only a subset of these fixed potential responses. We also assume that the potential responses of each unit do not depend on the treatments assigned the other units, which is referred to by Rubin as the "stable-unit-treatment-value assumption" (SUTVA)[ 22 ]. Figure 1 Decomposition of Total Effect of X on Y into Direct and Indirect Effects. The total average causal effect (ACE) of X on Y is achieved through two pathways, one which is termed "indirect'' because it operates through measured intermediate variable Z, and another that is termed "direct'' because it operates through no measured intermediates. The potential response variable for unit u at node Z is denoted by Z ux where index u identifies the individual unit and index x specifies the X value factually or counterfactually experienced by that unit. Given the deterministic model at the individual unit level, there are four possible patterns of response Z ux to input x that unit u can exhibit, and these have received various appellations in the literature, such as "doomed" for Z ux = 1 regardless of x, "causal" for Z ux = x, "preventive" for Z ux = 1-x, and "immune" for Z ux = 0 regardless of x[ 23 ]. These four patterns may be represented by potential response type index values of 1, 2, 3, and 4, respectively, such that each unit in the population is classified by one of these four index values. At the endogenous node Y, the counterfactual or potential response variable for unit u is denoted by Y uxz , where u identifies the individual unit and indices x and z specify the X and Z inputs to that unit. Conditional on the individual unit and conditional on the z input, there are four possible patterns of response Y uxz to input x: Y uxz = 1 regardless of x, Y uxz = x, Y uxz = 1-x, and Y uxz = 0 regardless of x. Therefore, each unit can be fully characterized by one of the 4 × 4 × 4 = 64 possible values of three indices, {ijk}, where index i specifies the Z ux response, index j specifies the Y ux0 response, and index k specifies the Y ux1 response. As an illustration, {123} refers to a unit in which Z will equal 1 regardless of the value taken by X. Under this naturally occurring outcome for Z, Y will equal 1-x. However, if Z were to be manipulated by external intervention to equal 0, then Y would equal x. In this way, the 64 possible potential response types for individual units in the population are symbolized by {ijk}; i = 1,...,4; j = 1,...,4; k = 1,...,4. We define q ijk to be the proportion of type {ijk} in the total population. Furthermore, because X is exogenous, the potential response types occur in these same proportions inboth X = 0 and X = 1 subpopulations. The set of all 64 q ijk proportions determines the causal behavior of the population in the context of the three observed variables (X, Y, Z) and potential confounding of the causal effects between them. The values of the 64 q ijk proportions, however, are not identified from the 8 observed proportions in the study population: Pr(Y = y, X = x, Z = z); x = 0,1; y = 0,1; z = 0,1. We make a further simplifying assumption of (strong) monotonicity for the remainder of this paper (see endnote 2). This assumption states that there are no individuals who exhibit preventive effects at either endogenous node. That is, for all units u and for z = 0,1 and x = 0,1: Z u0 ≤ Z u1 Y u0z ≤ Y u1z Y ux0 ≤ Y ux1 Since the binary values can be arbitrarily coded, the monotonicity of effects can be in any direction (i.e., preventive or causative, since reversing the coding is equivalent to interchangebetween subscript values 2 and 3 and between subscript values 1 and 4). This assumption reduces the number of potential response types in the population from 64 to 18, and may be reasonable on substantive grounds. For example, consider X to be assignment to cholesterol lowering drug cholestyramine versus placebo in the Lipid Research Clinics (LRC) Primary Prevention Trial[ 24 ], Z = 1 to be absence of hypercholesterolemia one year after initiation of the cholestyramine, and Y = 1 the absence of coronary heart disease (CHD) at follow-up. In this example, there are no individuals for whom assignment to cholestyramine (X = 1) will cause hypercholesterolemia (Z = 0), nor individuals for whom assignment to cholestyramine or absence of hypercholesterolemia will cause CHD (Y = 0). Note that monotonicity eliminates not only types {3jk}, {i3k} and {ij3}, but also types {i12}, {i14} and {i24}. This is why the assumption reduces the potential outcome patterns not merely to 3 × 3 × 3 = 27, but rather to (3 × 3 × 3)- (3 × 3) = 18. Complete descriptions of the 18 potential outcome types that occur under monotonicity are provided in the first seven columns of Table 1 . Table 1 Potential Response Type Characteristics Under Monotonicity Assumption (18 Response Types) a b c d Response of Y to fixing X to value: Response of Y to fixing X and Z to values: Contributes to: Potential Response Type Representation † X = 1 X = 0 X = 1 Z = 0 X = 0 Z = 0 X = 1 Z = 1 X = 0 Z = 1 Total Effect Direct Effect in (Z-stratum) Indirect Effect in (Z-stratum) {111} 1 1 1 1 1 1 {141} 1 1 0 0 1 1 {211} 1 1 1 1 1 1 {122} 1 0 1 0 1 0 + + (0,1) {241} 1 0 0 0 1 1 + + (0,1) {222} 1 0 1 0 1 0 + + (0,1) {411} 1 1 1 1 1 1 {422} 1 0 1 0 1 0 + + (0,1) {144} 0 0 0 0 0 0 {244} 0 0 0 0 0 0 {441} 0 0 0 0 1 1 {444} 0 0 0 0 0 0 {121}* 1 1 1 0 1 1 + (0) {221}* 1 0 1 0 1 1 + + (0) + (1) {421}* 1 0 1 0 1 1 + + (0) {142}* 1 0 0 0 1 0 + + (1) {242}* 1 0 0 0 1 0 + + (1) + (0) {442}* 0 0 0 0 1 0 + (1) * Unit-level interaction (interdependence) present because ( a-b ) ≠ ( c-d ) † Potential response type representation indices are: 1 = "doomed", 2 = "causal" and 4 = "immune" Index i of the {ijk}representation specifies the Z [X = x] response, index j specifies the Y [X = x; Z = 0] response (columns a and b ), and index k specifies the Y [X = x; Z = 1] response (columns c and d ). 3. Definitions of Causal and Associational Parameters of Interest The total average causal effect (ACE) of the treatment X in the population is the proportion of all individuals in the population who would experience outcome Y if they were treated, but not if they were untreated, without regard to Z. Given the monotonicity assumption, this effect is the sum of 8 of the 18 potential response type proportions in the population: ACE [X→Y] = average causal effect = Pr(Y = 1|SET[X = 1]) - Pr(Y = 1|SET[X = 0]) = (q 122 + q 241 + q 222 + q 421 + q 422 + q 221 + q 142 + q 242 ) The average causal (controlled) direct effect (ACDE) of the treatment X in the population is the proportion of individuals who would experience outcome Y if they were treated, but not if they were untreated, if Z were forced (SET) to have a specific value z (thus blocking any indirect effects). In general, there is no reason for this effect to take the same value if Z were forced (SET) to 0 as it would take if Z were forced (SET) to 1, and so for binary Z in our DAG there are two distinct average causal direct effects. Given the monotonicity assumption, these effects are the sums of 6 of the 18 potential response type proportions in the population: ACDE [X→Y] | SET[Z = 0] = average causal direct effect for Z forced (SET) to 0 = Pr(Y = 1|SET[X = 1,Z = 0]) - Pr(Y = 1|SET[X = 0,Z = 0]) = (q 122 + q 222 + q 422 + q 121 + q 221 + q 421 ) ACDE [X→Y] | SET[Z = 1] = average causal direct effect for Z forced (SET) to 1 = Pr(Y = 1|SET[X = 1,Z = 1]) - Pr(Y = 1|SET[X = 0,Z = 1]) = (q 122 + q 222 + q 422 + q 142 + q 242 + q 442 ) A manipulative definition of the total average causal indirect effect, ACIE [X→Y] , is not straightforward, and some authors assert that no general definition exists [e.g., [ 21 ], p. 165]. The usual interpretations granted to applications of effect decomposition methodology imply that analysts take ACIE [X→Y] to mean the proportion of all individuals who would experience outcome Y if they were treated, but not if they were untreated, but only via the pathway in which X has an effect on Z and then Z has an effect on Y. In this causal mechanism, therefore, external intervention to hold Z fixed will prevent X from having any effect on Y. Of the 18 potential response types that exist under the monotonicity assumption, clearly {241} corresponds to this conceptual definition. In units of this type, Z = X. But were Z to be blocked from occurring (i.e., SET to Z = 0) by external intervention, then Y = Z = 0, regardless of X. Alternatively, if Z were to be forced to occur (i.e. SET to Z = 1) by external intervention, then Y = Z = 1. For potential response types {242} and {221}, however, the common-sense meaning of an indirect effect may also apply, depending on the specific intervention applied to Z. Specifically, if the external intervention on the intermediate is SET[Z = 0], then potential response type {242} is an indirect type, whereas if the external intervention on the intermediate is SET[Z = 1], then potential response type {221} is an indirect type (Table 1 ). This is the ambiguity that has made it difficult to provide a general manipulative definition of the ACIE [X→Y] without prohibiting these interacting types, as we do in Section 5. We can also define the value of the total average causal effect (ACE) of the treatment X on the intermediate covariate Z, which is the proportion of individuals who would experience intermediate Z if they were treated, but not if they were untreated. Given the monotonicity assumption, this effect is the sum of 6 of the 18 potential response type proportions in the population: ACE [X→Z] = average causal effect of X on Z = Pr(Z = 1|SET[X = 1]) - Pr(Z = 1|SET[X = 0]) = (q 211 + q 241 + q 222 + q 244 + q 221 + q 242 ) Because the value of Y is determined through the joint effects of X and Z, it is also possible to define the effect of Z on Y as the proportion of individuals who would experience outcome Y if Z were forced (SET) to 1, but not if Z were forced (SET) to 0, conditional on X = x. In general, there is no reason for this effect to take the same value in the X = 0 subpopulation as it does in the X = 1 subpopulation, and so for binary X in our DAG there may be two distinct effects of Z on Y given strata of X. Given the monotonicity assumption, these effects are the sums of 6 of the 18 potential response type proportions in the population: ACE [Z→Y] | X = 0 = Average causal effect of Z on Y in the X = 0 stratum = Pr(Y = 1|SET[Z = 1], X = 0) - Pr(Y = 1|SET[Z = 0], X = 0) = (q 141 + q 241 + q 441 + q 121 + q 221 + q 421 ) ACE [Z→Y] | X = 1 = Average causal effect of Z on Y in the X = 1 stratum = Pr(Y = 1|SET[Z = 1], X = 1) - Pr(Y = 1|SET[Z = 0], X = 1) = (q 141 + q 241 + q 441 + q 142 + q 242 + q 442 ) Recall that because X is randomized, the potential response type distributionsare independent of X, meaning that the proportions over the total population are the same within the X = 1 and X = 0 subpopulations. We can define ACE [Z→Y] , the effect of Z on Y unconditionally, as the proportion of individuals who would experience outcome Y if Z were forced (SET) to 1, but not if Z were forced (SET) to 0, over the entire population. As this is not a stratum-specific quantity, there is only a single value, although this depends on the marginal distribution of X in the population [[ 21 ], eq 3.19]. By definition: ACE [Z→Y] = Pr(Y = 1|SET[Z = 1]) - Pr(Y = 1|SET[Z = 0]) = Pr(Y = 1, X = 1|SET[Z = 1]) - Pr(Y = 1, X = 1|SET[Z = 0]) + Pr(Y = 1, X = 0|SET[Z = 1]) - Pr(Y = 1, X = 0|SET[Z = 0]) Given that X is not affected by Z in the specified DAG, this can be re-written as: (Pr(Y = 1|X = 1,SET[Z = 1]) - Pr(Y = 1|X = 1,SET[Z = 0]) )Pr(X = 1) + (Pr(Y = 1|X = 0,SET[Z = 1]) - Pr(Y = 1|X = 0,SET[Z = 0]) )Pr(X = 0) = Pr(X = 1) ACE [Z→Y] | X = 1 + Pr(X = 0) ACE [Z→Y] | X = 0 As shown above, the ACE [Z→Y] | X = x terms are each comprised of the sums of 6 of the 18 potential response type proportions in the population, 3 of which are common across the two strata of X and 3 of which are unique to one or the other stratum, so that ACE [Z→Y] involves a weighted sum of 9 of the 18 potential response type proportions, with weights dependent upon the marginal distribution of X. The causal effects defined above are counterfactual, in that they involve hypothetical manipulation of the treatment or intermediate or both. The realized data are the risks that arise in the form of observed proportions in the population. We define R xz as the risk (proportion) of Y = 1 among those with X = x and Z = z, i.e., Pr(Y = 1|X = x, Z = z). With binary variables, exogeneity of X and the monotonicity assumption, these observable quantities are related to the latent response type proportions as follows: The observed risk values R xz are used to compute the associational estimates of effect (see endnote 3), as follows: The risk difference RD [X→Y] = R 1• - R 0• is the associational estimate of the total average causal effect of X on Y on the additive scale, where R x• indicates the risk under X = x collapsed over levels of Z, i.e., R x• = Pr(Z = 0|X = x)R x0 + Pr(Z = 1|X = x)R x1 . Because X is assumed to be randomized and therefore Pr(Z = z|X = x) = Pr(Z = z), RD [X→Y] equals the causal RD Pr(Y = 1|SET[X = 1] - Pr(Y = 1|SET[X = 0]. The direct risk difference DRD [X→Y] | Z = z = R 1z - R 0z is the associational estimate on the additive scale of the average causal direct effect of X on Y within the Z = z stratum. DRD [X→Y] | Z = z may be a biased estimate of the analogous causal quantity, i.e., DRD [X→Y] | Z = z is not necessarily equal to Pr(Y = 1|SET[X = 1,Z = z]) - Pr(Y = 1|SET[X = 0,Z = z]). RD [X→Z] = Pr(Z = 1|X = 1) - Pr(Z = 1|X = 0) is the associational estimate on the additive scale of the effect of X on Z. Because X is assumed to be randomized, however, this associational estimate equals the analogous causal quantity Pr(Z = 1|SET[X = 1]) - Pr(Z = 1|SET[X = 0]). RD [Z→Y] | X = x = R x1 - R x0 is the associational estimate on the additive scale of the effect of Z on Y within stratum X = x. Because randomization of X does not imply that the effect of Z on Y is unconfounded, it may be a biased estimate of the analogous causal quantity, i.e., RD [Z→Y] | X = x is not necessarily equal to Pr(Y = 1|SET[Z = 1], X = x) - Pr(Y = 1|SET[Z = 0], X = x) sRD [Z→Y] = Pr(X = 1)RD [Z→Y] | X = 1 + Pr(X = 0)RD [Z→Y] | X = 0 is the associational estimate on the additive scale of the effect of Z on Y standardized to the distribution of X. The associational estimate for the indirect effect is generally computed by one of two methods[ 25 ]: 1) by subtracting DRD [X→Y] | Z = z from RD [X→Y] , or 2) by multiplying RD [X→Z] by sRD [Z→Y] The first of these methods is the one more commonly applied in epidemiologic research, as represented for example by the expression for "Excess Risk Explained" in Szklo & Nieto[ 7 ]. Subtraction of DRD [X→Y] | Z = z from RD [X→Y] is also recommended in the social sciences methodology literature. For example, Stolzenberg writes: "Once the total and direct effects are calculated, indirect effects may be computed merely by subtracting the direct effect of an antecedent variable from its total effect. This subtraction procedure is applicable both to linear additive models ... and to nonlinear / nonadditive models." [[ 26 ] p. 483]. The second method follows from the path analysis rules of Wright [ 19 ], and this method also appears in the epidemiologic literature [e.g., [ 27 ]]. There will generally be two distinct estimates by the first method above, depending the level chosen for Z. This is a necessary consequence of the manipulative definition of the controlled direct effect, since it involves deactivation of the indirect pathway by preventing variation in Z, and so if Z is to be fixed to a unique value, this value must be specified. In the second method shown above, however, only the two components of the indirect pathway are involved, and so no explicit fixing of Z is specified. This leads to a single estimate of the direct effect, and so the two methods can only be consistent with one another when there is homogeneity of the ACDE over strata of Z (i.e., DRD [X→Y] | Z = 0 = DRD [X→Y] | Z = 1 ). The usual regression-based approach for the second method involves the regression of Z on X, followed by the regression of Y on both X and Z, and finally the multiplication of the coefficient estimate for X in the first model by the coefficient estimate for Z in the second model. This conditional estimation of the Z→Y effect in the second model is analogous to taking a weighted average over stratum-specific values as we have done for sRD [Z→Y] . Any presumed equivalence of the two approaches shown above by virtue of a homogeneity assumption for the stratum-specific estimates would often be unwarranted, as even under monotonicity it would generally require the additional restrictions that q 142 = q 242 = q 442 = q 121 = q 221 = q 421 = 0 (see endnote 4). 4. Absence of Confounding Since X is randomized in Figure 1 , there is no confounding of ACE [X→Z] or ACE [X→Y]. Furthermore, in our examples we wish to examine the most optimistic scenario in which there is no confounding between Z and Y. We therefore need to formally define exchangeability conditions that imply the absence of confounding. These conditions are a generalization of those provided in Robins and Greenland [[ 13 ], eq E1 and E2, p. 149]. We define 4 counterfactual parameters. The first two are the risks of outcome Y among those with X = x and Z = 1 that would have been observed had Z been forced (SET) to take the value 0 rather than the actually occurring value 1. The second set of counterfactual parameters are the risks of outcome Y among those with X = x and Z = 0 that would have been observed had Z been forced (SET) to take the value 1 rather than the actually occurring value 0. It is now possible to specify exchangeability conditions that are sufficient to guarantee that there is no confounding, meaning that associational measures and causal effects are equivalent[ 28 ]. The four equality conditions that guarantee the absence of confounding between the Z and Y nodes of the DAG are: R 11|SET[Z = 0] = R 10 R 01|SET[Z = 0] = R 00 R 10|SET[Z = 1] = R 11 R 00|SET[Z = 1] = R 01 These conditions assert that the risk that is observed among those with X = x and Z = 0 is the same risk that would have been observed among those with X = x and Z = 1 had Z been forced (SET) from 1 to 0, and that the risk that is observed among those with X = x and Z = 1 is the same risk that would have been observed among those with X = x and Z = 0 had Z been forced (SET) from 0 to 1. The first two sets of exchangeability conditions imply that DRD [X→Y] | Z = 0 is unconfounded, whereas the latter two sets of exchangeability conditions imply that DRD [X→Y] | Z = 1 is unconfounded. Under the general scenario in which stratum-specific direct effects may differ, all four conditions are needed to guarantee that there is no confounding in either stratum of Z for any arbitrary choice of effect contrast that may be constructed from the four component risks. 5. Example 1: A Restriction that Permits Valid Effect Decomposition We now impose an additional restriction which, as we will see in Section 6, is necessary for the general validity of the decomposition strategy described above. This restriction is that for no individual in the population may there exist both a causal effect of X on Y and a causal effect of Z on Y. For this condition to hold in general, it must be the case that Z does not modify the effect of X for any unit. This restriction, which can be characterized as the absence of unit-level synergism or interaction, implies homogeneity of the stratum specific direct effects ACDE [X→Y] | SET[Z = 0] and ACDE [X→Y] | SET[Z = 1] . Under the monotonicity assumption, this requires that 6 of the 18 types, namely {142}, {242}, {442}, {121}, {221}, and {421} be absent from the population. For example, potential response type {242} refers to a unit in which Z will equal x. When X = Z = 1, outcome Y will occur (Y = 1), and when X = Z = 0, outcome Y will not occur (Y = 0), leading to a unit-level casual effect of X on Y equal to (1-0) = 1. However, if Z were to be manipulated by external intervention (SET) to equal 0, then the unit-level effect of X on Y becomes (0- 0) = 0. That is to say, there is no direct effect at this controlled level of Z. In contrast, if Z were to be manipulated by external intervention (SET) to equal 1, the unit-level effect of X on Y remains (1-0) = 1. The direct causal effects of X on Y are heterogeneous for these 6 omitted potential response types because there is unit-level interaction; the value obtained by Y depends not only on the value taken by X, but also on the level to which Z is held by external manipulation. Homogeneity of the stratum-specific direct effects ACDE [X→Y] | SET[Z = 0] and ACDE [X→Y] | SET[Z = 1] also corresponds to absence of effect measure modification on the additive scale. When such unit-level synergism is prohibited, then it becomes possible to state an unambiguous definition of the average causal indirect effect (ACIE) of the treatment X in the population as the proportion of all individuals who would experience outcome Y if they were treated, but not if they were untreated, by virtue of the effect that X has on Z and then the effect that Z has on Y. In this mechanism, therefore, external intervention to hold Z fixed will prevent X from having the effect on Y, regardless of the specific value to which Z is SET. Given the restrictions, the average causal indirect effect is merely a single potential response type proportion in the population: ACIE [X→Y] = q 241 . The total effect is indeed decomposable into the sum of direct and indirect effects under this restriction, and in the absence of confounding may be estimated without bias. The decomposition is valid because the ACE [X→Y] reduces to the sum of only 4 proportions (i.e., q 122 , q 222 , q 422 , and q 241 ), since 4 of the previous 8 are restricted to be 0 (i.e., q 221, q 421, q 142 and q 242 ). The average causal direct effects (ACDE) of X on Y are the sums of 3 potential response type proportions (rather than 6), which are identical in the two strata: ACDE [X→Y] | SET[Z = 0] = ACDE [X→Y] | SET[Z = 1] = (q 122 + q 222 + q 422 ). Likewise, ACE [X→Z] is the sum of four proportions, ACE [Z→Y] = ACE [Z→Y] | SET[X = 0] = ACE [Z→Y] | SET[X = 1] is the sum of three proportions, and the observed risks R XZ are similarly restricted by deleting the prohibited interacting potential response types from the quotients shown above (Table 1 ). Consider data arising from a population of unit-level potential response type proportions q ijk as shown in Table 2 . This population satisfies the restrictions of monotonicity and absence of unit-level synergism. Simple addition of the proportions yields observed risks R XZ of R 11 = 0.9170, R 01 = 0.5377, R 10 = 0.6101 and R 00 = 0.2307. These observed risk values then determine the various associational estimates of effect. The total effect RD [X→Y] = (0.8763- 0.3117) = 0.5646. The observed stratum-specific risk differences DRD [X→Y] | Z = 0 = DRD [X→Y] | Z = 1 = (R 1z - R 0z ) = 0.3793. Likewise the observed risk difference and stratum-specific risk differences for the effect of Z on Y are also homogeneous, sRD [Z→Y] = RD [Z→Y] | X = 0 = RD [Z→Y] | X = 1 = 0.3070. The observed effect of X on Z, RD [X→Z] = 0.6036. Table 2 Example with No Interaction Permitted Potential Response Type Representation Prevalence in the Population {111} 0.0609 {141} 0.0810 {211} 0.1100 {122} 0.0710 {241} 0.1853 {222} 0.2873 {411} 0.0599 {422} 0.0210 {144} 0.0510 {244} 0.0210 {441} 0.0407 {444} 0.0110 Furthermore, the data in this example are constructed such that there is no confounding of the relation between Z and Y. To verify this property, we use the proportions in Table 2 to calculate the values of the counterfactual risks that would be observed under interventions on the intermediate Z. These are: R 11|SET[Z = 0] = 0.6101; R 01|SET[Z = 0] = 0.2307; R 10|SET[Z = 1] = 0.9170; and R 00|SET[Z = 1] = 0.5377. The absence of confounding is implied by the following set of equalities, the first two of which imply an absence of confounding of DRD [X→Y] | Z = 0 and the latter two of which imply an absence of confounding of DRD [X→Y] | Z = 1 : R 11|SET[Z = 0] = R 10 = 0.6101 R 01|SET[Z = 0] = R 00 = 0.2307 R 10|SET[Z = 1] = R 11 = 0.9170 R 00|SET[Z = 1] = R 01 = 0.5377 In this example, which was constructed to have no confounding and in which potential response types corresponding to unit-level synergism have been eliminated, the associational estimates of the total and direct effects are unbiased. That is, the true total average causal effect equals the observed risk difference (0.5646) and the homogeneous average causal direct effects equal the stratum-specific risk differences (0.3793). It only remains to show that the indirect estimate is valid and that the total effect is decomposable. The true average causal indirect effect in the table is the single potential response type proportion, ACIE [X→Y] = (q 241 ) = 0.1853. As described above, there are two common approaches for estimating the associational measure of the indirect effect. The first is to subtract DRD [X→Y] | Z = z from RD [X→Y] , which in this case yields (0.5646- 0.3793) = 0.1853. The second is to multiply RD [X→Z] by sRD [Z→Y] , which yields (0.6036 × 0.3070) = 0.1853. The estimation of direct and indirect effects and their decomposition from total effects is valid, as will always be the case with this set of assumptions. The justification for this assertion is trivial: this set of assumptions is sufficient to guarantee that the true total ACE is the sum of three ACDE type proportions and one ACIE type proportion, whereas in general, without these restrictions, the ACDE is not constrained to be a subset of the total ACE. We have demonstrated a valid and unbiased estimation of the portion of a total effect that is transmitted through a specified intermediate when there is both absence of confounding and absence of unit-level interaction. We next relax this second constraint in order to demonstrate that the decomposition analysis can then fail. 6. Example 2: Removing the Restriction of No Unit-Level Interaction Now we relax one assumption, the prohibition of unit-level interaction, which was operationalized in Section 5 by requiring that q 142 = q 242 = q 121 = q 221 = q 421 = q 442 = 0. Therefore we have, under the monotonicity restriction alone, 18 potential response types in the population. Consider data arising from a population of unit-level potential response type proportions q ijk as shown in Table 3 , which conform to this assumption, and additionally are constructed such that there is no confounding. Simple addition of the proportions yields observed risks R XZ of R 11 = 0.9580, R 01 = 0.3910, R 10 = 0.4180 and R 00 = 0.3170. These observed risk values then determine the various associational estimates of effect. The associational estimate of the total ACE is RD [X→Y] = (0.8166- 0.3470) = 0.4696. The observed stratum-specific risk differences are no longer constrained to be homogeneous: DRD [X→Y] | Z = 0 = (R 10 - R 00 ) = 0.1010 and DRD [X→Y] | Z = 1 = (R 11 - R 01 ) = 0.5670. The observed stratum-specific risk differences for the effect of Z on Y similarly need not be homogeneous: RD [Z→Y] | X = 0 = 0.0740 and RD [Z→Y] | X = 1 = 0.5400. sRD [Z→Y] will now depend on the observed marginal distribution of X. If values were assigned with equal probability, then sRD [Z→Y] = (0.5 × 0.0740) + (0.5 × 0.5400) = 0.3070. The observed effect of X on Z, RD [X→Z] , equals 0.3327. Table 3 Example with Interaction Permitted Potential Response Type Representation Prevalence in the Population {111} 0.1285 {141} 0.0100 {211} 0.1100 {122} 0.0100 {241} 0.0100 {222} 0.0200 {411} 0.0785 {422} 0.0210 {144} 0.0100 {244} 0.0210 {441} 0.0040 {444} 0.0110 {121} 0.0200 {221} 0.0200 {421} 0.0100 {142} 0.2269 {242} 0.1517 {442} 0.1374 To verify that, as in the previous example, the data in this example are unconfounded, the proportions in Table 3 are used to determine the values of the counterfactual risks that would be observed under interventions on the intermediate Z. These are: R 11|SET[Z = 0] = 0.4180; R 01|SET[Z = 0] = 0.3170; R 10|SET[Z = 1] = 0.9580; and R 00|SET[Z = 1] = 0.3910. The absence of confounding is implied by the following set of equalities, the first two of which imply an absence of confounding of DRD [X→Y] | Z = 0 and the latter two of which imply an absence of confounding of DRD [X→Y] | Z = 1 : R 11|SET[Z = 0] = R 10 = 0.4180 R 01|SET[Z = 0] = R 00 = 0.3170 R 10|SET[Z = 1] = R 11 = 0.9580 R 00|SET[Z = 1] = R 01 = 0.3910 Because X is randomized, the average causal effect of X on Y is identified by the observed associational measure of effect: RD [X→Y] = ACE [X→Y] = 0.4696. Furthermore, because we have established that there is no confounding, the average causal direct effect of X on Y is identified by the observed associational measure of effect: ACDE [X→Y] | SET[Z = z] = DRD [X→Y] | Z = z . In this example, ACDE [X→Y] | SET[Z = 0] = DRD [X→Y] | Z = 0 = 0.1010 and ACDE [X→Y] | SET[Z = 1] = DRD [X→Y] | Z = 1 = 0.5670. However, in this scenario in which the only assumption we have relaxed is to allow the presence of unit-level synergism, the total average causal effect is no longer decomposable into direct and indirect effects. The average causal indirect effect no longer has single unambiguously true value. If the external manipulation contemplated is to SET[Z = 0], then ACIE [X→Y] = (q 241 + q 242 ) = (0.0100 + 0.1517) = 0.1617. On the other hand, if the external manipulation contemplated is to SET[Z = 1], then ACIE [X→Y] = (q 241 + q 221 ) = (0.0100 + 0.0200) = 0.0300. As described above, there are two common approaches for estimating the associational measure of the indirect effect. The first is to subtract DRD [X→Y] | Z = z from RD [X→Y] , which in this case yields either (0.4696- 0.5670) = -0.0974 or (0.4696-0.1010) = 0.3686, depending on the stratum of Z, neither of which equals either of the corresponding true values of 0.0300 or 0.1617. The second approach is to multiply RD [X→Z] by sRD [Z→Y] , which yields (0.3327 × 0.3070) = 0.1021, a value that equals neither of the corresponding true values of 0.0300 or 0.1617, nor is it the weighted average formed from any meaningful set of weights. In this scenario, the estimation of direct and indirect effects and their decomposition from total effects is not valid. It is immediately apparent that once unit-level interaction is permitted, there are potential response types that contribute to the ACDE but which do not contribute to the total ACE, making it incorrect to view the ACDE as a partition of the total ACE. Likewise, there are potential response types that contribute to the ACDE in one stratum of the intermediate, but which contribute to the ACIE in the alternate stratum, making it incorrect to view ACDE and ACIE as adding together to sum to a total effect. Therefore, for the decomposition methodology to be reliable, there must be both absence of confounding and absence of unit-level interaction. Because the absence of unit-level interaction would be difficult to assert with any confidence in a real-world application, the practical utility of decomposition as an analytic strategy is doubtful. 7. Discussion The demonstration above would appear to be somewhat gloomy as regards the potential for analytic epidemiology to identify biologic pathways through the contrast of variously specified statistical models. Indeed, the situation is even more grim than stated above, because even the optimistic scenario in Example 1 (Section 5) relies on the linear causal contrast estimator (i.e., the risk difference). Epidemiologic applications, such as those recommended by Szklo and Nieto [[ 7 ], pp. 184–187] nearly always use ratio measures of effects, such as risk ratios, odds ratios and hazard ratios. For ratio contrasts, the total effect is not generally decomposable under any set of causal assumptions. Therefore, the recommended "% Excess Risk Explained", defined as a function of ratio parameters, will never have a causal interpretation and the inference generated will always be ambiguous. In Example 1 (Section 5), for instance, the crude observed RR = 2.81, the Mantel-Haenszel adjusted RR = 2.40, and the Szklo and Nieto "% Excess Risk Explained" therefore equals 22.8%, which does not equal the true proportion of the effect that is relayed though the intermediate, i.e., (ACIE [X→Y] / ACE [X→Z] ) = (q 241 / q 211 + q 241 + q 222 + q 244 + q 221 + q 242 ) = (0.1853 / 0.5646) = 32.8%. We note that a valid contrast between total and direct effects for ratio measures of effect was described by Joffe & Colditz[ 29 ], but that this does not correspond to a decomposition of effects because the authors did not assume that the ACDE was necessarily a partition of the total ACE. Even if one steadfastly utilized the risk difference as the causal contrast and justified the no-confounding assumption, in order to reliably decompose the effect, one would still have to believe that there are no units in the population for whom Z and X both affect Y. Under the sharp null hypothesis for the exposure effect, this might be plausible. That is, if X has no effect on Y for any unit, then it follows that there are no units in the population for whom both X and Z have an effect on Y. However an average causal effect equal to the null does not imply this condition. If one were able to assert the no-confounding assumption, then observing that the association parameter is equal to the null would imply that the average causal effect is null, but no observation would imply the absence of unit-level interaction. The observation of heterogeneity would be sufficient to reject the assumption, but the observation of homogeneity would have no implications for this assumption. Nevertheless, as a practical matter, the incidental balancing out of unit-level causal effects leading to homogeneity might be considered unlikely, and therefore as a feasible approximation, the observation of risk difference homogeneity under a substantively defensible assertion of no-confounding might be taken as a setting in which effect decomposition can be attempted with a modicum of credence. Several previous authors working with latent potential response models have commented on the non-decomposibility of total effects into direct and indirect effects, most notably Robins[ 30 ], Robins & Greenland[ 13 ] and Pearl [[ 21 ], pp. 126–131, 165]. What is perhaps surprising is that although many quantitative sociologists also utilize this same latent outcomes framework [e.g., [ 20 , 31 , 32 ]], there are to our knowledge no instances of this critique published in the social sciences literature. We find this surprising because effect decomposition is formally embraced in the sociological methodology literature as an essential inferential strategy in the context of structural modeling[ 25 , 33 ]. Indeed, rather than critique this approach, it is strenuously upheld, even for non-linear models[ 26 ]. We note that our specification of average causal direct effects in this manuscript corresponds to the controlled direct effect, which is to say, the proportion of individuals who would experience outcome Y if they were treated, but not if they were untreated, if Z were to be fixed to have a specific value z (thus blocking any indirect effects). Rather than impose through external intervention a uniform value of Z = z for all units, it is possible to define the average causal direct effect of X on Y that results from fixing Z to the value that would naturally occur under a specific single value of X, for example the unexposed level X = 0. This is referred to by Robins as the "pure direct effect"[ 34 ] and by Pearl as the "natural direct effect"[ 35 ]. It is noteworthy that this alternate definition does allow for the effect decomposition to hold more generally, and gives rise to additional concepts such as the "total direct effect", which is the difference between the total ACE and the pure (natural) indirect effect. Furthermore, analogously to the controlled direct effects formulation, which leads to as many direct effects as there are levels of intermediate Z, the pure (natural) direct effects formulation leads to as many direct effects as there are levels of exposure X. Although it allows for decomposition without the assumption of no unit-level interaction, the approach involving pure (natural) direct and indirect effects has a substantial deficiency. The exchangeability conditions shown above (Section 4) characterize confounding in relation to hypothetical but defined manipulations of the target population. That is, X and Z are controlled to specific values. Because the pure (natural) direct and indirect effects are defined based on intermediate Z being manipulated to an unobserved value that it would have taken under an exposure X value that may not have occurred, the exact nature of this intervention remains obscure. Because the hypothetical manipulation cannot be specified, the decomposed effects no longer have any possible relevance to a specific public health intervention or policy [[ 34 ], Section 3]. For example, if one were to estimate that the pure (natural) indirect effect through intermediate Z equals 50% of the total effect, one could not infer that 50% of the outcomes attributable to the exposure could be prevented by blocking Z from occurring. Controlled direct effects can be used to make statements such as "The effect that postmenopausal hormone therapy would have on breast cancer risk if we were to persuade every woman to receive a screening mammography is ...." No similarly practical statement could be made for a pure (natural) direct effect, however, which would correspond to something like "The effect that postmenopausal hormone therapy would have on breast cancer risk if every woman were to engage in the screening mammography behavior that she would have exhibited under the absence of treatment is ...." This latter statement obviously has no clear public health policy implications, since it requires a policy of fixing the intermediate to different values, some of which are unobserved (e.g., the screening behavior that a woman taking postmenopausal hormone therapy would have experienced had she not taken hormone therapy). In summary, the ubiquitous strategy of adjusting for one or more putative causal intermediates in order to estimate the portion of the effect that it mediated by this pathway, in epidemiology and in other fields, lacks a reliable foundation. There are highly constrained sets of assumptions which allow this strategy to be valid, but it is often difficult to know when, if ever, these assumptions are approximately satisfied. Previous critiques have focused on confounding between the intermediate and the outcome, but we show that even when there is no confounding, the total causal effect of treatment is not generally decomposable into direct and indirect effects. Valid estimation of the direct or indirect effects, or of the proportion of the total effect that is due to an intermediate variable, requires not only the assumption of no confounding, but also the assumed absence of unit-level synergism, the latter of which may be particularly difficult to assert in a real-world analysis. Furthermore, even under these conditions, the decomposition is only valid for the difference contrast as the measure of causal effect, not for ratio measures of effect such as risk ratios, rate ratios, hazard ratios or odds ratios. In circumstances when it is possible to estimate (controlled) average causal direct effects, these should not generally be interpreted as portions of the total average causal effect, nor should they generally be used to make any statement about the proportion of the effect that is attributable to the measured intermediate variable. List of Abbreviations Used ACE average causal effect ACDE average causal (controlled) direct effect ACIE average causal (controlled) indirect effect CHD coronary heart disease DAG directed acyclic graph DRD direct risk difference LRC Lipid Research Clinics RD risk difference RR relative risk or risk ratio sRD standardized risk difference SUTVA stable-unit-treatment-value assumption Competing Interests The authors declare that they have no competing interests. Endnotes 1. Because of the assumption of zero sampling error "proportions" (in the observed sample) and "probabilities" (in the source population)are interchangeable. 2. The characterization of this monotonicity assumption as strong is intended to distinguish it from a weaker stochastic monotonicity assumption that may be defined: Pr(Z 0 = 1) ≤ Pr(Z 1 = 1) Pr(Y 0z = 1) ≤ Pr(Y 1z = 1); z = 0,1 Pr(Y x0 = 1) ≤ Pr(Y x1 = 1); x = 0,1 where Z x is the random variable representing the potential response at Z to SET[X = x], and Y xz is the random variable representing the potential response at Y to SET[X = x, Z = z]. 3. Associational estimates are obtained from contrasts in the observed data, rather than being estimates of what would pertain under the hypothetical manipulation that is indicated by a SET operation. 4. Strictly, equivalence of the two approaches for specifying the indirect effect requires merely that (q 142 + q 242 + q 442 ) = (q 121 + q 221 + q 421 ), but this incidental cancellation would be difficult to anticipate, whereas the absence of some potential outcomes types is a more plausible form of background knowledge that an investigator could bring to the analysis. Authors' Contributions JSK led the writing, but all three authors contributed heavily to editing and revision. RFM devised the notational system and the numeric examples.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526390.xml

549555

Do hormonal contraceptives stimulate growth of neurofibromas? A survey on 59 NF1 patients

Background Neurofibromas are benign tumors of the peripheral nerves and hallmark of neurofibromatosis type 1 (NF1), a tumor suppressor gene syndrome. Neurofibromas mostly start developing at puberty and can increase in size and number during pregnancy. Expression of progesterone receptors has been found in 75% of the tumors. Many female NF1 patients are thus concerned about the possibility that hormonal contraceptives may stimulate the growth of their neurofibromas. Methods A survey was carried out on 59 female NF1 patients who are practicing or have practiced hormonal contraception to examine the effect of the various contraceptives on the growth of neurofibromas. Results Majority (53 out of 58) of patients who received oral estrogen-progestogen or pure progestogen preparations reported no associated tumor growth. In contrast, significant tumor growth was reported by two patients who received depot contraceptive containing high dose of synthetic progesterone. Conclusions Oral contraceptives do not seem to stimulate the growth of neurofibromas in NF1 patients. High doses of progesterone might stimulate the growth of neurofibromas and deserve more caution.

Background Neurofibromatosis type 1 (NF1) is a genetic disorder with an incidence of about 1 in 3000. Multiple neurofibromas are the most significant hallmark of NF1. These benign tumors of peripheral nerves mostly start developing at puberty and can increase in size and number during pregnancy [ 1 ]. Since decades, physicians involved in the care of NF1 patients are concerned by the dilemma if hormonal contraceptives containing estrogen and progestogen could stimulate the onset of new or the growth of the yet present neurofibromas. Recently, McLaughlin & Jacks [ 2 ] reported expression of progesterone receptors and estrogen receptors in 75% and 5% of 59 neurofibromas immunohistochemically, respectively. The authors thus inferred an important role of progesterone in neurofibroma growth and suggested that antiprogestins may be useful in the treatment of this tumor. Hormonal contraceptives contain synthetic progestogen which bind to the progesterone receptors. Depending on the formulation of the currently available preparation, combined oral contraceptives contain 0.02 to 0.05 mg synthetic estrogen and low doses of various types of synthetic progestogen. These kinds of contraceptives suppress the pituitary gonadotropin secretion and thus reduce the endogenous levels of estrogen and progesterone. The deficiency of endogenous estradiol is balanced by the exogenous supply of ethinylestradiol. Synthetic progestogen bind to progesterone receptors and thus compensate the deficiency of endogenous progesterone to certain extent. Progestogen-only preparations, such as the so-called progestogen-baby-pills, contain progestogen that is below the ovulation-inhibiting dose. This kind of pills only suppress the peak levels of the gonadotropins and thus reduce the estrogen and progesterone level to certain degree. In contrast, parenteral progestogen preparations (depot contraceptives) contain high doses of medroxyprogesterone acetate (150 mg) or norethisteron enanthat (200 mg). In the first days after the administration of these preparations, the blood concentration of progestogen is very high which decreases slowly over weeks. To examine the effect of hormonal contraceptives on the behaviour of neurofibromas, we carried out a survey on 59 female NF1 patients in this study. Methods NF1 was diagnosed according to the NIH criteria [ 3 ]. The protocol was approved by the institutional review board and all participants provided informed consent. A total of 110 female NF1 patients of the NF-Clinic Hamburg, Germany, were asked to fill out a question form (appendix in additional file 1 ) and some of them were interviewed personally by the authors. Only patients who have or had neurofibromas were included in this survey. The age of the included patients was between 18 and 80 years. Data collection was done from Aug. to Dec. 2003, except for one patient who changed preparation in Dec. 2003 and provided us new information recently. The major information acquired were: age of menarche, paramenia, contraceptive means and behaviour of their neurofibromas (appendix in additional file 1 ). Patients were asked to describe the increase of the growth of their neurofibromas as either slight, medium, significant or no (appendix in additional file 1 ). Results Among the total of 110 patients included in the survey, 69 were practicing or had practiced hormonal contraception. Sixty-three received oral estrogen-progestogen preparations, 3 had pure progestogen and one had been given a parenteral depot contraceptive containing very high dose of medroxyprogesterone acetate (150 mg) and norethisterone enanthate (200 mg). For two patients, the names of the preparations could no longer be recalled. Eight out of the 63 patients who used oral estrogen-progestogen could not recall whether there was any change in the behaviour of their neurofibromas in association with hormonal contraception and were excluded from further evaluation. Data from a total of 59 patients were thus available for evaluation (table 1 ). The period of hormonal contraception was between 3 months and 22 years among these 59 patients. Fifty three (91%) out of the 58 patients who used combined estrogen-progestogen preparation or pure progestogens were convinced that there was no tumor growth in association with the practiced hormonal contraception. Other five patients reported a slight increase in the size or/and number of their neurofibromas in the first few months of the hormonal contraception. These 5 patients used combined estrogen-progestogen preparation, containing 0.03 to 0.05 mg of estrogen and 0.125 to 2.5 mg synthetic progestogen (table 2 ). However, no tumor growth was reported by other 15 patients who also used the same preparation (table 2 ). One patient received a depot progestogen (Depot-Clinovir) and reported a strong growth of neurofibromas and intraspinal tumors right after begin of the hormonal contraception. In December 2003, after the closure of our data collection, one patient changed from a combined contraceptive to the depot contraceptive Depot-Clinovir. Recently, she reported experience of rapid growth of her neurofibromas since the change. Interestingly, her tumors had been stable during the two years she took combined contraceptive (Ministon, Leios). Of the 7 patients whose neurofibromas had increased in size and number upon hormonal contraception, two had been treated for paramenia as adolescents, while the other 4 were gynecologically normal. Discussion Our results suggest that in majority of cases, combined hormonal contraceptives containing estogen and progestogen do not seem to stimulate growth of neurofibromas in NF1 patients. The reported slight tumor growth in 5 cases may not necessarily be the consequence of the contraceptives since other patients who used the same preparation did not notice related tumor growth (table 2 ). Our finding seems to eliminate the previous uncertainty and excessive caution in using hormonal contraceptives which often mean exporing NF1 patients to more severe problems as undesired pregnancies with important effects on tumoral growth. The significant tumor growth associated with depot contraceptive in two cases suggests that high doses of medroxyprogesterone acetate and nerethisterone enanthate might stimulate the growth of neurofibromas in some cases. However, only two cases are far too few for further speculation. Additional reports and reviews regarding response of NF1 patients to this form of contraception will be very helpful. Our survey was done a posteriori and mostly by a questionnaire. The results thus report only subjective impressions of patients, not an objective and quantifiable judgment of the researcher. This is clearly a major limitation of the study. Use of cultured Schwann cells and fibroblasts from human neurofibromas as well as recently developed mouse models of neurofibromas may help to further dissect the role of progesterone in regulating neurofibroma growth. Conclusions Oral contraceptives do not seem to stimulate the growth of neurofibromas in most cases and thus may be used by NF1 patients. High doses of progesterone might stimulate the growth of neurofibromas and deserve closer observation. Competing interests The author(s) declare that they have no competing interests Authors' contributions ML has carried out the survey and prepared the preliminary version of the manuscript. LK has prepared and completed the manuscript. She is corresponding author. VM was responsible for the dianogis of the NF1 patients. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Questionary for patients Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549555.xml

539267

Cerebrospinal fluid levels of opioid peptides in fibromyalgia and chronic low back pain

Background The mechanism(s) of nociceptive dysfunction and potential roles of opioid neurotransmitters are unresolved in the chronic pain syndromes of fibromyalgia and chronic low back pain. Methods History and physical examinations, tender point examinations, and questionnaires were used to identify 14 fibromyalgia, 10 chronic low back pain and 6 normal control subjects. Lumbar punctures were performed. Met-enkephalin-Arg 6 -Phe 7 (MEAP) and nociceptin immunoreactive materials were measured in the cerebrospinal fluid by radioimmunoassays. Results Fibromyalgia (117.6 pg/ml; 85.9 to 149.4; mean, 95% C.I.; p = 0.009) and low back pain (92.3 pg/ml; 56.9 to 127.7; p = 0.049) groups had significantly higher MEAP than the normal control group (35.7 pg/ml; 15.0 to 56.5). MEAP was inversely correlated to systemic pain thresholds. Nociceptin was not different between groups. Systemic Complaints questionnaire responses were significantly ranked as fibromyalgia > back pain > normal. SF-36 domains demonstrated severe disability for the low back pain group, intermediate results in fibromyalgia, and high function in the normal group. Conclusions Fibromyalgia was distinguished by higher cerebrospinal fluid MEAP, systemic complaints, and manual tender points; intermediate SF-36 scores; and lower pain thresholds compared to the low back pain and normal groups. MEAP and systemic pain thresholds were inversely correlated in low back pain subjects. Central nervous system opioid dysfunction may contribute to pain in fibromyalgia.

Background Fibromyalgia (FM) is an enigmatic condition characterized by increased complaints of widespread pain with tenderness to palpation [ 1 ]. The tenderness is traditional tested by manually pressing over so-called tender points, but more recent studies have shown that the tenderness is generalized phenomenon [ 2 ]. The mechanisms responsible for the increase in the perception of pain in FM, and the variation of pain sensitivity in the general population are unclear. A similar continuum is seen with heat – induced pain. However, when subjects who report pain to a minimal stimulus (low pain threshold) were compared to subjects reporting less pain with the same stimulus (high pain threshold), there was enhanced functional magnetic resonance imaging (fMRI) responses in the low pain threshold group [ 3 ]. The differences in activation were greatest in the primary somatosensory cortex, anterior cingulate cortex and prefrontal cortex. These fMRI patterns suggest there may be a continuum within the population for pain thresholds, central cortical activation and verbalized pain perception. These results may be applicable to FM since the same brain regions have been identified in response to painful stimuli [ 4 ]. The pain present in FM may induce antinociceptive neural mechanisms with the release of opioid peptides. This hypothesis was tested by measuring opioid peptides in cerebrospinal fluid, and comparing these levels to systemic pain thresholds, subjective complaints, and quality of life measures in 3 sets of volunteers. FM and Chronic Fatigue Syndrome often overlap [ 5 ]. Chronic Fatigue Syndrome is characterized by severe fatigue associated with exertional exhaustion, pain symptoms, neurocognitive and sleep dysfunction [ 6 ]. Therefore, opioid levels were compared for FM and FM/ Chronic Fatigue Syndrome subsets. The second group had chronic low back pain (LBP) without FM or Chronic Fatigue Syndrome. These subjects have a chronic regional pain syndrome [ 7 ] and served as a positive control group. The negative control group was formed by healthy persons with no pain or fatigue. Two peptides were selected for measurements because they were involved in antinociceptive responses. Alternatively, dysfunction of their release could predispose to chronic pain. Preproenkephalin A is the precursor for leucine-enkephalin, methionine-enkephalin (Met-enk), Met-enk-Arg 6 -Gly 7 -Leu 8 , and Met-enk-Arg 6 -Phe 7 (MEAP) [ 8 ]. MEAP was elevated in many brain regions in inflammatory models of arthritis and gluteal carregeenan injection in rats [ 9 , 10 ]. Nociceptin, also known as orphanin FQ, was increased in the cingulate gyrus in rat chronic pain models [ 9 ]. Methods Subjects FM, LBP and Normal control subjects between the ages of 18 and 70 years were recruited to this IRB-approved protocol from rheumatology and orthopedics clinics, advertisements, and word of mouth. Normal subjects were pain and chronic fatigue free, had no diabetic, neurologic, inflammatory, autoimmune, or other chronic disorder that could predispose to pain, alterations in sensation, or known variations in cerebrospinal fluid composition. LBP inclusion criteria were (i) dominant pain complaint of low back pain, and (ii) imaging studies within the past 6 months. Exclusion criteria were: (a) evidence of a lumbar fracture or tumor to explain the pain, (b) any chronic illness that may affect functional status such as diabetes, cancer, chronic obstructive pulmonary disease, chronic inflammatory diseases, renal insufficiency or similar debilitating disorders, (c) previous back or neck surgery, and (d) FM or Chronic Fatigue Syndrome. FM subjects had a prior clinical diagnosis of FM including widespread pain affecting all 4 quadrants and the axial skeleton lasting at least 3 months that was not explained by any other chronic illness, and the presence of at least 11 of 18 tender points when manual, digital pressure of ~4 kg was applied [ 1 ]. All subjects were medication-free for at least 4 days prior to study. Subjects participated in a 1/2 day protocol that involved confirmatory history and physical examination, questionnaires, tender point examinations, and lumbar puncture. Questionnaires Systemic complaints questionnaire This self – report questionnaire containing 44 queries grouped into the following modules: (i) Fatigue; (ii) Musculoskeletal: morning stiffness, muscle pain, muscle spasms, dry eyes, dry mouth, fingers sensitive to the cold, fingers turn blue and/or white in the cold, swollen lymph nodes, swollen joints, fever; (iii) Chest: shortness of breath (SOB), SOB when hurrying on level ground or walking up a slight hill, SOB when walking with other people of own age on level ground, stop for breath when walking at own pace on level ground, SOB when washing or dressing, rapid heart rate, chest pain, irregular heart rate, palpitations; (iv) Headaches: migraine or tension type; (v) Neurological: numbness or tingling of hands or legs, inability to concentrate, problems with memory, dizziness; (v) Ear, Nose and Throat (ENT): problems with balance, hearing loss, ear pain, sensation of ear blockage or fullness, ringing in the ears, sinus pain; (vi) Bladder: urinary urgency, pelvic discomfort / pain / pressure, persistent bladder fullness after urination, dysuria; and (vii) Irritable Bowel Syndrome (Rome I criteria): abdominal pain relieved with bowel movement, abdominal pain with a change in frequency or consistency of stool, changes in stool consistency, changes in stool form (hard or loose/watery), changes in passing of stool, bloating or feeling of abdominal distention, passage of mucus, nausea or vomiting [ 11 ]. Subjects were asked to respond "Yes" if they had recurrent or chronic symptoms for more than 3 of the past 12 months. The sum of positive responses for each module and the total were determined. Subjects completed the MOS SF-36 [ 12 ]. The domains were Physical Functioning (PF), Social Functioning (SF), Role Limitation due to Physical Problems (PP), Role Limitation due to Emotional Problems (EP), Mental Health (MH), Energy / Fatigue (E/F), Pain (P), General Perception of Health (GP) and Change in Health (CH). Pain threshold and tender point examinations All subjects had pressure testing at 9 bilateral sites (18 total) using a hand held dolorimeter (algometer) with a 1 cm 2 rubber stopper making contact with the skin (Chatillon, etc.) [ 1 , 2 ]. The degree of pressure required to cause pain (pain threshold) was recorded at each site, and the number with pain induced by < 4 kg / cm 2 recorded. The average pressure causing pain was the Average Pain Threshold. FM and Normal subjects had manual digital pressure examinations of these points [ 1 , 2 ]. The number of points that were painful was recorded as Manual Tender Points. Cerebrospinal fluid (CSF) radioimmunoassays Lumbar punctures were performed using local anesthetic and 23G spinal catheters. Volumes of 4 to 8 ml were obtained, centrifuged, aliquoted, and immediately frozen at -80°C. Samples were shipped on dry ice to Dr. Lars Terenius for measurement of neuropeptides. Neuropeptides were extracted from 1 ml aliquots using C-18 SepPak cartridges, eluted, dried, and resuspended for validated radioimmunoassays for MEAP [ 13 , 14 ] and nociceptin [ 15 ] using the standard methodologies developed in their laboratory. Concentrations in samples were interpolated from parallel standard curves. There was insufficient CSF to use HPLC for precise peptide identification. Hence, immunoreactive materials (irm) were measured as MEAP-irm and nociceptin-irm. Statistics Geometric mean and 95% confidence intervals were determined for each neuropeptide, with arithmetic means and 95% CI's for all other variables. Differences between groups were assessed by ANOVA. Differences between means for each pair of groups were assessed by 2-tailed unpaired Student's t-tests with Bonferroni corrections for multiple comparisons. Significance was ascribed for p < 0.05. Results Demographics Lumbar punctures were performed on 14 FM (1 male), 10 LBP (5 male) and 6 Normal (2 male) subjects. The averages and ranges of ages for these 3 groups were similar (overall average 42.7 yr, 38.8 to 46.6; 95% CI). There were 4 African-Americans in the FM group, and 1 each in the LBP and Normal groups, and 1 Asian in the FM group. The remainder was Caucasian. Systemic complaints questionnaire Significant differences were found between FM and Normal results by 2-tailed unpaired Student's t-tests with p < 0.05 after Bonferroni corrections for multiple comparisons of this data. All of the FM subjects complained of fatigue (figure 1 ). Other highly prevalent individual symptoms in FM were morning stiffness, muscle pain and spasms, and difficulties concentrating on cognitive tasks. FM scores were higher than Normal for ENT and IBS (p < 0.05), Chest, Headache, Neurological and Bladder (p < 0.01), Fatigue and Musculoskeletal (p < 0.001) complaints. When compared to LBP, FM scores were higher for Neurological (p < 0.05), Chest (p < 0.01) and Fatigue (p < 0.001) complaints. Normal and LBP scores were not different. Chronic Fatigue Syndrome co-existed with FM in 7 of the 14 subjects compared to none in the LBP and Normal groups. IBS was present in 58% of FM, 20% of LBP and 17% of Normal subjects. SF-36 Domain Scores for the Normal group were all near the predicted values of 100 (figure 2 ). FM scores were significantly lower for Emotional Problems (p < 0.05), Social Functioning (p < 0.01), Physical Functioning, Role Limitation due to Physical Problems, Energy / Fatigue, Pain and General Perception of Health (p < 0.001), but not Mental Health or Change in Health. All Normal means were highly significantly greater (p < 0.001) than all LBP scores. FM scores were higher than LBP for Emotional Problems, General Perception of Health, and Change in Health (p < 0.05), Social Functioning (p < 0.01), and Mental Health and Pain (p < 0.001). FM and LBP were not different for Physical Functioning, Role Limitation due to Physical Problems or Energy / Fatigue. Pain thresholds and tender point counts Dolorimetry identified significantly lower pressure pain thresholds for FM (1.51 kg/cm 2 ) compared to the LBP (2.48 kg/cm 2 ; p < 0.01) and Normal (2.60 kg/cm 2 ; p < 0.001) groups (figure 3 ). The numbers of tender points determined by dolorimetry were 9.07 for FM, 3.67 for LBP, and 1.33 (p < 0.05 vs. FM) for Normal subjects (figure 4 ). Digital pressure identified more manual tender points in FM (13.00; 12.27 to 14.73) than Normal (4.67; 1.00 to 8.34; p < 0.001) groups. An average of 3.8 (1.90 to 5.70) more tender points were detected by manual examinations than by dolorimetry (p < 0.01). This difference has been attributed to higher anxiety and other psychometric variables in FM [ 16 ]. (Manual tender point counts were not recorded for LBP subjects.) Cerebrospinal fluid neuropeptide concentrations MEAP – irm The 3 groups had significantly different mean CSF concentrations (p = 0.0014, ANOVA). The normal volunteers had significantly lower geometric mean concentrations (26.3 pg/ml; 13.9 to 49.9) than the FM (101.7 pg/ml; 72.8 to 142.0; p < 0.01 vs. normal), and LBP (78.0 pg/ml; 51.1 to 119.0; p < 0.05 vs. normal) (figure 5 ). Co-morbid Chronic Fatigue Syndrome did not affect the MEAP – irm results, since the arithmetic means were 112 pg/ml for FM subjects with this syndrome (n = 7) and 124 pg/ml with FM alone (n = 6). There were no obvious relationships with age, gender or race. The small sample size precluded further statistical analysis of these variables. Nociceptin – irm was not different between FM (4.27 pg/ml, 3.22 to 5.66, n = 14), LBP (4.52 pg/ml, 3.12 to 6.55, n = 10) and Normal (5.65 pg/ml, 2.65 to 12.04, n = 6) groups. Systemic pain thresholds and MEAP – irm These 2 variables were correlated (Pearson's correlation coefficient of -0.38, p < 0.05; explained variance 0.15) when all subjects were examined as a single group (figure 6 ). This correlation was not found when the Normal and FM groups were examined by themselves. Normal subjects had higher thresholds and lower MEAP – irm concentrations. FM subjects had pain thresholds below 2.3 kg/cm 2 , which was coincidentally the lower 95% CI for the Normal group. MEAP – irm concentrations had a wide range in the FM subjects, but the geometric mean was significantly higher than for Normal subjects. Pain threshold and MEAP – irm concentrations did not have linear correlations in either the FM or Normal group. These data suggested that FM subjects were fundamentally different from Normal. When the pain threshold was below 2 to 2.3 kg/cm 2 , MEAP – irm levels increased approximately 4-fold compared to Normal. The combination of the widespread pain, systemic complaints, low pain threshold and high MEAP – irm concentrations in CSF was distinct from the low level of symptomatic complaints, normal (high) pain thresholds, and lower MEAP – irm levels found in the Normal group. The LBP group had a chronic regional pain syndrome, no fibromyalgia, normal pain thresholds and systemic complaints, but severe disability (SF-36 scores). Only the LBP group showed a linear correlation between pain threshold and MEAP – irm concentrations (figure 6 ). The parameters of this correlation were similar to that of the entire group. This was due to the overlap of some high pain threshold / low MEAP – irm LBP subjects with the Normal group, and low pain threshold / high MEAP – irm LBP subjects with the FM data. This continuum of pain threshold and MEAP – irm levels in LBP was different from the clustered FM and Normal datasets, and suggested a different mechanism of MEAP – irm regulation in LBP from FM. Discussion The Normal group had Systemic Complaints and SF-36 scores in the normal ranges, high pain thresholds, low numbers of manual and dolorimetry-derived tender points, and low CSF concentrations of MEAP – irm and nociceptin – irm. The LBP group was a positive control for chronic regional pain. Their Systemic Complaints scores, systemic pain thresholds and dolorimetry defined tender point counts were not significantly different from Normal. However, most of their SF-36 results were near zero indicating the worst level of impairment of the 3 groups. They were the only group to show a correlation of decreasing pain thresholds with increasing MEAP – irm concentrations. The continuum of MEAP – irm levels in the LBP group led to borderline significance for the comparison to Normal levels. Inclusion of LBP subjects with higher or lower pain thresholds may have shifted the MEAP – irm concentration distribution towards or away, respectively, from the Normal group results. This is important when comparing these data to those of other studies. For example, chronic sciatica patients did not have elevated MEAP – irm compared to controls [33]. However, severity was not graded as extensively as in our study. Some subjects may have had less severe low back pain than in our group. If so, then the linear correlation noted in FIGURE 6 would predict no significant difference from normal subjects. Conversely, female LBP subjects with the lowest pain thresholds and highest MEAP – irm levels may have been making a transition from chronic low back pain to fibromyalgia [ 7 ]. Half of our LBP group was male, introducing gender as a potentially confounding factor. The FM group's results were distinctly different from the Normal and LBP groups. FM had widespread pain complaints, the highest Systemic Complaints scores, the lowest pain thresholds, and highest numbers of tender points of the 3 groups. Their SF-36 scores were intermediate between LBP and Normal groups. Widespread pain, low pain thresholds, and high Systemic Complaints scores differentiated FM from Normal and LBP. CSF MEAP – irm concentrations were approximately 3-fold higher in FM than Normal (figure 5 ). This confirmed earlier findings [ 14 ] where a group of women meeting an older set of fibromyalgia criteria [ 17 , 18 ] had 34% higher CSF MEAP – irm concentrations than a group of 8 age-matched female control subjects [ 13 ]. None of the FM subjects in the earlier group required analgesics or other medications suggesting that their symptoms may have been milder than for our FM group. Our group contained 1 male and 13 females. In contrast, Lui et al. found MEAP concentrations (peptide identity confirmed by HPLC) that were 38% lower in FM than control subjects (p < 0.01) [ 19 ]. There was inadequate clinical data to compare the severity of complaints between these FM populations. These investigators also used a liquid-liquid peptide extraction method. The differences in FM severity, control groups, extraction procedures, and lack of sufficient CSF to identify precise peptides by HPLC [ 20 ] made it difficult to compare these sets of divergent results. Standardized measurements on CSF withdrawn from highly characterized and clearly defined subjects and controls will be required to resolve these inconsistencies. This is the first investigation to examine the potential effect of co-existing CFS on MEAP – irm levels in FM. This suggested that the mechanism(s) of CFS were probably independent of those responsible for the elevated MEAP – irm levels in FM. Unfortunately, CFS subjects without FM could not be simultaneously tested to determine if their MEAP – irm levels were normal (as would be predicted). Nociceptin – irm levels were the same in our three groups. The levels were about 10% of that found in women during labor [ 21 ]. It was unclear if the higher concentrations were due to pregnancy, neurohormonal adaptations during labor and delivery, or the effects of acute pain. Again, the absence of a control group comparable to our pain-free Normal group makes mechanistic comparisons difficult. Conclusions The Normal group had Systemic Complaints and SF-36 scores in the normal ranges, high pain thresholds, low numbers of manual and dolorimetry-derived tender points and low cerebrospinal fluid MEAP and nociceptin concentrations. The LBP chronic regional pain group had similar Systemic Complaints scores, systemic pain thresholds and dolorimetry-defined tender point counts. However, most of their SF-36 results were near zero indicating the worst level of impairment of the 3 groups. MEAP – irm was just significantly elevated compared to the Normal group, and was correlated to the systemic pain threshold. The FM group was distinct since they had widespread pain complaints, the highest Systemic Complaints scores, the lowest pain thresholds, and highest numbers of tender points of the 3 groups. Their SF-36 scores were intermediate between the LBP and Normal groups. MEAP – irm concentrations were significantly higher in the FM than Normal group. The co-existence of Chronic Fatigue Syndrome with FM did not alter the MEAP – irm concentrations. This suggested that Chronic Fatigue Syndrome mechanism(s) did not involve preproenkephalin dysfunction. Nociceptin – irm levels were not different between these groups, and were lower than previously reported results from pregnant women in labor. Significant differences in MEAP – irm concentrations from previous studies may be due to the highly controlled definition of patients in this study, selection of control groups, and differences in peptide extraction methods. Abbreviations CSF, cerebrospinal fluid; FM, fibromyalgia; fMRI, functional magnetic resonance imaging; irm, immunoreactive material; LBP, low back pain; MEAP, Met-enk-Arg 6 -Phe 7 ; SF-36, short-form of 36 questions with the following domains: PF, physical functioning; SF social functioning; PP, role limitation due to physical problems; EP, role limitation due to emotional problems; MH, mental health; E/F, energy / fatigue; P, pain; GP, general perception of health; CH, change in health; Systemic Complaints domains: MS, musculoskeletal; HA, headache; Neuro, neurological; ENT, ear, nose & throat; IBS, irritable bowel syndrome (Rome I criteria); SOB, shortness of breath. Competing interests The author(s) declare that they have no competing interests. Authors' contributions JNB compiled the results and wrote the manuscript. DJC oversaw the clinical investigations that were performed by JC. GW was responsible for the sample repository, shipping, and collection of data. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539267.xml

531613

A Taxpayer-Funded Clinical Trials Registry and Results Database

It already exists within the US Food and Drug Administration, argues a former clinical reviewer of psychotropic drugs at the FDA

Over the past several years, there has been growing concern about selective publication of clinical trial results [ 1 , 2 ]. The debate has intensified since New York State Attorney General Elliot Spitzer filed suit against GlaxoSmithKline on June 2, 2004, alleging that the company was hiding data regarding the efficacy and safety of selective serotonin reuptake inhibitors in pediatric patients with depression [ 3 ]. The two most frequently suggested remedies for the selective reporting of clinical trials results have been to register all clinical trials and to make their results publicly available. Registries have been called for at least as far back as 1974; hundreds have in fact already been established [ 4 ]. Shortcomings of registries include the fact that they are often not coordinated and that participation is often voluntary and—in cases where they are mandated by legislation—difficult to enforce. For example, ClinicalTrials.gov, a registry authorized by the Food and Drug Modernization Act of 1997, appears not to be comprehensive. One study found that, of 127 cancer protocols sponsored by pharmaceutical companies that met criteria for inclusion, only 48% were in fact submitted to the registry [ 5 ]. Thus, one can check a number of registries and still have little assurance that all the relevant trials of interest have been included. The public would benefi t from more freedom of information at the FDA (Illustration: Margaret Shear) Increasing the pressure on pharmaceutical companies to include more trials in registries, the International Committee of Medical Journal Editors has announced that, as a condition of considering a trial for publication, member journals will require its registration in a public trials registry [ 6 ]. Further, at the American Medical Association (AMA) Annual Meeting of the House of Delegates in June 2004, the AMA called on the Department of Health and Human Services to establish a comprehensive national registry. In September 2004, an AMA trustee testified in a United States Congressional hearing, outlining elements necessary to make such a registry effective [ 7 ]. Momentum for a comprehensive clinical trials registry is also building internationally [ 8 ]. In this essay, I argue that a highly valuable but underused registry and results database for US trials already exists within the Department of Health and Human Services, specifically within the Food and Drug Administration (FDA). New Drug Applications Before a pharmaceutical company can conduct a US trial that it intends to use in support of a new drug application (NDA), it must first register that trial with the FDA. Because the NDA forms the basis for marketing approval, it seems likely that the percentage of industry-sponsored trials that are registered with the FDA is very high. This registration takes the form of an investigational new drug (IND) application [ 9 ]. The IND contains a trial protocol; protocols for additional studies within the same clinical trials program are submitted as amendments to the IND. Later, when the sponsor has completed its clinical trials program and wishes to apply for marketing approval, it submits its NDA. The FDA then begins the NDA review process, during which a physician, a statistician, and a pharmacologist, among others, generate lengthy review documents [ 10 ]. These reviews not only address the sponsor's analyses of the data on pivotal studies, but they often also include reanalyses by the reviewers using raw data obtained from the sponsor. These analyses are conducted in adherence to the statistical methods set forth a priori in the original trial protocols. (By contrast, with most journal publications, it is usually not possible for the reader to verify whether what is presented as the main finding is consistent with the original hypothesis or whether it was a post hoc finding.) After the primary reviewers have written their reviews, shorter reviews are written by their superiors, with the process culminating in a decision about whether to approve the drug for the proposed indication. Paroxetine for Anxiety Disorders: Checking the Published Literature against the FDA Reviews A Cochrane systematic review of antidepressants for generalized anxiety disorder [ 15 ] lists only one double-blind placebo-controlled study of paroxetine [ 16 ], a positive study. A PubMed search reveals no additional double-blind placebo-controlled studies. In accessing the review from Drugs@FDA (approval date April 2001), we learn that there were in fact three pivotal double-blind placebo-controlled studies. One of these studies corresponds to the published positive study noted above. Of the remaining two studies, both apparently unpublished, one was positive while the other was marginally positive. Turning to the controlled-release formulation of paroxetine (Paxil CR) for panic disorder, a review article states in its abstract that the drug “demonstrated efficacy in three well designed studies in patients with panic disorder with or without agoraphobia” [ 17 ]. In reading the corresponding FDA statistical review, we verify that there were indeed three studies. However, the FDA statistical reviewer found that only one of these studies was strongly positive. A second study, judged “supportive” of efficacy, had a marginally significant ( p = 0.039) result on a secondary observed-cases analysis, but a nonsignificant ( p = 0.38) result on the primary efficacy analysis defined a priori. The third study was clearly negative, with p-values of 0.33 and 0.57 on the primary and secondary analyses, respectively. A Semi-Public Database This process occurs entirely outside of the public domain. However, in the interest of making the FDA more “transparent,” and in accordance with the Electronic Freedom of Information Act [ 11 ], the FDA has, for the past several years, posted selected NDA reviews for approved drug–indication combinations on the FDA Web site Drugs@FDA ( http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cfm ). These NDA review documents are much more detailed than the resulting package insert and often more detailed than corresponding journal publications. For example, while the clinical trials section of the package insert is typically a few paragraphs long, the efficacy portion of the clinical review usually runs tens of pages. Because the FDA is made aware of all studies that the sponsor plans to use in support of the NDA before they are conducted, and thus before there can be any selection based on outcome, these reviews cover not only studies that are positive (and more likely to be published in journals), but also studies whose outcome was negative or indeterminate. The sidebar gives an example of how NDA review documents at the FDA give valuable information about paroxetine for anxiety disorders. FDA Reviews for All Approved Drugs Should Be Made Public In the examples discussed in the sidebar, our having access to the FDA review documents allows us to become aware of, and see beyond, apparent publication bias. It is in the best interest of the public health for the FDA to make as many reviews available as possible. According to the FDA Web site, “As FDA continues to be one of the world's leading agencies in its emphasis on openness and transparency, it is aware that making even more information available to the public will further the Agency's mission to protect and promote public health and improve its credibility. For example, FDA has aggressively implemented the Electronic Freedom of Information Act…” [ 11 ]. Unfortunately, the availability of review documents on Drugs@FDA is sporadic. To take additional examples from psychiatry, NDA reviews have been posted on Drugs@FDA for some approved drug–indication combinations, such as fluoxetine for pediatric depression, and aripiprazole and quetiapine for schizophrenia. However, NDA reviews for many other drug–indication combinations have not been posted: the Prozac Weekly formulation of fluoxetine, clozapine for suicidal behavior in patients with schizophrenia or schizoaffective disorder, and quetiapine for mania, among others. A review on paroxetine for pediatric depression, the subject of Elliot Spitzer's suit against GlaxoSmithKline, is not posted. This is probably because this drug–indication combination was not approved; in fact, it is possible that GlaxoSmithKline did not file an NDA to be reviewed. However, I do not understand why, in cases where NDAs were both submitted and approved, such as the ones listed above, some reviews are posted while others are not. I therefore suggest that we increase access to the clinical trials registry and results database that already exist within the FDA. The agency could expand its implementation of the Electronic Freedom of Information Act and make all NDA reviews, at least for approved NDAs, available in the public domain. The act is written into the FDA portion of the Code of Federal Regulations as follows: “The Food and Drug Administration will make the fullest possible disclosure of records to the public, consistent with the rights of individuals to privacy, the property rights of persons in trade secrets and confidential commercial or financial information” [ 12 ]. Obstacles and Limitations There would surely be obstacles. The pharmaceutical industry would vigorously invoke Exemption 4 of the Freedom of Information Act, the exemption for trade secrets and confidential business information [ 13 ]. However, the FDA Freedom of Information Office already deals with confidential and proprietary information by redacting or editing it out of the review documents before making them available. Within the FDA's Freedom of Information Office, staffing would need to be greatly increased. Some oversight might be necessary to ensure that the taxpaying public has been granted the fullest possible access and that unwarranted redaction does not occur. Unless the Freedom of Information Act is modified, access would still likely be limited to approved NDAs. Data would remain unavailable for trials that did not lead to an approved NDA. It should be clarified that this resource does not compete with proposals by the AMA and other groups for clinical trial registries—rather, it complements them. The AMA has proposed the creation of a registry that is comprehensive in scope. The FDA's registry and results database are restricted to those trials aimed at supporting US marketing approval or a change in labeling in the US. While data from many studies conducted abroad are submitted to the FDA for this purpose, this is not the case for drugs for which the sponsor has elected not to seek approval for marketing in the US. Nor does the FDA review data from most trials funded by other US government agencies, such as the National Institutes of Health, or by foundations. And drug companies fund investigator-initiated trials that are often not registered with the FDA. To make the FDA review data more accessible and user-friendly, simple formatting changes would be needed. For those (few) reviews that are currently posted on Drugs@FDA, one can determine the indication being evaluated only after opening the document and paging through it. (Descriptive titles would be helpful, and these could be linked to ClinicalTrials.gov. Further, the trials reviewed could be identified with a unique international identifier, as promoted by the World Health Organization [ 14 ].) Despite the fact that the reviews are created in Microsoft Word and converted to PDF, the versions that appear on the Web site are no longer in a searchable text format. While the reviews tend to be well organized, the posted versions are difficult to navigate because there is no hyperlinked table of contents. In addition to having these formatting issues addressed, clinicians and patients might benefit from brief summaries, the writing of which might require the addition of new FDA staff. Conclusion Despite the limitations of the FDA's database, making it public is a strategy that could be implemented both rapidly and easily by building upon existing infrastructure. While we await the creation of a clinical trials registry and results database that is truly comprehensive, we already have at our disposal one that could serve as a trove of in-depth and unbiased information on many, if not most, drugs currently marketed in the US.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC531613.xml

517617

Applying Support Vector Machines for Gene ontology based gene function prediction

Background The current progress in sequencing projects calls for rapid, reliable and accurate function assignments of gene products. A variety of methods has been designed to annotate sequences on a large scale. However, these methods can either only be applied for specific subsets, or their results are not formalised, or they do not provide precise confidence estimates for their predictions. Results We have developed a large-scale annotation system that tackles all of these shortcomings. In our approach, annotation was provided through Gene Ontology terms by applying multiple Support Vector Machines (SVM) for the classification of correct and false predictions. The general performance of the system was benchmarked with a large dataset. An organism-wise cross-validation was performed to define confidence estimates, resulting in an average precision of 80% for 74% of all test sequences. The validation results show that the prediction performance was organism-independent and could reproduce the annotation of other automated systems as well as high-quality manual annotations. We applied our trained classification system to Xenopus laevis sequences, yielding functional annotation for more than half of the known expressed genome. Compared to the currently available annotation, we provided more than twice the number of contigs with good quality annotation, and additionally we assigned a confidence value to each predicted GO term. Conclusions We present a complete automated annotation system that overcomes many of the usual problems by applying a controlled vocabulary of Gene Ontology and an established classification method on large and well-described sequence data sets. In a case study, the function for Xenopus laevis contig sequences was predicted and the results are publicly available at .

Background Ongoing genome sequencing and recent developments in cDNA sequencing projects have led to an exponential rise in the amount of sequence information. This has increased the need for acquiring knowledge from sequences as to their biological function. Annotating a single sequence is the gateway to interpreting its biological relevance. However, the usefulness of these annotations is highly correlated with their quality. Accurate annotation has traditionally been maintained manually with the experience of individual experts and the experimental characterisation of sequences. However, the increasing gap between the amount of sequence data available and the time needed for their experimental characterisation demands computational function prediction in complementing manual curation [ 1 - 4 ]. Commonly, computational functional assignment is based on homologues identified from database searches [ 5 ]. Such an automated annotation process provides comparable results due to a uniform analysis of all query sequences across the same databases and the possibility of repeating the annotation to updated sequence data [ 6 ]. However, crucial aspects for consideration in automated annotation are i) the problems associated with the databases themselves: sequence errors, erroneous annotation due to spelling ambiguities, incomplete functional annotation, inconsistent functional annotation across databases, consistent but wrong annotation across databases, and ii) the problems associated with the inference, i.e. false positives, where an assignment is made on the basis of a wrongly inferred homology [ 3 , 7 , 8 ]. A number of excellent annotation systems have been developed to tackle these problems, e.g. RiceGAAS [ 9 ], GAIA [ 10 ], Genotator [ 11 ], Magpie [ 12 ], GeneQuiz [ 6 ], GeneAtlas [ 13 ] and PEDANT [ 14 ]. However, little has been done to quantify the annotation accuracy by defined benchmarks and establish a method to provide a confidence value for each annotation. The current annotation, written in a rich, non-formalised language also complicates this automated process. We addressed this problem by applying a controlled vocabulary from Gene Ontology (GO) [ 15 - 17 ]. GO provides consistent descriptions of gene products in a species-independent manner. The GO terms are organised in structured, controlled vocabularies (ontologies) to describe gene products in terms of their associated biological processes, cellular components and molecular functions. An increasing number of GO-mapped sequence databases make it possible to replace traditional database searches with GO-related searches. These include databases such as GenBank [ 18 ], SWISS-PROT [ 18 ], SwissPROT/TrEMBL [ 19 ], the TIGR Gene Index [ 20 ] and several other genome databases. Many annotation approaches have now been developed based on Gene Ontology. The uncharacterised sequences are searched across GO-mapped protein databases and assigned with GO terms of the best hits [ 21 , 22 ]. Jensen and co-workers used neural networks to predict specific subsets of GO terms [ 23 ]. Furthermore, Schung et al predicted GO terms by intersecting domain profiles [ 24 ]. The SwissPROT/TrEMBL entries were associated with GO terms by an automated process coupled with manual verification [ 19 ]. Text mining and similarity searches were combined to annotate SWISS-PROT and GenBank entries with GO terms [ 18 ]. However, these approaches were either applied to specific GO subsets or did not provide defined benchmarks and confidence values for their predictions. We have developed an automated system for large-scale cDNA function assignment, designed and optimised to achieve a high-level of prediction accuracy without any manual refinement. Our system assigns molecular function GO terms to uncharacterised cDNA sequences and defines a confidence value for each prediction. The cDNA sequences were searched against GO-mapped protein databases and the GO terms were extracted from the homologues. In the training phase, these GO terms were compared to the GO annotation of the query sequences and labelled correspondingly. We applied Support Vector Machines (SVMs) as the machine learning method to classify whether the extracted GO terms were appropriate to the cDNA sequence or not. In order to classify the GO terms we used a broad variety of elaborated features (attributes) including sequence similarity measures, GO term frequency, GO term relationships between homologues, annotation quality of the homologues, and the level of annotation within the GO hierarchy. To enhance the reliability of the prediction, we used multiple SVMs for classification and applied a committee approach to combine the results with a voting scheme [ 25 ]. The confidence values for the predicted GO terms were assigned based on the number of votes i.e. number of SVMs predicting particular GO term as correct. The performance of the system was benchmarked with 36,771 GO-annotated cDNA sequences derived from 13 organisms. It achieved 80% precision for 74% of the test sequences. We applied our annotation system to predict the function for Xenopus laevis , a widely studied model organism in developmental biology. Because many researchers are now focussing on the functional genomics of this organism, a demand exists for a quality annotation [ 26 ]. Therefore we applied our system to improve the quality and coverage of the existing annotation. We predicted the function for 17,804 Xenopus laevis contig sequences (from TIGR Gene Indices) yielding annotation with good confidence values for more than half of these sequences. Results General workflow of training and classification The classifier (SVM) needs to specify attribute values (features) for a broad list of samples and a class label for each of these samples. Through the training samples it learns the feature patterns and tries to group them according to their class labels. After training, the algorithm assigns class labels to new samples according to the class that they best match. We selected GO-annotated cDNA sequences for training the SVM classifier. The nucleotide sequences were searched against GO-mapped protein databases and GO-annotations were extracted from the significant hits. Then, each GO term obtained was utilized as a sample for the feature table. The sample GO terms were then labelled as either correct ("+1") or false ("-1") by comparing them to the original annotation. Note that we applied the relationships of the GO terms based on their graph structure: "Correct" was assigned not only if they were exact matches but also if the GO terms were related as either "parent" or "child" (Figure 1 ). Next, the samples were attached with their features or attributes, calculated from the BLAST [ 27 ] results. With this data, the classifier was trained to distinguish between the attribute patterns that contributed to class +1 (correct prediction of a GO term) and class-1 (false prediction). To predict the function of unknown sequences, the same procedure was applied as for the training sequences in order to obtain their GO terms and corresponding attribute values. According to these attribute values, the classifier assigned a class for every GO term of the BLAST hits (Figure 2 ). Datasets for training and testing SVM For training and testing the SVM, we selected 39,740 GO-annotated cDNA sequences from the following organisms: Saccharomyces cerevisiae (yeast), Drosophila melanogaster (fly), Mus musculus (mouse), Arabidopsis thaliana ( Arabidopsis ), Caenorhabditis elegans (worm), Rattus norvegicus (rat), Danio rerio (fish), Leishmania major ( Leishmania ), Bacillus anthracis Ame ( Bacillus ), Coxiella burnetii RSA 493 ( Coxiella ), Shewanella oneidensis MR-1 ( Shewanella ), Vibrio cholerae ( Vibrio ) and Plasmodium falciparum ( Plasmodium ) (Table 1 ). From these, 55.3% of the cDNA sequences were contributed by Arabidopsis , mouse and fly (22.1%, 18%, and 15.2% respectively). Prokaryotic bacteria ( Bacillus , Coxiella , Shewanella and Vibrio ) contributed 20.6% and the remaining 24.1% of the sequences came from rat, fish, worm, Plasmodium , Leishmania and yeast. Yeast and fly are purely manually annotated datasets. Where as Bacillus , Coxiella , Vibrio , Shewanella , Leishmania and Plasmodium are mostly manually, and the rest mostly automatically annotated datasets. Manual annotation tends to be conservative and sparse, since the GO terms are assigned only if the annotator is highly confident. Therefore, a GO term may be missed due to a poor definition of a false negative. To reduce this critical problems, yeast and fly annotations are accompanied by an "unknown molecular function" term for sequences with questionable further functions. To reduce false negatives, we discarded all sequences with these tags for training and testing (yeast: 2999 discarded out of 6355, fly: 8495 out of 14335). The cDNA sequences were searched across the protein databases covering a wide range of organisms from prokaryotes to eukaryotes and SWISSPROT. For 36,771 sequences we got hits with GO terms, contributing to 856,632 sample GO terms and yielding an average of 23.29 GO terms per query sequence (Table 1 ). These 856,632 samples were used to train our classifier. Generally, the number of GO terms per sequence was less for prokaryotes than for eukaryotes. Rat had the maximum number of GO terms per sequence (36.9), followed by fish (32.1) and worm (27.13). In contrast, Shewanella , Coxiella and Vibrio sequences had the lowest number of GO terms per sequence (10.78, 12.33 and 12.54, respectively). SVM training and testing SVM training We set up multiple classifiers by splitting the whole dataset (856,632 samples) into 99 equal subsets. Note that, amongst these 99 subsets, 96 contained data from a single organism and the remaining 3 from two organisms each. Subsequently, we built 99 classifiers with these subsets. Since the training sets were created organism-wise, the classifiers were trained from different ranges of data, based on purely manual annotation (yeast, fly), mostly automated annotation or a mixture of both. For training each of these classifiers, we performed a model selection (parameter optimisation by cross-validation; see Methods), which yielded varying accuracy values ranging from 78.81% to 96.03%, with an average accuracy of 85.11%. SVM testing To test the classifiers performance, we prepared 13 test sets (each set corresponding to a single organism) using the same 856,632 sample GO terms. The prediction quality of all 99 classifiers were assessed by an organism-wise cross-validation approach, i.e. for each organism (test set), we used all the classifiers for prediction except those that corresponded to the same organism. With this approach, we were able to simulate the annotation of a new organism. The number of classifiers used for predictions varied highly across organisms (maximum: Plasmodium and Leishmania , 98 classifiers; minimum: Arabidopsis , 74 classifiers). The quality of the predictions was estimated by comparing the predicted terms with the original annotation and the results were expressed in terms of precision and accuracy values (see Methods). The average-accuracy refers to the average of the accuracy values attained by all classifiers used for the prediction. The maximum average-accuracy was achieved for fly (81.51%), followed by yeast (80.50%), and the minimum for mouse (76.0%). Additionally, we compared the classification efficiency of the classifier derived from automatic annotation (mouse, worm and Arabidopsis ) with the manually annotated test sequences (yeast and fly). The prediction of the yeast and fly sequences with the 20 classifiers from the mouse sequences produced an average-accuracy of 79% and 80% respectively. Similar results were acquired with the 25 classifiers from Arabidopsis (79% and 80%). Likewise, the worm classifiers (11 classifiers) yielded the average-accuracy of 82% for yeast and 83% for fly. These values were comparable with the average-accuracy of 81% achieved by both, using yeast as test sequences against fly classifiers (16 classifier) and vice-versa (fly test sequences against yeast classifiers). Likewise, we classified the mouse test sequences against yeast classifiers (5 classifier) and fly classifiers yielding 69% and 71% average-accuracy respectively. Combining multiple classification results by the committee approach Though we already achieved a good accuracy with some of the classifiers, our intention was to improve the precision and, furthermore, to obtain confidence values for the predicted GO terms. To this end, we combined the predictions of multiple classifiers by the committee approach. If a classifier predicted a particular GO term as correct, it contributed a vote. Votes were collected from all classifiers and summed up to yield a final score value. If no vote supported a GO term as correct, it was assigned with the label "false". Otherwise, the number of votes provided a measure of the reliability. Figure 3 shows precision and accuracy versus the number of votes. If we made predictions with a minimum of one vote, we were able to achieve 43% precision and 59% accuracy. When the stringency was raised to 25 votes, a minimum of 25 votes was required to classify a GO term as correct, yielding an accuracy of 84% and precision of 75%. At a cut-off value of 74 votes, we attained 91% precision and 71% accuracy. A cut-off value of 94 votes gave 100% precision and 67% accuracy. Our accuracy reached a plateau at 20 votes. However, it decreased slightly for stringencies of more than 30 votes. Note, that this was due to the increasing number of false negatives. The relation between the precision and the number of votes (Figure 3 ) was used as a means of calibrating to assign the confidence values for new predictions. For each threshold value of the votes, we calculated the sensitivity and the false positive rate to obtain a Receiver Operating Characteristic plot (ROC; Figure 4 ). The graph shows that the classification performance was comparable for different classes of organisms like prokaryotes, single cell eukaryotes and multi-cellular eukaryotes, which reflect the organism-independent performance of our method. Note that for fish, worm, Plasmodium and Leishmania the classification performance was particularly good due to the low number but well characterised test sequences. We compared the prediction performance for GO terms annotated with the evidence code IEA (automated annotation) and non-IEA (manually verified annotation). All sequences from Bacillus , Coxiella , Vibrio , Shewanella , yeast, Leishmania , and Plasmodium were non-IEA annotated and 99.5% of the fly GO terms were non-IEA annotated. In contrast, all sequences from fish and worm were IEA annotated. The remaining test organisms were mostly IEA annotated (rat: 88%, Arabidopsis : 79.4%, and mouse: 69.5%). The classification performances revealed by the ROC plots were comparable between IEA and non-IEA annotated test organisms (Figure 4 ). Therefore, the classifier could reproduce the annotation of other automated systems as well as high-quality manual annotation. We were interested in the coverage of sequences with respect to the average precision of the annotations (shown in Figure 5 ). Considering 1 vote as the cut-off value, we obtained 52% average precision for 98% coverage. We obtained 80% average precision for 74% coverage (cut-off: 34 votes), and 90% average precision for 42% coverage (cut-off: 65 votes). These coverage values varied when regarding the test organisms individually. The coverage for different test organisms at 80% average precision were: fish 97%, Coxiella 89%, worm 88%, Vibrio 86%, rat 85%, Bacillus 83%, Plasmodium 81%, mouse 78%, Leishmania 76%, Shewanella 74%, Arabidopsis 69%, fly 66% and yeast 57%. Xenopus annotation We extracted all Xenopus laevis contig sequences from the TIGR Xenopus laevis Gene Index (XGI) [ 28 ] and got a total of 35,251 contig sequences, excluding singletons. We applied our method to predict functional GO terms for these contig sequences. We predicted the function for 17,804 sequences with an average of 12.16 GO terms per sequence. In total, 23.4% of all the GO terms were predicted with less than 50% confidence value, 51.5% of them were between 50% to 80% confidence and the remaining 25% with a predicted confidence value of above 80%. At 80% stringency (predicted if the GO term possessed a confidence value of 80% or more), we made predictions for 9,510 contig sequences including 55,994 GO terms, yielding on average 5.88 GO terms per sequence. To compare the functional abundance of the expressed genome across the organisms, we mapped the predicted GO terms (with at least one vote) to the high-level, i.e. more generalised or high-level terms of the molecular function ontology ("GO slim" for molecular function) [ 29 ]. These molecular function GO slim nodes were taken from the second level of the molecular function ontology. The distribution of higher-level GO terms were compared between Xenopus , fly, yeast and mouse (Figure 6 ). Note that some of the deeper-level terms had multiple paths. They were mapped to two or more higher-level nodes, so that the total sum of the higher-level nodes exceeded 100%. Comparison to the TIGR Xenopus annotation TIGR provides a GO mapping for Xenopus contigs (TIGR Xenopus laevis gene indices). We compared our annotation with the TIGR GO annotation for molecular function. From 35,251 contig sequences, TIGR annotated 5,444 contigs with a total of 16,432 molecular function GO terms. In contrast, our approach was able to predict function terms for 17,804 contigs, i.e. more than three times that of TIGR sequences. Our procedure did not annotate 295 contigs from the TIGR annotated contigs. For the remaining 5,149 contigs, 85% of all TIGR terms were found to be exact with those using our method; 3.2% of the TIGR terms were at a higher-level of the GO tree than our annotation, so in this case we provided annotation at a deeper level; in 0.9% of the cases our annotation was at a higher-level; 8.3% of the cases were completely different; and 0.6% of the TIGR terms were obsolete. We compared the quality of TIGR and that of our annotations by a raising stringency and found that when we applied a confidence threshold of 80% for our annotation, we lost 46.6% of the sequences. This included 1,492 sequences holding equivalent TIGR annotation or 27.4% of the total TIGR annotation. With this stringency, our system annotated 9,510 contig sequences, i.e. twice the TIGR annotation at this quality. We were interested in novel annotated sequences with the highest confidence values and found we could predict GO terms for 557 contigs with a confidence value of 100% (all votes matched). Interestingly, 192 of these lacked any GO annotation by TIGR. Out of these, 184 had got a descriptive TIGR annotation and the rest had not got any. Table 2 shows the novel annotation for these eight sequences. Our novel predictions are as follows: 1) TC212171 and TC196381 are predicted to display endopeptidase activity and more specifically serine-type peptidase activity (98% and 97% confidence respectively). 2) TC209487 and TC190605 are predicted to be aminopeptidases, however for the latter the more specific prediction of prolyl aminopeptidase activity is assigned with 86% confidence. 3) TC199713 is predicted as glutathione peroxidase at 100% confidence and TC194305 is annotated as protein kinase with the same confidence. 4) Both TC187949 and TC210151 are transmembrane receptors but the latter one is classified as frizzled receptor with 82% confidence. In most of these examples the functional assignment and associated confidence were recorded in multiple levels of granularity. Discussion In this paper, we presented an automatic annotation system that is able to cope with the expanding amount of biological sequence data. Our approach efficiently combines the ongoing efforts of Gene Ontology and the availability of GO-mapped sequences with a profound machine learning system. The GO-mapped databases provide annotation described in a controlled vocabulary and also a measure of reliability, as these GO entries are labelled with their type of origin. Furthermore, GO terms are structured hierarchically, which allow us a twofold use of the information: i) the level within the tree is taken as a classification criterion to distinguish low from high-level annotations during the learning procedure, and, ii) the hierarchical structure allows us to extend hits by slightly moving up and down within a restricted local area of the tree. This may overcome fluctuations of the annotation levels coming from varying annotation experts. Our annotation system exploits the different combinations of attributes and yields functional transitivity: SVM learning and prediction are organism-independent and comparable to manual annotation, which may be supported by the nature of the attributes we utilise. Subsets and overlaps are counted in a balanced fashion to avoid biases due to the complexity of an organism and a potentially correlated complexity of its sequences. The committee approach allows us to improve the prediction quality as well as to assign confidence values for the new predictions in a straightforward manner. Our classifiers performance is hardly limited by the varying quality of the training data, whether manual or automatic annotated. The prediction results of manually annotated test sets with the classifiers based on automated annotation as well as classifiers based on manual annotation were comparable. Regarding the outcome of the overall classifiers, we achieve consistency with existing annotation from automatic annotations. This is the less complex part of our work and shows a comparable efficiency of our system. Additionally, our system reproduces annotation of purely manually annotated datasets (fly, yeast, etc). However, the performance results for these datasets are low in terms of recall, i.e. 47.4% recall with 80% precision compared to 60.6% recall with the same precision of the complete test set. Note that manual annotation tends to be conservative and sparse, yielding stringent true positive definitions, whereas automatically annotated sequences may accumulate information to a greater extent. We were interested in annotating Xenopus since it is a familiar model organism. However, the sequences were not very well annotated. Our system was applied to annotate the Xenopus contig sequences from TIGR. Through our approach, we annotated 50.5% of all contig sequences available at present, and associated a confidence value for each prediction, yielding roughly three times more sequences as compared to the currently available GO annotation. However, the coverage of annotation to new organism like Xenopus is crucial. We were able to attain predictions for 50.5% of all Xenopus contig sequences (no singletons). This compares to the applied databases that contained 53% satisfactory annotation for their sequences (not regarding sequences with unknown function terms), and better than the organism specific databases (36%). Obviously, improving the quality and quantity of annotation within the available databases goes along with the coverage exploit of machine learning algorithms for new organisms. In future we want to extend our method with the information from other sources such as domain databases and protein family databases. Conclusions We developed an automated annotation system to assign functional GO terms to an unknown sequence. We used the well-established technique of Support Vector Machines (SVM) for the classification of correct and incorrect GO terms. Our approach benefited from the broad variety of potential attributes used for the functional transitivity and a vast amount of data used for training and validating. The committee scheme exploited in our system provided a means to assign confidence values in a straightforward manner. Our system performance was robust, organism-independent and reproduced the high-quality manual annotation. When applying it to Xenopus laevis contig sequences, we obtained a remarkably enhanced annotation coverage compared to the existing annotation. Methods Quality criteria for assessing the performance of the classifier We used the following statistical terms [ 30 , 31 ]. Accuracy was the rate of correct predictions compared to all predictions, Accuracy: = (TP + TN) / (TP + FP + TN + FN),     (1) where TP denotes true positives, FP false positives, TN true negatives and FN false negatives. Precision was the portion of true positives with respect to all positives, Precision: = TP / (TP + FP).     (2) Also used were sensitivity := TP / (TP + FN), specificity := TN / (FP + TN), and false positive rate := 1 - specificity. We defined the term "coverage-of-sequences" as the portion of query sequences for which the classifier delivers a prediction; " Precision-per-sequence" the (average) portion of correct GO terms for a single query sequence, with respect to all GO terms assigned to it. Note that these terms were defined within our model, i.e. a good "accuracy" meant good consistency with respect to our training and test sets. Defining the GO term relationships We focused on the molecular function terms from GO, because the information extracted from the gene products is usually more predictive for determining molecular functions than for biological processes or cellular components. The functional terms and their hierarchy were obtained from the web pages of the Gene Ontology Consortium [ 29 ] (version of June 2003). In our study, relationships "is-a" and "part-of" were not distinguished. Note, that the "part-of" relationship is rare in the molecular function ontology (26 out of 6521 child-parent relationships). The annotation level varies across databases depending on the curator's individual knowledge about the gene product. To consider varying levels of annotation in the databases for similar gene products, we traced the relationships to match GO terms of different granularity for the same function. To find a relationship between two terms, the whole path of a GO term was traced back to the root (the root is the "molecular function" node, GO:0003674). We defined the distance between two GO terms as the distance of the shortest path. GO terms are organised in directed acyclic graphs, i.e. a child (more specialised term) may have multiple parents (less specialised terms). Therefore, we defined single path and multiple path relationships. In the case of single path relationships, GO terms had only one possible path to the root. The relationship of the term GO 1 with respect to GO 2 was classified as "parent", "child", "sibling" or "different" (Figure 1 ) according to the following rules: GO 2 is a "parent" of GO 1 if their respective paths P 2 and P 1 intersect in such a manner that P 1 ⊂ P 2 ,     (3) P i denotes the set of nodes from GO i to the root GO 2 is a "child" of GO 1 if their paths P 2 and P 1 intersect such that P 1 ⊃ P 2 ,     (4) GO 2 is a " sibling" of GO 1 if a common parent exists with a distance of one to GO 1 and GO 2 (Figure 1E ). To avoid ambiguities for less differentiated terms, the sibling relationship was set only, if GO 1 and GO 2 were at least 5 nodes away from the root. The relationship "different" was set if none of the previously stated criteria was fulfilled. We could apply the single path relationship for most of the GO terms (3665 out of 5391). However, for the remaining 1726 terms more than one path to the root were found. For these cases we defined multiple path relationships and each path was considered individually. The single path relationship was applied to each possible pair of these paths (path for GO 1 and GO 2 , respectively) and is henceforth referred to as "path-pairs". This method could yield a list of several relations. To select the appropriate relation from this list, we considered the parent relationship to be most relevant, followed by the child relationship, and the sibling was considered least relevant. We implemented the following order: 1. The parent relationship was set if at least one of the path-pairs gave a (single path) parent relationship; 2. The child relationship was set if at least one of the path-pairs gave a child relationship. To avoid a bias due to an overwhelming number of path-pairs that did not match, we set a threshold: we considered this relationship only, if the number of path-pairs with no child relationship was equal or less than four times the number of path-pairs with child relationship; 3. The sibling relationship was set if at least one of the path-pairs gave a sibling relationship. We again set a threshold: we considered this relationship only, if the number of path-pairs with no sibling relationship was equal or less than twice the number of pairs with sibling relationship; 4. If none of these criteria could be applied, the relationship "different" was set. Note that we also implemented the hierarchy of these relations by tuning the stringencies for the fractions of path-pairs that must match (parent: no threshold, child: 1/4, sibling: 1/2). Data basis used for this study Since the function transitivity at the protein level is more reliable, we used GO-mapped protein databases for searching homologues. Gene association files were obtained via the Gene Ontology Consortium. By combining the gene association files with corresponding sequence databases we created the unified protein databases. The following organisms were used: yeast, fly, mouse, Arabidopsis , worm, rat, fish, Leishmania , Bacillus , Coxiella , Shewanella , Vibrio , Plasmodium , Oryza sativa , Trypanosoma brucei , and Homo sapiens . Apart from this, the SWISS-PROT database was also included [ 32 ]. For SVM training and testing we selected 39,740 cDNA sequences from 13 organisms. These cDNA sequences were collected from the following organisms: yeast, fly, mouse, Arabidopsis , worm, rat, fish, Leishmania , Bacillus , Coxiella , Shewanella , Vibrio and Plasmodium (see Table 1 ). Out of all the known cDNA sequences we extracted 39,740 with GO molecular function terms, discarding incompletely annotated ones, i.e. sequences assigned with the GO term "molecular function unknown" (GO:0005554). Computing the attributes Each cDNA sequence was searched across the protein databases, using BLASTX within the HUSAR system [ 33 ]. A query sequence was not searched within the database of their own organism. In case of SWISSPROT, hits corresponding to the query (cDNA) organism were filtered out. The BLAST files were parsed using the BLAST parser modules from W3H [ 34 ] and a low-stringent e-value cut-off of 0.01 was applied to yield a high number of possible hits. Multiple high scoring segment pairs were combined as described elsewhere [ 35 ] and used for computing the alignment features. GO terms for each database hit were extracted by considering only GO terms corresponding to the molecular function and by discarding GO terms that were prefixed with NOT (annotators state that a particular gene product is NOT associated with a particular GO term), or corresponding to "molecular function unknown" (GO:0005554). These steps reduced our dataset to 36,771 sequences, contributing to 856,632 samples. Each GO term that occurred in the hits represented a sample entry in the feature table. Below it will be referred to as "sample GO term". If a GO term occurred several times in the hits, it was considered only once. We defined 31 attributes for each GO term, representing 5 major classes of information (A)- E)): A) GO level and path: The GO structure was exploited to derive the first two attributes, A.1. GO level : the distance of the sample GO term to the root (molecular function node); A.2. GO path : the number of paths from the sample GO term to the root. B) Alignment quality criteria: These attributes are based on the BLAST alignments. For attributes B.1 - B.4, the best value for the corresponding attribute was taken, if a GO term occurred in more than one hit, B.1. Expectation value: the expectation value ("E-value") from BLASTX; B.2. Bit score : the bit score value provided by BLASTX; We wanted to award alignment length and quality by combining features. This was done with respect to the length of the query and the hits to offset biases due to different complexities of the query and subject organisms. Attributes B.3, B.4, C.3 and D.3 were obtained from initial trials with a small dataset (6270 cDNA sequences, data not shown) and applying parameter optimisation to distinguish the samples. B.3. Query coverage score (QC S ): Combined measure of alignment size and quality concerning the query sequence, QC S := (A L / Q L ) × (I + S),     (5) where A L denotes the alignment length, Q L the length of the query sequence, I the number of identities in the alignment, and S the number of positively contributing residues in the alignment; B.4. Subject coverage score (SC S ): as in B.3, however only with respect to the corresponding subject sequence (database hit), SC S := (A L / S L ) × (I + S),     (6) where S L denotes the length of the subject sequence; Additionally, we decomposed these attributes into the following further six attributes (B.5 - B.10). For these attributes, we considered the hit with the best coverage score if a GO term occurred in more than one hit (query coverage score for attributes B.5, B.7, B.9, and subject coverage score for B.6, B.8, B.10). B.5. Query percentage (QP C ): Percentage of coverage of the alignment region in the query sequence (with respect to QC S ), i.e. QP C := (A L / Q L ) × 100;     (7) B.6. Subject percentage (SP C ) Percentage of coverage of the alignment region in the corresponding subject sequence (with respect to SC S ), i.e. SP C := (A L / S L ) × 100;     (8) B.7. Query identity (QI): Percentage of identical residues in the BLASTX alignment (with respect to QC S ); B.8. Subject identity (SI): Percentage of identical residues in the BLASTX alignment (with respect to SC S ); B.9. Query similarity (QS): Percentage of similar or positively contributing residues in the alignment (with respect to QC S ); B.10. Subject similarity (SS): Percentage of similar or positively contributing residues in the alignment (with respect to SC S ). C) GO frequency related attributes: We extracted information about the frequency of GO terms in the hits by the following attributes: C.1. GO frequency (F G ): the number of hits that contained the sample GO term. C.2. Number of hits (T H ): the total number of hits for the query. C.3. Frequency score (F S ): the number of hits that contained the sample GO term. Unlike C.1, we limited this score to emphasize differences in queries with few hits: C.4. Species frequency : The number of organisms contributing to a sample GO term for a single query sequence; C.5. Total GO (T G ): total number of GO terms from all hits. C.6. Unique GO (U G ): as C.5, except, that GO terms occurring more than once (in the hits) were counted only once. D) GO frequency by considering relationships: For these attributes we applied the structure of the Gene Ontology graph. Not only perfectly matching terms were considered, but also their defined parents, children or siblings: D.1. Relative frequency for all (R A ): the relationships for the sample GO term with all GO terms that occurred in the hits were calculated. The sum of non-"different" relationships i.e. parent, child, or sibling was used for this attribute; D.2. Relative frequency for unique (R U ): similar to attribute D.1, with the exception that GO terms occurring more than once were counted only once. D.3. Relative frequency for all (limited) (R Alim ): same as attribute D.1, however this score was limited to emphasize differences of queries with few hits: D.4. Relative frequency for unique (limited) (R Ulim ): same as attribute D.2, however this score was limited to emphasize differences of queries with few hits: E) Annotation quality related attributes: Quality attributes were selected from the evidence codes provided by the gene association tables of the GO-mapped sequence databases. We selected 9 commonly used evidence codes (TAS, NAS, ISS, IPI, IMP, IGI, IEP, IEA, IDA), resulting in attributes E.1 to E.9. The entries of these attributes for each sample GO term were calculated by summing the occurrences of the corresponding evidence codes of all hits. Training and testing the classifier Before training, normalisation was performed. We normalised the attributes by taking the logarithm (log) and log of log if necessary. We used log values for 16 attributes (B.3-B.10, C.3, C.4, D.1, D.2, D.4 and E.1) and log of log for 8 attributes (B.2, C.1, E.2, E.4-E.8). Furthermore, we converted the attribute values into mean 0 and standard deviation 1 by applying the Z-transformation. The feature table contained 856,632 samples and 31 attributes. We split the dataset into 99 training subsets. Each subset comprised of approximately 1% of the samples i.e. 8,566 GO terms. This resulted in 96 organism specific subsets and 3 hybrid subsets. We applied the support vector machines in the implementation of LIBSVM [ 36 ], which supports a weighted SVM for unbalanced data. We used a higher penalty (5 instead of 1) for false positives (FP) for the model selection and also the training process to support a high specificity of the resulting classifiers. Also note, that our training set contained a high portion of negative samples (see Table 1 ) due to our relaxed E-value threshold. We utilised the radial basis function kernel and set the parameter epsilon (tolerance of termination criterion) to 0.01. The parameter C (regularisation term, cost for false classification) and gamma (kernel width) of the SVM were optimised using a grid search. The grid search determined the combination of C (log2-range: 13 to 15, step 1) and gamma (log2-range: 10 to 15, step 1) with the lowest classification error according to a five-fold cross validation such that each of the 99 data subsets was subdivided into a training set (90%) and a validation set (10%). The validation sets were used to estimate the parameters C and gamma for each of the 99 classifiers individually. Finally, the parameters from the classifier selection were applied to train each of the classifiers with 90% of each data set, respectively. The testing was based on the same 13 organisms and 856,632 GO terms corresponding to 36,771 sequences as described above. We performed the testing by an organism-wise cross-validation so that one organism was used as test set and the remaining ones as the training set. Data availability The annotation for Xenopus laevis contig sequences is downloadable at . We followed the standard GO annotation style (using Gene ontology guideline). The evidence code is always IEA. The confidence value is included for each GO term. Authors' contributions The main work was carried out by AV. RK and KG conceived the idea of the study. AV and RK drafted the manuscript. FS developed and JM applied the machine learning strategy. KG implemented the databases in SRS. RE and SS supervised the work. All authors participated in reading, approving and revising the manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517617.xml

522874

An Ambystoma mexicanum EST sequencing project: analysis of 17,352 expressed sequence tags from embryonic and regenerating blastema cDNA libraries

An EST database has been generated for the axolotl Ambystoma mexicanum . Analysis of this data has uncovered an unusual phylogenetic distribution of the cyclin dependent kinase inhibitor 1 gene family in amphibians.

Background The Caudata (tailed amphibians such as salamanders) are a major focus of work in vertebrate evolution and speciation [ 1 , 2 ]. The salamander is also an important vertebrate model organism for understanding regeneration, being one of the few vertebrates that is able to regenerate entire body structures such as the limb, tail and jaw as an adult. Despite the pivotal role of this animal order in research, comparatively little sequence information is available. In contrast, 458,413 nucleotide sequences exist for the Anura (frogs and toads). This high number is primarily attributable to large EST sequencing efforts for the model organisms for embryology - Xenopus laevis and Silurana tropicalis. A salamander EST project is particularly important as these organisms have extremely large genomes, making a genome project unwieldy and unlikely without specialized approaches such as methylation filtration [ 3 ]. Genome sizes range from 8.5 billion base pairs for Desmognathus monticola (seal salamander) to nearly 70 billion base pairs for Plethodon vandykei (Van Dyke's salamander) [ 4 ]. The ambystomatid Ambystoma mexicanum , a species important for studies in evolution, regeneration and development, has an estimated genome size between 21.9 billion and 48 billion base pairs [ 5 , 6 ] and measurements of its genome in centimorgans (cM) has yielded the largest size reported for a living vertebrate so far (7,291 cM [ 7 ]). In maize, another organism with a large genome, 60,000 sequence reads were required before genome sequencing of methylation-filtered genomic libraries generated significantly more gene sequence information than the available maize EST sequences [ 8 ]. Molecular evolution studies of salamanders have relied primarily on mitochondrial genes such as those for ribosomal RNAs and cytochrome c [ 9 ]. The lack of sequence information among the Caudata hinders the ability to perform sequence comparison with other important gene families. Furthermore, because of the lack of clones, the number of molecular markers available to study salamander embryology and regeneration is low. To address this gap in sequence availability we have generated a large gene sequence set for A. mexicanum . We chose this species because of its role in evolutionary, developmental and regeneration studies. A. mexicanum is easily bred in the laboratory, and animals can be obtained from a large, NSF-funded colony [ 10 ]. We have sequenced inserts from two cDNA libraries, one produced from dorsal regions of stage 18-22 embryos, consisting primarily of neural tube, somite and notochord. The second library was constructed from day-6 regenerating tail blastema tissue. By sequencing from these two sources, our goal was to obtain sequences of transcripts involved in organizing and regenerating the primary body axis. Here we describe the EST gene set, provide an example of molecular phylogenetic analysis of one gene from this collection, and describe the database created for organizing the A. mexicanum EST information. This database is also being implemented for EST sequences from a full-length X. laevis cDNA library, and for sequences from a Canis familiaris EST project. Results Assessment of library and EST sequence quality To generate a diverse set of sequences involved in organizing and regenerating the primary body axis, two independent cDNA libraries were used for sequencing. One was derived from dorsal regions of stage 18-22 embryos containing neural tube, somite and notochord - called the 'neural tube' library - the other from 6-day post-amputation regenerating tail blastema. From 18,432 sequencing attempts 17,522 high-quality sequences were obtained after Phred analysis [ 11 ]. All sequences are 5' reads of the inserts. Of 17,522 high-quality, single-pass sequencing runs, 32 clones contained no insert and 137 sequences were below 32 base pairs (bp). These sequences were excluded from further analysis (32 bp representing the lower limit for assembly of a sequence using TIGR-assembler), yielding 17,352 clones for final analysis. The neural tube library was the origin of 7,469 sequences and the blastema library of 9,883 sequences (Table 1 , and see Materials and methods). As shown in Figure 1a , the average sequence read length peaked between 500 and 600 nucleotides with an average length of 510 nucleotides and a maximum of 871. The blastema and neural tube libraries were unnormalized and unamplified. We assessed library quality and diversity on the basis of the number of redundant clones in the library. Redundancy was estimated by performing BLASTN searches [ 12 ] against all clones sequenced. After sequencing 10,752 clones of the blastema library 42% of the sequences were still unique, and 50% of clones were still singlets after sequencing 7,680 clones from neural tube, indicating that both libraries display high diversity. EST assembly into contigs To identify ESTs belonging to the same open reading frames (ORFs), sequences were assembled into contigs using TIGR-Assembler version 2 [ 13 ]. The 17,353 sequences assembled into 6,594 contigs, of which 217 were less than 100 nucleotides long and excluded from further analysis. A total of 6,377 contigs was therefore left for final analysis (Table 1 ). Of these, 4,561 contigs contained a single clone. The average contig length of the remaining dataset was 616 nucleotides (Figure 1b ). Other than singlets, most of the contigs consisted of two ESTs (884 contigs, Figure 1c ). The largest contigs included cytochrome c oxidase subunit I (469 ESTs), 12S rRNA (445 ESTs), nuclear factor 7 Zn-binding protein A33 (332 ESTs), type II keratin (274 ESTs), keratin (211 ESTs) and cytoplasmic beta-actin (206 ESTs) (Table 2 ). Comparison to existing A. mexicanum genes in NCBI: 6,000 new contig sequences A total of 1,134 ESTs were available from A. mexicanum in the National Center for Biological Information (NCBI) EST databases prior to this work, most of which originate from a sequencing effort of the Voss laboratory ([ 14 ] and S.R. Voss, D. King, N. Maness, J.J. Smith, M. Rondet, S.V. Bryant, D.M. Gardiner, and D.M. Parichy, unpublished work (NCBI-accession numbers BI817205-BI818091); see also [ 15 ]). We examined to what extent our EST dataset overlapped with the sequences available to date. Only 600 of the ESTs in the public database identified one of our contigs in a BLASTN search as a homolog; in 85% of cases, the E-value was below 1E-50 and the sequences can be considered as potentially identical. Existing ESTs in the database largely originate from regenerating limb (S.R. Voss, D. King, N. Maness, J.J. Smith, M. Rondet, S.V. Bryant, D.M. Gardiner and D.M. Parichy, unpublished work). There was, however, only a slight bias of matching contigs to regenerating blastema (49%) as compared to neural tube (44%). Seven percent of identified contigs were found in both libraries. These results mean that our EST data enriches the existing sequence resource of A. mexicanum with approximately 6,000 new gene sequences. BLAST analysis of A. mexicanum contigs to assign homologies To identify putative homologies to known proteins, we subjected the contigs to BLASTX searches against the non-redundant protein database (NR, NCBI) where a cutoff E-value of 1e-05 was used for parsing output files. In our annotation, we used an E-value of 1e-20 as an upper limit to assign significant homology. We note that this does not imply that such sequences are true orthologs. In addition, in cases where no significant homology was found, we used an E-value limit of 1e-05 to designate weak homology. We find this additional category of 'weak homology' useful for data mining. As most contigs do not represent full-length sequences, it is possible that only a highly divergent region of a gene sequence is available in our collection. The category of weak homology allows us to find potential homologs in such situations. For example, the BLAST search for contig Am_4671 yielded the GenBank entry NP_004055, cyclin-dependent kinase inhibitor 1B ( Homo sapiens ), as the top hit with an E-value of 4e-07. This assignment was based on the carboxy-terminal 120 amino acids of the protein, which represents the less conserved region. When we isolated a full-length clone for Am_4671 from our library, we could confirm that it is indeed the axolotl ortholog of cyclin-dependent kinase inhibitor 1B (p27 Kip1 ), as discussed later. Taken together, a total of 3,718 (58%) sequences shared homology with a protein from selected model organisms in the non-redundant database and could be assigned a putative identity. The E-value distribution of the top hits in the non-redundant database is shown in Figure 2a . Of the contigs, 11% matched a protein with an E-value below 1e-99 and are therefore likely to be true orthologs. Seventy percent of the contigs found a hit with an E-value between 1e-20 and 1e-99 and were assigned significant homology. Finally, 19% of contigs had a first hit with an E-value between 1e-19 and 1e-05 and were assigned weak homology to a protein from the non-redundant database. For annotating our database, these top hits from human, mouse ( Mus musculus ), rat ( Rattus norvegicus ), frog ( X. laevis ), zebrafish ( Danio rerio ), fugu ( Takifugu rubripes ), fruitfly ( Drosophila melanogaster ), mosquito ( Anopheles gambiae ), worm ( Caenorhabditis elegans ), newts and the yeast species Saccharomyces cerevisiae , Schizosaccharomyces pombe and Candida albicans were collected and the closest homolog from the above species was used to assign a putative identity. To estimate how many of the clones are full length we examined the BLAST alignments for the position of the alignment in respect to the database sequence. Of the 3,718 sequences with homologs, 1,107 (29.8%) could be aligned in the amino terminus (with the alignment starting before position 10). As the library was poly(dT) primed, many of these clones are likely to represent full-length inserts. Of these 199 (5.4%) could be aligned from the amino terminus to the carboxy terminus and are potential full-length sequences. Forty percent of our EST sequences did not generate a significant hit in the non-redundant protein database. The availability of additional sequence databases including complete genome sequences from several organisms allowed us to expand our BLAST searches to identify all possible homologs to the A. mexicanum contigs. With the remaining set of contigs, we first performed BLASTN searches against the nucleotide non-redundant (NT) database and BLASTX searches against the EST database. Finally, we performed BLASTX searches against the fugu and human proteomes. In all cases, an E-value of 1e-05 was used to assign potentially homologous sequences. Sequences in the NT database identified an additional 134 contigs and a further 220 contigs found a hit in the EST databases. A homolog was found for 3,340 (52%) contigs in the fugu proteome and 3,698 (58%) contigs shared homology with a protein from the human proteome. In total, an additional 468 contigs identified a homolog in the selected databases beyond the original assignment from the non-redundant protein database (Figure 2b ). Gene sequences with no identifiable homology No homologous sequence could be found for 2,191 (34%) contigs in any of the databases searched. Because the library was poly(dT) primed, many of these sequences could represent 3' untranslated regions (3' UTRs). We determined that 953 sequences (43% of non-homologous contigs) contained no ORF and were therefore potential untranslated regions. Thirty of the sequences shared homology to an existing A. mexicanum clone from the EST database (Table 3 ). The complete list of unique ESTs can be downloaded from [ 16 ]. Assignment of the A. mexicanum dataset to common Gene Ontology terms From the homologous proteins found, contigs were assigned a biological process, molecular function and cellular component from the Gene Ontology (GO) database [ 17 ]. The closest annotated homolog in the GO database was used, using an E-value of 1e-20 as a cutoff, for assigning these categories. A biological process could be assigned to 2,156 contigs (34% of all contigs and 58% of those sharing a homolog in the non-redundant database); 2,186 contigs (34% and 59%, respectively) were assigned a molecular function; and 2,198 contigs (34% and 59%, respectively) could be assigned a cellular component. The most abundant molecular function assigned was 'death receptor interacting protein', followed by 'peptidase', the highest-ranking biological process were 'biological process unknown' and 'proteolysis/peptidolysis' and the most abundant cellular components assigned were the 'actin cytoskeleton' and 'transcriptional repressor complex'. The largest fraction of the contigs was assigned a cellular process in the GO category biological process (87% of annotated contigs) (Figure 3a ). We split the biological processes further into different categories: the most abundant categories were 'protein metabolism/modification' (18% of assigned contigs); 'housekeeping functions/metabolism' (17%); 'intracellular transport' (15%); 'cell cycle/proliferation' (13%); 'RNA metabolism' (13%); 'intracellular signaling' (8%); and 'DNA metabolism/repair' (5%) (Figure 3a , Table 4 ). A list of annotated contigs is downloadable from [ 16 ]. Common SMART and PFAM domains in the A. mexicanum dataset To identify potential domains in the axolotl contigs, we performed RPS-BLAST searches against the conserved domain database (CDD, NCBI) [ 12 , 18 ] using the default cutoff E-value of 0.01. A total of 2,199 (34.5%) contigs had a known protein domain in either the CDD or the SMART or PFAM databases. A detailed list of common protein domains identified in our dataset is given in Table 5 . Among the protein domains identified were homeobox domains such as HOX, PAX and Prox1, eight helix-loop-helix (HLH) domains, RNA-binding domains such as KH and RRM, 69 kinase domains, metal- and lipid binding domains and domains involved in cell-cycle control and ubiquitination (RING fingers, HECT domains, three cullin domains and 12 cyclin domains). Many of these domains were annotated for the first time in a sequence from A. mexicanum . We also compared the occurrence of those domains in other vertebrate species. For most of the common protein domains, only a fraction were found in our dataset; many of these are quite abundant compared to X. laevis or Gallus gallus . The RNA-binding domains KH and RRM especially showed high abundance in our contigs. A complete list of domains is downloadable from [ 16 ]. We assigned cellular functions to the identified domains and analyzed the output according to the functional distribution of contigs (Figure 3b ). The most abundant domains were found in the category 'intracellular transport'; this is due to redundant annotations of small GTPases. The second largest fraction belonged to 'RNA-binding and metabolism', followed by 'DNA-binding and transcriptional control'. In silico differential display of A. mexicanum contigs in blastema and neural tube Regeneration versus development We were interested to see if there were strong differences in the sequence representation of the libraries that reflect the different biological processes taking place in each tissue. To this end, we compared the representation of ESTs in the two libraries. This type of in silico differential display has been performed for ESTs in the NCBI collection, and, as with the NCBI differential display data, we have assessed the statistical significance of the differences using Fisher's exact test. A total of 104 contigs met the cutoff value of 0.005 in Fisher's exact test and can therefore be considered differentially expressed. Table 4 provides a detailed comparison of EST representation categorized according to their biological process annotation. Considering the biological properties of the blastema tissue versus the neural tube tissue, we were particularly interested in differential display results of gene sequences that had been assigned to the biological functions of RNA metabolism (as an indicator of an high proliferation index), cell cycle and proliferation and differentiation. The blastema library was produced from tail tissue that was in the process of forming the blastema progenitor cells for regeneration. Blastema formation involves dedifferentiation of mature cells, and entry into rapid cell cycles. In contrast, the neural tube library contains tissue undergoing cell specification and differentiation, such as neurogenesis and somitogenesis. Although these embryonic tissues are still proliferating, the proliferation index of the cells from neural tube should be lower than from blastema. RNA metabolism A total of 168 contigs annotated under RNA metabolism (127 when normalized to the ratio of sequenced ESTs from blastema and neural tube) were more frequently sequenced or uniquely sequenced in blastema (6% of assigned contigs, 2.6% of all contigs). This group included RNA metabolism, RNA processing, splicing, editing, nuclear export, binding, catabolism, cleavage, capping, rRNA modification, rRNA transcription and tRNA aminoacetylation. Forty-five contigs assigned a process in RNA metabolism were upregulated or unique in neural tube (2% of assigned and 0.7% of all contigs). After Fisher's exact test analysis, 24 of the clones were considered differentially regulated in the two libraries; 22 out of the 24 contigs were enriched or unique in blastema (Table 4 ). Cell cycle and proliferation 126 contigs (95 when normalized to sequencing ratios) were assigned as cell-cycle genes (5% of assigned contigs and 1.5% of total contigs) and were more frequently sequenced or uniquely sequenced in the blastema library, compared with 52 in the neural tube library (2.5% and 0.8%, respectively). This category included regulation of mitosis, mitosis, cell-cycle regulation, regulation of cyclin-dependent kinase (CDK) activity, cell proliferation, DNA replication, M phase, mitotic spindle checkpoint, mitotic spindle assembly, chromosome segregation and cytokinesis. As an example, 10 different types of cyclins were found, from various stages of the cell cycle. Seven of the contigs found in cell-cycle regulation met the cutoff criteria of statistical significance in Fisher's exact test. Five out of the seven contigs were more highly represented or unique in blastema (Table 4 ). Differentiation Whereas proliferation-associated genes were found with a higher sequence representation in the blastema library, genes that had been electronically annotated as involved in 'cell differentiation' had a higher representation in the neural tube library. A total of 28 contigs were electronically assigned the biological process 'differentiation'. After Fisher's exact test, five contigs showed differential regulation in this group. Three out of the five contigs were found in neural tube (Table 4 ). Taken together, these results indicate that the two cDNA libraries have differences in sequence representation that appear to correlate with the physiological processes taking place in the two tissues. Gene families involved in cell-cycle control and development in the A. mexicanum dataset As mentioned earlier, the Mexican axolotl is an important model organism for a number of reasons. First, it is the premier vertebrate model for studying regeneration. Second some aspects of caudate development, for instance mesoderm involution and notochord formation, more closely resemble those found in higher vertebrates than do those in other amphibian embryological models such as X. laevis [ 19 ]. Finally, the axolotl has interesting developmental features, particularly in relation to metamorphosis. The axolotl undergoes 'cryptic metamorphosis', which is defined by its existence in a perrenibranchiate state and retaining some larval features into adulthood (for instance gills, larval skin morphology, caudal fins). The animals become sexually mature in this state, and develop only small rudimentary lungs. So far, very few markers are available to study these processes in this organism. We examined our dataset for genes that are potentially useful for studying regeneration features or developmental processes. To this end, we analyzed our data for genes that are either involved in regulating the cell cycle - as would be expected for the highly proliferative tissue of a regenerating body structure - or could play an essential role during development and metamorphosis from the larval to the adult stage. A list of genes that could be assigned to either cell-cycle regulation or development is shown in Table 6 . Among the genes involved in cell-cycle regulation were A-, B- and E-type cyclins, cyclin-dependent kinase 4 (Cdk4), Polo kinase, the kinase inhibitor p27 Kip1 , the protein phosphatase Cdc25A, as well as the anaphase-promoting complex (APC) activator proteins Cdc20 and Cdh1. Representing genes involved in developmental processes, we found transcription factors such as HoxA2, B12, C4 and C8, Pax6, as well as Cdx1 and Cdx2. Furthermore we found several genes for proteins that are part of the transforming growth factor-beta (TGF-β) signaling pathway, such as TGF-β, bone morphogenetic protein 1 (BMP-1), BMP and activin membrane-bound inhibitor, activin receptor type II, as well as the transcription factors Smad5 and Smad8. Genes for proteins such as Smad8 and BMPs might be of especial interest to the research field of embryonic development, as they have been associated with mesoderm involution [ 20 ]. Other important developmental genes that could be found in our dataset include those for Wnt5 and Wnt8, Sonic hedgehog, retinoblastoma binding protein 2, beta-catenin, as well as Frizzled 2, 5 and 7. Finally, it has been shown that the thyroid hormone receptor pathway has an essential role in the timing of metamorphosis in A. mexicanum [ 21 - 23 ]. We identified the protein TRIP12 (thyroid hormone receptor interacting protein 12), which is a HECT-domain-containing ubiquitin ligase and could have an essential role in regulating thyroid hormone response during development and/or metamorphosis. Phylogenetic analysis of the CDKN1 gene family in vertebrates: amphibians contain an unusual CDKN1 family member The EST collection will provide rich data for the phylogenetic comparison of particular genes. Cell cycle and cell differentiation are cellular functions that have been modified in various organisms through evolution and it will be interesting to understand the evolutionary basis of such changes. Here we analyze a particularly interesting gene family, the CDKN1 family of cell-cycle regulators which inhibit cell-cycle progression by binding to and inactivating CDKs. As a starting point for phylogenetic analysis, the mitochondrial 12S ribosomal RNA gene from our collection resulted in the expected tree, with the anuran amphibian X. laevis and the caudate A. mexicanum grouping together compared to other vertebrates such as fish, birds and mammals (Figure 4a ). Next, we constructed an unrooted phylogenetic tree to compare members of the cyclin B family - cyclins B1, B2 and B3. The sequences of each family member formed strictly separate groups, with the A. mexicanum and X. laevis cyclin B1, B2 and B3 genes grouping with their vertebrate orthologs (Figure 4b ). In contrast, we obtained a quite different picture when we examined the CDKN1 family. In most vertebrates, this family consists of three members: p21 (CDKN1A), p27 Kip1 (CDKN1B) and p57 (CDKN1C). In X. laevis , however, only a single family member called p28 Kix1 (also called p27 Xic1 ), which shows unusual sequence features compared to the p27 sequences from any other vertebrate species, had been described in the literature [ 24 , 25 ]. We wondered whether A. mexicanum harbored the 'canonical' p27 Kip1 or a p28 Kix1 similar to that of Xenopus . We initially searched our A. mexicanum data for CDKN1 orthologs and, in contrast to Xenopus , we found a bona fide p27 Kip1 sequence that clusters closer to vertebrate p27 Kip1 sequences compared to the Xenopus p28 Kix1 (Figure 4c,d ). Considering this interesting finding, we then undertook a more complete analysis of the CDKN1 family in vertebrates by searching for CDKN1 family members in several databases: the sequenced genomes from human, mouse, rat, fugu or zebrafish, the recently released genome sequence of X. tropicalis , the X. laevis EST collection, the zebrafish and fugu genomes, and a complementary A. mexicanum and A. tigrinum EST set generated by Putta et al . [ 26 ]. This data mining revealed two striking features about the distribution of CDKN1 family members among vertebrates (Table 7 ). First, the p28 Kix1 orthologs were only found in amphibians ( X. tropicalis , X. laevis , A. mexicanum , A. tigrinum tigrinum ). We were not able to identify a p28 Kix1 -like gene in any other database. These p28 orthologs group as a distinct branch in an unrooted phylogenetic tree (Figure 4c,d ). These data so far suggest that the p28 family is a CDK inhibitor that is specific for amphibians. With new genome sequence data being released, it will be interesting to see whether the most closely related lineage of birds contains a p28-like gene or whether this gene family is found solely in amphibians. Second, CDKN1B (p27 Kip1 ) and CDKN1C (p57) were present in the A. mexicanum databases but were not found in either X. laevis or X. tropicalis , which have far more EST and genome sequence information (Table 7 , Figure 4c,d ). While it is not possible to conclude definitively that Xenopus species lack these genes, the current data are highly suggestive of such a scenario. We examined in depth the phylogenetic relationships of the CDKN1 family members among vertebrates by constructing unrooted phylogenetic trees, either using the most conserved, amino-terminal 88-amino-acid domain, which includes the functionally important Cdk2-interaction region, or the entire coding sequence. Analysis of the amino terminus showed that while A. mexicanum p27 and p57 clearly grouped with their respective orthologs from other vertebrates, the p28 Kix1 proteins from axolotl and the two Xenopus species clustered as a group distinct from any of the other CDKN1 families (Figure 4c ). The p28 Kix1 family showed a closer relationship to p57 than to other CDKN1 members, branching off close to the p57 family. Phylogenetic analysis using the entire coding sequence of the CDKN1 genes, which includes the Cdk2- and PCNA-binding site, resulted in a closer grouping of p28 with the p27 branch (Figure 4d ). In both cases, however, the p28 family clearly formed a separate group from the other CDKN1 families. The Ambystoma mexicanum EST database A relational database with a web-based front end was created to store, navigate and annotate analyzed contigs. The main object of the database is the annotated sequence contig, which contains information about its length, putative identity, computationally calculated expression profile, GO annotation, homologous proteins and identified domains, as well as number and identity of ESTs that build the contig (Figure 5a ). The Gene Identifier (GI) and GO annotation can be modified by the administrator. To circumvent the problem of split contigs, we introduced a super-contig, to which related contigs can be assigned. Furthermore, the administrator can modify the relationship of EST to contig manually. All protein and domain alignments, as well as the assembly of the EST sequences of a contig are stored and can be viewed by the user. On the contig main page, three homologs at most from selected species are shown, with a full list of homologs from selected species displayed on the protein information page (Figure 5c ). To make use easier, an image of the identified domains with the beginning and end base pair of the alignment is shown on the contig page. Individual ESTs can be accessed via the contig page, including their length, storage information, quality information and available trimmed EST-sequence (Figure 5b ). Some of the main advantages of this database are: first, the direct links to source databases such as the NCBI sequence database, GO database, CDD, and the Smart and Pfam databases for identified domains; second, direct visualization of source data such as sequence alignments of contigs to homologs and domains, as well as alignments of EST assemblies; third, easy retrieval of sequences for further analysis like BLAST-searching; fourth, user-specific annotation of contigs; and fifth, easy manipulation and editing of contig annotations. The database will be available from [ 27 ]. Discussion The salamander, and in particular the species A. mexicanum, represents an important vertebrate organism for evolutionary, developmental and regeneration studies. The salamanders provide an essential amphibian counterpoint to the anurans such as X. laevis , displaying distinct embryology and other physiological features. For example, mesoderm involution during gastrulation and subsequent notochord formation is distinctive between A. mexicanum and X. laevis . The characteristics of mesoderm involution in A. mexicanum more closely resemble those found in other vertebrates [ 19 ]. This and other evidence indicates that A. mexicanum and other urodele amphibians are likely to have retained more ancestral features in common with the 'primitive' tetrapod compared to X. laevis , which appears to be more derived. It is interesting that we observed such segregation on the sequence level of the CDKN1 family. X. laevis appears to have a highly unusual make-up of CDKN1 family members. So far, CDKN1A (p21) and the highly derived p28 Kix1 are the only CDKN1 family members found in both X. laevis and X. tropicalis . In contrast, the ambystomatids appear to have all the members of the CDKN1-family - including p28 Kix1 - assuming that the p21 gene is missing purely as a result of lack of sequence information. In addition, our data suggest that p28 is an amphibian-specific variant of the CDKN1 family. Two major questions arise from these data: first, does the amphibian-specific p28 fulfill a cellular function that is unique to this phylogenetic lineage; and second, does the genotypic difference in the gene set of the CDKN1 family in the two amphibian species account for the macroscopic differences observed in developmental mechanisms. The fact that the CDKN1 family is an essential regulator of the cell cycle opens new possibilities for experimental research along these lines. Given the estimates in the number of genes present in the human genome (20,000-50,000) [ 28 ], we estimate that our EST contig set (6,377) contains between 10 to 25% of the total number of genes in the axolotl. While the database is not yet complete, it represents a significant proportion of the axolotl transcriptome. Further sequencing efforts, including an NIH-funded EST sequencing project for the axolotl [ 26 ], will enlarge the current dataset to provide a comprehensive gene sequence resource for this organism. Our analysis indicates that the majority of A. mexicanum genes are homologous to genes present in other vertebrates. Sixty-six percent of contigs gave a significant match in either the non-redundant protein or nucleotide databases, the EST databases or the human and fugu protein databases. Thirty-four percent of contigs could not be assigned a homolog in any of the searched databases, and 44% of those could not be assigned a coding sequence and are therefore considered to be part of the UTR. Nineteen percent of the contigs seem to represent novel genes that have not been found in any other organism so far. The expressed sequence tags generated in this study also provide a large source of sequence information for developmental and regeneration studies. For example, an examination of the database yielded 194 genes involved in cell proliferation, including pivotal cell-cycle genes such as those for Cdc2, 10 different cyclin family members, Cdk4 and p27. A search for developmental molecules involved in intercellular communication yielded Wnt8, Wnt5B, FGF receptor 4a (FGFR4a), Sonic hedgehog, BMP receptor (BMPR) and BMP-1, while a search for homeodomain-containing proteins yielded 11 members, including Cdx1, Cdx2, HoxA2, HoxC8 and HoxB13. The ESTs were derived from two cDNA libraries, stage 18-22 embryonic neural tube/notochord/somite tissue, and day-6 regenerating tail tissue. The embryonic library represents a developmental stage where tissue specification is occurring, whereas the blastema library represents a tissue that is undergoing dedifferentiation, rapid proliferation and cell respecification. Accordingly, we find differences in transcript representation in the two libraries. The blastema library is particularly enriched in cell-cycle genes and RNA metabolism genes, presumably reflecting the high proliferative index of the early regenerating blastema. Conclusions This set of 17,352 ESTs from A. mexicanum was generated to provide a comprehensive sequence dataset for the community of biologists. Forty percent of genes could still be found in singlets, which reflects a high diversity of sequences in our cDNA set. Annotation of the assembled contigs revealed a substantial difference in gene representation in the two sequence libraries, reflecting their biological source - regenerating blastema being in a highly proliferative state and embryonic neural tube being a tissue undergoing differentiation. Sequence analysis of assembled contigs revealed that 64% of genes had a putative homolog in other species; 19.4% of the contigs contained a putative coding sequence and can be considered novel genes. From this, we conclude that A. mexicanum does not contain an unusually high number of organism-specific genes. The CDK inhibitor family CDKN1 was selected for comparative phylogenetic analysis. Unlike the frogs X. laevis and X. tropicalis , ambystomatids most probably contain all members of the CDKN1 family, including the amphibian-specific protein p28 Kix1 /p27 Xic1 , which shows unusual sequence divergence compared to CDKN1 members in other vertebrate species. Such data would support the contention that A. mexicanum is closer to a basal tetrapod compared to X. laevis . The EST sequences and annotated contigs presented in this paper will be a publicly available and useful resource for research in various fields. Materials and methods Plasmid cDNA library construction Total RNA was purified using Trizol (Invitrogen) from 6-day regenerating tail blastemas and from neural tube-somite-notochord-containing tissue dissected from stage 18-22 A. mexicanum embryos. Total RNA quality was assessed by determining the relative brightness of the 28S:18S rRNA bands (2:1). For library construction mRNA was purified and size fractionated, then poly(dT)-primed cDNA was synthesized and directionally cloned into the Not I- Sal I sites of the pCMVSport6 vector. DNA was transformed by electroporation into EMDH10B-TONA bacteria (library construction performed by Invitrogen). Two separate, unnormalized libraries were produced. The blastema library contained an average insert size of 1.67 kb and 2.67 × 10 7 independent transformants and the neural tube library had an average insert size of 1.5 kb and 1.9 × 10 7 transformants. From each library 100,000 clones were arrayed into 384-well plates (Resource Zentrum/Primary Database, Berlin, Germany). Sequencing For sequencing, single-pass reads from the 5' end of the library inserts were performed using a custom-designed SP6 primer: GCACATTAGGCCTATTTAGGTGACA. DNA from bacterial library clones was amplified using the Templiphi reaction, based on φ29 rolling-circle replication of DNA (AP Biotech). Briefly, approximately 0.5 μl of bacterial glycerol stocks were picked up using 96-pin plastic replicators (Genetix) and centrifuged into 96-well PCR plates. Five microliters of denaturing buffer was added, and samples heated to 95°C for 3 min. After cooling, 5 μl Templiphi enzyme was added and samples incubated overnight in a 30°C incubator. The Templiphi reaction provides two advantages for large-scale sequencing projects on capillary sequencers. First, the reaction proceeds to an endpoint where all nucleotide is incorporated, yielding uniform quantities of DNA from varying amounts of starting bacteria (or DNA). Second, the rolling-circle reaction results in large pieces of DNA that, in contrast to plasmid DNA, do not enter the capillary and interfere with the sequencing run. For sequencing reactions, the DNA preparation was diluted fivefold with distilled water. Sequencing reactions were performed using the DYEnamic ET Dye terminator kit diluted twofold with DYEnamic ET dilution buffer (AP Biotech). Five microliters of DNA was added to 5 μl of sequencing reaction mix with primer and cycled 30 times under the following conditions: 95°C 20 sec, 60°C 1 min. Sequencing was performed on a MegaBACE 1000 (AP Biotech). Runs were either performed at injection: 3 kV 60 sec, run: 8 kV 120 min, or injection: 3 kV 60 sec injection, run: 3 kV 360 min. Analysis of library quality The redundancy of the arrayed libraries was tested by performing BLASTN searches [ 12 ] against all sequenced ESTs from the two libraries. Hits against clones other than the query with an E-value lower than 1e-50 were considered for clustering. Submission of ESTs to NCBI GenBank The sequences were submitted to GenBank. After quality control, individual ESTs were used to search the non-redundant protein database (release of July 2004) using the program BLASTX from the standalone NCBI-BLAST package [ 12 ]. For annotation of sequenced ESTs, the top hit of the BLAST output was used, whereby an E-value of 1e-20 was used for significant similarity and an E-value of 1e-05 was used as a cutoff value for weak similarity. Analysis and assembly of sequence data Quality control of sequenced ESTs was performed using the program Phred [ 11 ] using a cutoff of 20 for trimming low-quality regions, and vector trimming was performed using the program cross-match [ 11 ]. (We note here that the arbitrary Phred score reflects the likelihood of a false base. A Phred score of 20 indicates that in 1 out of 100 trials (10 2 ), the base would be false, 30 would reflect a wrongly sequenced base in 1 of 1,000 trials (10 3 ), and so forth.) Sequence and contig files can be downloaded at [ 16 ]. The resulting high-quality sequences were assembled into sequence contigs with the program TIGR-Assembler version 2 [ 13 ]. Alignment of contigs was performed with the program ClustalW with the settings Gap Opening 5 and Gap Extension 85 [ 29 ] or Cap3 [ 30 ], when ClustalW could not correctly assemble the sequences. Assembled contigs were used to perform BLAST searches (BLASTX, BLASTN from NCBI-BLAST [ 12 ]) against the non-redundant protein sequence database (release of November 2003), human and fugu protein databases and the NCBI EST database, all downloaded from the NCBI. Domain searches were done with RPS-BLAST against the conserved domain database (CDD [ 18 ]) from the NCBI. BLAST and domain-search output files were parsed for homologous sequences, whereby an E-value of 1e-05 was used as a cutoff for BLASTN and BLASTX searches against the sequence databases and the default cut-off of 0.01 was considered to yield significant homology to conserved domains from CDD. A gene identifier was assigned to those contigs that showed reliable homology to a sequence in the non-redundant database (E-value cutoff of 1e-20 for significant similarity and 1e-05 for weak similarity). Potential untranslated regions were identified using the program ESTScan [ 31 ]. Electronic annotation of contigs Based on the GO annotation of the closest annotated homolog, contigs were assigned a molecular function, biological process and cellular component from the GO database [ 17 ]. To this end, the GenBank annotation files from the GO database were downloaded and parsed for the gene identifier (gi) numbers of previously identified homologs. The cutoff for annotating an A. mexicanum contig was an E-value of 1e-20. Isolation of the full-length p27 Kip1 gene from the EST sequence Two EST sequences of the p27 Kip1 gene were sequenced in the EST collection but neither were full-length sequences. To isolate the full-length sequence, 200,000 clones of our arrayed blastema and neural tube libraries were screened by PCR. Briefly, the bacterial library clones in each 384-well plate were pooled, mini-prepped and arrayed into 96-well plates (RZPD, Berlin), resulting in 576 DNA pools. These DNA pools were screened by PCR using the custom SP6 primer (GCACATTAGGCCTATTTAGGTGACA) as a forward primer and a gene-specific p27 reverse primer (TGATTTCCAATGGCTGGTTT). Fifty nanograms of DNA from each pool was used for PCR reactions and PCR cycling was performed at the following conditions: 94°C 2 min, 30 cycles of 94°C 15 sec, 65.5°C 30 sec, 72°C 90 sec, followed by 72°C 7 min). The largest positive band (1.1 kb) was gel purified and sequenced on an ABI377 machine using the SP6 primer. Phylogenetic analysis Multiple sequence alignments were done with the program ClustalX [ 32 ] using standard parameters. Phylogenetic analysis of mitochondrial 12S rRNA was done using the programs dnadist, phylogenetic analysis of the cyclin B family and the CDK inhibitor family (CKI family) was done using protdist, both from the Phylip package [ 33 ]. Trees were calculated with the program fitch from the same software package, using 100 iterations. For the CKI family, only the amino-terminal, CDK-inhibitory domain or the full-length sequences were used for construction of a phylogenetic tree. For the cyclin-B family, only the region overlapping in A. mexicanum contigs was used for tree construction. Trees were displayed using the program nj-plot [ 34 ] for the mitochondrial 12S rRNA tree and unrooted [ 34 ] for the CKI- and cyclin B families. Database design A relational database was created using the open source software MySQL as the database server to store and navigate through resulting sequence contigs and annotations. Scripts connecting the web-based front end to the database were written in the programming language Python.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC522874.xml

526192

Psychosocial correlates with depressive symptoms six years after a first episode of psychosis as compared with findings from a general population sample

Background Depression is frequently occurring during and after psychosis. The aim of this study was to analyze if the psychosocial characteristics associated with depression/depressive symptoms in the late phase of a first episode psychosis (FEP) population were different compared to persons from the general population. Methods A questionnaire was sent out to all individuals six years after their FEP and to a general population sample. Depressive symptoms were recorded using a self-rating scale, the Major Depression Inventory. Results Formerly FEP persons had a higher representation of depressive symptoms/depression, unemployment, financial problems and insufficient social network. Depressive symptoms/depression were found to be associated with psychosocial problems. An age and gender effect was found in the general population, but not in the FEP sample. When the psychosocial characteristics were taken into account there were no association between having had FEP and depressive symptoms. Conclusions The association between having been a FEP patient and depressive symptoms/depression disappeared when negative social aspects were taken into account.

Background Major Depression is the one of the most common psychiatric disorders and is frequently occurring in persons with psychotic disorders. Up to 25% of individuals with psychosis have this condition at some point during the course of their illness [ 1 ]. Insufficient social network, unemployment, living alone, financial problems and low social class are among the reported risk factors for depression in prospective studies [ 2 - 5 ]. Studies of characteristics associated with depression in psychosis have reported various results. Associations have been reported with negative as well as positive symptoms, medications and neuroleptic induced movement disorders [ 6 - 9 ]. In a study by Baynes et al, depressive symptoms were explored in a population of 120 patients with stable, chronic schizophrenia living in the community [ 10 ]. Patients who perceived themselves to have poor social support were more likely to be depressed. They proposed that a similar mechanism for the etiology of depression might exist in schizophrenia as in non-schizophrenic persons. The most commonly used depression self-rating scales in population studies were developed before the introduction of DSM-III, e.g. the Beck Depression Inventory (BDI) and the Zung Self-rating Depression Scale (Zung-SDS) [ 11 , 12 ]. The Major Depression Inventory (MDI) used in the present study was developed by Bech et al based on the DSM-IV symptoms of Major Depression and ICD-10 moderate to severe depression [ 13 , 14 ]. The MDI includes symptom thresholds as well as duration criteria. The internal and external validity has been reported to be higher than for Zung-SDS [ 14 ]. In order to evaluate screening scales there is a need for a "golden standard" and several studies has used Schedules for Clinical Assessment in Neuropsychiatry (SCAN) [ 15 , 16 ]. SCAN incorporates the 10 th edition of the Present State Examination and the reliability has been reported to be good [ 17 ]. The authors of the scale have reported the sensitivity to be 0.90 and the specificity 0.82 when validating the MDI versus SCAN in a clinical setting [ 18 ]. The aim of the present study was to analyze if the psychosocial characteristics associated with depressive symptoms in persons six year after a first episode of psychosis (FEP) differed from the associations found in a population sample. Methods Population from the Parachute project The Parachute project is a Swedish multi center study of FEP. It includes all persons from the catchment area who for the first time sought psychiatric help for psychotic symptoms during 1996–1997. Persons who were 18–45 years old and without a dominating substance abuse or diagnosed brain disorder were included in the study. The catchment area covers approximately 20% of the Swedish population. The project integrates an epidemiological approach with intensive psychosocial and medical treatment of a cohort of first episode psychotic patients. It includes a large-scale system of "need adapted treatment" [ 19 ], which includes high degree of psychosocial support, lowest optimal antipsychotic medication, participation of families and treatment in normalized and integrated settings. The participating patients were followed for a period of six years with assessments of psychiatric, psychological, social and economic aspects. The project, which is a controlled study, is described in detail in a previous paper [ 20 ]. Of the 175 patients included in the project from the start 133 were followed during the complete six-year period. The questionnaire included in this study was sent out to the 133 persons six years after their first psychotic episode and 57.1% (76 persons) participated. Those who participated did not differ in age, gender, country of origin or psychiatric diagnosis from those who did not participate. Population from the PART study The PART study is a longitudinal population study of risk factors and social consequences of mental ill health in the Stockholm County. 19 744 Swedish citizens aged 20–64 years registered as living in the county of Stockholm were randomly selected in 1998–2000. This represents 1.8% of the population in this age group in this area. 10 442 persons (53%) participated in the study. The personal identification number (that all Swedish citizens have) of participants as well as non-participants were linked to the following official registers: income and wealth, sick leave, hospital discharge register (including diagnoses) and disability pension. Participation was related to female gender, higher age, higher income and education, being born in Sweden, and having no psychiatric diagnoses in the hospital discharge register or in the disability pension register. The odds ratios for associations between gender, income, country of origin, education and having a psychiatric diagnosis previously according to the registers, were similar among participants and non-participants (Lundberg et al, manuscript). Participants having had a diagnosis of psychotic disorder in the hospital discharge register were excluded from this study (n = 20). Questionnaire A questionnaire was sent out to the population included in the PART study. The questions included risk and protective factors for mental illness as well as psychiatric symptoms scales [ 21 ]. The same questionnaire was replicated in the follow up of the FEP group. The following variables were used in the present study • Demographic characteristics: age, gender and country of origin (Sweden/other). • Financial problems included the availability to get 14 000 SEK (1 797 USD) within a week, if necessary. • Working life: The persons were divided in the following two groups: employed/students and unemployed/disability pension/sick leave/early retirement pension. • Social network: An instrument developed by Unden & Orth-Gomer [ 22 ] was used. This instrument is developed from ISSI, the Interview Schedule for Social Interaction [ 23 ] Two sub scales were used, AVAT-availability of attachment and AVSI-availability of social integration. The score was calculated according to the authors of the scale. • Depressive symptoms: The Major depression Inventory was used strictly according to the authors of the scale [ 13 , 14 ]. The scale covers the ICD-10 as well as the DSM-IV symptoms of depression. It contains 12 items, but functionally it has 10 items since two of them contain sub items (restless/subdued and reduced/increased appetite). Each item gives a score from 0–5 based on the following answers: at no time, some of the time, slightly less than half of the time, slightly more than half of the time, most of the time and always. The MDI can be used as a scale for measuring the severity in which the total score is calculated giving a theoretical score from 0–50. A score of 26 is considered pathological according to the authors of the scale. Statistical analysis Simple factorial ANOVA'S were performed using being in the PART population or not as the dependent variable and age, gender and country of origin as covariates. Pearson's correlation was used to see if the demographic and psychosocial variables correlated with depressive symptoms. These analyses were performed separately in the PART and FEP populations. Additionally multiple regression analyses were performed with the scores on the Major Depression Inventory and a cut off score of 26 or more as the dependent variables. All variables were entered simultaneously. Being in the PART population or not was entered as a variable. Results The demographic and psychosocial characteristics of the two populations are presented and compared in table 1 . Persons in the FEP population more often had unemployment/disability pension/sick leave/early retirement pension (F = 110.26, df 1, p < 0.001) and financial problems (F = 30.06, p < 0.001). Additionally fewer of them had a sufficient social network ((AVSI; F = 52.26, df 1, p < 0.001) and (AVAT; F = 39.18, df 1, p < 0.001)). They also had higher score on the Major Depression Inventory (F = 25.69, df 1, p < 0.001). The symptoms within the Major Depression Inventory were also analyzed separately. The only symptom that was equally distributed was sleep disturbances; all other symptoms were more common in the FEP population. Table 1 Demographic and psychosocial variables in the FEP population and in the general population sample. The statistical analyses were controlled for age, gender and country of origin (Sweden/other). General pop. sample, n = 10 425, %(n) FEP pop. n = 76, %(n) Female gender 55.5(5 798) 50.0(38) Not born in Sweden 10.7(1 120) 15.8(12) Unemployed † 9.7(1 009) 38.2(29)* Financial difficulties 14.9(1 553) 40.8(31)* Mean (95%CI) Mean (95%CI) Age 41.4 (41.2–41.7) 28.5 (26.8–30.3) AVAT (availability of attachment) 17.7 (17.6–17.7) 16.0 (15.1–16.9)* AVSI (availability of social integration) 16.2 (16.2–16.3) 12.4 (11.4–13.4)* Major Depression Inventory 8.8 (8.6–9.0) 15.8 (12.7–18.9)* † Including disability pension/early retirement pension/sick leave, *p < 0.001 When using a MDI cut-off score of 26, 25.8% (17 persons) had a score above the cut-off in the FEP population and 8.0% (783 persons) in the general population sample. Table 2 presents the correlations between the demographic and psychosocial characteristics and the total score of the MDI. In the general population sample all demographic and psychosocial variables were found to be associated with the total score of the MDI. In the FEP population not born in Sweden and being younger were not associated. The gender associations were different in the two samples, while female gender was associated in the general population sample male gender was associated in the FEP population. Table 2 Correlations between the demographic, psychosocial characteristics and total score of the Major Depression Inventory in the FEP population and in the general population sample. General pop. sample N = 10 425 FEP pop. N = 76 Not born in Sweden 0.11*** 0.12 Age -0.13*** -0.01 Female gender 0.13*** -0.21* Unemployment 0.14*** 0.23* Financial problems 0.27*** 0.35** AVAT (availability of attachment) -0.32*** -0.48*** AVSI (availability of social integration) -0.31*** -0.41*** *p < 0.1, **p < 0.01,***p < 0.001 Separate multiple regression analyses were performed in the two samples with the Major Depression Inventory total score as the dependent variable. In the general population sample the adjusted R square was 0.21 (SE 8.7) and in the FEP population 0.28 (SE 11.4). In addition an analysis was made were general population/FEP population was inserted as a variable. The result is presented in table 3 , and shows that the correlation between being a person having had a FEP and higher scores on the Major Depression Inventory no longer was present when the other variables were taken into account. Adjusted R square for this analysis was 0.21 (SE 8.7). A similar regression was performed entering depressed/not depressed using a cut-off score of 26 and the result is also presented in table 3 . Adjusted R square for this analysis was 0.12 (SE 25.5) Table 3 Multiple regression analyses with total score on the Major Depression Inventory and a MDI cut off at 26 as the dependent variables. MDI Score Beta Stand. MDI Cut-off Beta Stand. FEP pop. vs. General pop. sample 0.00 0.00 Not born in Sweden 0.04* 0.04* Age -0.16* -0.07* Female gender 0.13* 0.08* Unemployed 0.10* 0.09* Financial problems 0.10* 0.09* AVAT (availability of attachment) -0.15* -0.19* AVSI (availability of social integration) -0.24* -0.09* *p < 0.001 Discussion The main finding of this study was that the association between having suffered a FEP and self-reported depressive symptoms/depression six years later disappeared when a negative social situation was taken into account. Not surprisingly, persons with a previous FEP had a higher representation of unemployment, financial problems and insufficient social network, which have been reported in other studies [ 24 - 26 ]. In the general population there was an age and gender effect, females and younger age had an overrepresentation of depressive symptoms. This was not seen in the FEP follow up group where age had no effect and being a male was slightly over represented. This is in agreement with a study by Zisook et al [ 9 ]. Not born in Sweden was associated with depressive symptoms in the general population sample, but not in the FEP population. This could have been due to low numbers in the FEP population. The non-participation rate was high in the general population and persons with severe psychiatric disorders most likely did not respond to the enquiry. However, the associations between gender, income, country of origin, education and having a previous psychiatric diagnosis was similar among participants and non-participants. The FEP group also had a high non-participation rate, although the distribution of age, gender, country of origin and psychiatric diagnoses were similar among participants and non-participants. The general population sample was an urban population while the FEP population was from areas all over Sweden, which might have affected the result. The strengths of the study were that the FEP group was a total population, followed over six years and having received treatment according to a "need-adapted approach". Moreover, the instrument in use, the MDI has been reported to have a higher internal validity than Ham-D 17 and Zung-SDS [ 13 , 14 ]. The sensitivity and specificity have been above 0.80 when the MDI was compared to clinical interviews using SCAN [ 18 ]. The results of this study fully agree with the goals of WHO [ 27 ]: Links need to be established between mental health services and various community agencies at the local level so that appropriate housing, income support, disability benefits, employment and other social service supports are mobilized on behalf of patients and in order that prevention and rehabilitation strategies can be more effectively implemented. Following these recommendations would most likely decrease the rates of depressive symptoms in former FEP person with a secondary positive effect on their quality of life in general. Conclusions Having had a first episode psychosis six years earlier had no association with depressive symptoms/depression when a negative social situation was taken into account. Competing interests The authors declare that they have no competing interests. Authors' contributions YF, SL and JC participated in the data collection. YF performed the statistical analysis. All three authors wrote and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526192.xml

517825

The Program of Gene Transcription for a Single Differentiating Cell Type during Sporulation in Bacillus subtilis

Asymmetric division during sporulation by Bacillus subtilis generates a mother cell that undergoes a 5-h program of differentiation. The program is governed by a hierarchical cascade consisting of the transcription factors: σ E , σ K , GerE, GerR, and SpoIIID. The program consists of the activation and repression of 383 genes. The σ E factor turns on 262 genes, including those for GerR and SpoIIID. These DNA-binding proteins downregulate almost half of the genes in the σ E regulon. In addition, SpoIIID turns on ten genes, including genes involved in the appearance of σ K . Next, σ K activates 75 additional genes, including that for GerE. This DNA-binding protein, in turn, represses half of the genes that had been activated by σ K while switching on a final set of 36 genes. Evidence is presented that repression and activation contribute to proper morphogenesis. The program of gene expression is driven forward by its hierarchical organization and by the repressive effects of the DNA-binding proteins. The logic of the program is that of a linked series of feed-forward loops, which generate successive pulses of gene transcription. Similar regulatory circuits could be a common feature of other systems of cellular differentiation.

Introduction A fundamental challenge in the field of development is to understand the entire program of gene expression for a single differentiating cell type in terms of an underlying regulatory circuit. This challenge can be met in part through recent advances in transcriptional profiling, which have made it possible to catalog changes in gene expression on a genome-wide basis ( Brown and Botstein 1999 ). However, most systems of development involve multiple differentiating cell types, complicating the challenge of deciphering the program of gene expression for individual cell types. Also, many developmental systems are insufficiently accessible to genetic manipulation to allow genome-wide changes in gene expression to be understood in detail in terms of an underlying regulatory program. An understanding of how a cell differentiates from one type into another requires both a comprehensive description of changes in gene expression and an elucidation of the underlying regulatory circuit that drives the program of gene expression. Here we report our efforts to comprehensively catalog the program of gene expression in a primitive system of cellular differentiation, spore formation in the bacterium Bacillus subtilis, and to understand the logic of this program in terms of a simple regulatory circuit involving the ordered appearance of two RNA polymerase sigma factors and three positively and/or negatively acting DNA-binding proteins. Spore formation in B. subtilis involves the formation of an asymmetrically positioned septum that divides the developing cell (sporangium) into unequal-sized progeny that have dissimilar programs of gene expression and distinct fates ( Piggot and Coote 1976 ; Stragier and Losick 1996 ; Piggot and Losick 2002 ; Errington 2003 ). The two progeny cells are called the forespore (the smaller cell) and the mother cell. Initially, the forespore and the mother cell lie side by side, but later in development the forespore is wholly engulfed by the mother cell, pinching it off as a cell within a cell. The forespore is a germ cell in that it ultimately becomes the spore and, upon germination, gives rise to vegetatively growing cells. The mother cell, on the other hand, is a terminally differentiating cell type that nurtures the developing spore but eventually undergoes lysis to liberate the fully ripened spore when morphogenesis is complete. The entire process of spore formation takes 7–8 h to complete with approximately 5 h of development taking place after the sporangium has been divided into forespore and mother-cell compartments. Much is known about the transcription factors that drive the process of spore formation, and in several cases transcriptional profiling has been carried out to catalog genes switched on or switched off by individual sporulation regulatory proteins ( Fawcett et al. 2000 ; Britton et al. 2002 ; Eichenberger et al. 2003 ; Feucht et al. 2003 ; Molle et al. 2003a ). Here we have attempted to go a step further by comprehensively elucidating the program of gene expression for a single cell type in the developing sporangium. For this purpose we focused on the mother cell and its 5-h program of gene expression. Gene expression in the mother cell is governed by five positively and/or negatively acting transcription factors. These are the sigma factors σ E and σ K and the DNA-binding proteins GerE, GerR (newly characterized in the present study), and SpoIIID. The appearance of these regulatory proteins is governed by a hierarchical regulatory cascade of the form: σ E →SpoIIID/GerR→σ K →GerE ( Figure 1 A) in which σ E is the earliest-acting factor specific to the mother-cell line of gene expression ( Zheng and Losick 1990 ; results presented herein). The σ E factor is derived from an inactive proprotein, pro-σ E ( LaBell et al. 1987 ), whose synthesis commences before asymmetric division ( Satola et al. 1992 ; Baldus et al. 1994 ), but whose continued synthesis becomes strongly biased to the mother cell after asymmetric division ( Fujita and Losick 2002 2003 ). Proteolytic conversion to mature σ E takes place just after asymmetric division ( Stragier et al. 1988 ) and is triggered by an intercellular signal transduction pathway involving a secreted signaling protein that is produced in the forespore under the control of the forespore-specific transcription factor σ F ( Hofmeister et al. 1995 ; Karow et al. 1995 ; Londono-Vallejo and Stragier 1995 ). Transcriptional profiling has established that σ E turns on an unusually large regulon consisting of 262 genes, which are organized in 163 transcription units ( Eichenberger et al. 2003 ; results presented herein). Among the targets of σ E are the genes for the DNA-binding proteins SpoIIID and GerR ( Kunkel et al. 1989 ; Stevens and Errington 1990 ; Tatti et al. 1991 ; Wu and Errington 2000 ; results presented herein). SpoIIID is both a negatively acting protein that switches off the transcription of certain genes that have been activated by σ E and a positively acting protein that acts in conjunction with σ E -containing RNA polymerase to switch on additional genes, including genes involved in the appearance of σ K ( Kroos et al. 1989 ). Figure 1 The Mother-Cell Line of Gene Transcription (A) Gene transcription is governed by a hierarchical regulatory cascade that involves gene activation and gene repression. The σ E factor turns on a large regulon that includes the genes for GerR and SpoIIID. These DNA-binding proteins, in turn, block further transcription of many of the genes that had been activated by σ E . SpoIIID is also an activator, and it turns on genes required for the appearance of pro-σ K . The conversion of pro-σ K to mature σ K is governed by a signal emanating from the forespore as represented by the squiggle. Next, σ K activates the subsequent regulon in the cascade, which includes the gene for the DNA-binding protein GerE. Finally, GerE, which, like SpoIIID, is both an activator and a repressor, turns on the final regulon in the cascade while also repressing many of the genes that had been activated by σ K . The thickness of lines represents the relative abundance of genes activated (arrows) or repressed (lines ending in bars) by the indicated regulatory proteins. (B) The regulatory circuit is composed of two coherent FFLs linked in series and three incoherent FFLs. In the first coherent FFL, σ E turns on the synthesis of SpoIIID, and both factors act together to switch on target genes, including genes involved in the appearance of σ K . Likewise, in the second coherent FFL, σ K directs the synthesis of GerE, and the two factors then act together to switch on target genes (X 4 ). The σ E factor and SpoIIID also constitute an incoherent FFL in which SpoIIID acts as a repressor to downregulate the transcription of a subset of the genes (X 2 ) that had been turned on by σ E . Similar incoherent FFLs are created by the actions of σ E and GerR (X 1 ) and by σ K and GerE (X 3 ), with GerR and GerE repressing genes that had been switched on by σ E and σ K , respectively. The AND symbols indicate that the FFLs operate by the logic of an AND gate in that the output (either gene activation or a pulse of gene expression) requires the action of both transcription factors in the FFL (see Mangan and Alon 2003 ). For example, σ K and GerE are both required for the activation of X 4 genes, whose induction is delayed compared to genes that are turned on by σ K alone. Similarly, both σ E and the delayed appearance of GerR are anticipated to create a pulse of transcription of X 1 genes. The appearance of σ K is a critical control point that involves multiple levels of regulation: transcription, DNA recombination, and proprotein processing. SpoIIID both activates the transcription of the 5′ coding region for σ K (spoIVCB) and that for a site-specific DNA recombinase (spoIVCA) ( Kunkel et al. 1990 ; Halberg and Kroos 1994 ) that joins the 5′ coding sequence to the 3′ coding region by the excision of an intervening sequence of 48 kb called skin ( Stragier et al. 1989 ). Finally, the product of the intact coding sequence is an inactive proprotein, pro-σ K ( Kroos et al. 1989 ), whose conversion to mature σ K (as in the case of pro-σ E ) is governed by a complex, intercellular signal transduction pathway involving a secreted signaling protein that is produced in the forespore under the control of the forespore-specific transcription factor σ G ( Cutting et al. 1990 , 1991a ; Lu et al. 1990 ). The signal transduction pathway helps to coordinate the appearance of σ K in the mother cell with the timing of events taking place in the forespore. The σ K factor turns on an additional gene set that includes the gene for GerE ( Cutting et al. 1989 ), a DNA-binding protein that is responsible for activating the final temporal class of genes in the mother-cell line of gene expression ( Zheng et al. 1992 ). Other than the case of σ E , little was previously known about the full set of genes, whose transcription is governed by the five regulators in the mother-cell line of gene expression—indeed, nothing at all in the case of GerR, whose function had previously been uncharacterized. Here we present evidence indicating that the program of mother-cell-specific gene transcription involves the activation of at least 383 genes (242 transcription units), representing 9% of the genes in the B. subtilis genome. We explain the pattern of transcription of each of these genes in terms of the action of the five regulatory proteins that govern the mother-cell program of gene transcription. Our results reveal that the program chiefly consists of a series of pulses in which large numbers of genes are turned on and are then turned off shortly thereafter by the action of the next regulatory protein in the hierarchy. Evidence is also presented that this repression is critical for proper morphogenesis. Finally, we show that the mother-cell program of gene transcription can be understood in terms of a simple regulatory circuit involving a linked series of feed-forward loops (FFLs) that are responsible for generating pulses of gene transcription. We propose that this regulatory circuit will serve as a model for understanding other programs of cellular differentiation. Results Transcriptional Profiling Our strategy for elucidating the mother-cell program of gene transcription was to carry out transcriptional profiling at hourly intervals during sporulation at 37 °C, starting just after asymmetric division and ending before the time at which lysis of the mother cell had commenced. At each time point, RNA from cells mutant for the transcriptional regulator that was maximally active at that time interval was compared against RNA from cells mutant for the next transcription factor in the hierarchy or, in the case of the last regulatory protein in the hierarchy, GerE, against RNA from wild-type cells. Thus, at hour 2.5, RNA from cells mutant for σ E (strain PE437) was compared against RNA from cells (strain PE436) that were wild type for σ E but mutant for the next regulatory protein in the sequence, SpoIIID. Likewise, at hour 3.5, RNA from cells that were mutant for SpoIIID (strain PE456) was compared against RNA from cells that were mutant for σ K (strain PE452). (Strains PE456 and PE452 were additionally mutant for σ G to eliminate indirect effects of the presence or absence of SpoIIID on the activity of the forespore-specific transcription factor. Although SpoIIID has no direct effect on σ G , the absence of negative feedback on several σ E -controlled genes [see below] in the strain mutated for spoIIID could have had indirect consequences on σ G activity.) Likewise, at hour 4.5, RNA from cells that were mutant for σ K (strain PE455) was compared against RNA from cells mutant for GerE (strain PE454). Finally, at hours 5.5 and 6.5, RNA from cells mutant for GerE was compared against RNA from wild-type cells (PY79). Three transcriptional-profiling analyses were carried out for each of these time points, using three independent preparations of RNA from each of the two cultures of cells that were being compared against each other. The complete dataset for these experiments is presented in Table S1 , and transcriptional profiles for representative genes are displayed in Table 1 . Table 1 Transcriptional Profile of Representative Genes a Ratios of relative RNA levels in sigE + versus sigE mutant b Ratios of relative RNA levels in spoIIID + versus spoIIID mutant c Ratios of relative RNA levels in gerR + versus gerR mutant d Ratios of relative RNA levels in sigK + versus sigK mutant e Ratios of relative RNA levels in wild type versus gerE mutant In addition to the four previously known members of the hierarchical regulatory cascade, one of the genes in the σ E regulon is inferred to encode a previously uncharacterized DNA-binding protein YlbO ( Wu and Errington 2000 ; Eichenberger et al. 2003 ). Additional transcriptional-profiling experiments were carried out to assess the function of this putative regulatory protein. Updating the σ E Regulon We previously reported that the σ E regulon is composed of 253 genes, organized in 157 transcription units. Since then two additional σ E -controlled genes, yjcA ( Kuwana et al. 2003 ) and ctpB (yvjB) ( Pan et al. 2003 ), have been identified. These genes were found to be transcribed in a σ E -dependent manner during sporulation in our previous analysis, but they were not significantly induced in cells engineered to produce σ E during growth and hence had not been included in our original list of σ E -controlled genes. In addition, results presented here (see below) show that one gene, ypqA, and two operons, yhcOP and yitCD, that are chiefly under the control of σ K , are also transcribed, albeit at a low level, in a σ E -dependent manner. These and other considerations (see below) bring the current total number of genes in the σ E regulon to 262 and the total number of transcription units to 163 ( Table 2 ). Table 2 Genes Activated in the Mother-Cell Line of Gene Expression a Includes four genes (two transcription units) that were also repressed by SpoIIID and one gene (one transcription unit) that was also activated by SpoIIID b Only includes genes that were strongly dependent upon SpoIIID for expression. Fifteen genes (11 transcription units) that were partially dependent on SpoIIID are in the “activated by σ E ” category c Only includes genes that were strongly dependent upon GerE for expression. Twenty-eight genes (12 transcription units) that were partially dependent on GerE for expression are in the “activated by σ K ” category d Numbers include genes and transcription units that were transcribed under the control of σ E as well as σ K This updated description of the σ E regulon does not include genes and transcription units that are additionally strongly dependent upon SpoIIID for their transcription because our previous transcriptional-profiling experiments were performed with a strain that was mutant for SpoIIID. SpoIIID is a DNA-binding protein that acts in conjunction with σ E -containing RNA polymerase ( Kroos et al. 1989 ; Kunkel et al. 1989 ; Halberg and Kroos 1994 ). Therefore, as a starting point for the present study, we investigated the influence of SpoIIID on the global pattern of σ E -directed transcription. As we shall see, this analysis revealed ten genes (representing eight transcription units) that were strongly dependent upon SpoIIID for expression and were not expressed under the control of σ E alone , bringing the present total number of genes in the σ E regulon to 272 and the total number of transcription units to 171 ( Table 2 ). SpoIIID Is Both a Repressor and an Activator of Genes Whose Transcription Is Dependent Upon σ E Transcriptional profiling revealed that SpoIIID had profound effects on the global pattern of σ E -directed gene transcription. As many as 181 genes were found to be downregulated in the presence of SpoIIID. Of these, 148 had previously been identified as being activated in a σ E -dependent manner, at least 112 of which (representing 62 transcription units) were bona fide members of the σ E regulon (that is, they met multiple criteria for being under the direct control of σ E ) (see Table S2 ). Therefore, a principal function of SpoIIID is to inhibit the transcription of a substantial proportion (greater than 40%) of the genes whose transcription had been activated by σ E prior to the appearance of SpoIIID. Members of the σ E regulon that are downregulated by SpoIIID are colored green in Figure 2 A. Figure 2 Location of Genes in the σ E and σ K Regulons and Their Regulation by DNA-Binding Proteins (A) The σ E regulon and its modulation by SpoIIID and GerR. The first gene of each σ E -controlled transcription unit identified by transcriptional profiling is indicated. In the inner circle, genes repressed by SpoIIID are green, and genes repressed by GerR are blue. In the outer circle, genes partially dependent on SpoIIID for expression are orange, and genes strongly dependent on SpoIIID are red. Underlined are SpoIIID-controlled genes for which SpoIIID binding to their upstream sequences has been demonstrated biochemically. Genes unaffected by SpoIIID or GerR are indicated in black. (B) The σ K regulon and its modulation by GerE. The first gene of each σ K -controlled transcription unit identified by transcriptional profiling is indicated. In the inner circle, genes repressed by GerE are green. In the outer circle, genes partially dependent on GerE for expression are orange, and genes strongly dependent on GerE are red. Genes unaffected by GerE are indicated in black. SpoIIID not only repressed many genes in the σ E regulon but also stimulated or activated the transcription of many others. At least 70 genes were identified whose transcription was upregulated by SpoIIID ( Table S2 ), but in many cases these genes were not members of the σ E regulon, and the effect of SpoIIID could have been indirect. Examples are seven genes (cysK , cysH, cysP, sat, cysC, yoaD, and yoaB) from the S-box regulon ( Grundy and Henkin 1998 ) and two genes (argC and argJ) from the arginine biosynthesis operon ( Smith et al. 1989 ). In other cases, however, SpoIIID stimulated or activated the transcription of genes that had been reported to be under the control of σ E . Thus, 13 (asnO, cwlJ, proH, proJ, spoIVCA, spoIVCB, spoVK , yhbB, yheC, yheD, yknT, yknU, and yknV) of the genes whose transcription was upregulated by SpoIIID had previously been assigned to the σ E regulon, and four others (mpr, ycgM, ycgN, and yqfT) were known to be under σ E control but had not met all of the criteria for assignment to the σ E regulon ( Eichenberger et al. 2003 ). In two of these 17 cases (spoIVCA and spoVK), the dependence on SpoIIID was almost complete, whereas in the other 15 the dependence was partial. Our analysis revealed eight additional genes (cotF, cotT, cotV, cotW, lip, ydcI, yheI, and yheH) that were almost completely dependent on SpoIIID for their transcription and that are likely to be under the dual control of σ E and SpoIIID. Thus, in addition to repressing at least 112 members of the σ E regulon, SpoIIID activates the transcription of 25 other members of the regulon, representing 19 transcription units. The 15 σ E -transcribed genes (11 transcription units) whose expression was partially dependent upon SpoIIID are indicated in orange in Figure 2 B, and those whose expression was completely dependent on the DNA-binding protein are indicated in red (ten genes; eight transcription units). Evidently, then, SpoIIID plays a pivotal role in the mother-cell line of gene expression, negatively or positively affecting the transcription of many members of the σ E regulon. It was therefore important to determine whether the genes so affected were direct targets of the DNA-binding protein. For this purpose, we used three complementary approaches to identifying binding sites for SpoIIID: biochemical analysis by gel electrophoretic mobility-shift assays (EMSAs) and DNAase I footprinting, in vivo analysis by chromatin-immunoprecipitation in combination with gene microarrays (ChIP-on-chip), and the identification of SpoIIID-binding sequences by computational analysis. Biochemical Identification of SpoIIID-Binding Sites We selected 18 of the newly identified SpoIIID-regulated genes for EMSA analysis, mostly on the basis of the importance of their role in sporulation. As positive controls, we subjected two previously known targets of SpoIIID, bofA and spoIVCA ( Halberg and Kroos 1994 ), to EMSA analysis, and as negative controls three Spo0A-regulated genes ( Molle et al. 2003a ), abrB, racA, and spoIIGA ( Figure 3 A). SpoIIID exhibited binding to the upstream sequence of all 18 of the selected genes ( Figure 3 B). In some cases (those of asnO, gerM, spoIVA, spoIVFA, ybaN, ycgF, yitE, ykvU, and ylbJ ) additional shifted bands were detected at high concentrations of SpoIIID, which may indicate the presence of two or more SpoIIID-binding sites with distinct binding affinities. Figure 3 Gel Electrophoretic Mobility-Shift Analysis of SpoIIID Binding DNA fragments of interest were amplified by PCR, gel-purified, and end-labeled using [γ- 32 P]-ATP and polynucleotide kinase. Purified SpoIIID was added at increasing concentrations (0 nM for lanes 1 and 5, 50 nM for lane 2, 100 nM for lane 3, and 200 nM for lane 4) and incubated at room temperature for 30 min before loading on to a nondenaturing gel containing 6% polyacrylamide. With the exception of (D), the DNA fragments corresponded to the upstream regions of the indicated genes. See Materials and Methods for the identity (coordinates) of the specific DNA sequences used in the analyses. (A) Gel shifts for known targets of SpoIIID ( bofA and spoIVCA ), representing positive controls, and genes ( abrB, spoIIGA, and racA ) under the control of another DNA-binding protein (Spo0A), representing negative controls. (B) Gel shifts for genes identified as possible targets of SpoIIID by transcriptional profiling. (C) Gel shift for cotE . Expression of cotE from its P2 promoter is strongly dependent on SpoIIID. No binding of SpoIIID to the upstream sequence for cotE is observed, suggesting that the effect of SpoIIID on transcription from the P2 promoter is indirect. (D) Gel shifts for chromosomal regions strongly enriched for SpoIIID binding as judged by ChIP-on-chip analysis. For each region, four consecutive DNA fragments of approximately 400 nucleotides in length were analyzed. In addition, we also subjected the upstream region of cotE to EMSA analysis ( Figure 3 C). The cotE gene is transcribed from two promoters: a σ E -controlled promoter called P1 and a second promoter called P2 that strongly depends on SpoIIID ( Zheng and Losick 1990 ). It had been assumed that transcription from P2 is under the dual control of σ E and SpoIIID, but EMSA analysis failed to reveal a binding site for SpoIIID, and other work presented below indicates that transcription from cotE P2 is governed by σ K rather than by σ E . We conclude that the SpoIIID dependence of cotE P2 is an indirect consequence of the dependence of σ K synthesis on SpoIIID. To obtain further evidence for direct interaction by SpoIIID and to investigate the mechanism by which SpoIIID inhibits transcription, we subjected the promoter regions of three genes (spoIID, spoIIIAA, and spoVE) identified as being under the negative control of SpoIIID to DNAase I footprinting analysis. SpoIIID protected two regions in the upstream sequence of spoIID from DNAase I digestion ( Figure S1 ). One region (extending from positions −10 to −28 on the top strand and from −18 to −35 on the bottom strand) overlapped with the −10 element of the σ E promoter, and the other (extending from −33 to −52 on the top strand) overlapped with the −35 element. The binding site for SpoIIID also overlapped with the promoter in the case of spoIIIAA, in this case protecting a single sequence that included the −35 element (extending from −21 to −45 on the top strand and from −30 to −48 on the bottom strand). Finally, the regulatory sequence of spoVE exhibited two binding sites, one (extending from +16 to −1 on the bottom strand) that was located in the vicinity of the predominant σ E -controlled promoter (P2) for this gene and another further upstream, overlapping with a secondary promoter (P1) (extending from +13 to −7 on the top strand). Thus, repression of the promoters of spoIID, spoIIIAA, and spoVE by SpoIIID is likely to be a direct consequence of the binding of the sporulation regulatory protein to the promoter in such a way as to compete with binding by σ E –RNA polymerase. SpoIIID Binds to Some Sites that Do Not Correspond to Genes under Its Control ChIP-on-chip analysis was carried out as described in Materials and Methods and previously ( Molle et al. 2003a , 2003b ), using DNA–protein complexes from formaldehyde-treated cells at hour 3 of sporulation. After sonication, SpoIIID–DNA complexes were precipitated with antibodies against SpoIIID. Next, after reversal of the cross-links, the precipitated DNAs were amplified by PCR in the presence of cyanine 5-dUTP. In parallel, total sonicated DNA from the formaldehyde-treated cells (i.e., DNA that had not been subjected to immunoprecipitation) was similarly amplified, but in the presence of cyanine 3-dUTP. The two differentially labeled DNAs were combined and hybridized to the same batch of DNA microarrays that were used for the transcriptional-profiling experiments. Transcriptional profiling was carried out with three independent preparations of formaldehyde-treated cells, twice with two of the preparations and once with the third, for a total of five analyses. An enrichment factor was calculated for each gene, representing the enrichment of that gene by immunoprecipitation relative to DNA that had not been subjected to immunoprecipitation, and the entire dataset is displayed in Table S3 . Thirty-one genes, corresponding to 26 regions of the chromosome, were found to be enriched by immunoprecipitation by a factor of two or greater. Only seven of the regions (cotF, lip, spoIIIAF, spoVD, ycgF, yhbH, and ykvI) identified by the ChIP-on-chip analysis were in close proximity to a gene that was differentially expressed in the SpoIIID transcriptional-profiling experiments. Thus, in only a small number of cases did ChIP-on-chip analysis support the idea that a gene under SpoIIID control was a direct target of the DNA-binding protein. Our interpretation of these findings is that ChIP-on-chip is less sensitive for detecting SpoIIID-binding sites than it is for the B. subtilis DNA-binding proteins CodY ( Molle et al. 2003b ), Spo0A ( Molle et al. 2003a ), and RacA ( Ben-Yehuda et al. 2003 ). Likely contributing to this decreased sensitivity is the fact that SpoIIID is present in only one of the two chromosome-containing compartments (the mother cell) of the sporangium and that its concentration is low (∼1 μM; Zhang et al. 1997 ). While providing support for only a small proportion of the herein identified targets of SpoIIID regulation, ChIP-on-chip analysis, nonetheless, proved to be revealing. Specifically, we found that SpoIIID bound to many regions of the chromosome that did not correspond to genes under its negative or positive control. Were these regions bona fide SpoIIID-binding sites? To address this question, we subjected five regions that were most enriched for SpoIIID-binding (albE – albF, dctR – dctP, tenI – goxB – thiS, treA – treR–yfkO, and yfmC–yfmD) to EMSA analysis ( Figure 3 D). Given that SpoIIID was not exerting a transcriptional effect in these regions, we reasoned that the sites to which SpoIIID was binding might not reside in upstream regulatory regions and could instead be located in coding sequences. We therefore scanned across each of the five chromosomal regions by EMSA using successive DNA fragments of about 400 bp in length. The results showed that each of the five regions contained more than one binding site for SpoIIID and that some of these binding sites were indeed located within protein-coding sequences. (The presence of more than one binding site in each region may have facilitated their detection by the ChIP-on-chip analysis.) We conclude that SpoIIID binds to some sites on the chromosome at which it does not function as a transcriptional regulator. Conceivably, it plays an architectural role in the folding of the chromosome in the mother cell in addition to its role as a transcriptional regulator. Moqtaderi and Struhl (2004) have similarly found that in Saccharomyces cerevisiae the RNA polymerase III transcription factor TFIIIC binds to sites where binding of other components of the RNA polymerase III machinery is not detected and where the transcription factor does not activate transcription. Identification of Putative SpoIIID-Binding Sites by Bioinformatics As a final, computational approach to identifying direct targets of SpoIIID, we used the Gibbs sampling algorithm BioProspector to identify conserved motifs in sequences upstream of genes under the control of SpoIIID ( Liu et al. 2001 ). Initially, we limited our search to 40 regions where SpoIIID binding had been confirmed by biochemical analysis. BioProspector was used to find the best 35 motifs across several different widths (6–12 bp) under the restriction that every sequence had to contain at least one site. Each of these motifs was separately used as a starting point for BioOptimizer ( Jensen and Liu 2004 ) and applied to an expanded dataset that included the 89 upstream sequences for all SpoIIID-controlled genes (not just those analyzed by EMSA or footprinting). BioOptimizer optimized both the set of predicted sites and the motif width, as detailed in the Materials and Methods section. BioOptimizer was required to identify at least one binding site in the sequences that had been confirmed by EMSA but was unrestricted for the sequences for which a binding site had not been confirmed biochemically. The optimized motif was 8 bp in length and identified at least one putative SpoIIID-binding site in 60 of the 89 upstream sequences that were analyzed (see Table S2 ). Figure 4 shows that the logo for the optimized motif (B) was similar to a consensus sequence (A) that was derived independently using 12 previously reported binding sites (for the genes bofA, cotD, spoVD, spoIVCA, and spoIVCB; Halberg and Kroos 1994 ; Zhang et al. 1997 ) and five sites herein identified by DNAase I footprinting. Figure 4 Consensus Sequences for SpoIIID, σ K , and σ E Consensus sequences are displayed as sequence logos ( Schneider and Stephens 1990 ). The height of the letters in bits represents the information content at each position (the maximum value is two bits). (A) Consensus binding sequence for SpoIIID as derived from 17 SpoIIID-binding sites mapped by DNAase I footprinting ( Halberg and Kroos 1994 ; Zhang et al 1997 ; results presented herein). (B) Consensus binding sequence for SpoIIID obtained by compilation of 68 putative SpoIIID-binding sites identified as common motifs by BioProspector and BioOptimizer analysis in sequences upstream of genes identified by transcriptional profiling or within regions identified by ChIP-on-chip analysis. (C) Consensus binding sequence for SpoIIID obtained by MDscan analysis of the sequences of 26 SpoIIID-binding regions identified by ChIP-on-chip analysis. (D) Consensus promoter sequence for σ K -containing RNA polymerase obtained from the compilation of 58 sequences identified as common motifs in regions upstream of σ K -regulated genes by a BioProspector/BioOptimizer computational approach ( Jensen and Liu 2004 ). Positions 1–5 on the horizontal axis correspond to the −35 element and positions 21–30 to the −10 element. The optimal spacing between the two regions is 15 bp (± 1 bp). (E) Consensus promoter sequence for σ K -containing RNA polymerase obtained from the compilation of 23 previously mapped ( http://dbtbs.hgc.jp/ ; Helmann and Moran 2002 ) and 18 newly identified σ K -controlled promoters identified by transcription start site mapping. (F) Consensus promoter sequence for σ E -containing RNA polymerase obtained from the compilation of 62 σ E -controlled promoters identified by transcription start site mapping ( Eichenberger et al. 2003 ). Positions 1–8 on the horizontal axis correspond to the −35 element, and positions 21–30 to the −10 element. The optimal spacing between the two regions is 12 bp (± 1 bp). In an independent computational approach, we sought to identify a conserved motif in the 26 regions that had been identified by ChIP-on-chip analysis, which likely represent the strongest binding sites for SpoIIID. We used Motif Discovery scan(MDscan) ( Liu et al. 2002 ) for this analysis, which is designed to identify conserved motifs in sequences that have been ranked according to their enrichment factor in ChIP-on-chip experiments. The resulting sequence logo is displayed in Figure 4 C. Whereas it is largely similar to that obtained from the BioProspector/BioOptimizer analysis ( Figure 4 B), there is one notable difference: The first position of the binding motif corresponds almost exclusively to a guanine in the sites identified by ChIP-on-chip analysis. The presence of a guanine at this position could be characteristic of high-affinity sites for SpoIIID binding. In conclusion, SpoIIID negatively or positively influences the transcription of over half of the members of the σ E regulon, and a combination of complementary approaches leads us to believe that it does so for many of the genes so identified by direct interaction with their promoter regions. In the case of genes under the negative control of SpoIIID, the mechanism of this repression probably involves steric interference as the inferred binding sites for SpoIIID were generally found to overlap with the expected binding sites for RNA polymerase. No such overlap was generally observed in the case of genes under the positive control of SpoIIID. GerR (ylbO), a Second Negative Regulator of the σ E Regulon The spoIIID gene is not the only member of the σ E regulon that appears to encode a DNA-binding protein. The inferred product of ylbO exhibits significant similarity to members of the basic leucine zipper family of transcription factors and is, in particular, 52% similar to RsfA ( Wu and Errington 2000 ), a regulator of σ F -controlled genes in the forespore line of gene expression. To study a possible role for ylbO we investigated the effect of a null mutation of the gene on sporulation and on σ E -directed gene expression. As noted previously, the mutation has no effect on the production of heat-resistant spores, but we have now discovered that the mutation causes a conspicuous defect in the capacity of the spores to germinate, as judged by their impaired ability to reduce 2,3,5-triphenyltetrazolium chloride (see Materials and Methods ). We therefore rename ylbO as gerR (in keeping with the nomenclature for germination genes in B. subtilis [ Setlow 2003 ]). We also carried out transcriptional profiling using RNA collected at hour 3.5 of sporulation from cells of a strain (PE454) that was wild type for GerR and from cells of a newly constructed strain (SW282) that was mutant for GerR. Both strains were also mutant for the next transcription factor in the hierarchical cascade, σ K . No genes were identified whose transcription was dependent on GerR, but 139 genes were found that were downregulated in a GerR-dependent manner by a factor of two or greater (see Table S1 ). Among the downregulated genes were 14 members of the σ E regulon. Nine of these members (colored blue in Figure 2 A) were known not to be under SpoIIID control (cypA, kapD, spoIIM, spoIIP, ybaS , yfnE , yfnD, yhjL, and yqhV), whereas the remaining five (phoB, spoIIIAA, spoIIIAB, spoIVCA, and ydhF) were also under the control of SpoIIID. We selected three of the putative targets of GerR for further analysis. The promoter sequences of spoIIM and yqhV were fused to the coding sequence of β-galactosidase and introduced into the chromosome at the amyE locus and a previously constructed fusion of lacZ to spoIIP (amyE :: spoIIP - lacZ) was obtained from P. Stragier (Institut de Biologie Physico-Chimique, Paris). The results, shown in Figure 5 , confirmed that GerR had a pronounced negative effect on the level of expression of all three fusions. Figure 5 Repression of σ E -Controlled Genes by GerR Culture samples from strains PE551 (solid triangles, amyE ::P spoIIM –lacZ ), SW312 (open triangles, amyE ::P spoIIM –lacZ, Δ gerR ), PE511 (solid squares, amyE :: spoIIP–lacZ ), PE568 (open squares, amyE :: spoIIP–lacZ, Δ gerR ), PE553 (solid diamonds, amyE ::P yqhV –lacZ ), and PE558 (open diamonds, amyE ::P yqhV –lacZ, Δ gerR ) were collected at indicated intervals after the start of sporulation in Sterlini–Mandelstam medium and analyzed for β-galactosidase activity. An example of σ E -controlled genes that are under the dual negative control of GerR and SpoIIID is the eight-cistron spoIIIA operon ( Illing and Errington 1991 ). As we have demonstrated, GerR is responsible for repressing yqhV, which is located just upstream of the spoIIIA operon. Given the absence of an apparent transcriptional terminator at the end of the gene, σ E -directed transcription from yqhV is likely to read into spoIIIA, which is also transcribed from its own σ E -controlled promoter located in the intergenic region between yqhV and the operon. Thus, by repressing yqhV, GerR would inhibit read-through transcription into spoIIIA . Indeed, our transcriptional-profiling analysis revealed a small negative effect of GerR on spoIIIA transcription. Meanwhile, SpoIIID acts at the promoter for the spoIIIA operon to inhibit it from being used by σ E -RNA polymerase. Thus, maximum repression of spoIIIA is evidently achieved by the combined action of GerR and SpoIIID, each acting to block different promoters. Finally, we note that GerR inhibited the expression of a large number of genes that do not belong to the σ E regulon. Interestingly, many of these genes are organized in large clusters, such as azlB–azlC–azlD–bnrQ–yrdK–gltR, albA–albB–albC–albD, yefA–yefB–yefC–yeeA–yeeB–yeeC, yjcM–yjcN–yjcO, and yydB–yydC–yydD–yydG–yydH–yyd –yydJ . The genes found in these clusters rarely belong to a single transcription unit and are sometimes transcribed in opposite directions (either convergently or divergently). In summary, transcription of genes in the σ E regulon is in part self-limiting. The σ E factor induces the synthesis of two proteins, GerR and SpoIIID, that act to switch off other genes in the regulon, thereby preventing their continued transcription during the next stage of the mother-cell line of gene expression. The σ K Regulon Next, we used two complementary transcriptional-profiling approaches to identify genes under the control of σ K , an RNA polymerase sigma factor that follows SpoIIID in the hierarchical regulatory cascade. In one approach, we sought to identify genes that were upregulated during sporulation in a σ K -dependent (but not a GerE-dependent) manner. In the other approach, we sought to identify genes whose transcription was artificially activated in cells engineered to produce σ K during growth. For this approach we used a strain in which the coding sequence for the mature form of the transcription factor (σ K is normally derived by proteolytic processing from an inactive proprotein [ Kroos et al. 1989 ]) was under the control of an inducible promoter (see Materials and Methods ). Ninety-five genes were identified that were induced both during growth and sporulation in a σ K -dependent manner. Eight additional genes (cotA, cotE, cotM , gerE, gerPA, yfhP, yjcZ , and ykuD) that had previously been assigned to the regulon on the basis of gene-specific analysis were added to the tally, bringing the total to 103 (and representing 63 transcription units). These eight genes were cases in which we did not obtain a statistically significant score in one or the other of the two transcriptional-profiling approaches or for which a signal was not obtained for technical reasons (e.g., the strain used was mutant for gerE and yjcZ had not been annotated when the arrays were built). The list of 103 did not include σ K -controlled genes whose transcription additionally and strongly required the DNA-binding protein GerE. Some (28) of these 103 genes were also transcribed under the control of σ E (see Table 2 ), leaving a total of 75 genes that were newly activated during sporulation under the control of σ K . As we shall see, when genes that were strongly dependent on GerE are included (41 genes, five of which were also expressed under the control of σ E ), the size of the regulon increases to 144 genes (103 + 41) organized in 94 transcription units ( Table 2 ). A map of the σ K regulon is displayed in Figure 2 B and a detailed list of the genes in the regulon is presented in Table S4 . Identification of Promoters Controlled by σ K Using Bioinformatics and Transcriptional Start Site Mapping As a further approach to assessing our assignments to the σ K regulon, we used BioProspector and BioOptimizer to obtain a consensus sequence for promoters under the control of the sporulation transcription factor. The computational approach was complicated by the fact that the program had to find a two-block motif, with the first block corresponding to the −35 element and the second block to the −10 element separated by a gap of fixed length (+/− one nucleotide). The dataset consisted of 76 upstream sequences (the upstream sequences of transcription units that were strongly dependent on GerE were not included). The optimized motif with the best score identified 58 promoters and was composed of a five-nucleotide-long −35 element and a ten-nucleotide long −10 element, separated by a gap of 14–16 nucleotides ( Figure 4 D; Jensen and Liu 2004 ) To assess the validity of the predicted consensus sequence for σ K promoters, we mapped the transcription start sites of 18 of the newly identified targets of σ K by 5′ rapid amplification of complementary DNA ends–PCR (RACE–PCR). The results of the mapping experiments are displayed in Figure S2 . The newly identified σ K promoters were combined with the promoter sequences of 23 previously mapped σ K promoters to obtain an updated consensus sequence corresponding to a total of 41 promoters ( Figure 4 E). The logo for σ K promoters whose start sites had been mapped was very similar to the logo obtained by the BioProspector/BioOptimizer procedure (see Figure 4 D). Moreover, out of the 41 confirmed σ K promoters, the correct promoter was identified in 24 cases, with no prediction being made in 15 cases and an incorrect prediction in just two cases. All of the predicted sites are listed in Table S4 . The σ E and σ K factors are highly similar to each other, and the promoters they recognize are also very similar. The availability of updated logos for both categories of promoters based on the nearly complete regulons for both regulatory proteins provided an opportunity to revisit the issue of how the two regulatory proteins discriminate between their two classes of cognate promoters. A comparison of the motif recognized by σ K to that recognized by σ E ( Figure 4 F) reveals that both classes of promoters share identical −10 sequences and that the −35 elements differ by a single base pair: a cytosine in the fourth position of σ K -controlled promoters versus a thymine at the corresponding position in σ E -controlled promoters. These results reinforce the findings of Tatti et al. (1995) who identified glutamine 217 of σ E as the contact residue for the base pair at position 4. The two proteins are identical to each other in the region inferred to interact with the −35 element except for the presence of arginine instead of glutamine at the corresponding position in σ K . Moreover, replacing glutamine 217 with arginine was found to confer on σ E the capacity to recognize σ K -controlled promoters ( Tatti et al. 1995 ). The high similarity between the two classes of promoters also helps to explain why some σ K -controlled promoters are also recognized by σ E , but our bioinformatics analysis does not allow us to explain why some promoters are recognized exclusively by one or the other sigma factor and others are not. GerE Is Both a Repressor and an Activator of Genes Whose Transcription Is Dependent upon σ K The last regulator in the mother-cell line of gene expression is the DNA-binding protein GerE ( Cutting et al. 1989 ). Genes under GerE control were identified by transcriptional-profiling experiments carried out at two times (5.5 h and 6.5 h) late in sporulation. Strikingly, as many as 209 genes were downregulated in the presence of GerE at one or both time points, with many more genes being downregulated at the later time point (201 versus 61; see Table S1 ). Some of these downregulated genes (55) were members of the σ K regulon, with 29 being downregulated at the earlier time point and an additional 26 at the later time point. Thus, GerE is responsible for inhibiting the expression of 53% of the genes in the σ K regulon, but its repressive effects are not limited to genes under σ K . We note that the gene coding for σ K is itself repressed by GerE, which would be expected to curtail further synthesis of the mother-cell sigma factor late in sporulation. Thus, GerE has a wide impact in inhibiting gene transcription late in the process of spore maturation, including many genes in the preceding regulon of σ K -activated genes. At the same time, GerE is also an activator that stimulated or switched on the transcription of as many as 65 genes by hour 5.5 and 71 genes by hour 6.5. Of these, 41 were strongly dependent upon GerE for their expression and hence were not identified as members of the σ K regulon. Leaving aside genes that were members of both the σ E and σ K regulons (five), we see that GerE is responsible for turning on an additional 36 genes (27 transcription units) in the final phase of the mother-cell line of gene expression ( Table 2 ). Evidence that SpoIIID-Mediated Repression Is Required for Sporulation As we have seen, a striking feature of the mother-cell line of gene expression is that many of the genes activated by one transcription factor are turned off by the next-appearing regulatory protein in the cascade. Thus, most of the genes that are turned on by σ E are subsequently repressed by GerR or SpoIIID. Likewise, many of the genes activated by σ K are, in turn, downregulated by GerE. In the case of GerR, a mutant lacking the regulatory protein produced spores that were defective in germination. Hence, proper morphogenesis depends on the capacity of GerR, which appears to act exclusively as a repressor, to turn off genes under its control. The case of SpoIIID is more complex because in addition to its role as a repressor this DNA-binding protein is also an activator of two genes, spoIVCA and spoIVCB, that are essential for sporulation because of their role in the synthesis of σ K ( Halberg and Kroos 1994 ). To investigate the role of SpoIIID-mediated repression in spore formation, we created a construct in which a copy of the intact pro-σ K coding sequence, sigK, was introduced into the amyE locus, thereby bypassing the requirement for the spoIVCA -encoded recombinase, which is normally needed for creating sigK by a chromosomal rearrangement ( Stragier et al. 1989 ), and for spoIVCB, the 5′ portion of the coding sequence that participates in the rearrangement. In our construct, the insertion of sigK at amyE was under the direction of a σ E -controlled promoter that is not dependent upon SpoIIID for its activation (the promoter for spoIVF; Cutting et al. 1991b ). The amyE ::P spoIVF –sigK construct was introduced into spoIVCB mutant cells to create strain BDR1663. Even though pro-σ K was expected to be synthesized somewhat prematurely in BDR1663, the appearance of mature σ K remained subject to the pathway governing the proteolytic processing of pro-σ K and hence would have occurred at the normal time ( Cutting et al. 1991a ). Indeed, cells harboring the amyE ::P spoIVF –sigK construct sporulated as efficiently as the wild type and did so in a manner that did not depend on the presence of spoIVCB ( Table 3 ). We conclude that bypassing the requirement for SpoIIID in σ K synthesis does not measurably affect sporulation efficiency. Table 3 Systematic Inactivation of SpoIIID-Activated Genes a Sporulation efficiency is defined as the number of heat-resistant spores in a sporulating culture of the mutant strain divided by the number of heat-resistant spores present in a sporulating culture of the wild-type (PY79) strain grown in parallel However, when the amyE ::P spoIVF –sigK construct was introduced into cells harboring a spoIIID mutation (generating strain BDR1666), sporulation efficiency was still reduced by about a 100,000-fold compared to the wild type ( Table 3 ). This result reinforces the findings of Lu and Kroos (1994) , who showed that sporulation was impaired in spoIIID mutant cells even in the presence of a construct that allowed pro-σ K to be produced in a SpoIIID-independent manner. A possible explanation for these results is that, in addition to its role in σ K synthesis, SpoIIID is required for the synthesis of some other unidentified protein or proteins that are needed for sporulation. To investigate this possibility, we systematically inactivated all of the newly identified SpoIIID-activated transcription units ( Table 3 ). With three exceptions, those of spoVK, asnO, and ycgM, the resulting mutants sporulated at levels comparable to that of the wild type. In the case of spoVK, asnO, and ycgM, evidence suggests that each is transcribed in both a SpoIIID-dependent and a SpoIIID-independent mode. Thus, spoVK is transcribed from both a σ E -controlled (P1) and a σ K -controlled (P2) promoter, and it is known that P1 is dispensable for sporulation ( Foulger and Errington 1991 ). Experiments based on the use of cells engineered to produce σ K during growth indicate that asnO is capable of being transcribed under the direction of σ K . Finally, it has been shown that ycgM is induced during the early stages of sporulation under the control of Spo0A ( Molle et al. 2003a ), and so at least some YcgM protein should be present in a spoIIID mutant. Besides, complete inactivation of ycgM resulted in a sporulation defect that is less severe than that observed for strain BDR1666. These results do not rule out the possibility that SpoIIID activates the transcription of one or more genes in addition to spoIVCA and spoIVCB that are needed for sporulation. Nevertheless, the simplest interpretation of our findings is that the strong sporulation defect of strain BDR1666 is due to a failure in gene turn off rather than gene activation. Discussion The Mother-Cell Line of Gene Transcription Is a Hierarchical Regulatory Cascade That Is Subject to Successive Negative Regulatory Loops Our results reveal the almost complete program of gene transcription for a single differentiating cell type, the mother-cell compartment of the B. subtilis sporangium. The mother cell is a terminally differentiating cell that ultimately undergoes lysis (programmed cell death) when its contribution to the maturation of the spore is complete. Its program of transcription is played out over the course of about 5 h and, as we have shown, involves the activation in a cell-type-specific manner of 383 genes, which are grouped together in 242 transcription units. This corresponds to 9% of the 4,106 annotated protein-coding genes in the B. subtilis genome. The transcription of these 383 genes is orchestrated by five developmental regulatory proteins: two RNA polymerase sigma factors, σ E and σ K , and three DNA-binding proteins, SpoIIID, GerE, and a previously uncharacterized regulatory protein, GerR. The five regulatory proteins are organized in a hierarchical regulatory cascade of the form: σ E →SpoIIID/GerR→σ K →GerE. The earliest-acting regulatory protein in the cascade, σ E , turns on the transcription of 262 genes (163 transcription units), including the genes for GerR and SpoIIID. GerR and SpoIIID, in turn, acting as repressors, downregulate further transcription of almost half of the genes in the σ E regulon. In addition, however, SpoIIID, acting in conjunction with σ E -containing RNA polymerase, turns on the transcription of ten genes (eight transcription units), including genes involved in the appearance of σ K . Next, σ K activates 75 additional genes (44 transcription units). Among the members of the σ K regulon is the gene for the final regulatory protein in the cascade GerE. Strikingly, GerE represses the transcription of over half of the genes that have been activated by σ K while switching on 36 additional genes (27 transcription units), the final temporal class in the mother-cell line of gene transcription. Thus, the program of gene expression is driven forward by its hierarchical organization as well as by the successive, repressive effects of the DNA-binding proteins, which inhibit continued transcription of many genes that had been activated earlier in the cascade. Indeed, evidence presented herein is consistent with the idea that repression by GerR and SpoIIID contributes to proper sporulation, modestly in the case of GerR, and perhaps more significantly in the case of SpoIIID. The Mother-Cell Line of Gene Transcription Is Governed by a Linked Series of Coherent and Incoherent FFLs Transcription networks are based on recurring circuit modules, one of the most common of which is the FFL ( Milo et al. 2002 ; Shen-Orr et al. 2002 ; Mangan and Alon 2003 ). FFLs are simple circuits involving two regulatory proteins in which one (the primary regulatory protein) governs the synthesis of the other and both then control the expression of a set of target genes. Certain types of FFLs known as type 1 are particularly prevalent because of their favorable biological properties ( Shen-Orr et al. 2002 ). In type-1 FFLs, the primary regulatory protein acts positively on the synthesis of the second. The mother-cell line of gene transcription is based on two kinds of type-1 FFLs known as a “coherent” and “incoherent.” In coherent type-1 FFLs, both regulatory proteins act positively on target genes, whereas in incoherent type-1 FFLs, the primary regulatory protein acts positively and the second acts negatively. Using this nomenclature, we see that the hierarchical regulatory cascade that governs the mother-cell line of gene transcription is a circuit composed of two coherent type-1 FFLs linked in series ( Figure 1 B). Thus, σ E turns on the synthesis of SpoIIID, and both transcription factors then act jointly to turn on target genes, including genes involved in the appearance of σ K . The FFL is acting by the logic of an AND gate in that both σ E and SpoIIID are required for the expression of target genes. This first FFL is linked in series to a second coherent type-1 FFL in which σ K turns on the synthesis of GerE, and the two transcription factors then collaborate to activate the transcription of target genes (the terminal temporal class of gene transcription in the mother cell). Once again this is an AND gate in that both σ K and GerE are required for the activation of target genes. Simulation studies show that coherent type-1 FFLs have the property of being persistence detectors in which the activation of target genes depends on the persistence of the primary regulatory protein (σ E and σ K ) and “rejects” situations in which the primary regulatory protein is present only transiently in its active form ( Mangan and Alon 2003 ). The mother-cell line of gene transcription is also governed by three incoherent type-1 FFLs, involving SpoIIID, GerR, and GerE, each acting in this context as repressors. Thus, σ E turns on the synthesis of SpoIIID, which in turn represses a subset of the genes that have been turned on by the primary regulatory protein. The σ E factor similarly turns on the synthesis of GerR, which then represses a largely nonoverlapping subset of the genes that have been activated by σ E . Finally, the σ K factor turns on the synthesis of GerE, which then acts to downregulate the transcription of many of the genes that have been switched on by σ K . Simulations have shown that incoherent type-1 FFLs have the property of producing a pulse of gene transcription ( Mangan and Alon 2003 ). Incoherent type-1 FFLs also operate by the logic of an AND gate in that pulses of gene transcription require the action of both the activator and the delayed appearance of the repressor. Viewing the mother-cell line of gene transcription in terms of an interconnected series of FFLs reveals an underlying logic to the mother-cell program of gene expression. The use of coherent type-1 FFLs to drive the activation of successive sets of genes and the ordered appearance of regulatory proteins may help to minimize noise and to ensure that each temporal class of gene activation is tightly tied to the persistence of the previously acting regulatory proteins in the sequence ( Mangan et al. 2003 ). Meanwhile, the use of incoherent type-1 FFLs to switch off the transcription of genes in previously activated gene sets helps to generate pulses of gene transcription in which certain genes, whose products may only be required transiently during differentiation, are transcribed over a limited period of time. Indeed, as we now consider, genes with related functions are often transcribed coordinately in a pulse, the timing of which corresponds to the function of their products. Coordinated Expression of Functionally Related Genes The mother-cell program of gene expression is characterized, as we have seen, by pulses of gene expression in which different sets of genes are successively switched on and then switched off. In some cases, these pulses correspond to the expression of genes with related functions ( Table 4 ). This can be most clearly seen with the gene set that is activated by σ E and repressed by SpoIIID or GerR, which includes genes involved in engulfment, cortex formation, and the appearance of σ G and σ K . Thus, three genes that are responsible for driving engulfment, spoIID ( Lopez-Diaz et al. 1986 ), spoIIM ( Smith and Youngman 1993 ; Smith et al. 1993 ), and spoIIP ( Frandsen and Stragier 1995 ), are coordinately activated by σ E and then repressed by SpoIIID (in the case of spoIID ) or by GerR (in the case of the other two). Likewise, all of the σ E -controlled genes that are known to be required for spore cortex formation ( cwlD, dacB–spmAB, spoIVA, spoVB, spoVD, spoVE, yabPQ, ykvUV, ylbJ, and yqfCD; Piggot and Losick 2002 ; Eichenberger et al. 2003 ) are repressed by SpoIIID. Yet another example is the eight-gene spoIIIA operon, which is involved in the activation of σ G in the forespore ( Stragier and Losick 1996 ). The operon is transcribed from two σ E -controlled promoters, one located immediately upstream of the operon and one preceding the next upstream gene, yqhV . As we have shown, both promoters are turned off shortly after their activation; one by SpoIIID and the other by GerR. Table 4 Functional Categories a In cases where the transcription start site has been mapped by primer extension, the corresponding σ factor is indicated in bold characters b In cases where binding of the transcription factor has been confirmed by DNAase I footprinting or EMSA analysis, the corresponding regulator is indicated in bold characters Particularly illuminating is the case of the five σ E -controlled genes involved in the appearance of σ K : bofA, spoIVCA, spoIVCB, spoIVFA, and spoIVFB . Two of these genes (spoIVCA and spoIVCB) are involved in the synthesis of the proprotein precursor, pro-σ K , whereas the remaining three (bofA , spoIVFA, and spoIVFB) are involved in the conversion of the proprotein to mature σ K ( Cutting et al. 1991b ; Ricca et al. 1992 ). Interestingly, bofA , spoIVFA, and spoIVFB are repressed by SpoIIID, whereas spoIVCA and spoIVCB are switched on by SpoIIID, in this context acting as an activator. Hence, and ironically, genes involved in the processing of pro-σ K are expressed in a pulse that precedes the time of activation of the genes involved in the synthesis of the substrate for processing. How can we explain these seemingly anomalous observations? BofA, SpoIVFA, and SpoIVFB are integral membrane proteins that form a complex in the mother-cell membrane that surrounds the forespore ( Resnekov et al. 1996 ; Rudner and Losick 2002 ). Evidence indicates that they initially localize to the cytoplasmic membrane that surrounds the mother cell and then reach their final destination by diffusion to, and capture at, the outer forespore membrane ( Rudner et al. 2002 ). Such a diffusion-and-capture mechanism requires that the synthesis of BofA, SpoIVFA, and SpoIVFB takes place prior to the completion of engulfment since the outer membrane surrounding the forespore has become topologically isolated from the cytoplasmic membrane once engulfment is complete. Conversely, no such restriction applies to pro-σ K (a peripheral membrane protein) whose synthesis is delayed (by virtue of being under the positive control of SpoIIID) relative to that of the integral membrane proteins. Strikingly, and in extension of these observations, a high proportion of σ E -controlled genes that encode proteins with predicted transmembrane segments are negatively regulated by SpoIIID and GerR. We speculate that many of these genes encode proteins that localize to the outer forespore membrane and do so by a diffusion-and-capture mechanism. Hence their synthesis is restricted to the time prior to the completion of engulfment. By contrast, σ E -controlled genes that are unaffected by SpoIIID and GerR, or are activated by SpoIIID, rarely encode proteins with predicted transmembrane segments (see Table S2 ). As a final example of the coordinate expression of genes with related function we consider the case of cwlC and cwlH, which are switched on in the terminal phase of differentiation under the positive control of GerE ( Kuroda et al. 1993 ; Smith and Foster 1995 ; Nugroho et al. 1999 ). The cwlC and cwlH genes encode cell-wall hydrolases that are responsible for the lysis of the mother cell when morphogenesis is complete so that the mature spore can be liberated from the sporangium. It is of crucial importance that mother-cell lysis not take place prematurely, and thus it makes sense that genes involved in this process are among the last genes to be turned on in the mother-cell line of gene expression. Some Functionally Related Gene Classes Exhibit Heterogeneous Patterns of Gene Expression Many of the genes in the mother-cell line of gene expression are known or inferred to be involved in metabolism, assembly of the spore coat, or the synthesis of coat-associated polysaccharides (see Table 4 ). Interestingly, not all of the genes in these categories are coordinately expressed. Rather, genes in all three categories exhibit heterogeneous patterns of expression. Thus, among genes inferred to be involved in metabolism, some, such as members of the yngJIHGFE operon, which are expected to govern lipid catabolism, and members of the yjmCD–uxuA–yjmF–exuTR operon, which are expected to direct hexuronate synthesis ( Mekjian et al. 1999 ), are expressed early in development, whereas other genes, such as the members of the yitCD and yitBA – yisZ operons, which are inferred to be involved in phosphosulfolactate synthesis ( Graham et al. 2002 ), are expressed late in development. Sulfolactate is indeed known to be a major component of the dry weight (5%) of mature spores of B. subtilis but is not found in spores of B. megaterium and B. cereus ( Bonsen et al. 1969 ). Consistent with these observations, the genome of B. cereus lacks an ortholog of the yitCD operon. Interestingly, the gene for asnO, which encodes an asparagine synthetase ( Yoshida et al. 1999 ), is under the positive control of three of the five mother-cell-specific transcription factors (σ E , σ K , and SpoIIID) and the negative control of GerE, and hence its expression is maintained until very late in development. Of special interest are genes involved in the assembly of the coat, the most conspicuous morphological feature of the mature spore. The coat is a complex, two-layered structure that creates a protective shield around the spore and is composed of at least 30 proteins ( Driks 2002 ; Kuwana et al. 2002 ; Takamatsu and Watabe 2002 ; Lai et al. 2003 ). The earliest-acting protein in the formation of the coat is SpoIVA, which creates a substratum around the outer forespore membrane upon which assembly of the coat takes place ( Roels et al. 1992 ; Stevens et al. 1992 ; Driks et al. 1994 ; Price and Losick 1999 ). In keeping with its early role in the assembly process, the gene for SpoIVA is switched on early in the mother-cell line of gene expression under the control of σ E and is then turned off by the action of SpoIIID. The σ E factor also turns on the genes for at least five other coat proteins that play important roles in coat assembly ( cotE, cotH, safA, spoVM, and spoVID; Piggot and Losick 2002 ) , but expression of these genes persists longer than that for spoIVA as none of these is repressed by SpoIIID. In fact, cotE and cotH continue to be expressed at even higher levels later in development under the control of σ K , eventually being downregulated by GerE. In the case of cotE, Li and Piggot (2001) have shown that transcription from its σ E -dependent promoter P1 ceases before the activation of σ K . Interestingly, certain temporal classes of mother-cell-specific genes are particularly enriched in coat protein genes. For instance, almost half of the σ E -controlled genes that are strongly or partially dependent on SpoIIID for expression (i.e., ten out of 25; C. F., P. E., and R. L., unpublished data) code for coat proteins. Similarly, our preliminary cytological data (C. F., P. E., and R. L., unpublished data) indicate that many of the newly identified σ K -controlled genes encode coat-associated proteins. In addition to being composed of many different proteins, the coat is composed of polysaccharides. Playing an important role in the synthesis of these polysaccharides is the 11-gene sps operon, the longest of the 236 mother-cell-specific transcription units identified in this study. The sps operon is transcribed from a σ K -controlled promoter, which we have mapped to a site just upstream of the first gene in the operon, spsA . Transcription from this promoter is enhanced by the appearance of GerE but is not dependent upon it. Thus, expression of genes involved in the biosynthesis of spore-coat polysaccharides persists until the very late stages of sporulation, in keeping with the idea that these polysaccharides are a component of the outer surface of the spore. Nevertheless, some genes in the sps operon are switched on early in sporulation under the control of σ E , most likely from a second promoter located upstream of the seventh gene in the operon, spsG . Hence spsG and the genes downstream of it exhibit a protracted pattern of expression that persists throughout the entire process of differentiation. The sps operon may not be the only set of genes involved in the synthesis of coat-associated polysaccharides. We have identified several paralogs of members of the operon that contribute to the mother-cell line of gene expression. These include genes in the yfnED operon, which is switched on by σ E , downregulated by GerR, and turned on again by σ K . Another example is the yfnHGF operon, which is under the positive control of σ K and GerE. Yet another example is a paralog of spsJ, yodU–ypqP, which is activated under the dual control of σ K and GerE. Interestingly, in the strain used in this study (PY79), yodU and ypqP actually correspond respectively to the 5′ end and the 3′ end of a single gene. However, in strain 168, the gene formed by yodU and ypqP is interrupted by the prophage of the large temperate phage SPβ, thereby greatly separating ypqP from the sporulation promoter that would otherwise direct its transcription. It would be interesting to investigate whether the interruption of the yodU – ypqP gene by SPβ influences the polysaccharide composition of the spore coat. The Mother-Cell Line of Gene Transcription in Other Endospore-Forming Bacteria Endospore formation has been documented in many species of the low G+C group of gram-positive bacteria ( Stragier 2002 ). Two distantly related genera in that group, Bacillus and Clostridium, are able to sporulate, whereas several genera that are phylogenetically closer to Bacillus, such as Listeria and Staphylococcus, do not sporulate. Remarkably, in genome regions of otherwise high conservation (synteny) to corresponding regions in B. subtilis, sporulation genes are missing from Listeria ( Eichenberger et al. 2003 ) and Staphylococcus. It is likely that the common ancestor of all of these genera was an endospore-forming bacterium and that sporulation genes were deleted over time from genera that had adapted alternative modes of survival in their ecological niche or host in a manner that did not involve the need for a robust resting state. To investigate further the evolutionary relatedness of the mother-cell differentiation program among endospore-forming species, we searched for the presence of orthologs of B. subtilis genes in the mother-cell line of gene expression in the genome sequences of the following species: B. anthracis (Ames strain) ( Read et al. 2003 ), B. cereus (ATCC14579) ( Ivanova et al. 2003 ), B. halodurans ( Takami et al. 2000 ), and Oceanobacillus iheyensis (HTE831) ( Takami et al. 2002 ); Listeria monocytogenes and L. innocua ( Glaser et al. 2001 ); and Clostridium acetobutylicum (ATCC 824) ( Nolling et al. 2001 ) and C. perfringens (strain 13) ( Shimizu et al. 2002 ) (Tables S2 and S4 ). First, we searched for orthologs of the mother-cell-specific transcription factors. Interestingly, whereas genes for σ E , σ K , and SpoIIID were present in the Bacillus and Clostridium species, GerE ( Stragier 2002 ) and GerR were absent from Clostridium, suggesting that significant differences exist in the mother-cell programs between the two genera, especially during the terminal (GerE-controlled) phase of gene expression. Nonetheless, in cases when a transcription factor was conserved between Bacillus and Clostridium, the protein domains involved in nucleotide-sequence recognition were also highly conserved, indicating that the consensus binding sequences that we described here are likely to be conserved among many, if not all, endospore-forming bacteria. For instance, the glutamine residue that recognizes the specificity determinant in the −35 element of σ E -controlled promoters is absolutely conserved in all of the available σ E protein sequences, and the corresponding arginine is conserved in all of the available σ K protein sequences. In addition to differences in the presence of certain mother-cell regulatory proteins (e.g., GerR and GerE) among endospore-forming species, the gene composition of the individual regulons also varies in a species-specific manner. In general, genes in the σ E regulon appear to be more highly conserved than genes in the σ K regulon. For instance, approximately 75% of the B. subtilis σ E -controlled transcription units have orthologs in B. anthracis and B. cereus, whereas only 50% of the σ K -controlled transcription units do. Similarly, close to 40% of the B. subtilis σ E -controlled transcription units are present in Clostridium, but only about 20% of the σ K -controlled transcription units are present. An appealing explanation for the lower level of conservation among σ K regulons is that genes switched on late in the mother-cell line of gene expression are enriched for genes encoding components of the outer surface of the spore—proteins that are likely to undergo the greatest evolutionary adaptation to the ecological niche in which a particular species is found. Indeed, experiments involving the use of atomic force microscopy reveal that the surfaces of the spores of the closely related species B. subtilis, B. anthracis, and B. cereus exhibit quite distinctive landscapes ( Chada et al. 2003 ). Conclusions We have provided a comprehensive description of the program of gene transcription for a single differentiating cell type and have shown that this program is governed by a regulatory circuit involving the action of five transcriptional control proteins acting as activators or repressors or both. The underlying logic of the circuit is that of a linked series of coherent and incoherent type-1 FFLs involving two-way combinations of the five regulatory proteins. The circuit is expected to create pulses of gene transcription in which large numbers of genes are switched on and subsequently switched off. We anticipate that type-1 FFLs linked in series are likely to be a common feature of programs of cellular differentiation in a wide variety of developing systems. Materials and Methods Strains All strains used here are derivatives of the wild-type strain PY79, with the exception of the σ K overproducing strain, which is a derivative of strain 168. Strains PE436 and PE437 ( Eichenberger et al. 2003 ), PE452, PE454, PE455, PE456, and SW282 were used for transcriptional profiling under conditions of sporulation. PE452 was obtained by transformation of strain RL560 to MLS resistance with chromosomal DNA from strain MO1027 ( spoIVCB :: erm, a gift from P. Stragier, Institut de Biologie Physico-Chimique, Paris) ( sigG :: cat; a derivative in the PY79 background of strain MO479.2 [ Karmazyn-Campelli et al. 1989 ]). PE454 was generated by transformation of PY79 to chloramphenicol resistance with chromosomal DNA from strain RL16 ( gerE :: cat; Cutting and Mandelstam 1986 ). PE455 is the result of the transformation of strain PE454 with chromosomal DNA from strain MO1027. PE456 was obtained by transformation of PE452 to spectinomycin resistance with chromosomal DNA from strain PE239 ( spoIIIDΔ :: spc; Eichenberger et al. 2001 ). SW282 was generated by transformation of PE454 to spectinomycin resistance with chromosomal DNA from strain PE316 ( ylbOΔ :: spc; Eichenberger et al. 2003 ). Strain SI01, which was used for overproduction of σ K , was created by double cross-over recombination at the amyE locus, following transformation with XhoI-digested plasmid pMFNsigK. Strains PE511, PE551, PE553, PE558, PE568, and SW312 were used for β-galactosidase activity assays. PE511 is a derivative in the PY79 background of strain MO1533 ( amyE :: spoIIP–lacZ cat; Frandsen and Stragier 1995 ). PE551 and PE553 were obtained by double cross-over recombination of XhoI-digested plasmids pPE72 and pPE74, respectively, into PY79 and selection for chloramphenicol resistance and spectinomycin sensitivity. SW312, PE568, and PE558 were generated by transformation with chromosomal DNA from strain PE316 to spectinomycin resistance of strains PE551, PE511, and PE553, respectively. Strains PE529 (yheDΔ :: erm), PE534 (mprΔ :: tet), PE538 (yqfTΔ :: kan), PE548 (lipΔ :: erm), PE549 (cotTΔ :: spc), RL3231 (ycgMΔ :: tet), PE566 (cotVWΔ :: erm), PE569 (ydcIΔ :: tet), and PE570 (yheIΔ :: spc) were generated with the technique of long-flanking homology PCR ( Wach 1996 ). The sequence of the primers used for gene inactivation is available upon request. Strain PE563 (cotFΔ :: cat) was obtained by transformation of PY79 to chloramphenicol resistance with chromosomal DNA from RL654 ( Cutting et al. 1991c ). PE564 is the equivalent of strain MO1057 ( spoIVCA :: erm; a gift from P. Stragier) in the PY79 background. Strains RL2391, RL2396, and RL2519 are from Eichenberger et al. 2001 and RL2393, RL2559, RL2570, RL2571, and RL2581 are from Eichenberger et al. 2003 . Strain BDR107 is a derivative of PY79 harboring spoIVCB :: erm from strain MO1027. BDR362 corresponds to RL75 ( spoIIID :: erm; Kunkel et al. 1989 ). P spoIVF pro – sigK spc from pDR191 was introduced into the amyE locus of BDR107 to generate BDR1663. Genomic DNA from BDR1663 was used to transform BDR362 to spectinomycin resistance to generate BDR1666. Plasmids Plasmid pMFN20 was constructed by cloning a PstI (blunted)-EcoRI (blunted) 1.3-kb Neo r cassette from pBEST501 ( Itaya et al. 1989 ) and a 7.2-kb DNA fragment from plasmid pMF20 (M. F., unpublished data), amplified with primers SI1 (5′-ATGGATGAGCGATGATGATATCCGT-3′) and SI2 (5′-AACTATTGCCGATGATAAGCTGTC-3′). The pro-less sigK gene was constructed using recombinant PCR ( Wach 1996 ). A DNA fragment containing the upstream region of the sigK gene was amplified from chromosomal DNA of strain 168 using primers SI3 (5′-CCC AAGCTT TTAGTATGCTGCTTACC-3′) and SI4 (5′-AAG GCATTGTTTTTCACGTA CATCGTCACCTCCACAAAAGTAT-3′) (restriction site is underlined; the sequence complementary to primer SI5 is italicized). The other primer set, primer SI5 (5′-TACGTGAAAAACAATGC-3′) and primer SI6 (5′-CGC GGATCCT TCTGCATTATTTCCCC-3′), was used for PCR amplification of chromosomal DNA of strain 168 isolated from cells 6 h after initiation of sporulation to generate the rearranged sigK gene. The two PCR products were fused by PCR to generate the pro-less sigK gene. Plasmid pMFNsigK was constructed by cloning the pro-less sigK fragment into pMFN20 between HindIII and BamHI. Plasmids pPE72 (amyE ::P spoIIM –lacZ cat) and pPE74 (amyE ::P yqhV –lacZ cat) were obtained as follows. Primers PE857 (5′-CGTCTAGCC GAATTC CAGCACACCATCTTTCAGCACACA-3′) and PE858 (5′-CGTATCCCG GGATCC CGCCCGCTCTAGTGATTTGATTTA-3′) were used to amplify the promoter sequence of spoIIM from PY79 chromosomal DNA by PCR (restriction sites are underlined), whereas primers PE860 (5′-CGTCTAGCC GAATTC GGCTTGTTGTAAACGTGCCGTTCT-3′) and PE861 (5′-CGTATCCCG GGATCC CCCATTTTTTTATATGATATGCTCT-3′) were used for the PCR amplification of the promoter sequence of yqhV. The PCR products were gel-purified (QIAquick Gel extraction kit; Qiagen, Valencia, California, United States) and digested with EcoRI and BamHI. In parallel, vector pDG1661 ( Guerout-Fleury et al. 1996 ) was similarly digested with EcoRI and BamHI and treated with shrimp alkaline phosphatase (USB). The digested vector and PCR fragments were purified (QIAquick PCR purification kit; Qiagen) and ligated using T4 DNA ligase. Ligations were transformed in DH5α cells to ampicillin resistance, and plasmids pPE72 and pPE74 were recovered by alkaline lysis. Primers odr261 (5′-GGCAAGCTTGTGGAGGTGACGATGGTGACAG-3′) and odr262 (5′-GGCGGATCCTGCGGGAGGATTATAAGTCAAG-3′) were used to amplify pro–sigK from pSK6 ( Kunkel et al. 1990 ). The PCR product was cloned into pdr77 ( Rudner and Losick 2002 ) between HindIII and BamHI to generate pdr191 (amyE ::P spoIVF – pro – sigK spc). Growth and sporulation conditions Strains used for transcriptional profiling, β-galactosidase activity assays, and ChIP-on-chip experiments were grown in hydrolyzed casein medium at 37 °C to an A 600 nm of 0.6. Pellets obtained by centrifugation were suspended in Sterlini–Mandelstam medium ( Sterlini and Mandelstam 1969 ; Harwood and Cutting 1990 ) and placed in a shaking water bath at 37 °C. Samples were collected at the indicated times after resuspension. A fresh colony of the σ K overproducing strain SI01 was grown in 5 ml of Penassay broth (Difco Laboratories, Detroit, Michigan, United States) overnight at 30 °C. Next, 50 ml of LB broth with and without 10 mM of xylose was inoculated with 1 ml of the overnight culture. Cells were grown by incubation at 37 °C with shaking and harvested 2 h after induction (A 600 nm of 0.8) for extraction of RNA. Transcriptional profiling DNA microarrays were generated as described by Britton et al. (2002) . RNA preparation, sample labeling, and hybridization procedures were performed as described by Eichenberger et al. (2003) . Expression data were obtained from three independent experiments for SpoIIID, GerR, σ K , and GerE. Our statistical analysis procedure, described in detail by Conlon et al. (2004) , was performed separately for each set of microarrays for SpoIIID, GerR, σ K , and GerE. Normalization of each slide was performed using an iterative rank-invariant method. A Bayesian hierarchical model incorporating experimental variation was used to combine normalized slides across replicated experiments. A Markov chain Monte Carlo implementation of the model with 4,000 iterations produced a posterior median estimate of the log-expression ratio for each gene, and the corresponding Bayesian confidence interval. Genes were scored for the posterior probability of a positive log-expression ratio. Genes with scores above or equal to a threshold of 0.95 were determined to be upregulated in an experimental condition, and genes with scores below or equal to a threshold of 0.05 to be downregulated. Finally, genes in the upregulated category with a nonlogarithmic expression ratio inferior to a threshold of 2.0 and, similarly, genes in the downregulated category with an expression ratio superior to a threshold of 0.5 were not included, unless indicated otherwise (Tables S2 and S4 ) in the list of differentially expressed genes. The data are available online in MIAME-compliant format at http://mcb.harvard.edu/losick and were also deposited in the Gene Expression Omnibus database under the accession number GSE1620. Overexpression and purification of SpoIIID protein SpoIIID protein was overproduced by the T7 promoter overexpression system of Escherichia coli. The SpoIIID protein expression plasmid was constructed by amplifying the corresponding region from PY79 chromosomal DNA using primers 5′-TACATATGCACGATTACATCAAAGAG-3′ and 5′-CCCTCGAGCGATTGCTGAACAGGCTC-3′. The PCR fragment was digested by NdeI and AvaI and ligated into the NdeI/AvaI-digested vector pET22b (Novagen, Madison, Wisconsin, United States) to generate the SpoIIID protein expression plasmid (pETIIID). The plasmid was transformed into strain BL21 (DE3). Cells carrying pETIIID were grown at 37 °C in 2 l of LB containing 100 μg/ml of ampicillin to an A 600 nm of 0.6, at which point T7 RNA polymerase synthesis was induced by the addition of IPTG to a final concentration of 1 mM. Cells were harvested 5 h later by centrifugation. The pellet was resuspended in 40 ml of binding buffer (50 mM Tris-HCl [pH 8.0], 500 mM NaCl, 5 mM imidazole) and disrupted by sonication. Cell debris was removed by centrifugation at 20,000 g for 30 min and the supernatant was loaded on 1 ml of Ni 2+ –NTA agarose resin (Qiagen) equilibrated with binding buffer. SpoIIID was eluted with a 20-ml imidazole gradient from 5 to 500 mM in binding buffer. Peak fractions were pooled and dialyzed against TGED buffer. The amount of protein was determined with a Bio-Rad (Hercules, California, United States) protein determination kit with BSA as the standard. The purified SpoIIID protein was tested in an in vitro transcription system using reconstituted σ E –RNA polymerase and spoIID as template (data not shown). Gel EMSAs DNA fragments of interest were obtained by PCR amplification of PY79 chromosomal DNA (see Supporting Information for the description of primers used), gel-purified (QIAquick Gel extraction kit), and end-labeled with T4 polynucleotide kinase in the presence of [γ– 32 P]-ATP for 30 min at 37 °C. After labeling, the fragments were purified with the QIAquick PCR purification kit. Prior to loading on a 6% polyacrylamide–0.16% Bis-0.5X TBE gel that had been prerun at 200 V for 1 h, DNA fragments were preincubated in gel shift-binding buffer (10 mM Tris-HCl [pH 7.5], 50 mM NaCl, 1 mM EDTA, 5% glycerol, 1 mM DTT, and 50 μg/ml BSA) for 30 min at room temperature with various amounts of purified SpoIIID protein (0 nM, 50 nM, 100 nM, and 200 nM). The gel was run for 1 h at 200 V, dried, and exposed to Kodak X-OMAT film (Kodak, Rochester, New York, United States). The following DNA fragments were used in the analysis: abrB, spoIID, spoIIG, and racA ( Molle et al. 2003a ), bofA (from nucleotide 29439 to 30030), spoIVCA (from 2654348 to 2654008), asnO (from 1156248 to 1156617), cotE (from 1774088 to 1774414), cotF (from 4165851 to 4166189), cotT (from 1280431 to 1280108), gerM (from 2902355 to 2902020), spoIVA (from 2387164 to 2386826), spoIVFA (from 2856840 to 2856657), spoVB (from 2828500 to 2828872), spoVK (from 1873045 to 1873470), yabP (from 67986 to 68297), ybaN (from 160708 to 160468), ycgF (from 333921 to 334278), yitE (from 1174319 to 1174179), ykvU (from 1448417 to 1448728), ylbJ (from1571288 to 1570927), ypjB (from 2361797 to 2361463), yqfC (from 2616344 to 2616016), yqfZ (from 2587752 to 2588021), albE–albF 1 (from 3838793 to 3839220), albE–albF 2 (from 3839214 to 3839656), albE–albF 3 (from 3839637 to 3840082), albE–albF 4 (from 3840067 to 3840481), dctR–dctP 1 (from 498913 to 499280), dctR–dctP 2 (from 499263 to 499554), dctR–dctP 3 (from 499531 to 499913), dctR–dctP 4 (from 499912 to 500356), tenI–goxB–thiS 1 (from 1242878 to 1243331), tenI–goxB–thiS 2 (from 1243322 to 1243755), tenI–goxB–thiS 3 (from 1243749 to 1244115), tenI–goxB–thiS 4 (from 1244063 to 1244366), treA–treR–yfkO 1 (from 852742 to 853109), treA–treR–yfkO 2 (from 853071 to 853409), treA–treR–yfkO 3 (from 853387 to 853810), treA–treR–yfkO 4 (from 853797 to 854209), yfmC–yfmD 1 (from 825948 to 825585), yfmC–yfmD 2 (from 825607 to 825255), yfmC–yfmD 3 (from 825266 to 824884), yfmC–yfmD 4 (from 824903 to 824489). DNAase I footprinting DNAase I footprinting was carried out as described by Fujita and Sadaie (1998) . Antibodies for SpoIIID The C-terminus of the SpoIIID protein was overproduced by the T7 promoter overexpression system of E. coli and used as an antigen for the production of anti-SpoIIID antibodies. The SpoIIID C-terminus protein expression plasmid was constructed by amplifying the region from PY79 chromosomal DNA using primers, 5′-GAAGCTAGCATGATTAACCCCGACTTGGCAAACG-3′ and 5′-GAACTCGAGCGATTGCTGAACAGGCTCTCCTT-3′. The PCR fragment was digested by NheI and XhoI and ligated into the NheI/XhoI-digested vector pET21b (Novagen) to generate the SpoIIID C-terminus protein expression plasmid pMF213. Overexpression and purification of the protein are described in a previous section. The anti-SpoIIID antibodies were prepared by Covance Research Products (Denver, Pennsylvania, United States) and were highly specific as judged by Western blot analysis, which revealed only a single cross-reacting species. Chromatin immunoprecipitation in combination with gene microarrays (ChIP-on-chip) Three hours after resuspension of PY79 cells in Sterlini-Mandelstam medium at 37 °C, cross-links were generated by treatment with formaldehyde (1% final concentration) for 30 min. The rest of the procedure was identical to the one described by ( Molle et al. 2003a , 2003b ). The data analysis for the ChIP-on-chip experiments was carried out using the Resolver statistical package (Rosetta, Seattle, Washington, United States). Experiments were normalized and combined for enrichment-factor determination (Rosetta Resolver). An enrichment factor for a given gene represents the ratio of immunoprecipitated DNA to total DNA. It was considered significant when higher than 2 and with an associated p -value lower than 0.001. BioProspector/BioOptimizer BioProspector ( Liu et al. 2001 ) is a stochastic motif-discovery program used to find conserved subsequences of fixed width in a set of DNA sequences, based on a statistical motif-discovery model reviewed in Jensen et al. (2004) . The program can also be used for motifs consisting of two conserved blocks connected with a variable-length gap of unconserved nucleotides, and BioProspector can also be forced to find sites in every input sequence. Since BioProspector is a stochastic algorithm, more than one possible motif can be found, and since the program requires the motif width to be fixed, several different fixed widths should be used in the usual case where the motif width is not known. Thus, we collected the top five Bioprospector motifs under a range (6–12 bps) of seven fixed widths, giving a total of 35 putative motifs. BioOptimizer ( Jensen and Liu 2004 ) is an optimization program designed to improve the results of each discovered BioProspector motif and to score each motif so that the “best” putative motif can be selected out of the 35 we discovered. The scoring function used is the exact log-posterior density of the Bayesian motif-discovery model given in Jensen and Liu (2004) . Starting from the set of sites predicted by BioProspector, the scoring function is optimized by accepting the addition of new motif sites or removal of current motif sites only if these changes increase the score. BioOptimizer also has the flexibility to allow the motif width to vary, so that the “best” width can also be determined. As well, BioOptimizer can be restricted to force particular sequences in the dataset to contain at least one site while leaving other sequences unrestricted. This property was utilized in our SpoIIID motif search, where a subset of sequences has additional biochemical evidence that they contain at least one SpoIIID-binding site. Having found an optimal motif with our combined BioProspector/BioOptimizer procedure, we implemented an additional scanning procedure to find more potential SpoIIID sites. Using the estimated proportion of nucleotide k in position j of the motif (θ^ j,k ) and the estimated proportion of nucleotide k in the background (θ^ j,k ) provided by our optimal motif, we scanned all upstream sequences to see if there were additional sites that matched our discovered motif closely but were not strong enough to be detected by the motif-discovery procedure. In each sequence, for each potential starting position i, we had a potential site S i = r i , r i +l , …, r i + w −l ) , for which we compute the following score: We considered the site in each sequence with the largest Strength value to be the best candidate as an additional site. If a sequence already contained a site found by our motif-discovery procedure, we would expect that this same site would be the one with the largest Strength value. For any sequence that did not have an optimal site found by the motif-discovery procedure, this scanning procedure gave us new site predictions. However, for any new sites found by the scanning procedure, one must be cautious about the strength of these sites, since the procedure found sites in each sequence regardless of how well those sites matched our optimal motif. Therefore, we also calculated a p -value for each site by comparing the Strength value calculated for that site to the Strength value calculated for 10,000 random sequences. Only sites with low p -values were considered as potential sites. With the Bonferroni correction for multiple comparisons, we considered only sites with p -values less than 0.000183. MDscan analysis of the SpoIIID-binding motif We used the word-enumeration algorithm “MDscan” ( Liu et al. 2002 ) to identify motifs in sequences most enriched by immunoprecipitation experiments. In this algorithm, it is assumed that the most enriched sequences have stronger motif signals than the remaining sequences. MDscan first identifies oligomers of width w ( w -mers) in the top sequences, which are used as seed oligomers. Motif matrices are constructed for each seed oligomer using all similar segments from the top sequences. Segments are defined to be similar if they share at least m matched positions, with m determined so that the probability that a pair of randomly produced w -mers are m -matches is less than 0.15%. The resulting motif matrices are evaluated using the following semi-Bayesian scoring function: where x m is the number of segments aligned in the motif, p ij is the frequency of base j at motif position i, and p o (s) is the probability of generating segment s from the background model. The top distinct highest scoring motifs are defined as candidate motifs. These motifs are refined using the remaining sequences, by adding new w -mers to the matrix if the score is increased. The motifs are further refined by reexamining all segments of the motif matrix and removing segments if the motif score is increased. We first ranked by enrichment ratio the 26 regions of the chromosome that were enriched by immunoprecipitation by a factor of 2 or greater. We used the top 20 regions as the top sequences, with the remaining six sequences used for refinement. The 26 regions were used as the background sequences, and we reported 30 candidate motifs. We first searched for motifs of width w = 8. In using alternative widths ( w = 7, 9, and 10), and alternative definitions of top regions (15–25), the top reported motif was similar to that for width 8. As reported in Liu et al. (2002) , MDscan is tolerant of different top sequence definitions (∼3–20), and of moderate ranking errors. Germination assays Tests for germination using 2,3,5-triphenyltetrazolium chloride overlay were carried out as described in Nicholson and Setlow (1990) . Strains mutant for GerR (PE316) were compared to wild-type cells (PY79), and strains mutant for GerE (strain PE454) or CotE (strain RL322; Driks et al. 1994 ) were described as negative controls. All strains were sporulated in DSM. Heat activation was performed in a 65 °C oven for 3 h. Measuring β-galactosidase activity β-galactosidase activity assays were carried out as previously described ( Miller 1972 ; Harwood and Cutting 1990 ). Promoter mapping by 5′ RACE–PCR The 5′ end of several σ K -controlled mRNAs was determined by the RACE–PCR procedure ( Frohman 1994 ; Price et al. 2001 ). Total RNA was extracted from strains PE454 (sigE + , sigK + ) and PE455 (sigE + , sigK − ) and analyzed as described by Eichenberger et al. (2003) . Measuring sporulation efficiency Strains were grown to exhaustion in DSM for 30 h at 37 °C and assayed for heat resistance as previously described by van Ooij et al. (2004) . Supporting Information Figure S1 DNAase I Footprinting of SpoIIID Binding to the Promoters of spoIID, spoIIIA, and spoVE (A) Radioactive DNA fragments were incubated with no protein (left lane) or with 400 nM of SpoIIID protein (right lane) and then subjected to DNAaseI footprinting. A chemical sequencing ladder was used as a marker (not shown). Protected regions are indicated by a bar. (B) Position of SpoIIID-binding sites. The nucleotide sequence upstream of the transcriptional start site (+1) is shown for spoIID, spoIIIA, spoVE -P1, and spoVE -P2. The boundaries of the region protected from DNAase I digestion by SpoIIID are indicated by bars. The bold letters identify the sequences within the protected regions that match with the SpoIIID consensus sequence. (1.72 MB PPT). Click here for additional data file. Figure S2 Mapping of Transcription Start Sites by 5′ RACE–PCR The underlined uppercase bold letters identify the 5′ ends of mRNAs from σ K -controlled genes as determined by RACE–PCR. Also indicated are the corresponding −35 and −10 regions (uppercase letters in bold), the ribosome-binding site (double underlining), and the translation start site (uppercase letters). RNA collected from strain PE454 (sigE + sigK + ) and strain PE455 (sigE + , sigK − ) was used for the determination of transcription start sites. In four cases indicated with an asterisk (yfnE, yhcO, yitC, and ypqA), an identical transcription start site was identified for strains PE454 and PE455, which is interpreted as evidence that the promoters for these three transcription units are recognized both by σ E and σ K . In all of the other cases, a transcription start site was obtained only with RNA collected from strain PE454. (22 KB DOC). Click here for additional data file. Table S1 Mother-Cell Gene Expression (1.5 MB XLS). Click here for additional data file. Table S2 Effect of SpoIIID and GerR on the expression of genes in the σ E regulon (351 KB XLS). Click here for additional data file. Table S3 ChIP-on-chip data for SpoIIID (332 KB XLS). Click here for additional data file. Table S4 Effect of GerE on the expression of genes in the σ K regulon (238 KB XLS). Click here for additional data file. Accession Numbers The Swiss-Prot ( http://www.ebi.ac.uk/swissprot/ ) accession numbers for the gene products discussed in this paper are σ E (P06222), σ G (P19940), σ K (P12254), BofA (P24282), CodY (P39779), GerE (P11470), GerR (O34549), RacA (P45870), Spo0A (P06534), SpoIIID (P15281), SpoIVFA (P26936), and SpoIVFB (P26937). The GenBank ( http://www.ncbi.nlm.nih.gov/Genbank ) accession numbers for the genes discussed in this paper are abrB (BSU00370), albA (BSU37370), albB (BSU37380), albC (BSU37390), albD (BSU37400), albE (BSU37410), albF (BSU37420), argC (BSU11190), argJ (BSU11200), asnO (BSU10790), azlB (BSU26720), azlC (BSU26710), azlD (BSU26700), bnrQ (BSU26690), bofA (BSU00230), cotA (BSU06300), cotD (BSU22200), cotE (BSU17030), cotF (BSU40530), cotH (BSU36060), cotM (BSU17970), cotT (BSU12090), cotV (BSU11780), cotW (BSU11770), ctpB/yvjB (BSU35240), cwlC (BSU17410), cwlH (BSU25710), cwlJ (BSU02600), cypA (BSU26740), cysC (BSU15600), cysH (BSU15570), cysK (BSU00730), cysP (BSU15580), dctP (BSU04470), dctR (BSU04460), exuR (BSU12370), exuT (BSU12360), gerE (BSU28410), gerM (BSU28380), gerPA (BSU10720), gltR (BSU26670), goxB (BSU11670), kapD (BSU31470), lip (BSU31470), mpr (BSU02240), phoB (BSU05740), proH (BSU18480), proJ (BSU18470), racA/ywkC (BSU37030), safA (BSU27840), sat (BSU15590), spoIID (BSU36750), spoIIGA (BSU15310), spoIIIAA (BSU24430), spoIIIAB (BSU24420), spoIIIAF (BSU24380), spoIIM (BSU23530), spoIIP (BSU25530), spoIVA (BSU22800), spoIVCA (BSU25770), spoIVCB (BSU25760), spoIVFA (BSU27980), spoIVFB (BSU27970), spoVD (BSU15170), spoVE (BSU15210), spoVID (BSU28110), spoVK (BSU17420), spoVM (BSU15810), spsA (BSU37910), spsG (BSU37850), spsJ (BSU37830), tenI (BSU11660), thiS (BSU11680), treA (BSU07810), treR (BSU07820), uxuA (BSU12340), ybaN (BSU01570), ybaS (BSU01590), ycgF (BSU03090), ycgM (BSU03200), ycgN (BSU03210), ydcI (BSU04780), ydhF (BSU05730), yeeA (BSU06760), yeeB (BSU06770), yeeC (BSU06780), yefA (BSU06730), yefB (BSU06740), yefC (BSU06750), yfhP (BSU08620), yfkO (BSU07830), yfmC (BSU07520), yfmD (BSU07510), yfnD (BSU07310), yfnE (BSU07300), yfnF (BSU07290), yfnG (BSU07280), yfnH (BSU07270), yhbB (BSU08920), yhbH (BSU08980), yhcO (BSU09160), yhcP (BSU09170), yheC (BSU09780), yheD (BSU09770), yheH (BSU09720), yheI (BSU09710), yhjL (BSU10550), yisZ (BSU10910), yitA (BSU10920), yitB (BSU10930), yitC (BSU10940), yitD (BSU10950), yitE (BSU10960), yjcA (BSU11790), yjcM (BSU11910), yjcN (BSU11920), yjcO (BSU11930), yjmC (BSU12320), yjmD (BSU12330), yjmF (BSU12350), yknT (BSU14250), yknU (BSU14320), yknV (BSU14330), ykuD (BSU14040), ykvI (BSU13710), ykvU (BSU13830), ylbJ (BSU15030), ylbO/gerR (BSU15090), yngE (BSU18210), yngF (BSU18220), yngG (BSU18230), yngH (BSU18240), yngI (BSU18250), yngJ (BSU18260), yoaB (BSU18540), yoaD (BSU18560), yodU (BSU19810), ypqA (BSU22240), ypqP (BSU21670), yqfT (BSU25120), yqhV (BSU24440), yrdK (BSU26680), yydB (BSU40220), yydC (BSU40210), yydD (BSU40200), yydG (BSU40170), yydH (BSU40160), yydI (BSU40150), and yydJ (BSU40140). Microarray data were deposited in the Gene Expression Omnibus database under the accession number GSE1620, where they are accessible at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE1620 .

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517825.xml

517824

Mechanism of Prion Propagation: Amyloid Growth Occurs by Monomer Addition

Abundant nonfibrillar oligomeric intermediates are a common feature of amyloid formation, and these oligomers, rather than the final fibers, have been suggested to be the toxic species in some amyloid diseases. Whether such oligomers are critical intermediates for fiber assembly or form in an alternate, potentially separable pathway, however, remains unclear. Here we study the polymerization of the amyloidogenic yeast prion protein Sup35. Rapid polymerization occurs in the absence of observable intermediates, and both targeted kinetic and direct single-molecule fluorescence measurements indicate that fibers grow by monomer addition. A three-step model (nucleation, monomer addition, and fiber fragmentation) accurately accounts for the distinctive kinetic features of amyloid formation, including weak concentration dependence, acceleration by agitation, and sigmoidal shape of the polymerization time course. Thus, amyloid growth can occur by monomer addition in a reaction distinct from and competitive with formation of potentially toxic oligomeric intermediates.

Introduction Many proteins of diverse sequences, structures, and functions form morphologically similar β-sheet–rich fibrillar aggregates commonly referred to as amyloid ( Kelly 1998 ; Dobson 2001 ). Amyloid formation is associated with a range of disorders, including neurodegenerative diseases such as Alzheimer's and Parkinson's, and the self-propagating nature of amyloids is thought to underlie prion inheritance. Despite the importance of this process, many questions remain about how amyloid fibers form and grow ( Goldberg and Lansbury 2000 ; Zerovnik 2002 ; Ross et al. 2003 ; Thirumalai et al. 2003 ). While reminiscent of other protein polymerization processes, such as those of actin and tubulin, amyloid formation in most cases does not seem to be well described by simple nucleated polymerization models ( DePace et al. 1998 ; Serio et al. 2000 ; Padrick and Miranker 2002 ; Zerovnik 2002 ; Ross et al. 2003 ; Thirumalai et al. 2003 ). Efforts to decipher the underlying mechanism of amyloid conversion have been greatly complicated by the near ubiquitous presence of smaller, oligomeric aggregates during fiber formation and growth ( Serio et al. 2000 ; Bitan et al. 2003 ; Caughey and Lansbury 2003 ; Souillac et al. 2003 ). These oligomers vary widely in morphology, and include spherical, protofibrillar, and annular structures. A growing body of evidence suggests that certain oligomers may be the toxic species that gives rise to amyloid disease ( Caughey and Lansbury 2003 ). Furthermore, it remains an open question whether the fibers themselves are toxic, neutral, or even protective in some cases. However, despite great interest in these oligomers, it is unknown whether they are critical intermediates for amyloid formation or if fibers can form in their absence ( Goldberg and Lansbury 2000 ; Ross et al. 2003 ; Scheibel et al. 2004 ). The yeast prion state [ PSI + ], which results from self-propagating aggregation of the translation termination factor Sup35 and leads to a nonsense suppression phenotype, provides an excellent system for studying amyloid fiber formation and prion propagation ( Tuite and Cox 2003 ). Prion inheritance in vivo is mediated by a glutamine/asparagine-rich N-terminal domain (N) and, to a lesser extent, a charged middle domain (M) ( Bradley and Liebman 2004 ). In vitro the NM domain forms self-replicating amyloid fibers ( Glover et al. 1997 ; King et al. 1997 ), and when introduced into yeast these fibers initiate the [ PSI + ] prion state ( Sparrer et al. 2000 ; King and Diaz-Avalos 2004 ; Tanaka et al. 2004 ), establishing that the amyloids are in fact the infectious prion element underlying [ PSI + ]. De novo NM polymerization is characterized by a long lag phase followed by a cooperative conversion into amyloid. The lag phase can be eliminated by addition of preformed NM fibers. Oligomers similar to those seen during polymerization of other amyloidogenic proteins have been observed during NM fiber formation and even have been seen localized proximal to fiber ends ( Serio et al. 2000 ). The kinetic role of these oligomers, however, remains poorly understood ( Scheibel et al. 2004 ); for example, does amyloid growth occur by capture of oligomeric intermediates at fiber ends, as suggested in earlier studies ( Serio et al. 2000 )? Here we use a combination of kinetic studies designed to report on specific, well-defined steps in the polymerization reaction together with direct single-molecule fluorescence measurements to explore the mechanism of formation and growth of the infectious NM amyloids. Results We first examined the distribution of oligomeric species present during NM amyloid formation using analytical ultracentrifugation (AUC). Equilibrium AUC indicated the material was predominantly monomeric, giving fitted average molecular masses of 28.5 ± 1.7 kDa (calculated monomer mass is 29.6 kDa), with no appreciable concentration dependence from 1.5 to 5.8 μM ( Figure 1 A and 1 B). By velocity AUC, which is better suited to resolve a small population of a larger oligomer ( Schuck et al. 2002 ), the vast majority of the material fit well to a single peak in a sedimentation coefficient distribution obtained by direct boundary modeling [c(s)] ( Schuck et al. 2002 ) with a sedimentation coefficient of 1.9 S ( Figure 1 C). The lack of other detectable peaks is significant, as we were readily able to detect the presence of small amounts of a larger protein complex (10% GroEL) added to an NM sample ( Figure 1 D). Thus, on the time scale of the 5–20 h needed for the AUC analysis, which exceeds the polymerization reaction times in our present studies, the large majority of NM is monomeric. This has allowed us to examine fiber formation in the absence of significant off-pathway aggregation, which was seen to obscure the underlying kinetics of other amyloid systems ( Rhoades et al. 2000 ; Padrick and Miranker 2002 ; Souillac et al. 2003 ). However, our AUC data did not address whether a rare oligomeric species serves as a critical on-pathway intermediate for fiber growth. Figure 1 NM Is Predominantly Monomeric (A) Oligomeric state of NM prior to assembly was assessed by equilibrium AUC. Above, raw plot of absorbance versus radial position for 5.8 μM NM at equilibrium, with best-fit line for a single species. Below, residuals from the fit shown above. (B) Equilibrium AUC data from 1.5, 3.8, 5.2, 5.5, and 5.8 μM samples of NM were fit to a single-species model. Shown is fitted molecular mass versus concentration. For reference, dashed lines are shown at the calculated monomer mass (29.6 kDa) and dimer mass (59.2 kDa). (C) Distribution of absorbance versus sedimentation coefficient at the indicated concentrations was obtained from velocity sedimentation data using Sedfit ( Schuck et al. 2002 ). Inset is a magnification of the main peaks. (D) A small amount of a larger complex could be resolved by velocity sedimentation. Distribution of absorbance versus sedimentation coefficient for each of 3 μM NM and 2.7 μM NM plus GroEL (with absorbance equivalent to that of 0.3 μM NM) is shown. Inset is a magnification for sedimentation coefficients between 15.6 S and 28.4 S. To explore this possibility, we looked at the concentration dependence of the initial rate of growth of soluble NM onto the ends of a well-defined amount of preformed fibers. If a rare oligomer is critical for fiber growth, then mass action dictates that the quantity and rate of formation of such oligomers will be highly concentration dependent. As a consequence, the initial rate of polymerization should depend strongly on the concentration of soluble NM. Using a thioflavin T binding assay that allows continuous measurement of amyloid formation, we measured the rate of amyloid growth after mixing a quantity of soluble NM with a known amount of preformed nuclei. We found that the initial rate of fiber growth was directly proportional to the concentration of soluble NM over a range of 0–1 μM ( Figure 2 A). This linear dependence of fiber growth on NM concentration was not dependent on the method used to monitor growth ( Figure 2 B and 2 C). We also looked at the dependence of the polymerization rate on seed concentration. As expected, this rate was directly proportional to concentration of fiber ends ( Figure 2 D– 2 F), and importantly, the rate was still linear up to a concentration of seed at which half of the soluble material was polymerized in less than 3 min. Therefore, NM in solution is in rapid equilibrium with conformations competent to add to fibers. Figure 2 Kinetics of NM Fiber Growth Support a Monomer Addition Model (A) Initial rate of polymerization versus concentration of NM. Soluble NM at the indicated final concentrations was mixed with sonicated fibers (2.5 μM fibers at 5% of final volume) and polymerization was followed by a continuous thioflavin T assay. Rates shown were determined by the initial slopes of polymerization curves. (B) Initial rate of polymerization versus concentration of soluble NM in the presence of sonicated seed (1% of final volume) measured by a discrete thioflavin T binding assay. Error bars throughout represent the standard deviation of at least three measurements. (C) Polymerization of NM labeled with Alexa-647 at a C-terminal cysteine was monitored by the quenching of Alexa-647 fluorescence. The indicated concentrations of soluble NM were mixed with sonicated fibers (4% of final volume), and the initial rate of polymerization was measured. (D) Initial rate of polymerization versus concentration of seed. Soluble NM (2.5 μM final concentration) was mixed with the indicated quantity of sonicated seed. Highly fragmented fibers were used to maximize the absolute rate. Rates were measured as in (A). (E) Initial rate of polymerization versus concentration of seed as measured by discrete thioflavin T binding assay. Initial soluble NM concentration was 2.5 μM. (F) Initial rate of polymerization versus concentration of seed as measured by Alexa-647 fluorescence quenching. Initial soluble NM concentration was 200 nM. Interestingly, at higher NM concentrations (>10 μM), the rate of fiber elongation shows a weaker-than-linear dependence on NM concentration. Accumulation of off-pathway aggregates, which we can observe at high [NM], could contribute to such an effect. However, in the presence of moderate levels of denaturant (100 mM guanidine hydrochloride [GuHCl] and 100 mM urea), NM remained monomeric up to at least 20 μM ( Figure 3 A), and under these conditions we still observed that the rate of polymerization becomes largely independent of the concentration of NM ( Figure 3 B). This suggests that a conformational rearrangement of NM after binding to fiber ends becomes rate limiting at high NM concentrations (see Figure 5 C). As would be expected if unconverted NM is occupying fiber ends, even at the highest NM levels the rate of polymerization remains linearly dependent on the amount of seed added ( Figure 3 B, inset). This conformational rearrangement may be analogous to the locking step observed in Aβ polymerization ( Esler et al. 2000 ; Cannon et al. 2004 ), although here it occurs on a faster time scale. Estimates from growth rates measured by atomic force microscopy (AFM) indicate that the rearrangement occurs within ∼1 s (see Materials and Methods ), a time similar to that seen for the folding of comparably sized β-sheet–rich proteins ( Reid et al. 1998 ). Figure 3 Evidence that a Conformational Conversion following NM Monomer Binding to Fiber Ends Becomes Rate Limiting at High [NM] (A) NM is predominantly monomeric at concentrations up to 20 μM in the presence of a moderate level of denaturant (100 mM urea and 100 mM GuHCl). Oligomeric state was assessed by velocity AUC as in Figure 1 C for NM at 10, 15, and 20 μM. (B) Soluble NM was mixed with sonicated fibers in the presence of 100 mM urea and 100 mM GuHCl to minimize off-pathway aggregation at higher [NM], and the initial rate of polymerization was measured as in Figure 2 A. Inset, rate of polymerization of 20 μM soluble NM versus seed percentage (v/v). The continued linear dependence of rate on fiber ends indicates that at high [NM], fiber ends remain limiting. Figure 5 Effect of Fiber Fragmentation on Polymerization Kinetics (A) De novo NM polymerization (1.0–12 μM) followed by a continuous thioflavin T binding assay. (B) Lag time (measured as time to 5% completion of polymerization) versus NM concentration for the polymerizations shown in (A). (C) Schematic of nucleated polymerization with fragmentation. At low concentrations fiber growth is limited by monomer binding, whereas at high concentrations conformational conversion after binding becomes limiting. (D) The de novo polymerization data shown in (A) were fit with a linearized model ( Ferrone 1999 ) that includes nucleation, fiber growth by monomer addition, and fragmentation, assuming the indicated sizes for the smallest stable amyloid species (nucleus size). Plotted are the residuals (best-fit value minus observed data). Each block (labeled at the left by the nucleus size used) represents the residuals from simultaneously fitting all of the data. Each line within the block displays the individual residuals from a single concentration (1 μM [top] to 12 μM [bottom]). Residuals are color coded according to the color key at the right, with red and green indicating large errors and yellow indicating a good fit. Time varies from time zero (farthest left) to the time of 3% completion of polymerization. Below (NP) are residuals from fitting the same data using a simple nucleated polymerization model. Note the systematic deviations from the model for large nucleus sizes and for NP. A model in which growth of NM fibers is largely limited by the encounter rate of monomer and fiber end still has two puzzling features of the polymerization process to explain. First, like many other amyloid formation reactions, NM polymerization is dramatically accelerated by agitation ( DePace et al. 1998 ; Serio et al. 2000 ) ( Figure 4 A), which has been hypothesized to increase the fiber growth rate by breaking up off-pathway aggregates or speeding up diffusion of large on-pathway oligomers ( Serio et al. 2000 ). Neither of these explanations applies to a predominantly monomeric solution where fibers grow by monomer addition. Previous attempts to understand agitation were complicated by the multistep nature of the polymerization process, which includes nucleation, growth, and other steps. We specifically looked at the effect of agitation on the fiber growth step using two identical seeded reactions, one agitated and one undisturbed. Remarkably, we found that seeded polymerizations initially proceeded at exactly the same rate with or without agitation ( Figure 4 B). However, after extensive growth of the added seeds (∼20 min) the rate of the agitated reaction began to accelerate markedly ( Figure 4 B, inset), whereas unagitated reactions slow down as the pool of monomeric NM becomes depleted. Figure 4 Agitation Causes Fiber Fragmentation (A) Dependence of de novo polymerization reactions on degree of agitation. Polymerization in a microplate with shaking every minute (line only) was followed by thioflavin T fluorescence. Polymerization in a test tube disturbed only by pipetting to take samples for measurement (diamonds) was measured by Congo Red binding. Polymerization in a microplate with absolutely no agitation (multiple samples were started in parallel and no sample was measured more than once) (squares) was measured by Congo Red binding. (B) Effect of agitation on elongation rate. Identical reactions with 5 μM soluble NM and 2% (v/v) seeds were grown with (triangles) or without (circles) agitation. The seeds were made fresh and sheared by passing through a 25-gauge needle ten times. Polymerization was assayed by discrete thioflavin T measurements. Inset, identical reactions followed for 100 min. (C) Effect of agitation on fiber lengths. Long fibers were grown from sonicated seeds in the absence of agitation and then subjected to agitation (end-over-end rotation in a 2-ml tube). AFM images were taken after 0, 1, and 15 h of rotation. Each image is 5 μm by 5 μm. (D) Effect of agitation on seeding efficacy. Fibers from the samples imaged in (C) were used to seed polymerization of 10 μM soluble NM, and the initial rate of polymerization was measured as in Figure 2 A. (E) De novo NM polymerization was measured by continuous thioflavin T fluorescence, and relative fiber number was computed as a function of time (see Materials and Methods ). Inset, fiber number versus time was fit to an exponential curve for time = 0 to time = 100 min. At a given NM concentration the rate of polymerization is proportional to seed concentration, and therefore the acceleration seen in the agitated reaction implies that the number of fiber ends is increasing with time, perhaps due to the ability of agitation to fragment long fibers. To test this possibility, we prepared long fibers by allowing prolonged growth of NM onto preformed seeds in the absence of agitation, and then subjected the fibers to end-over-end rotation. Prior to rotation, the fibers were very long and had a weak ability to seed polymerization of monomeric NM. After 1 h of rotation, we saw many more fibers of much shorter length, and the seeding activity had increased 10-fold. Further rotation (14 h) produced still shorter fibers and higher seeding activity ( Figure 4 C and 4 D). Interestingly, a reanalysis of the kinetics of de novo NM polymerization, based on the fact that the rate of polymerization is linearly dependent on the amount of seed present and the amount of monomer remaining (see Figure 2 ), indicates that the number of seeds increases exponentially during NM polymerization ( Figure 4 E). Thus, while agitation does not affect the rate of NM addition to fiber ends, it accelerates polymerization by increasing the number of ends through amyloid fragmentation. The second unusual feature of NM polymerization that must be reconciled with a model of monomer addition is the weak dependence (less than first order) of the length of the lag time on the concentration of NM in unseeded polymerizations ( DePace et al. 1998 ; Serio et al. 2000 ). This finding, together with the sigmoidal curve shape involving a pronounced lag phase followed by an abrupt increase in the rate of polymerization, provided key evidence against a simple nucleation-polymerization model ( DePace et al. 1998 ; Serio et al. 2000 ; Padrick and Miranker 2002 ), which is characterized by an initially parabolic (t 2 ) time course ( Ferrone 1999 ). We similarly observed a lag phase that depends on approximately the 0.4 power of initial concentration ( Figure 5 A and 5 B). In other systems, this weak concentration dependence has been attributed to an accumulation of large off-pathway species whose formation is competitive with on-pathway processes ( Rhoades et al. 2000 ; Serio et al. 2000 ; Padrick and Miranker 2002 ; Souillac et al. 2003 ). However, in our experiments NM is predominantly monomeric, so the observed weak concentration dependence is not a simple consequence of the accumulation of off-pathway aggregates. Previous work from Ferrone (1999) using a linearized model demonstrated that addition of fragmentation to a nucleation-polymerization process could lead to sigmoidal polymerization curves. We explored whether fragmentation could also explain the weak concentration dependence using both numerical integration and direct fitting of the data to a linearized model. We modeled a simple polymerization process involving nucleation, growth, and fragmentation ( Figure 5 C). Numerical integration confirmed the expected strongly sigmoidal curve shape, but we also found that fragmentation greatly reduced the concentration dependence. For example, a nucleus size (the number of monomers in the smallest stable amyloid species) of six gives a third-order concentration dependence in nucleated polymerization but only approximately a first-order dependence with fragmentation, and the apparent concentration dependence decreases further for smaller nucleus sizes. We additionally tried fitting data to the linearized analytical solution which is valid for early parts of the polymerization ( Ferrone 1999 ; we restrict ourselves to the first 3%) and only requires fitting two parameters. Fixing nucleus size and fitting polymerization curves at seven concentrations, we obtained residuals of fits at a series of potential nucleus sizes ( Figure 5 D). Although more direct measurements are needed to define the exact size of the minimal stable seed, both approaches suggest a small nucleus size (three monomers or smaller), consistent with recent observations on polyglutamine polymerization ( Chen et al. 2002 ). The advent of single-molecule fluorescence technology enabled an independent and more direct way to test whether fibers grow by monomer addition. Previous work ( Inoue et al. 2001 ) established that fluorescent NM fibers attached to a microscope slide could be grown and visualized using epifluorescence. Here we examined fiber growth using total internal reflection fluorescence microscopy (TIRF), which allows single-molecule detection. We attached Cy5-labeled NM fibers to a slide through a biotin-streptavidin linkage and added a solution of Cy3-labeled soluble NM ( Figure 6 A). Working at a label concentration of 133 nM (200 nM or more total soluble NM) to minimize background fluorescence, we could readily detect addition of individual Cy3 fluorophores at the ends of Cy5-labeled fibers. Two observations indicate that we were monitoring growth events mediated by fiber ends. First, after extended time (∼1 h) at our working concentration, bright Cy3 fluorescence (consisting of many labeled NM molecules) accumulated specifically at fiber ends ( Figure 6 B). Second, Cy3 addition events at fiber ends were long lived (55% of spots at fiber ends remained visible in the second frame measured 15 s later, whereas in the absence of fiber ends fewer than 20% of spots remained). Figure 6 Fundamental Unit of Addition for NM Fiber Growth Determined by TIRF (A) Schematic of experimental setup. Cy5-labeled fibers (red) were attached to the microscope slide via biotin-streptavidin linkage. Cy3-labeled NM (green) was added in solution. (B) TIRF image of fibers (red) grown by addition of 200 nM NM (67% labeled with Cy3 [green]) for approximately 1 h. Cy3 fluorescence at fiber ends in this image represents multiple addition events. (C) Schematic of expected results if oligomers (top, trimer model shown for example) or monomers (bottom) are added to fiber ends. Graphs at the right show simulated data of fraction of additions versus fluorescence intensity of addition. Note that for the monomer model, the simulated intensities are the same whether 9% or 67% of the soluble NM is labeled. (D) Observed data: fraction of observed events versus fluorescence intensity of events. Intensities of Cy3 spots appearing at fiber ends were measured. (E) Intensities of Cy3 spots appearing at fiber ends (circles) and intensities of Cy3 spots that bleached in a single step (squares). By following the intensities of single fluorescent addition events, we could directly determine the size of the unit of addition. Oligomer addition (unlike monomer addition) predicts that the fluorescence intensity of a single addition event should depend strongly on the fraction of NM that is fluorescently labeled ( Figure 6 C). We prepared NM at both 67% (133 nM label, 200 nM total soluble NM) and 9% (89 nM label, 966 nM total soluble NM) labeling efficiency. Intensities of Cy3 spots that appeared at fiber ends were then measured, and for each degree of labeling, a histogram of intensities was created. Strikingly, we observed that the distribution of intensities was independent of the degree of labeling ( Figure 6 D). Furthermore, the intensity distribution of the fiber addition events was comparable to that of NM-Cy3, which was confirmed to be monomeric by single-step photobleaching experiments ( Figure 6 E). Together these data establish that amyloid growth is occurring by the addition of NM monomers onto fiber ends. Discussion Amyloid formation is a ubiquitous feature of polypeptides. However, compared to protein folding reactions, for which a wide range of biophysical, structural, and analytical approaches have provided detailed information on the pathways by which native states are obtained, very little is known about the underlying steps by which amyloid fibers assemble and grow. In large part this is due to the complexity of the amyloid formation reaction, which can involve a wide spectrum of on- and off-pathway intermediates. In the present study, we have used a combination of kinetic analysis designed to look at specific mechanistic steps and single-molecule fluorescence to determine how amyloid fibers of Sup35, the prion determinant of the yeast state [ PSI + ], form and grow. A key finding is that Sup35 NM amyloids grow efficiently by the addition of monomers to fiber ends. We also establish that monomer addition, in combination with fiber fragmentation, accurately predicts the otherwise puzzling features of de novo polymerization kinetics. While it is still possible that oligomers could add to the ends of fibers if allowed to accumulate, we find that monomer addition is rapid and efficient. This monomer addition mechanism can account for the generation of the [ PSI + ] translation read-through phenotype in vivo. While division of Sup35 prion particles appears to depend on cellular chaperones including Hsp104 ( Ness et al. 2002 ; Osherovich et al. 2004 ), both genetic and biochemical evidence suggests their growth is an Hsp104-independent process ( Ness et al. 2002 ; Shorter and Lindquist 2004 ). Additionally, with approximately 200 seeds in a cell ( Cox et al. 2003 ) and the second-order rate constant we observe (approximately 2 × 10 5 M −1 s −1 ), soluble Sup35 would have a half-life of about 3 min, which is comparable to the time scale of Sup35 translation and much faster than the doubling time of yeast (90 min). Thus, the monomer growth mechanism would lead to the depletion of Sup35 characteristic of the [ PSI + ] state (see Materials and Methods ). Moreover, the properties of Sup35 polymerization we observe seem particularly well suited to explain the prion phenotype in yeast. De novo nucleation of new fibers is extremely slow, but existing fibers grow rapidly and their fragility may allow them to be divided easily. These features are appropriate for a protein-based switch that is bistable ([ PSI + ] and [ psi − ] are both stable states) and whose aggregated state must be amplified exponentially to keep pace with cell division. Many kinetic features seen for NM polymerization are shared by other amyloidogenic proteins ( Uversky et al. 2001 ; Chen et al. 2002 ; Padrick and Miranker 2002 ), suggesting that monomer addition may represent a mechanism of amyloid growth common to other fibers. Whether other amyloids do in fact grow by monomer addition and how the role of oligomeric species in the polymerization process correlates with toxicity remain important open questions. Many of the analytical approaches and experimental techniques used in this work should be directly applicable for exploring these issues in other systems. A broader understanding of the role of oligomeric species in the formation and growth of amyloids will be critical for determining the physiological effects of amyloidogenesis as well as guiding efforts to counteract amyloid toxicity. For example, the conclusion that amyloid growth and oligomer formation can occur in distinct, competitive reactions may help explain the poor correlation between formation of visible aggregates and toxicity in neurodegenerative diseases of protein misfolding ( Saudou et al. 1998 ; Cummings et al. 1999 ; Goldberg and Lansbury 2000 ; Wittmann et al. 2001 ). More speculatively, the finding that fiber growth does not require oligomeric intermediates raises the possibility that agents designed to promote direct fiber formation, by disfavoring oligomer formation, may help prevent the accumulation of potentially toxic oligomeric intermediates. Materials and Methods Reagents Cy3 mono maleimide Gold and Cy5 mono maleimide Gold were purchased from Amersham (Little Chalfont, United Kingdom). Alexa Fluor 647 C 2 maleimide was purchased from Molecular Probes (Eugene, Oregon, United States). Streptavidin was purchased from Molecular Probes. Biotinylated BSA and thioflavin T were purchased from Sigma (St. Louis, Missouri, United States). Biotin-PEAC 5 -maleimide (6-{N′-[2-(N-Maleimido)ethyl]-N-piperazinylamido}hexyl D -biotinamide, hydrochloride) was purchased from Dojindo (Gaithersburg, Maryland, United States). Alkali-soluble casein (5%) was purchased from Novagen (Madison, Wisconsin, United States). Protein expression Sup35 residues 1 to 254 (NM) C-terminally tagged with 7×-histidine were purified as reported previously ( DePace et al. 1998 ; Tanaka et al. 2004 ). Fluorescence labeling NM with a single cysteine inserted after the polyhistidine tag was labeled by Cy3 mono maleimide Gold, Cy5 mono maleimide Gold, or Alexa Fluor 647 C 2 maleimide (10 equivalents) in 25 mM sodium phosphate buffer containing 6 M GuHCl and 0.1 mM TCEP (pH 8.0) at 4 °C overnight with 67% efficiency. Efficiency of labeling was determined from absorbance at 275 nm and at the excitation maximum of the dyes, correcting for absorbance of the dye at 275 nm. Biotinylated NM was made analogously using biotin-PEAC 5 -maleimide (2.5 equivalents) with 1 mM TCEP, obtaining about 20% efficiency. AUC Both velocity and equilibrium AUC were performed in a Beckman Optima XL-A analytical ultracentrifuge using an An60Ti-Rotor at 20 °C. Protein was in buffer C (5 mM potassium phosphate and 150 mM sodium chloride [pH 7.4]). Velocity sedimentation was analyzed at a speed of 40,000 rpm at NM concentrations of 3.7, 6.2, and 12.9 μM (determined by absorbance at 275 nm). Data were collected with three replicates at radial steps of 0.003 cm and scans every 9 min. Data were analyzed with Sedfit using the c(s) method ( Schuck et al. 2002 ). Equilibrium sedimentation was analyzed at a speed of 23,300 rpm for concentrations of 1.5, 3.8, 5.0, 5.5, and 5.8 μM. Data were analyzed with Sedphat ( Schuck 2003 ) by fitting each curve individually to a single species model. Sedfit and Sedphat are available at http://www.analyticalultracentrifugation.com . Polymerizations For continuous thioflavin T assay measurements (see Figures 2 A, 2 D, 3 B, 4 A, 4 D, 4 E, 5 A, 5 B, and 5 D), concentrated stocks of NM stored in either 6 M GuHCl (fluorescently labeled NM and NM used for Figures 2 C, 2 F, and 6 ) or 4 M GuHCl and 4 M urea (NM used for all other experiments) were diluted at least 200-fold into buffer C. In order to compare polymerizations done at different concentrations of NM, residual denaturant concentrations were equalized for all samples in each experiment. Reactions consisted of 100 μl of protein in buffer C plus 100 μl of 25 μM thioflavin T in 50 mM glycine (pH 8.5). Seed was added immediately before observation. Fluorescence was monitored in a 96-well fluorescence plate reader (Molecular Devices, Sunnyvale, California, United States; 442 nm excitation and 483 nm emission). Reactions were carried out at 25 °C either without (for seeded reactions) or with (for unseeded reactions) 3 s of shaking between each measurement. Unseeded polymerizations were done in the presence of a small amount of soluble casein (0.02%), as empirically that condition gave flatter plateaus at the end of the reaction. Discrete thioflavin T assay measurements (see Figures 2 B, 2 E, and 4 B) were performed as continuous measurements, except that polymerization proceeded in buffer C in the absence of thioflavin T. At indicated time points, a 100-μl aliquot of polymerization reaction was mixed with 100 μl of thioflavin T for measurement. Reactions were either undisturbed or rotated end-over-end in a 2-ml tube as indicated. For polymerizations followed by Alexa Fluor 647 fluorescence (see Figures 2 C and 2 F), the fluorescence of a 200-μl volume containing the indicated concentration of labeled NM and the indicated quantity of sonicated fibers was measured over time in a 96-well fluorescence plate reader (Molecular Devices; 650 nm excitation and 670 nm emission). Fluorescence was partially quenched (approximately 56% of fluorescence lost) upon polymerization. Rates of polymerization were determined by taking the initial slope of the polymerization curve divided by the total change in fluorescence multiplied by the concentration of the sample. Wells in the microplate were blocked with 5% casein and washed with water prior to polymerization experiments to minimize adsorption of NM to the sides of the wells. Seeds used were produced by polymerization of 2.5 μM NM followed by sonication with a Fisher Scientific Sonic Dismembrator (Model 500) fitted with a microtip for 60 s (for all experiments except that shown in Figure 4 B) or sheared by passing through a 25-gauge needle ten times (for Figure 4 B). Sheared rather than sonicated fibers were used for examining the effect of agitation on fiber growth because longer fibers were seen to break more easily than shorter fibers. AFM AFM samples were prepared and analyzed essentially as previously described ( DePace and Weissman 2002 ). All images were taken using tapping mode on a Digital Instruments Multimode AFM, Nanoscope IIIa controller, and Micromasch NSC15 tips. For fiber images (see Figure 4 C), 20 μl of fibers (either 2.5 or 5.0 μM total NM) was deposited on mica disk for 20 s. The disk was then washed twice with 160 μl of distilled deionized water and aspirated until dry. For estimating fiber growth rates (see below), new growth of unlabeled NM off of preformed fibers made from NM tagged with an HA epitope was measured as previously described ( DePace and Weissman 2002 ). The HA epitope allows the initial seeds, but not new growth, to be labeled with antibody after the fibers are deposited on the mica. Computer simulation, curve fitting, and computation Numerical integration was performed using Euler's method implemented in a C program (code available upon request). The following is a general description of the strategy used for numerical modeling. A set of differential equations was written to account for the following steps: formation of a new stable nucleus from a defined number (n) of monomers, growth of fibers by monomer addition (limited by the rate of encounter of monomer and fiber end), and production of new fibers by fragmentation of existing fibers. Fiber growth was modeled to be irreversible (i.e., fibers do not depolymerize) ( DePace and Weissman 2002 ), because experimentally the critical concentration for growth appeared to be very small (less than 50 nM), if one exists. Fragmentation was modeled as a length-dependent process (long fibers break more easily than short fibers), and to simplify calculations, it was assumed that no fiber breaks to yield a fragment as small or smaller than the size of the smallest stable nucleus. This is likely to be true for the large majority of fragmentation events because of the tendency of fibers to break closer to their center rather than their ends and because of the low propensity of short fibers to break ( Hill 1983 ). Concentrations of monomers and fibers of each length from n to 3,016 monomers were modeled as a function of time. An upper bound for fiber length was necessary to make the calculations possible, and the total of 3,016 was chosen because increasing this bound had no measurable impact on results. Variables and parameters used for modeling were the following: t = time; x = concentration of monomers; y i = concentration of fibers containing i monomers; y = concentration of all fibers of length n or greater; n = the number of monomers in the smallest stable species; BreakDep = the power to which the rate of fragmentation of a fiber depends on its length. For all simulations this value was set to 3, based on theoretical estimates for polymers ( Hill 1983 ), although the results were robust to changes in this parameter. The BreakDep parameter was included to model the length dependence of fragmentation: Here, j represents fiber length (the number of monomers in a fiber) and the sum is taken over all values of j greater than or equal to (i + n + 1). The z above(i) term accounts for fragmentation of all fibers long enough to break and give a resulting fiber of length i. The sum begins at j = (i + n + 1) because no smaller fiber could break to give a fragment of size i, leaving another fragment of at least size (n + 1). Differential equations for modeling contain the following terms: k nuc = effective fiber nucleation rate constant; k growth = fiber elongation rate constant; k break = fiber fragmentation rate constant. The equations are as follows. For monomer concentration: The k nuc *n* x n term accounts for the loss of n monomers during the formation of each stable nucleus. Stable nuclei are modeled to form at a rate of k nuc * x n . The k growth * x * y term accounts for the loss of one monomer for each fiber elongation event; each fiber grows at a rate of k growth * x and there are y total fibers. For the smallest stable species: The k nuc * x n term accounts for the spontaneous formation of new nuclei. The k growth * x * y n term accounts for the disappearance of fibers of length n as they grow into longer fibers. For short fibers (of length less than or equal to twice the size of the nucleus plus one monomer): Fibers of length i are created by elongation of shorter fibers at a rate of k growth * x * y i−1 and lost by growth into longer fibers at a rate of k growth * x * y i , accounting for the first term. The second term accounts for formation of fibers of length i by fragmentation of longer fibers (see description of z above(i) above). There is no loss of fibers of these lengths from fragmentation because they are assumed to be too short to break. For fibers longer than twice the size of the nucleus plus one monomer: The first two terms are the same as for the short fibers. The additional term accounts for fragmentation of fibers of length i into shorter fibers. This term is arrived at in the following way: k break is a rate constant for the fragmentation process, y i is the number of fibers of this length, (i − 2*n − 1) is the number of points along the length of the fiber where it could break, and i BreakDep accounts for the tendency of longer fibers to break more easily than shorter fibers (see the description of BreakDep above). There are (i − 2*n − 1) places the fiber could break because there are (i − 1) monomer-monomer interfaces in a fiber of length i and breaking at any of the n interfaces closest to either end of the fiber would give one fiber of size n or smaller. Curve fitting of experimental data to the linearized model of Ferrone (1999) was performed using a Levenberg-Marquardt least squares fitting method implemented in MATLAB (The Mathworks, Natick, Massachusetts, United States). Data was fit to the equation y = A(cosh (B t ) − 1), where t is time and A and B are parameters to be fit with the restriction that AB 2 scales as [NM] n+1 (where n is the number of monomers in the smallest stable species). Relative fiber number (see Figure 4 E) was computed in the following way: If z = total amount of NM in amyloid, and z final = total amount of NM in amyloid at the end of the reaction, then relative fiber number = (d z /d t )/(z final − z ). Values for z were measured by thioflavin T fluorescence (continuous thioflavin T assay). The equation used for relative fiber number comes from the observation (see Figure 2 ) that the bulk growth rate (d z /d t ) is proportional to both fiber concentration (relative fiber number) and soluble NM concentration (z final − z ). The relative fiber number versus time was fit to an exponential ( y = Ae b t ) by nonlinear least squares regression as described above (see Figure 4 E). TIRF Single-molecule imaging was performed using objective-type TIRF illumination configured on a Zeiss Axiovert 200M (Carl Zeiss, Inc., Zurich, Switzerland), and controlled by the QED in vivo software package (Media Cybernetics, Silver Spring, Maryland, United States). Images were acquired digitally with a Mega-10 intensified CCD camera (Stanford Photonics, Stanford, California, United States) and analyzed using ImageJ (National Institutes of Health, Bethesda, Maryland, United States) and MATLAB software. Samples were analyzed in glass-bottom microwell dishes (MatTek, Ashland, Massachusetts, United States; catalog #P35G-1.5-14-C) which were prepared by application of 60 μl of biotinylated BSA (1 mg/ml) for 20 min followed by washing with buffer, application of 60 μl of streptavidin (0.2 mg/ml) for 20 min, washing, application of 100 μl of casein (5%) for 30 min to 2 h, washing, application of 40 μl of Cy5-labeled fibers for 10 min, and finally washing with buffer. The fibers were prepared in a 300-μl reaction with 1% (v/v) seed (sonicated from a 2.5 μM NM reaction). The sonicated seed was grown for 15 min with NM at 2.5 μM (5% biotinylated NM, 20% Cy5-labeled, and 75% unlabeled). The seed was diluted 2-fold and sheared with a 200-μl pipet tip before application to the slides. The slides were kept in buffer until use. Buffer was removed and Cy3-labeled NM (200 nM, 67% labeled; or 955 nM, 9% labeled) was added to the slide immediately before viewing. For addition studies, a Cy5 image was taken to locate the fiber ends and then images were taken every 15 s in the Cy3 channel to look for events at fiber ends. Estimate of maximum fiber growth rate AFM analysis established that NM fibers grow at an average rate of 100 nm/min at a soluble NM concentration of 2.5 μM and can grow at least a factor of two faster at higher concentrations. Estimating one monomer per 3 nm of length in a fiber (from the approximate monomer volume and a fiber diameter of 4.5 nm from AFM) gives a rate of at least one per second. This is a conservative estimate (the real rate may be faster) because larger fiber diameters have been measured by electron microscopy ( Kishimoto et al. 2004 ), indicating that fibers may have more than one monomer per 3 nm of length. Estimating a half-life for Sup35 in a cell We observe a second-order rate constant for fiber elongation of approximately 2 × 10 5 M −1 s −1 , and estimating 200 seeds ( Cox et al. 2003 ), giving an approximate seed concentration of 20 nM in a yeast cell, would result in a half-time of approximately 3 min. We observe Sup35 to be relatively stable against proteolysis within cells, so it would need to be replenished with a half-time of 90 min dictated by the doubling time of yeast. Excluded volume effects from the high concentration of proteins in cytoplasm may further increase the polymerization rate, as we found that 25% ficoll 70 increases polymerization rates by a factor of two or three. Also, 2 × 10 5 M −1 s −1 may be an underestimate of the second-order rate constant if we underestimated the number of monomers per unit length in a fiber. We also note that fiber growth in the cell is not likely to be limited by conformational rearrangement after binding of monomer to fiber end because the concentration of Sup35 in a cell is in the neighborhood of 1 μM ( Sparrer et al. 2000 ). Supporting Information Protocol S1 Overview of Approach and Techniques Used (590 KB DOC). Click here for additional data file. Accession Numbers The GenBank accession number for Sup35p is NP_010457.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517824.xml

548139

Priorities of health policy: cost shifting or population health

Background This paper is an edited version of an invited paper submitted to the Australian Health Care Summit on 17–19 August 2003. It comments upon the policies which have dominated recent debate and contrasts their importance with the importance of five issues which have received relatively little attention. Methods Policy is usually a response to identified problems and the paper examines the nature and size of the problems which heave led to recent policy initiatives. These are contrasted with the magnitude and potential cost effectiveness policies to address the problems in five areas of comparative neglect. Results It is argued that recent and proposed changes to the financing and delivery of health services in Australia have focused upon issues of relatively minor significance while failing to address adequately major inequities and system deficiencies. Conclusion There is a need for an independent review of the health system with the terms of reference focusing attention upon large system-wide failures.

1 Introduction The theme of this paper is that recent and proposed changes to the financing and delivery of health services in Australia have focused upon issues of relatively minor significance while failing to address adequately major inequities and system deficiencies. An intriguing question – not discussed in the paper – is how such drastic failures could continue year after year with little comment and no decisive policy commensurate with the magnitude of the problems. The political-sociological answer to this question is undoubtedly complex and contentious. The failure may (or may not) be attributed to understandable, even unavoidable obstacles arising from Australia's social history. Nevertheless it is important to be aware that these failures exist. The current state of our health services could justifiably be described as a 'silent crisis'; service delivery is highly inequitable and inefficient; patients are dying unnecessarily and avoidable medical errors are imposing huge financial and human costs on the community. While this occurs, health policy at the political level has focussed upon cost shifting between the States and Commonwealth, between public and private sectors and between the well off and the poorer members of society. Reforms addressing the larger issue have progressed at glacial speed, relative to what is achievable, or they have stalled altogether. In Section 2 there is a brief overview of the economists' analytical framework in order to introduce two preliminary issues, viz, the role of social values in health system reform and the constraints created by the limited availability of resources. Two commonly made but wrong inferences from this latter constraint are discussed. Section 3 is concerned with recent policy and, in particular, the changes to private health insurance (PHI) which have been introduced since July 1997. In contrast with the relatively inconsequential (and possibly negative) impact of these policies, five major problems are outlined in Section 4, each of which has received insufficient or no attention. Policy implications and options for future policy are discussed and highlighted throughout the paper. In the final section I argue that the optimal health system – entirely public, largely private or one of the myriad combinations between these polar options – is the system which is most likely to address systemic failures. The most important changes to achieve the optimal system may have less to do with the public-private mix of services, or even funding, than with the extent to which these failures are addressed within any of the conceivable health systems of the future. This, in turn, may depend upon the willingness to create appropriate economic and other incentives. 2 Resources, values and the economic framework The discipline of economics provides a framework for the analysis of options which is based upon a comparison of costs and benefits. If social welfare is to be maximised then the logic of the framework must be adopted explicitly or implicitly. The framework focuses attention upon the benefits which might be obtained when resources are used in a particular way, and the benefits which might have been obtained if they had been used somewhere else – the (opportunity) cost of using those resources in the chosen way. Social wellbeing is maximised when the benefits exceed the (opportunity) costs in every setting and on every margin where choice is possible. While this statement is tautologically true, the focus upon choice highlights two important facts. First, choices generally do exist; the economy is flexible and choices are driven by individual and social preferences. Technical inevitabilities are rarely encountered. Second, and more fundamentally, it is necessary to define 'benefits'. In principle, the abstract framework is consistent with an almost unlimited number of value systems. For example, in the context of an intensive care department with limited capacity, benefits and costs might be measured by lives saved and lives lost. In this simple example the cost benefit formula would translate into a policy of providing ICU beds to those most likely to live. Social objectives in the health sector are clearly more complex than in most other settings and the nexus between objectives, policies and the optimal health system is more problematical. Nevertheless, an important conclusion from this framework (which needs constant repetition) is that there is not a single 'best' health system; rather, there are various options which are more or less consistent with different social goals. This conclusion is illustrated in Table 1 using two highly simplified but archetypal social objectives, viz, the egalitarian desire for equal access to health or health services and the social objective of maximising individual choice. As shown, the first of these objectives is more easily achieved through a compulsory public system with defined benefits and constrained choice. The second objective is most likely to be achieved in a less constrained and more competitive private system which responds to individual preferences, as described and generally prescribed by economic theory for less complex markets and social objectives. Table 1 The relationship between choice, values and the optimal health system Objectives/Social Values Option which maximises likelihood of success Equalise access, outcome Universal (monopoly) Public Insurance/Financing Maximise choice; diversity + safety net Optimise the max of these 2 objectives Pure private (competitive) scheme Both of the above Mixed public-private scheme Conclusion 1: The form of the optimal health scheme depends upon social objectives and disagreement about these translates into differences with respect to the funding and delivery of health services. While the two objectives in Table 1 are archetypes, they broadly correspond with two important but conflicting 'world views'; that is, with different ethical beliefs about the appropriate supply and financing of health services. The social values underpinning the competitive market model are well articulated and well labelled. Its 'liberal' or 'libertarian' value system emphasises the importance of individual responsibility and freedom of choice. It is the prevailing value system in many aspects of life in a democratic society and there is commonly a presumption that, in the absence of some compelling argument, liberty and choice should be maximised. In economic theory this objective is equivalent to the goal of maximising 'utility' and the Welfarist theory of 'Social Welfare'. Even those expounding liberal values, however, generally believe that some constraint upon choice, in the form of compulsory taxation, is justifiable to finance a limited number of public goods and that at least basic medical services should be provided for the medically and financially indigent. With this 'world view' fairness generally equates with a vertical redistribution of income to help the most needy. In contrast, the value system underpinning the public model is less clearly articulated (at least in Anglo Saxon countries). The financing and provision of services to the entire population is often characterised as 'middle (and upper) class welfare' and contrasted with the less intrusive 'safety net welfare' which is all that is required to help those who cannot help themselves. This interpretation of egalitarianism does not, however, correctly represent the values which underpin the public health insurance system. These are nicely described in a report of a commission of enquiry into Canadian Medicare as follows: 'Canadian Medicare is far more than just an administrative mechanism for paying medical bills. It is widely regarded as an important symbol of community, a concrete representation of mutual support and concern... it expresses the fundamental equality of Canadian citizens in the face of death and disease... as the Premier of Ottawa pointed out... "There is no social program that we have that more defines Canadianism"' [ 1 ]. The social value or world view embodied in this quotation does not correspond with the simple notion of assistance for the indigent. Rather it corresponds with a desire to 'remove health and health care from the economic reward system' in the same way as all citizens are, in principle, given equal protection by the law. A close analogy is the desire to have public parks which may be accessed by all members of the community without payment. The objective is not a redistribution of income or the provision of a safety net. Rather, with this view, access to public parks is one of the consequences of belonging to the community; it is a shared benefit and, as such, engenders a feeling of sharing, participation and belonging. The concept is closely linked to the notion of 'social capital' which 'accumulates' with an increase in communal sharing and participation. Pay parks are possible, but a fully informed community might reject this option. Its citizens may wish to live in a community where the Arts flourish, where its sportsmen and women are a source of national pride, where parks are free for all citizens, where an acceptable standard of living is guaranteed after retirement and where all citizens have access to the same range of medical services. In European countries the term 'solidarity' is used to describe this value system. Unfortunately, in English, there is no commonly used and understood word for the concept. ('Communitarianism' is the closest translation.) The consequence of this is a degree of confusion in the expression of social values as both sides of the debate attempt to appropriate the word 'equity' to support their world view. It is clearly desirable that the debate should not be derailed by linguistic ambiguities. Conclusion 2: Health policy should be informed by a careful evaluation of the social values held by different groups of the community with respect to different elements of the health system. While it is ultimately the responsibility of the government to decide which of these values should be embodied in policy, it is desirable for the government's decision to be informed by evidence concerning the community's values and the strength of preferences for different values systems by different groups in the community. In his influential book 'The Power of Public Ideas' Robert Reich emphasises the importance of 'discovery', or the process of getting the public to articulate these values [ 2 ]. To date, neither this nor research into health related social values has been carried out satisfactorily in Australia. Conclusion 3: The choice between public and private funding of health services depends upon social values and, in particular, the strength of liberal-libertarian versus solidarity-communitarian values as they apply to the health sector. Returning to the first defining characteristic of the economic framework, the emphasis upon choice conflicts with two commonly held technocratic beliefs about the inevitability of particular 'problems'. First, the common perception that the country cannot, or shortly will not be able to afford health services is unambiguously false, at least in the foreseeable future. In the USA, per capita expenditures are about double the Australian level and the US Health Care Financing Agency (HCFA) has projected a doubling of these expenditures relative to GDP by 2030. This is technically possible. The relevant question is whether or not we obtain commensurate benefits from these expenditures and if, as a society, we chose these benefits in preference to the benefits foregone. Taking an extreme example, if Australians could spend 25 percent of GDP upon health services this option would probably be embraced enthusiastically if it resulted in an illness free life expectancy of 120. Optimal expenditures are entirely a function of the benefits we obtain and are not driven by technological imperatives. Conclusion 4: There is no immediate limit to the optimal level of health expenditures. It is technically possible to increase present expenditures very significantly. The optimal level depends upon the costs and benefits of the various health services. Increased health expenditures should be enthusiastically embraced if they improve health and health related objectives sufficiently. Technocratic arguments asserting the economic impossibility of increased health spending or increased public funding are unambiguously wrong. At best they are based upon unstated political/ideological assumptions and not economic arguments. A similar argument applies to the second non-problem – the impossibility of funding health services through public taxation. Arguments of the form 'the Government can't afford to pay' are also unambiguously false. The country which can afford to finance health expenditure from private health insurance can also afford to pay an equivalent amount through taxation and some have argued that PHI is, itself, a form of privatised tax. The government share of the health bill is smaller in Australia than in most developed countries. Likewise, taxation is relatively low. This implies that Australia could significantly raise its level of public funding without exceeding the tax burden which is presently experienced in most comparable countries. More fundamentally, however, the form of financing for health services is flexible and is again a matter of social choice. It is likely that this choice will be influenced by the relative costs and benefits arising from the choice, but in the health sector the known costs and benefits associated with public and private health care are not compelling. Privately funded health care is often a little more expensive, but countries with a strong preference for liberal-libertarian values might sensibly opt for a relatively larger private scheme even if it is more expensive. The principle of paying more for what is wanted should not be controversial! Conclusion 5: The balance between public and private sources of revenue for a health service should be determined premanually by the social philosophy of the country. There are no compelling technical or economic constraints on the freedom of sound choice. 3 Recent policy issues Public debate has recently focused upon three 'problems', namely the high and rising cost of pharmaceuticals, the declining rate of bulk billing by General Practitioners (and particularly amongst health care card holders) and the declining number of people purchasing private health insurance. The comments below do not purport to be an exhaustive analysis of these subjects but are included to contrast the subject matter of the policy debate with the more substantive problems which will be discussed in the following section. Pharmaceuticals and PHI are discussed more fully in Richardson and Segal [ 3 ]. Pharmaceuticals Pharmaceuticals are included in the Pharmaceutical Benefits Scheme (PBS) after a detailed review of their effectiveness and cost effectiveness (see Salkeld et al 1998 for a description [ 4 ]). This process does not, by itself, reduce expenditures. Rather, it ensures that drugs whose effectiveness is low in relation to their cost will not be adopted. Expenditures will be lower if the manufacturers of relatively cost ineffective drugs reduce their prices to increase the likelihood of their inclusion in the PBS. However, this may be offset (more or less) if drug companies increase the price of new highly cost effective drugs as they know that the PBAC is aware of their cost effectiveness. Cost effective drugs may also be overused if doctors prescribe them for purposes or at thresholds not tested before their introduction. Partly for this reason the PBAC has sometimes negotiated a price-volume trade-off – if drug use exceeds the initial expectation then the agreed price is lowered. These measures have not contained pharmaceutical costs. In view of rising drug prices in other countries it is likely that this is primarily or entirely attributable to the high cost of new generation drugs and is not attributable to policy. Nevertheless, expenditures have risen and in recent years copayments have been progressively increased in order to reduce government expenditures through the PBS. However it is not clear that the use of copayments, and particularly in a single sector, will have an overall beneficial effect. First, as demonstrated by the Rand Experiment copayments reduce demand somewhat (elasticities are low but not insignificant) [ 5 ]. There is a disproportionate effect upon low income households [ 6 ] which, in this case, implies low income non health care card holders. There is little evidence that lower income patients will discriminate between effective and less effective drugs and at least one study suggests that, perversely, the greatest impact will be upon life saving drugs which have a relatively small immediate impact upon symptoms [ 4 , 7 ]. Secondly, relatively larger copayments in one sub-sector violates a fundamental principle for achieving allocative efficiency; viz, a 'level (financial) playing field' between alternative products/interventions. Violation of this condition increases the likelihood that less cost effective services may be used because of the distorted price signals. Allocative efficiency depends upon relative prices rather than the existence of copayments per se. More specifically, high pharmaceutical prices at the point of service will encourage the use of potentially more expensive interventions including hospital services if a copayment results in the deterioration of a patient's health. Thirdly, the effects of copayments on national health expenditures will be relatively small. Wealthy individuals will simply pay the copayment and health care card holders will be largely shielded from them. The most important effect of the copayment is that it will shift costs from the taxpayer to the patient; that is, its major effect will be upon the distribution of income with the healthy-wealthy taxpayer gaining at the expense of the less healthy-less wealthy. In 1960, pharmaceutical costs represented an estimated 22.3 percent of health expenditure. By 2002 they were 13.5 percent of the total. The comparison indicates that the composition of expenditures may vary significantly through time and, taken out of context, focus upon a single sub-sector may be misleading. Cost control requires a full system perspective and there is no optimal level of pharmaceutical expenditures which is independent of the cost of alternative services. The problem is, in part, attributable to the program structure of delivery and financing which treats pharmaceuticals as an independent program as distinct from an input into an overall treatment regime. Conclusion 6: Reliance on copayments for the control of pharmaceutical expenditures is probably an inappropriate policy as it violates an important principle for achieving allocative efficiency. A more global approach to health policy analysis and reform is needed. Bulk billing Since the introduction of Medicare the percentage of GP attendances bulk billed rose from 52.5 percent in 1984/85 to a maximum of 80.6 percent in 1996/97, then fell to a low of 66.5 percent in December 2004. The measures discussed below appear to have arrested this trend and by March 2004 bulk billing had risen to 68.7 percent of GP attendances [ 8 ]. The chief problem the reforms were designed to overcome was the possibility that the falling level of bulk billing may have jeopardised access to health services by pensioner/health care card holders and other low income households. Importantly, data did not exist to demonstrate that this was, indeed, a problem and that bulk billing or service use by this group had declined by the average or near average percentage. That is, the existence and quantitative significance of the 'problem' were not documented. The recent 'Medicare Plus' package introduced a potentially important structural change. Rebates for bulk billed/health care card holders were separated from the rebate to the general public thereby allowing the separate treatment of the two groups. To encourage the bulk billing of the first group the benefit was increased and bulk billing doctors were permitted, for the first time, to direct-bill the Health Insurance Commission for general patients while simultaneously charging an 'over the counter' extra payment, a measure widely perceived as encouraging an increase in fees as patients will commonly equate the relatively small copayment with the total fee. These measures have two important effects. First, the probable increase in general fees allows the benefit payable for GP bulk billed pensioners/health care card holders to be at a lower level than would occur than without the effective cross subsidy from increased general fees. Secondly, the government can preserve 'equity' (ie bulk billing the needy) by adjusting only the pensioner/health care card holder benefit, while simultaneously allowing a deterioration of the general rebate. Cost shifting would have been further facilitated by an earlier agreement to allow private health insurers to reimburse medical expenses above the schedule fee as part of an agreement with doctors to remove out-of-pocket costs. But this proposal was rejected by the Senate. In sum, there is now a structure which facilitates cost shifting from the public to the private sector. There are two important caveats to this conclusion. First, a rebate structure which facilities additional cost shifting does not imply that government will use this option in the short or long run. Secondly, the effect of additional cost shifting will be mitigated by an additional element of the Medicare Plus package, namely, the introduction of a 'safety net' which reimburses 80 percent of out-of-pocket, out-of-hospital costs including billing above the schedule fee once family expenses reach $1,000 (or $500 for a health care card holder's family). The long run effect of this latter cost off-set is difficult to assess. As reported in the announcement of Medicare Plus the expected cost of the safety net in its first three years would average $89 million p.a. as compared with likely out of pocket medical expenses of about $1,400 million. in 2004/05 (extrapolated AIHW) [ 9 ]. Even allowing for a probable cost overrun the amount is not large. Nevertheless, the 'safety net' offers additional protection to those who would be most affected by additional cost shifting; that is, this policy also helps create a structure where equity, defined in terms of need, may be reconciled with increased cost shifting. However, the most significant feature of these changes to the reimbursement formula for medical expenses is that they deal with relatively 'small order' issues. Under Medicare, pensioners/health care card holders with significant health problems have access to emergency and outpatient facilities at public hospitals. Poor access to these will increase inconvenience and queuing and, in an unknown number of cases, result in poorer health. But this more serious outcome is likely to be relatively infrequent as the majority of this group have always been bulk billed while others will have satisfactory access to hospital based care. Conclusion 7: Proposed and actual policies towards pharmaceutical expenditures and bulk billing have had a common element. Both represent immediate or potential cost shifting from the government to the public and any net reduction in societal expenditure because of a reduction in patient initiated services will have potentially adverse effects upon the health of the 'near poor'. However, the financial effects of the policies are relatively small and the indirect adverse effect upon health is likely to be correspondingly small. Private health insurance (PHI) The role and regulation of PHI after the introduction of compulsory national health insurance has been ambiguous and anomalous. Since Medicare was introduced as the vehicle for achieving fairness in the financing and delivery of health care it has never been clearly stated why PHI should be closely regulated to achieve fairness in he sub-group of the population which elects to purchase PHI and why it is regulated in a way which inhibits effective competition. But these are the effects of Australia's community rating and reinsurance requirements respectively. In the last one and a half decades there has been an ongoing concern that declining levels of private health insurance have had an adverse effect upon the public hospital system and that 'PHI reform' has been needed to solve this problem. The argument is summarised in the following (constructed) quotation: 'Private health insurance was caught up in a downward spiral caused by the adverse selection identified in the Productivity Commission report [ 10 ]. 'Rising premiums triggered by increasing costs induced low risk members to withdraw. Without their cross subsidy of high cost members, premiums were forced to rise again, which further increased premiums, which induced further departures, etc. " PHI is primarily purchased to cover the costs of private hospitalisation. Consequently, as PHI declined through time, fewer patients have been able to afford the cost of private hospitals, and this has created an excess demand for public hospital beds. The increasing length of queues is a confirmation of this problem ". While plausible, the account of history in the second half of this argument is wrong. Between 1985/86 and 1999/00 private hospitals increased their share of admissions by 32.4 percent from 25.9 to 34.3 percent of the total. The share of bed days rose from 21.9 to 28.1 percent of the total or by 28.3 percent [ 3 ]. These simple and readily available statistics unambiguously contradict the conventional wisdom propagated by the media and some politicians. It is true that queues in public hospitals have increased, but this is attributable to the increasingly draconian budget caps upon public hospitals throughout the 1990s. That is, queuing in the public sector has been primarily a result of supply side and not demand side factors. (Excess demand for beds cannot explain the closure of wards that occurred in many public hospitals!) The simplicity of the statistics contradicting the conventional wisdom calls into question the analytical capacity of many media commentators (and some independent commentators)! Conclusion 8: Media analysis of the relationship between PHI and queuing in the public system highlights an important system failure; viz, the failure of the media to exercise rudimentary critical skills in their analysis of PHI and hospital queuing. Even if PHI membership had fallen sufficiently to have reduced the share of private hospital admissions by almost a third, the balance of public and private hospital separations in 1999/00 would have been similar to the share in 1985/86. This suggests that it was the events summarised in the first half of the quotation which were of concern at the government and departmental level and that PHI membership and the public-private mix of health financing and delivery were explicitly or implicitly the real target of health policy. In response to the 'problem' of falling PHI and the demand side problem for public hospitals (which did not exist!), the government introduced three enduring policy changes. In July 1997 individuals with an income above $50,000 and families with a combined income above $100,000 who did not purchase PHI became liable for a 1 percent tax surcharge upon their incomes. In December 1998 a 30 percent rebate on PHI premiums was introduced. (In August 2004 a selective increase in the rebate was foreshadowed); and in September 1999 lifetime community rating was enacted which has the effect of reducing future premiums for those who have held PHI from the age of 30. Discounts apply at any age between 30 and 64; 30 is the age of maximum discount. Failure to enrol by this age will result in a higher premium if PHI is subsequently purchased. The evidence demonstrates that these policies, and particularly the last policy, have succeeded in increasing PHI membership. However it has created an industry which, along with the platypus and echidna could be Australia's entrants into the world's strange but true contest. Australia would almost certainly win – even without our egg laying marsupials. First, the surcharge results in a negative price for the wealthy. Individuals and families above the income threshold avoid an increasingly large tax payment as their income rises; that is, if they purchase PHI they will have a greater income at the end of the year than the individuals and families above the threshold who do not buy PHI. This is analogous to supporting the automobile industry by placing a surcharge on wealthy families who fail to buy an Australian car. It would be difficult to find any other support scheme which uses the income tax system to coerce the purchase of a particular product. It would be equally difficult to find a produce where the price is negative. Secondly, (and predating the recent legislation) the use of PHI to cover hospital bills generally results in a greater, not smaller, out of pocket expense. Public hospitalisation is free. Those with PHI commonly pay a copayment. (There is a perverse equity in these two anomalous outcomes. The wealthy are paid to take PHI but financially penalised if they use it!) Third, lifetime community rating also involves a leap of faith. The 30 year old must believe that public policy will remain stable for 50 years – a belief that should have been challenged when both parties announced changes in the 2004 election campaign. However lifetime community rating also has a bizarre dimension. Insurance is generally purchased to reduce risk and uncertainty. In the health sector these arise because of the risk and uncertainty of ill health and the cost of medical care. Prior to lifetime community rating this uncertainty depended upon possible events in the following 1–3 years. After the change it depended upon the next 1–30 years. Events in 30 years are more uncertain than events in 3 years and, consequently, the legislation increased risk and uncertainty. The perceived risk was amplified by the publicly financed marketing of private insurance. Predictably, people responded to the increased uncertainty by buying more insurance. Thus, to encourage the uptake of insurance the government increased the very thing insurance is designed to reduce, viz, risk and uncertainty. A final anomaly which predates the recent PHI reforms is that private health insurance in Australia leaves the consumer with residual risk. Particularly for ancillary services, most benefits are capped and out of pocket expenditures rise with the price of the service. This pattern of benefits is exactly the opposite of the structure of benefits which will maximise the patient's 'expected utility' [ 11 ]. Taken together, the changes introduced since July 1997 have created an extraordinarily complex and perverse set of incentives. The ethical basis of the free market and the liberal-libertarian model is that choices should be determined by individual's preferences in relation to real (opportunity) costs. The surcharge subverts this process and coerces choice and the strength of the coercion is based upon economic class. There is no justification for this in the economic theory of the efficient market. Despite this conclusion, it is a legitimate function of government to determine the balance between public and private delivery financing and the distribution of health care costs. As described earlier the preference for private sector funding and provision ( albeit in the context of compulsory core insurance) is consistent with a legitimate and defensible world view, viz, the liberal-libertarian belief that in a free society individuals should be encouraged to take responsibility for their own lives. In particular, the 30 percent subsidy is the orthodox approach to encouraging an industry which has a special claim for protection and, private health insurance does not compete on a 'level playing field': it competes with a public sector which is free at the point of service. There are, however, two important caveats. First, and as noted above, the measures taken have destroyed any nexus between potential benefits and the price paid by wealthy individuals. Secondly, the 'product' is unlike usual insurance where the benefit – a payment after an adverse event – does not impinge upon other individuals. PHI is purchased to avoid queues and to select the best possible doctor. With fixed capacity, the avoidance of queues by one person imposes a longer queue on another person – there is a 'negative externality'. Selection of the best doctor reduces the access to these doctors by public patients. Consequently, the important debate should be about the 'right' to purchase preferential care at the expense of those who do not have PHI . In a liberal democracy there is a presumption that individuals may spend their own income as they wish. For the individual there is probably no more important context for exercising this 'right' than in the context of preserving life and its quality. There is, therefore, a head-on-head conflict between the liberal-libertarian 'right' of the individual to spend his or her income on health care and the communitarian-solidarity based 'right' of each individual in a community to have equal access to high quality medical care. The latter goal must necessarily be achieved by imposing some constraint upon income-based preferential care to a particular group in the community. In Australia the constraint has been a de facto financial penalty for seeking priority care. Until the advent of engineering of negative prices for the wealthy, the purchase and use of PHI resulted in 'double payment' for some services, first via taxation and the Medicare Levy and secondly by premium paid for PHI. This 'penalty' still exists for lower income individuals and households. Conclusion 9: The public debate over Private Health Insurance has been misleading. The contentious issues do not only concern the most effective way of ensuring access to health care (with the erroneous presumption that public monies not spent on health services represents wasted resources). Rather, the contentious issues include the 'right' or otherwise of the individual to spend their own income on whatever they wish without coercive financial penalties and the consequences for the remainder of the community when one group jumps the queue and has priority access to the most experienced doctors. Distributional effects of recent policies PHI policy has been consistent with the other policies discussed above in one important respect. The policies are likely to have a relatively small effect upon the delivery of health services. Rather, they are about financing care and the public-private balance in the health system. The balance, in turn, affects the distribution of health care costs between different households. Copayments shift costs from the government to patients. As government payments are met by progressive taxation, copayments redistribute the cost of health care from wealthy-healthy taxpayers to the less healthy, less wealthy. Additionally, as copayments have a disproportionate effect upon the use of services by low income households, the proportion of the government subsidy returned to high income households rises as copayments rise. The redistributive effects of PHI are more complex. Low income households which purchase PHI unambiguously pay more for hospital and health services. Their taxes and the Medicare surcharge are unchanged and the purchase of PHI leaves them out of pocket. For higher income households, which are liable for the PHI surcharge, the effects are conceptually ambiguous as they depend upon the assumption made about the surcharge in a counterfactual world in which PHI did not result in a surcharge exemption. The surcharge was created specifically to permit an exemption for those who purchased PHI and the removal of the exemption might therefore be accompanied by the removal of the surcharge. Finally, and as noted earlier, the proposed changes to bulk billing represent a structural change which will facilitate the transfer of medical insurance costs from the public to the private sector. Conclusion 10: Recent and foreshadowed legislative policy initiatives with respect to bulk billing, copayments and PHI, all concern the financing of health services. Recent policy has been introduced with measures to mitigate the impact upon the most needy and, in the case of medical rebates, in a way which accommodates the popularity of bulk billing. Nevertheless, a common feature is that each of the proposed or implemented policies assists with the creation of a structure which facilitates the transfer of expenditure from the public to the private sector. If this occurs, it will reduce the cross subsidy from healthy wealthy to unhealthy less wealthy households. This effect occurs immediately with copayments. The transfers are more complex in the case of PHI. 4 Five major problem areas The five issues below have two common elements. First, they are directly concerned with health services and not the distribution of costs and incomes. Second, they have been almost ignored in the public policy debate despite being problem areas where there are quantitatively large inefficiencies and a corresponding potential to increase health outcomes and/or increase the overall cost effectiveness of the system. Efficacy, effectiveness and social objectives In common with all other health systems many, and probably most, of the services provided in Australia have not been adequately evaluated and there is no ongoing process for the elimination of cost ineffective services. In 1987 Chassin et al [ 12 ] estimated that between one third and one half of coronary services in his study were 'inappropriate' in the sense that they had no beneficial, or a detrimental affect upon health. An additional one third to one half of the procedures considered had equivocal benefits. Likewise, Brook [ 13 ] estimated that 51 percent of angiography and 42 percent of coronary artery bypass graft procedures were unnecessary. Other studies by the US Health Care Financing Agency and the OECD have likewise concluded that only a small number of services have been evaluated for efficacy [ 14 ]. In Australia Segal [ 15 ] demonstrated that in the context of diabetes the cost of obtaining a life year varied from $70,000 (drug therapy) to $2,400 for behavioural programs. Comprehensive diabetes care was estimated to have a negative cost per life year; ie the program saved life and saved cost. With respect to this issue, the Australian record is comparatively good. It has led the world with the introduction of mandatory economic evaluations for the drugs and services to be subsidised through the PBS and the Medical Benefits Scheme respectively. However the failure of other countries does not indicate that Australian procedures are satisfactory. The overwhelming majority of the services which were accepted before the introduction of mandatory economic evaluations have not been assessed and, with the passage of time, there is a need for the reassessment of services and drugs. A related problem is that drugs or services which are efficacious when used appropriately in a random control trial may be used inefficiently with inappropriate patients who do not have the clinical indications of the patients in the trial. Gold standard medical care requires that cost effective therapy, drawing upon evidence-based medicine, should be the norm and not the exception in medical practice. Cost effectiveness should be determined by a broad ranging comparison of options form across the full range of potential interventions; that is, comparator interventions should take account of substitute services from across the entire health sector. This does not presently occur. Guidelines for the PBS require a comparison between drugs and exclude comparison with lifestyle/behaviour change programs. Conclusion 11: The scale of present evaluation activities is inadequate. In an industry absorbing 9 percent of the GDP – the country's largest industry – there should be ongoing and large scale evaluation and re-evaluation of the cost effectiveness of the services provided. Evaluation should be based upon a comparison with the full spectrum of substitute services. A failure to do this almost certainly ensures that there will be widespread and significant allocative inefficiency in the level and mix of services. Implementation of evidence based practice will inevitably be resisted by service providers whose practice and income may be adversely affected. Consequently, provider education for a 'culture change' is likely to be a slow and ineffective vehicle for change. In contrast, once the desired practice pattern has been determined, financial incentives to encourage this form of practice may be implemented relatively quickly and do not involve direct coercion. The economics literature is replete with examples where such incentives have significantly altered provider behaviour [ 16 - 19 ]. Australian examples include the use of GP incentives to bulk bill; to undertake rural practice and to increase child immunisation. DRG based financial incentives were spectacularly successful in increasing hospital throughput in Victoria in the early 1990s [ 20 ]. Conclusion 12: Reimbursement formula for service provision should include financial incentives at all levels of service delivery for the use of the most cost effective therapies. The more general principle which should, but does not, permeate the market for health services is that the willingness to pay for services should vary with social objectives, a principle which is successfully incorporated in simple competitive markets when social and individual objectives are identical. As one example where this principle has recently and successfully been used, if society wishes to encourage bulk billing (which clearly benefits patients but lessens provider control over their incomes) then the benefit for services that are bulk billed should be increased relative to the rebate for other services and the differential increased until the target of bulk billing is achieved. Likewise, if society wishes providers to adopt evidence based medicine, hospitals to install clinical pathways, some procedures to be encouraged and others discouraged, then payment for the desired service or process should be increased relative to the reward when these services or activities do not occur. At the level of the individual service, recent research has demonstrated health-related social objectives which are often broader than the minimisation of morbidity and mortality, for example, preferential treatment for particular age groups, particular classes of patients and health benefits [ 21 , 22 ]. Presently these additional 'social preferences' are ignored. This is unsurprising as the research to identify and measure them is still underway. However, where these preferences are significant and consistent, payments should eventually be adjusted to ensure that they are achieved. Conclusion 13: The payment of service providers should incorporate the principle that society should pay for what it wishes to have in accordance with its social objectives, rather than what it is given. This implies the need for ongoing enquiry into health related social objectives (the numerous objectives loosely grouped under the heading of 'equity'). Practice variations In 1982 John Deeble and I used data from the first full year of compulsory health insurance, (Medibank), to determine the level of service use in each of Australia's statistical divisions [ 23 ]. An example of the results is shown in Table 2 . Huge variations in service use were detected even between the relatively large geographical units used in the study. Within these units small area variation would have further increased the discrepancies. These differences do not appear to have diminished with time. Thus, for example, Robertson and Richardson [ 24 ] found startling variation in the use of well-defined hospital procedures even after standardisation for age, sex and population. In this study data were collected for a two year period for each of Victoria's statistical sub-divisions. Procedural rates were expressed as a percentage of the rates which would be expected from the State average service use per age-sex cohort and from the demographic characteristic of the Statistical sub-division (SSD). Results are summarised in Figure 1 . The bar, lines and circles give an indication of the frequency distribution of the procedure utilisation rates (ie the 25 and 75th percentiles (bars), two standard deviations (lines) and outlying SSDs). Table 2 Practice Variations 1976 Consultations per capita Statistical Division GP/(GP) Q(Spec) Total Sydney 5.1 2.3 7.4 Tasmania 3.1 1.3 4.4 Darwin 1.1 0.5 1.6 Sydney/Darwin 4.6 4.6 4.6 Figure 1 Standardised Rate Ratios for Various Operations in the Statistical Local Areas in Victoria, Compared to the Rate Ratios for All Victoria, Source: Richardson 1999 p198. The results reveal an 8 to 10 fold variation in service use. Part of this is attributable to the random variation that would be expected because of the uncertainty of the episodes of ill health. Using state-wide data the ratio of actual to expected variance was calculated. This is reported in the column of numbers to the left of the bar diagram. For the first procedure, coronary angiography, the observed variance was 13.4 times greater than expected; ie actual variance was 1,440 percent of the expectation. This extraordinary result is reproduced for all of the procedures examined. A second example of uneven service delivery was also published by the same authors. This related to the likelihood of obtaining a high tech procedure – angiography or revascularisation – after a heart attack. Results shown in Table 3 indicate that in the 14 days following the heart attack, men and women admitted to a private hospital were 2.2 and 2.27 times more likely to receive angiography than their counterparts at a public hospital. They were 3.43 and 3.86 times more likely, respectively, to undergo revascularisation (coronary artery bypass surgery angioplasty, stent). These discrepancies did not diminish significantly in the following 12 months. The same study identified statistically significant differences in the likelihood of a procedure between men and women, young and old, urban and rural populations. Table 3 Ratio: (Likelihood of procedure after admission to a private hospital) ÷ (Likelihood of procedure as a public patient) Angiography Revascularisation Within 14 days Men 2.20 3.43 Women 2.27 3.86 Within 12 months Men 2.16 2.89 Women 2.22 2.84 Taken together these studies suggest that there is a highly erratic pattern of service delivery across Australia and between social groups. One of three conclusions is inevitable. Some groups are under-serviced; some groups are over-serviced or both of these problems occur to different sub-groups of the population. This indicates both allocative inefficiency (more health could be obtained with a redistribution of existing resources) and significant inequity for those with poor access to health services or a different form of inequity for those persuaded to undergo procedures where risks exceed likely medical benefits. Government has shown almost no interest in this type of result. There has been recognition of an 'urban-rural' discrepancy but, 20 years after the demonstration of a more complex pattern than a simple urban-rural dichotomy, remarkably little has been achieved. Conclusion 14: Significant variation in the use of services has been allowed to continue more than two decades after it was identified. Small area variation across Australia almost certainly reflects significant allocative inefficiency and an inequitable access to health services. There appears to have been relatively little interest in this problem. Lack of coordination The health system at present consists of a series of financially independent programs ('silos'), which are poorly linked to other health services or programs. There appears to be near universal agreement that a sensible health scheme cannot be built upon the current Federal-State division of financing, responsibilities and powers or upon the existing Commonwealth program structure. In the context of a recent review of hospital care, the Tasmanian Government accepted a recommendation for the pooling of all health and aged care resources [ 25 ]. However implementation of the recommendation requires Commonwealth support which may or may not be forthcoming. The fact that a simple solution exists for this most elementary problem but, to date, has not been seriously addressed, represents a fundamental failure of government in Australia. Two case studies are used below to illustrate what gold standard allocative efficiency would imply for the use and coordination of services. They indicate the distance that health services reform must travel in Australia before we have gold standard delivery. The first case study is real. A Seattle-based 'pure' Managed Care company, Ethix, was asked to establish a health scheme for a small town close to Seattle. Routine surveillance of the medical claims over the first two years of the new scheme highlighted an anomaly. There were excessively large numbers of youths receiving surgery for spinal injuries. Further investigation found that the problem was attributable to a toboggan run on the outskirts of the town which had a tree stump half way down the slope. Youths were crashing into the stump and damaging their spine. The health scheme paid for a bulldozer to remove the stump [ 26 ]. Medicare does not pay for bulldozer services. However, in the circumstances described here it should do so. More generally, the vignette illustrates two of the characteristics of gold standard delivery, namely, routine data surveillance to locate problems with any aspect of social or medical behaviour which might be modified to improve health and, secondly, the flexibility of funding which is needed to adopt the most cost effective solution to the identified problems. In contrast, in the Australian health scheme a problem of this sort would not be detected by the health system. It is possible that a similar type of accident could fall under the jurisdiction of an occupational health and safety or traffic accident authority. Otherwise there would be neither the will nor the means to respond. If such a problem was eventually identified the typical response would be accusation, blame shifting and, possibly, litigation. The cause of this problem is related to the practice of Management by Objectives, an aspect of managerialism, which encourages local rather than system-wide thinking. Interestingly, Peter Drucker, one of the early advocates of MBO, in recent years has publishes warnings about its over-use and limitations. Conclusion 15: A key challenge is to establish a single payer (for each person) with flexibility and incentives to purchase the most cost effective services. Services should not be determined by historical program boundaries and rigid budgets. There has been a serious failure by government to address this fundamental issue. The achievement of this relatively simple reform is necessary (but not sufficient) for the achievement of a range of system reforms. The second case study is a hypothetical scenario constructed by Duckett to illustrate gold standard system coordination and processes [ 27 ]. 'A woman with dizziness is concerned about her health. She rings the State call centre which advises her to visit her local health team . She is able to see the GP quickly who asks her a series of questions from the relevant research based protocol and undertakes a clinical examination. The GP emails the results to a local specialist... who orders some further investigations consistent with the state research based care path ... Advice of (an) impending admission is automatically conveyed electronically to the GP and the social worker in the referring health team . The social worker contacts the hospital to discuss discharge planning ... The specialist... suggest(s) a number of sources for information about the patient's condition . The patient contacts the call centre for further information ... The case is randomly selected by the hospital audit committee for quality review. The committee suggests some slight changes to the state-wide protocol committee .' p204. The key elements of this scenario relate to information access and transferral. It illustrates the role of evidence based medicine, routine service review, the adaptation of protocols, universal electronic transfer of all information and the absence of incentives to depart from best practice. Parts of this scenario correspond with Australian practice. But the events in italics would be unusual. It appears to be serendipitous whether a particular problem of a particular person and in a particular part of Australia results in a response which even partially mirrors the gold standard response in this scenario. Conclusion 16: Commonwealth and State authorities should mandate practices which improve the coordination of services. To facilitate this there should be universal use of electronic data systems for patient notes and information transfer. Evidence based protocols and clinical paths should be adopted when available and relevant. Feedback and error learning should be a routine part of the system. All of these desirable features are presently difficult because of the fragmentation of the system which is, in part, attributable to the failure to establish a single funder for all health services received by an individual. Quality of care Results from the 1995 'Quality in Australian Health Care Study' (QAHCS) suggest that the quality of health care in Australia is a problem which overshadows all others. In the initial study, reported by Wilson et al [ 28 ] medical records for more than 40,000 admissions to 28 hospitals in NSW and SA were individually examined to determine whether or not an adverse event (AE) was associated with the admission (prior to or during the episode of hospitalisation). A judgement was made concerning the consequences of the AE and whether or not it might have been avoided. By extrapolating results the authors estimated that about 470,000 admissions were associated annually with an AE and that these would have resulted in 18,000 deaths and 50,000 cases of permanent disability. In a subsequent report Runciman et al [ 29 ] estimated that in 50 percent of the AEs in the QAHCS had a high preventability score. Sixty percent of deaths could have been avoided. In this latter study, incidence and not prevalence scores were reported as part of the effort to standardise the methodology with an earlier Harvard Medical Practice Study (HMPS) reported by Brennen et al [ 30 ]. This reduced the annual rate of AEs to 10.6 percent of admissions. From the response to these events (discussed below) it appears likely that many have been unable to appreciate the scale of the carnage implied by the report. If the results from the original QAHCS are not discounted then medical errors have been responsible for the death of more Australians per annum than the average annual death rate of Australian soldiers in World War 1 (15,800). Permanent disabilities per annum approximate the annual rate of casualties in The Great War (62,500). Unlike The Great War, the problem of AEs has probably been ongoing for 50 years or more. In Table 4 below the death rate from AEs is compared with mortality rates from other causes which are of particular social concern. To be conservative and to take account of undetected bias, the AE rate reported in a 1995 study is reduced by 50 percent. Preventable deaths are assumed to be 50, not 60, percent of deaths associated with an AE. The resulting, conservative number of deaths from AEs in 1999 were about 40 percent higher than the number of deaths from AIDS, suicide, motor vehicle accident, accidental falls, homicide, drowning and poisoning combined. In Table 5 some equivalent events are listed. The conservative estimate of the unnecessary death rate is about the same as would occur if the Bali bombing occurred every week of the year, year after year. Table 4 Perspective: Selected Causes of death, 1999 Cause No of deaths AIDS 122 Suicide (intentional self harm) 2,492 Motor vehicle accidents 1,741 Accidental falls 520 Homicide 300 Accidental drowning/submersion 278 Poisoning by drugs/medications 1,015 Subtotal 6,468 All deaths from adverse events (1) 9,000 All preventable adverse events (2) 4,500 (1) 50 percent of reported estimate (2) assuming 50 not 60 percent are preventable Table 5 Events equivalent to avoidable AE deaths* Cause 1 in 10 customers in restaurants poisoned each year: annual deaths 4,500 13 Jumbo jets crash each year, each with 350 Australian passengers killed 45 Bali bombing type attacks, each with 100 Australians killed each year 'September 11' every 8 months: only Australians die *assuming preventable deaths = 25% QAHCS (1995) While those affected by adverse events will, on average, be older and frailer than those who died in Bali or during World War 1, the magnitude of the problem is still staggering. Considering the reaction to the Bali bombing it might have been expected that the publication of the QAHCS would have caused a seismic shock, with the public demanding immediate, comprehensive reform and government passing urgent legislation to mandate any achievable system reform which ameliorated the problem. If the system failures which preceded the Bali bombing warrant a Parliamentary inquiry, equivalent interest might have been expected with respect to a problem responsible for an equivalent number of deaths every week. An effectively unlimited budget might have been approved for the upgrading of quality and safety. In contrast, the actual reaction to the report must constitute one of the more puzzling episodes of Australia's social history. At all levels the response was sedate, cautious and incremental and there appears to have been greater concern that our health system might be perceived as unsafe than with the fact that it actually is unsafe. The level of activity in the 5 years following publication of QAHCS suggests that the results may not have been truly believed. But no subsequent study was funded to validate or refute the results. The QAHCS results were not ignored. As summarised in Table 6 an advisory body was immediately established which evolved into the Australian Council for Safety and Quality in Health Care (ACSQHC). Its activities are summarised in five successive annual reports. In addition, the Australian Health Care agreement between the Commonwealth and State governments allocated budgets of $680 million and $785 million for quality assurance activities for the periods 1998–2003 and 2003–2008. In each of the States sub-committees and working groups were created which, along with the ACSQHC have resulted in a very large number of reports, publications, some legislation and local initiatives. The ACSQHC alone lists 35 publications [ 31 ]. State activities are summarised in ACSQHC, [ 32 ]. These initiatives have resulted, inter alia , in moves to tighten up hospital accreditation processes; to monitor adverse events more closely; to improve consumer participation in the evaluation of health care; to encourage health professionals to report adverse events; to improve health information technology; to establish practice guidelines, etc. The large number of funded projects are described and listed in ACSQHC, [ 33 ]. Table 6 Response to QAHCS 1995 QAHCS published 1996 Taskforce established on QAHC 1998 Health ministers ask Advisory Group for report 1998 Interim Report 1999 Report of the National Expert Group 'Actions identified by the taskforce... need to be implemented at all levels of the health system ...' 1999 Australian Council for Safety and Quality in Health Care Established Despite the high level of activity, the importance, priority and sense of urgency reflected appear more appropriate for an (important) ongoing reform process in an already well-operating system. There is a yawning gulf between this and what the public would undoubtedly demand if our TV and news services were reporting deaths on the scale of the September 11 New York disaster very two months accompanied by about three times this number of injuries. The philosophy of the reform process appears to be summarised by the ACSQHC when it approvingly quotes Berwick as saying that 'there are no quick fixes. We must re-examine all that we do', p14 [ 31 ]. As a description of history the 'no quick fixes' statement is correct. The 1999 report – 4 years after the publication of the QAHCS recommended that 'actions identified by the taskforce... need to be implemented at all levels of the health system'. By 2001 NSW had passed legislation which, inter alia , created the Institute of Clinical Excellence. By 2002 – 6 years after publication – the Victorian Quality Council Plan had been established. In 2003 the updated version of this plan set as its goals, inter alia , the establishment of a framework, the involvement of consumers and education. Almost 10 years after the report, the ACSQHC called for mandatory participation of all hospitals in a process of assessment [ 31 ] and (with possible irony or ill concealed frustration) the Chairman of the ACSQHC argues in the 2004 Annual Report that 'Action must be taken without untimely delay where culpability is clear', p5 [ 34 ]. Between publication of the QAHCS and the Health Ministers' request for a report in 1995, 13,500 Australian would have died and 37,500 become permanently disabled. By the time the 1999 report was recommending the implementation of various policies at least 18,000 would have died. By the time of the NSW legislation cumulative national deaths would have reached 27,000. The Victorian Strategic Plan to develop a framework was published after the death of at least 31,500. It is scarcely surprising that in 1999 an editorial in the Medical Journal of Australia commented that: 'Welcome though (various initiatives) are, the pace of change nevertheless seems slow given the stark message of the original QAHCS study four years ago... 50,000 Australians suffer permanent disability and 18,000 die at least in part as a result of their health care', p404–50 [ 35 ]. By 2002 Siddons could still comment that: 'On the 10 th anniversary of the study year, the most striking outcome has been the paucity of reform currently exhibited at the coalface of tertiary health care', p823 [ 36 ]. The inadequacy of the response is surprising as the evidence suggests that a reduction in AEs would be spectacularly cost effective. The interim report from the national expert group estimated the potential savings from preventable adverse events in 1995/96 would be $4.17 billion per annum . Consequently, expenditures of this amount could be justified if they eliminated the unnecessary AEs or if the cost of the achievable 50 percent reduction in AEs was less than $2 billion per annum . In contrast with the view that little could be done quickly there are a number of examples where, prima facie , very significant and effective change could have been/can be rapidly implemented. The chief distinguishing feature of each of the suggestions below is that they involve legislation and regulatory enforcement which appears to be inconsistent with the apparent emphasis upon persuasion and voluntary culture change in many of the present activities. Mandated options include the suggestions below. Accreditation For decades health professionals have believed that a significant number of small hospitals are dangerous. However there has been no decisive action. With full knowledge of the QAHCS results, hospital accreditation remains voluntary in all States except Victoria. There will clearly be self-selection. Low quality hospitals will opt not to seek accreditation and poorly qualified doctors will seek out these hospitals. Universal accreditation could be mandated. Multiple accreditation teams could have the power to randomly inspect hospitals or units within hospitals and close those judged to be dangerous – as occurs with restaurants with sub-standard hygiene. It is unclear whether or not present accreditation is sufficiently rigorous to reduce avoidable adverse events significantly. There appears little reason why the accreditation process should not itself be reviewed to ensure that credentialed hospitals satisfy rigorous safety standards in their facilities and procedures. Doctor Accreditation Patterns of private practice patterns are already subject to scrutiny in Australia. But the chief purpose is to detect medical fraud. Legislation could require the examination of practices to detect those which deviate significantly from evidence based guidelines constructed by the relevant Royal Colleges. When there is a known relationship between the small number of procedures carried out by a doctor and negative outcomes, as occurs with surgery, critical annual procedure rates may be established which trigger the provision of information to the doctor, the mandatory review of the practice and finally the dis-accreditation of the doctor for the conduct of these procedures. While it is true that some doctors take on the hard cases partial standardisation for case complexity is possible and this problem would obviously be taken into account by those conducting the review. The appropriate systems could be established in 1 to 2 years. Mandatory Disclosure and Error Learning It is not compulsory for hospitals or doctors to register AE and routinely provide feedback to facilitate error learning. This means that the most important vehicle for improving quality and reducing patient risk is not compulsory. The opportunity for error leaning is almost certainly under-utilised in a large number of hospitals and probably ignored by the doctors whose performance most needs monitoring. Legislation might ensure the universality of this critical system reform. The adverse events register could and should be linked to doctors and appropriate threshold levels installed which sequentially trigger information feedback, review, and finally dis-accreditation of the doctor. It is likely that the first of these steps will be sufficient to effect satisfactory change. Despite this, 9 years after publication of the QAHCS the chairman of the ASCQHC notes that 'we have insufficient accurate data for fully appreciating the current size of the multiple causes of this problem ... we need the data from multiple sources, including incident monitoring systems, routine administration data sources and the use of screening tools to practically identify areas that may cause harm.' (p5) [ 34 ]. Protection from litigation The published research on 'high reliability organizations' suggests that it is wise to separate inquisitorial from punishment processes, such as dis-accreditation. Adverse events are unlikely to be reported if there is a financial incentive to hide the AE. For this reason legislative protection of doctors from the financial outcome of litigation is probably a prerequisite for a successful and comprehensive system of error learning. The consequences for a doctor associated with AE should be based upon medical criteria and uncoupled from the social mechanism for compensating patients. Legal protection alone is unlikely to improve the quality of the information used for error learning. Rather, it should be part of a package of requirements which includes penalties for the failure to report an AE. Information transfer Patient notes are still transferred within hospitals using 19th Century clipboards. It is known that this commonly causes potentially lethal errors. The mandated use of (long available) electronic forms of transmission could alert staff to the risk of inappropriate procedures, the administration of conflicting drugs or the failure to administer a drug. Likewise X-ray films are often misplaced or lost. The consequences may be lethal. Legislation could ensure the use of digital technology to ensure immediate access of results. New wireless technologies make it possible for roving staff – doctors and other professionals – to have constant access to text and basic technical data. As with electronic note taking this might take 1 to 2 years to fully install in all Australian hospitals, clinics and nursing homes. Hospital systems Hospital systems in Australia are commonly ramshackle or antique. There are no required pathways or mandatory discharge criteria. There are no internal or external financial incentives for the optimal treatment of patients. These changes are more complex but could be installed comfortably in 4–6 years. Conclusion 17: There is no reason why much of the health system should have missed the IT revolution which has transformed other parts of the community. In relation to the size of the AE problem, the cost of implementing late 20 th Century information technology throughout the health system is likely to be small relative to the human and financial cost of AEs averted. Minimum staffing requirements There is no regulation which links on site expertise and the complexity or riskiness of the procedures which may be undertaken in a hospital. For example, it is possible for a hospital to permit significant surgery but have no on-site medical practitioner post-operatively. This potentially lethal practice could be proscribed and minimum staffing ratios implemented within the 1–2 weeks needed to reschedule staff or the location of procedures. It was not until 2003 that the ACSQHC released a paper considering issues of staff rostering, skill mix, staff numbers, staff supervision and team functioning [ 33 ]. While endorsing the AMA (voluntary) code of practice, [ 37 ] it comments – almost 8 years after the QAHCS – that 'responsibility for improving the management of staffing variables cannot (ie should not but still is being) left to individuals. It is a governance responsibility...' pii (words in brackets added)] [ 38 ]. Queuing The airline industry operates a highly efficient computerised system of booking and queuing which may be accessed by travel agents throughout the world. By this standard most hospital queuing systems – if they exist – are rudimentary. It is possible and desirable for different hospitals in the public hospital system to be interconnected to provide patients or their doctors with available times for treatment city, state or nation-wide. Queuing and scheduling can and should be operated using publicly known criteria. In the USA there is a nation-wide system for matching patients with available organs. (There is, interestingly a prioritisation criterion which, for reasons of equity, assigns compatible organs on the basis of need, not prognosis.) Australia has no such system. Conclusion 18: Queuing should be regulated by a nation-wide prioritising system based upon explicit criteria. This would increase efficiency and, for the patient, increase choice and certainty. The private sector would have greater difficulty in operating such a system because of the patient's attachment to a particular doctor. Private health insurance organisations could, however, offer a similar service to patients who are willing to accept treatment from contracted providers. Australia's technologically conservative PHI organisations appear uninterested in these initiatives. Information and system audit The suggestions above and, to date, the majority of the reforms contemplated in reports, represent process measures of success. However, their objective is to reduce adverse errors and for this reason, record analysis of the form conducted by the QAHCS should be an ongoing feature of the system. The QAHCS research was relatively expensive, but these costs are infinitesimal in relation to the importance of the surveillance, the costs, the morbidity and deaths averted. Conclusion 19: A policy of persuasion and culture change is an insufficient response to a problem of the magnitude of the AE epidemic. As a matter of highest priority, legislation should be passed requiring extensive system regulation and reform to reduce the incidence of adverse events. The required changes should be fully funded (or they will not occur) and should include the use of state of the art IT, minimum staffing requirements and system feedback on both individual hospital and individual doctor performance. Evaluation and random audit of all hospitals and health care providers and the reporting of all AEs should be mandatory. Doctors should be given comprehensive protection against the results of successful litigation. The review of doctor performance should be uncoupled from compensation for the patient. Public information There is no legitimate reason for information relating to hospitals and individual doctors to be withheld from the public. There is a persuasive argument that the public has a right to such information. Additionally public awareness and the likely response from the public to such information are likely to accelerate the pace of reform. There are persuasive reasons for providing information more generally concerning the performance of all parts of the system, including individual care providers. Patient choice in the absence of information is a charade. Failure to forewarn patients that the hospital of their (doctors') choice has a substandard safety record and that capacity exists in hospitals with a better record could, and arguably should, be regarded as professional negligence. There can be little doubt that, if consulted, the public would overwhelmingly endorse the need for this information. More generally, the provision of information is an effective method for effecting change and it is unlikely that the pace of reform would have been as sedate if the public were properly aware of the safety record of various health care providers. One argument against this option is that the provision of information might result in a loss of confidence in hospitals and doctors. However, the argument that the public should be kept in ignorance to engender unjustified confidence is, at best, dubious and if this ignorance allows an inadequate policy response then it is additionally harmful. In some states of the USA, and most notably New York, severity adjusted mortality rates are available for every hospital and for every doctor. This has not resulted in a significant change in the pattern for public demand but it has galvanised doctors and hospital staff to successfully review and upgrade their procedures [ 12 ]. League tables have recently been introduced in England to allow doctors and patients to evaluate the performance of particular hospitals [ 39 ]. From late 2004 the performance of individual surgeons will probably be available [ 40 ]. Conclusion 20: Information is a key element in achieving system reform and for the efficient operation of all markets, private or public. All institutions and individuals should receive rapid feedback on their performance. Information should be routinely collected and also obtained from random audit of hospitals. In particular, information should be publicly available with respect to the safety record of hospitals and providers of medical care. Financial incentives As discussed earlier, financial incentives are one some of the most effective, non-coercive ways of achieving desired outcomes. There has been very limited use of this powerful instrument and the financing of medical services has generally been perceived as a reward for providers doing what they select to do rather than as an opportunity for influencing what is done. This is an important missed opportunity. Importantly, financial incentives are non-coercive and avoid the head-to-head conflict between 'clinical autonomy' and the 'patient's right to evidence based medicine' which may accompany direct regulation. Conclusion 21: Financial incentives should be used flexibly throughout the health system to induce behaviour which minimise AEs. Governance and ownership of the problem A key theme of this paper has been that government policy has focused almost exclusively upon financial issues and that there has been little concern from within the legislature with health services and health. This is possibly the root cause of the government failure. Activities from within the various bureaucracies have largely been bureaucratic. In its 2004 review of State initiatives the ACSQHC reports that Queensland now has 'a program to develop a workforce culture that values a multi-disciplinary evidence based approach to improvement'; in 2003 the NT implemented an 'overarching quality committee'; NSW had 'involved the appointment of patient safety managers'. The ACT now has a framework which 'provides explicit lines of accountably'; Victoria now 'has a policy of an open and transparent approach to the provision of information' etc (p8) [ 34 ]. There is no reference to national legislative action to enforce safety. The ACSQHC itself appears frustrated with the scale of the national effort when it comments that 'over the last year with public failures reported in some Australian hospitals it is clear that a small program of national investment and development needs to be matched by the will, skill and capacity of stakeholders....' (p5). On the same page it comments that 'Council is actively working to further develop data sources, but cannot do so without continued support by all governments' (p5). Almost 10 years after the 1995 report the ACSQHC comments that 'future work ... needs to consider what governance and responsibilities are required at all levels of the health care system to ensure the provision of safe health care' (p84). Conclusion 22: Despite the unnecessary death of between 40,000 and 80,000 Australians since the publication of the QAHCS, Australian government has not taken ownership of the problem of adverse events. As a result, those attempting to effect change have been forced to 'coax and cajole' Options to 'mandate and enforce' have been largely ignored. Disregard of evidence The last conclusion relates to the need to generate and disseminate information relating to system safety. The arguments for this apply more generally to health system data. Australia has a wealth of administrative databases which could be employed to investigate and improve system performance. As noted earlier, service use is very uneven across Australia. To the extent that this violates the usual notion of 'equity' this information should be an important input into policy formulation. It does not appear to be used for this purpose outside NSW. However the quality of the data also presents other opportunities. The effect of different treatments, such as the varying rates of angiography and revascularisation observed between public and private hospitals could be traced through time if record linkage were possible. This would allow an assessment of downstream costs, mortality and morbidity associated with the two patient groups. There are clearly many opportunities for longitudinal research of this sort. A further option which may be piggy-backed on administrative data is the routine provision of information to different groups of patients who have been identified by their service mix. Information of this sort is probably a highly effective way of 'empowering patients'; that is, enabling patients, and particularly those with a chronic disease, to take greater control of their disease management. These developments have not occurred for a variety of reasons. First, for an $80,000 million industry, research funding for 'product development and marketing' – health services research – is astonishingly small. In the USA six Federal agencies alone spent $US 1,658 million in 2002 upon HSR. The agencies and their expenditures are detailed in Table 7 . Significant US funding is also obtained form the US network of Foundations which does not exist in Australia. Benchmarking against the Federal agencies alone, at an exchange rate of $US 0.70 = $AUS 1.0 (0.65) and scaling these expenditures down in relation to the size of the US and Australian economies, Australia should be spending about $AUS 120 million on HSR. Australia does not currently spend a fraction of this amount. As a major initiative, the NHMRC is to provide $10 million per annum for HSR – or about 7.7 percent of the US Federal benchmark. Table 7 US expenditure upon Health Services Research Federal Agency Expenditure ($US millions) Agency for Health Care Research and Quality 300 National Centers for Health Statistics 127 Extra Mural Prevention Research CDC 18 Centers for Medicare and Medicaid Services 55 Veteran's Health Administration 371 National Institute of Health 787 Source: Annual reports A second and possibly related reason is that there is no dedicated instrumentality, similar to the AIHW, which has taken 'ownership' of the need to provide and periodically to review the need for information generated by HSR. Funding is currently inadequate but also ad hoc. Third, concern over the confidentiality of records has been elevated to such a level that easy and routine data linkage to observe the outcome of different service patterns does not seem to be a possibility. For example, access to Australia-wide, de-identified public hospital records requires the separate consent of all States and Territories as well as the cooperation of the Commonwealth Department of Health or AIHW. Data linkage to determine the consequences of different treatment patterns – who lives and who dies – is so difficult that the research is effectively proscribed. It is extremely doubtful that this concern in the bureaucracy over privacy would reflect the preferences of a well-informed population. Patients almost certainly suffer and die because of the interpretation and implementation of our confidentiality laws in a way which seriously inhibits the capacity of the public, researchers and private and public agencies to investigate the outcome of system performance and differences in individual treatments. The bureaucratic fetish with confidentiality does not occur in the USA where the risk of litigation is significantly greater than in Australia. In some states data relating to hospital and doctor mortality rates are regularly published and in California unit record hospital data is available on CD in university libraries. Conclusion 23: Routinely collected administrative data should be fully used to monitor system performance. In particular it should be employed to monitor equity of access to services regularly and to provide disease-related information to population groups identified as having particular needs and interests. A statutorily independent national institute for health services research should be established whose terms of reference require the achievement of these objectives. 5 Discussion and conclusion There are various options for the macro reform of the health system and the corresponding reform of financial incentives and the roles and responsibilities of the various players. In particular, Dick Scotton has cogently argued for the adoption of Managed Competition [ 41 - 43 ]. A partial movement in this direction could be achieved by transferring responsibility for the purchase of health services to the various health regions [ 44 ]. There has been no attempt to review this large topic here. Rather, the paper has reviewed the 'micro' elements of such reform. The most appropriate 'macro' model for the health system is the model which maximises the likelihood of implementing satisfactory solutions to the numerous problems facing the system including those that have been discussed above. The chief conclusion from this paper is that the 'health care debate' and recent policy 'reform' has focussed upon issues which are best described as 'very small order' as judged by their likely effect upon either health or the cost effectiveness of the health system. They have been primarily concerned with dollars, not health and, more specifically, with the distribution of the cost between the public and private sectors. None of the policies discussed in Section 3 of the paper is likely to increase cost effectiveness. Cost shifting from the PBS to the public creates differential copayments which encourages allocative inefficiency. PHI reforms have created an industry with bizarre financial incentives and is a spectacular example of negative micro economic reform. The common feature of the three policies is that each moves the health system in a direction which is more consistent with the liberal/libertarian world view in which responsibility is transferred to the individual and away from the community. While PHI helps the individual to avoid financial decisions at the point of service, the individual is responsible for the purchase of the insurance and for the payment of (net) premiums. Copayments are the most direct method for shifting responsibility to the individual users of health services. Bulk billing for pensioners/health care card holders preserves the safety net which is needed for the achievement of equity as commonly defined by this world view. Over the longer term the pursuit of these values would redistribute income to the healthy, wealthy and away from the unhealthy unwealthy which is the antithesis of the communitarian/solidarity value system. The vehicle for the transfer is both a decline in community financed expenditures and a corresponding decline in taxation. As discussed in Section 2, decision making with respect to social objectives is the legitimate role of government. However, good economic and social policy seeks to achieve these objectives in a way that is cost effective. The reforms discussed here have not achieved this. In contrast with these policies, the five neglected areas discussed in Section 4 deal with problems which are 'large order issues' as judged by their likely impact upon health and the cost effectiveness of the health system. Also contrasting with the first group, these issues are relatively 'value neutral' as judged by either the communitarian or libertarian perspectives. The failure to address satisfactorily these issues is attributable to neglect, not the dominance of a particular ideology. This failure over a very long period itself refects a failure in the governance structure and a failure to identify and act upon opportunities for quantitatively large system improvements. The failure is probably replicated in most other developed countries, reflecting the complexity of the health sector. But benchmarking against similarly impaired systems does not alter the fact that there are opportunities for significantly improving the community's health which have been largely ignored. The reasons why this has occurred has not been discussed here and, to a greater or lesser extent, these failures have probably been replicated in most other developed countries, reflecting similar social and technical histories. Competing interests The author(s) declare that they have no competing interests.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548139.xml

549596

Open Access to Trials Register

I find the arguments raised by the PLoS Medicine editors very useful [ 1 ] as I had not considered that a scientific community would tolerate barring access to registers of trials. It leaves huge gaps for exploitation by privileged groups. It is not only colleagues in research and allied professions who need access but the global community, including members of the public wherever they live, those who participate in trials and those who will be on the receiving end of their outcomes. The annual reports of research ethics committees (RECs) are supposedly in the public domain after approval by Strategic Health Authorities in the UK. But very few members of the public know of their existence or how to access them. Approaches to individual committees even now can meet with varied reactions, from suspicious, defensive, or hostile—reluctantly sending one report, quizzing as to which organisation the enquirer belongs to and why they should want one—to extremely welcoming of interest and discussion. The annual reports should be easily accessible online by now, surely, but they are not. The activities of RECs and information on what research is being carried out in the name of society as a whole largely remain hidden from public view. There is no information about public access on COREC (Central Office for Research Ethics Committees; www.corec.org.uk ) or OREC (Office for Research Ethics Committees; www.orecni.org.uk ). COREC has not been open about dealing with issues of concern raised with them in the past. They do state that public interest is welcome now, so it would show a real commitment to making research activity more open if they would show support for totally open access to a register and to promote that through their Web site.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549596.xml

548688

Differences in need for antihypertensive drugs among those aware and unaware of their hypertensive status: a cross sectional survey

Background Lack of antihypertensive use among hypertensive individuals is a major public health problem. It remains unclear as to how much of this lack of treatment is because of failure to diagnose hypertension or failure to initiate drug treatment for those with a diagnosis of hypertension. The primary aim of this study was to determine the proportion of those untreated individuals who would be recommended to start drug therapy for control of blood pressure among those aware or unaware of their diagnosis of hypertension. Methods The Canadian Heart Health Surveys (1986 – 1992), a national, cross-sectional descriptive survey (n = 23 129), was used to determine the proportion of individuals who were untreated, yet satisfied the 2004 Canadian hypertension guidelines for initiating drug therapy. Patients were divided into subgroups of those aware and unaware of having a diagnosis of hypertension according to self reported awareness from the survey. Results Of those with untreated hypertension (= 140/90 mmHg), only 37% were aware of their diagnosis. 74% of untreated individuals aware of their diagnosis of hypertension would require drug therapy, compared to 57% of those who were unaware. Of those >65 years of age, 52% of aware individuals needed drug therapy whereas only 34% of unaware elderly would need drug treatment. Conclusion In both unaware and aware subgroups, the majority of patients with untreated hypertension would benefit from antihypertensive drug therapy according to the 2004 Canadian Hypertension recommendations. The proportion of untreated patients that still need drug therapy was higher among those who were aware compared to those who were unaware. This finding suggests that the major gap in hypertension control may be in initiating drug therapy rather than in diagnosing hypertension. Further studies are needed to confirm these results to ultimately help strategize public health efforts in controlling hypertension.

Background Antihypertensive drug therapy can reduce cardiovascular morbidity and mortality by 25–30% [ 1 - 3 ] in those with hypertension. The 2004 Canadian hypertension guidelines [ 4 ] recommend pharmacotherapy in those patients where there is proven cardiovascular benefit from randomized controlled trial evidence. However, many of these 'at risk' patients with hypertension remain untreated even though they have a higher risk for cardiovascular events [ 5 ]. This lack of treatment is a major public health problem as hypertension is a modifiable risk factor. It remains unclear as to how much of this lack of treatment is because of failure to diagnose hypertension or failure to initiate drug treatment for those with a diagnosis of hypertension. Further understanding would help to strategize public health care initiatives in controlling high blood pressure. The primary aim of this study was to determine the proportion of those untreated individuals who would be recommended to start drug therapy for control of blood pressure among those aware or unaware of their diagnosis of hypertension. Methods From the 2004 Canadian Hypertension Guidelines, pharmacotherapy is recommended if the diastolic blood pressure is > 100 mmHg; the systolic pressure is > 160 mmHg; or the diastolic blood pressure is >90 mmHg in the presence of cardiovascular heart disease or risk factors. To determine the proportions of individuals who satisfied the guidelines for initiation of drug treatment for hypertension, we used the Canadian Heart Health Surveys (1986 – 1992). This national, cross-sectional descriptive survey (n = 23 129) contains data on individual blood pressure measurements using standardized technique as well as information on cardiovascular risk profile, antihypertensive medication use and self reported awareness of hypertension. Further details of this survey have been published elsewhere [ 6 ]. We then calculated for both aware and unaware subgroups, the number of untreated hypertensive individuals that would be recommended drug therapy/ those untreated, hypertensive individuals. Patients with diabetes mellitus were excluded from the analysis because they require antihypertensive drug therapy at a much lower threshold. Descriptive statistics were used to calculate proportions and 95% confidence intervals in this study. Results Of those with untreated hypertension (= 140/90 mmHg), only 37% (n = 897 786) were aware of their diagnosis. Among those aware and unaware of their diagnosis of hypertension, 35% vs. 36% were female; 14% vs. 12% had any cardiovascular disease and 31% vs. 75% were under the age of 65 years respectively. The proportions of those who would be recommended drug therapy among those who are aware and unaware are presented in Table 1 . Table 1 demonstrates that compared to the unaware subgroup, a greater proportion of those aware of their hypertensive status needed drug therapy but were untreated. Among both aware and unaware groups, the largest proportion needing drug therapy was less than age 65 years. Of note, among those who were unaware, there was a considerably lower proportion that needed drug therapy over the age of 65 years. Table 1 Proportions of those aware and unaware who require drug therapy (95% CI) Awareness of hypertensive status >65 years 18–64 years Total Aware 71 477/ 138 342 51.7% (51.5–51.9%) 592 186/ 759 444 78% (77.9–78.1%) 663 663/ 897 786 73.9% (73.8–74%) Unaware 129 398/ 376 861 34% (33.8–34.2%) 738 542/ 1 143 872 64.6% (64.5–64.7%) 867 940/ 1 520 733 57.1% (57–57.2%) Discussion In both unaware and aware subgroups, the majority of patients with untreated hypertension would benefit from antihypertensive drug therapy according to the 2004 Canadian Hypertension recommendations. This analysis extends the findings from other studies [ 5 , 7 , 8 ] by determining those 'at risk' individuals that would be recommended pharmacotherapy based on awareness of their hypertensive status. The proportion of untreated patients that still needed drug therapy was higher among those who were aware compared to those who were unaware. Specifically, among the elderly, most of the patients who are unaware that they are hypertensive would not actually be recommended to take drug therapy. These study findings suggest that a significant gap in hypertension control may be initiating drug therapy among those known to have hypertension. Possible barriers to initiating drug therapy may arise from patients, physicians [ 9 ] or the health care system. There are several limitations with this study. First, the Canadian Heart Health Survey data are 12 years old and there is some evidence to suggest that prescription rates of antihypertensive drugs have improved since then [ 10 ]. However, this data set represents the best population data available on patients with hypertension in Canada since it is of high quality and captures a large cohort of Canadians. Another limitation is that the awareness or unawareness of hypertensive status is based on self report and not from physician records. Conclusion This study found a considerable number of both unaware and aware patients with untreated hypertension require drug therapy. However, the majority of those who needed drug therapy were actually aware that they had a diagnosis of hypertension compared to those who were not aware. This finding suggests that the major gap in hypertension control may be in initiating drug therapy rather than in diagnosing hypertension. Further studies are needed to confirm these results to ultimately help strategize public health efforts in controlling hypertension. Competing interests The author(s) declare that they have no competing interests. Authors' Contribution NK, DW and NC contributed to the design of the project. NK contributed to data analysis. NK, DW and NC contributed to writing the manuscript and for substantive, intellectual editing contributions. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548688.xml

549541

Spontaneous rupture of giant gastric stromal tumor into gastric lumen

Background Gastrointestinal stromal tumors (GIST) constitute a large majority of mesenchymal tumors of the gastrointestinal (GI) tract, which express the c-kit proto-oncogene protein, a cell membrane receptor with tyrosine kinase activity. GI stromal tumors of the stomach are usually associated with bleeding, abdominal pain or a palpable mass. Case presentation A 75-year-old male presented with upper abdominal pain and palpable mass. Computed tomographic (CT) scan of the abdomen showed a large mass arising in the posterior aspect of fundus, body, and greater curvature of the stomach. Second day after the admission, there was significant reduction in the size of the tumor, clinically as well as radiologically. Endoscopic biopsy showed large bulge in fundus and corpus of the stomach posteriorly with an opening in the posterior part of the corpus, and biopsy from the edge of the opening reveled GIST. Patient underwent curative resection. Conclusion Spontaneous ruptured of giant gastric stromal tumor is very rare presentation of stomach GIST. Thorough clinical examination and timely investigation can diagnose rare complication.

Background Gastrointestinal stromal tumors (GIST) are the most common form of mesenchymal tumors arising from the gastrointestinal (GI) wall, mesentery, omentum or retroperitoneum that express the c-kit proto-oncogene protein [ 1 ]. This expression of c-kit distinguishes GIST from true leiomyomas, leiomyosarcomas, and other mesenchymal tumors of the GI tract [ 1 , 2 ]. Stomach (60–70%) and small intestine (20–30%) is the most common site for GIST [ 2 ]. Approximately 10–30% of patients with GIST may be asymptomatic. Stomach and small intestinal stromal tumors are usually associated with abdominal pain, GI bleeding or palpable mass. Around 30% of all GISTs are malignant and liver is the most common site for metastasis. Surgical resection is the primary treatment of GIST [ 3 ]. The 5-year survival following curative resection ranges from 20–80% [ 3 - 7 ]. Imatinib mesylate, tyrosine kinase inhibitor, is the first effective drug with response rate of 54% in the treatment of metastatic GIST. We report here a case of GIST which presented with rupture in to the gastric lumen. Case presentation A 75-year-old diabetic male presented with dull upper abdominal pain of one-week duration. He noticed swelling in left upper abdomen. There was no history of vomiting, fever or gastrointestinal bleeding. He had no significant medical or family history and was non-smoker and non-alcoholic. Physical examination showed a 14 × 10 cm mass palpable in epigastrium and left hypochondrium with minimal intrinsic mobility. Routine biochemical investigations were normal. Ultrasonogram and CT-scan of the abdomen showed large heterogeneous mass of 13 × 10 cm extending from the tail of pancreas to anterior pararenal space, lesser sac to gastrosplenic ligament enveloping the posterior aspect of fundus, body and greater curvature (Figure 1 ). One day after the admission, examination showed reduction in the size of palpable mass to 8 × 6 cm size which was not associated with aggravation of the symptoms. Ultrasonography of the abdomen was repeated which showed reduction in the diameter of mass to 8 × 8 cm. Upper endoscopy showed large bulge in fundus and corpus of the stomach posteriorly with an opening in the posterior part of the corpus with edematous margin with dissemination of serous fluid and necrotic material in to the stomach (Figure 2 ). Fluid analysis was normal for CEA and CA 19-9. Biopsy taken from the edge of the opening showed bundles of spindle cells with elongated nuclei and tumor cells (Figure 3 ) and was strongly positive for CD117 immunohistochemical examination, diagnostic of gastrointestinal stromal tumor (Figure 4 ). At laparotomy a large tumor was seen arising from the posterior wall of stomach measuring 8 × 8 cm, which has ruptured into the gastric lumen, and was infiltrating the upper pole of spleen, anterior capsule of pancreas and mesocolon. He underwent total gastrectomy and splenectomy with esophagojejunostomy, and segmental transverse colectomy. Histopathology of resected specimen showed large spindle cell GIST with >5/50 HPF (high-power field) mitotic activity. Postoperative period was uneventful. Postoperatively he was put on imatinib mesylate 400 mg once daily. Patient is asymptomatic on follow up for 11 months. Figure 1 CT -Scan showed large heterogeneous mass of 13 × 10 cm size extending from the tail of pancreas to anterior prararenal space, lesser sac to gastrosplenic ligament enveloping the posterior aspect of fundus, body and greater curvature. Figure 2 Upper endoscopy picture shows opening in the posterior part of the corpus with edematous margin with dissemination of serous fluid and necrotic material in to the stomach. Figure 3 Photomicrograph showing upper endoscopy biopsy specimen of gastric GIST showing multiple spindle cells with eosinophilic cytoplasm and ovoid to elongated nuclei Figure 4 Photomicrograph of biopsy specimen with immunohistochemical staining for CD117 Discussion GI stromal tumors express c-kit protein also known as CD 117, and is considered as highly specific marker that differentiates GIST from other mesenchymal tumors such as leiomyomas [ 8 - 10 ]. The majority of GISTs occur in the stomach (60–70%) and small intestine (20–30%) [ 9 ]. GIST arises from the stomach, presented with abdominal pain, GI bleeding or palpable mass. Around 20–30% of GISTs detected during surgery for intestinal obstruction or bleeding [ 9 ]. Among the diverse clinical presentation of stomach GISTs, spontaneous ruptured in to peritoneal cavity lead to peritonitis [ 11 ], extragastric growth [ 12 ], complicated with hiatus hernia [ 13 ], ruptured of gastric stromal tumor with cystic degeneration presenting as hemoperitoneum [ 14 ], gastric stromal tumor with myxoid degeneration [ 15 ] have been reported in the literature. Our patient had giant gastric GIST, which ruptured into the stomach with dissemination of its necrotic tissue in the stomach. There was no aggravation of the symptoms. Abdominal examination revealed reduction of the size of tumor from 13 × 10 cm to 8 × 6 cm size. Ultrasound abdomen also confirmed 40% reduction in the size of the tumor. Upper endoscopy biopsy from the edge of the opening, posterior wall of the stomach was suggestive of GIST. Since most of the GISTs are submucosal and grow endophytically, preoperative tissue diagnosis is difficult. In our patient biopsy from the edge of the rent and deep inside the opening confirmed diagnosis of GIST preoperatively. Patient underwent complete tumor resection, with uneventful postoperative period. Assessment of malignant potential of a primary GIST lesion is difficult in many cases; even as small as less than 2 cm size GIST has certain malignant potentials [ 9 ]. Currently, there is no single prognostic factor that can be used alone to predict tumor behavior. Biological behavior of tumor also depends on the location; for example, GISTs arising from the small bowel or colon are generally associated with a less favorable outcome than those arising in the stomach [ 16 ]. Radiological and surgical factors that have been used to determine malignancy include invasion to adjacent organs; omental or peritoneal seeding; tumor recurrence after surgical resection; or distant metastasis [ 8 , 17 ]. Pathological factors that determine malignancy are tumor size, mitotic activity, nuclear pleomorphism, degree of cellularity, nuclear to cytoplasmic ratio, and mucosal invasion [ 18 ]. Both mitotic activity and tumor size have been identified as the most important factors predicting malignant behavior [ 3 , 9 , 19 ]. In our patient, most of the factors like size of the tumor, local invasion and high mitotic activity indicated high malignant potential. Imatinib mesylate, competitive inhibitors of certain tyrosine kinases including the intracellular kinases ABL and BCR-ABL fusion proteins present in some leukemia's, and platelet-derived growth factor receptors [ 20 ], is the first effective drug in the treatment of metastatic GIST. Demetri et al [ 21 ] had reported a response rate of 54%, with median time to response is about 13 weeks, in patients with either inoperable or metastatic GISTs treated with a daily dose of 400 mg or 600 me with follow-up of at least 6 months. Even in patients with large tumor, response to imatinib mesylate can occur rapidly [ 22 ]. The optimum dose of imatinib mesylate in the treatment of GIST is not yet known. Toxicity increases with increasing dose, with maximum tolerated dose was 800 mg taken for 8 weeks [ 22 ]. Role of imatinib mesylate in the treatment of malignant GIST after curative response is still under investigation. Rational of giving imatinib mesylate in our patient following curative resection, was presence of a very large tumor (>10 cm) with local invasion and high mitotic activity (>5/50 HPF). Conclusions Spontaneous ruptured of giant gastric stromal tumor is very rare presentation of stomach GIST. Clinical examination and timely investigations can diagnose this rare complication. Competing interests The author(s) declare that they have no competing interests. Authors' contributions RM, AJ : Preparation of manuscript NR : performed upper endoscopy and biopsy and helped in preparation of draft manuscript. SOV, SS, PD : surgical management, manuscript revision for scientific content PD, BV : Revision of manuscript and preparation of final manuscript. All authors read and approved the final version of the manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549541.xml

516045

Similar promotion of Aβ1-42 fibrillogenesis by native apolipoprotein E ε3 and ε4 isoforms

The apolipoprotein E ε4 allele contributes to the genetic susceptibility underlying a large proportion (~40–60%) of typical, sporadic Alzheimer disease. Apolipoprotein E deficient mice made transgenic for human apolipoprotein E ε4 accumulate excess cerebral amyloid when compared to similarly prepared mice expressing human apolipoprotein E ε3. Therefore, it is important to search for relevant interactions(s) between apolipoprotein E ε4 and Aβ in order to clarify the biological role for apolipoprotein E ε4 in Alzheimer disease. Using a thioflavine T (ThT)-based assay, we have investigated the effects of native human apolipoprotein E isoforms on the kinetics of Aβ fibrillogenesis. No obvious profibrillogenic activity was detected in Aβ 1-40 -based assays of any native apolipoprotein E isoform. However, when ThT assays were repeated using Aβ 1-42 , modest, but statistically significant, profibrillogenic activity was detected in both apolipoprotein E ε3- and apolipoprotein E ε4-containing media and was similar in magnitude for the two isoforms. These data demonstrate that native apolipoprotein E possesses "pathological chaperone"-type activity for Aβ: in other words, the data indicate that a chaperone-like misfolding reaction can occur between native apolipoprotein E and Aβ. However, the equipotent activities of the apolipoprotein E ε3 and ε4 isoforms suggests the possibility that either extended co-incubation of apolipoprotein E and Aβ, or, perhaps, the inclusion in the reaction of other fibrillogenesis-modulation co-factors (such as metal ions, or inflammatory mediators such as reactive oxygen species, α 2 -macroglobulin, apolipoprotein J, etc.) may be required for modeling in vitro the apolipoprotein E-isoform-specific-regulation of extracellular Aβ accumulation that occurs in vivo . Alternatively, other events, such as differential apolipoprotein E-isoform-mediated clearance of Aβ or of apolipoprotein E/Aβ complexes may underlie apolipoprotein E-isoform-dependent Aβ accumulation.

Background Genetic-neuropathological correlation indicates that the apolipoprotein E type ε4 isoform specifies increased cerebral [ 1 , 2 ] and cerebrovascular [ 3 ] accumulation of amyloid β-protein (Aβ). In addition, the apolipoprotein E ε2 isoform can apparently prevent the expression of clinical Alzheimer-type dementia which is otherwise typically associated with amyloidogenic mutations in the amyloid-β protein precursor [ 4 ]. Since the apolipoprotein E ε4 allele contributes to the genetic susceptibility underlying a large proportion (~40–60%) of typical, sporadic Alzheimer disease, it is important to search for relevant interactions(s) between apolipoprotein E ε4 and Aβ in order to clarify the biological role for apolipoprotein E ε4 in Alzheimer disease. Currently proposed mechanisms include differential activities of apolipoprotein E isoforms in modulating Aβ fibrillogenesis [ 5 - 7 ] and/or Aβ clearance [ 8 , 9 ]. Many studies of apolipoprotein E modulation of Aβ fibrillogenesis have utilized denatured apolipoprotein E, purified from the serum of human apolipoprotein E homozgotes following extraction in organic solvents [ 10 ]. While providing a convenient source of pure apolipoprotein E protein, this preparation does not represent native apolipoprotein E as it exists in vivo . Using a thioflavine T (ThT)-based assay [ 11 ] we have investigated the effects of native human apolipoprotein E isoforms on the kinetics of Aβ fibrillogenesis. Methods Synthetic Aβ 1-40 or Aβ 1-42 (Keck Foundation Protein Facility, Yale University, New Haven CT) was freshly prepared for each assay at a concentration of 16 mg/ml in distilled, deionized water and diluted with phosphate-buffered saline (PBS) to generate a 5 mg/ml working solution. The "aggregation step" consisted of a reaction mixture containing 8 μl Aβ peptide (1 mg/ml final conc) and 32 μl of either apolipoprotein E ε3- or ε4-containing conditioned medium or control conditioned medium from SV40 empty vector-transfected cells. For the investigation of native apolipoprotein E preparations, apolipoprotein E isoforms were generated in the conditioned medium of stably-transfected SV40-apolipoprotein E ε3-, or SV40-apolipoprotein E ε4-, expressing Chinese hamster ovary (CHO) cells (CHO cells lack detectable endogenous apolipoprotein E; data not shown). All conditioned media were prepared using Dulbecco's minimal essential medium supplemented with 0.2% (wt/vol) bovine serum albumin only (no fetal bovine serum). Apolipoprotein E isoform levels were determined by quantitative immunoblotting of conditioned medium and apolipoprotein E-containing serum standards, the latter having been kindly provided by Dr. Petar Alaupovic of the Oklahoma Medical Research Foundation (Oklahoma City OK). Conditioned medium apolipoprotein E concentrations were then standardized using control medium conditioned by SV40 empty vector-transfected cells as diluent, yielding a final concentration of apolipoprotein E of 14 μg/ml, within the range of that reported in human cerebrospinal fluid. Since the final concentration of Aβ peptide was 1 mg/ml, the Aβ/apolipoprotein E stoichiometry (molar ratio) was ~500:1, suggesting models for the Aβ/apolipoprotein E interaction based either on a "catalytic" "pathological chaperoning" model of apolipoprotein E action on Aβ, or with a "seeding" model of Aβ folding. Detailed biochemical characterization of this native apolipoprotein E preparation has been reported [ 9 ]. The "aggregation step" fibrillogenesis reaction [ 11 ] was incubated at 37°C until the time of the ThT fluorescence measurement, which was performed from 1 to 7 days later. For the "measurement step" [ 11 ], 960 μ1 of 10 μM ThT (Nakarai Chemicals, Kyoto, Japan) in 50 mM phosphate buffer (pH 6.0) was added to the "aggregation step" reaction mixture. Within 30 minutes after addition of ThT, fluorescence was measured with a Millipore Cytofluor (Bedford MA) in each of five successive 200 μl aliquots of the reaction mixture, using an excitation filter of 450 nm and an emission filter of 482 nm, and a temperature of 25°C. In order to standardize the ThT assay in our Laboratory, we performed studies of Aβ fibrillogenesis following 1–7 day incubations of Aβ 1-40 , either in physiological phosphate buffer alone or in the presence of metal ions (Zn 2+ , Fe 2+ , or A1 3+ ; 1 mM final conc [ 12 ]. ThT fluorescence and ultrastructural features were measured daily (not shown). Profibrillogenic activities of the metal ions tested were in agreement with a published report [ 12 ] (e.g., Al 3+ stimulated ThT fluorescence of Aβ 1-40 by 3.6 ± 1.1- to 5.7 ± 1.4-fold; p < 0.01), indicating that our Aβ preparations were capable of metal ion-induced fibrillogenesis. Metals were not present during assessment of profibrillogenic effects of apolipoprotein E isoforms. Results and discussion No obvious profibrillogenic activity was detected in Aβ 1-40 -based assays of any native apolipoprotein E isoform (Table 1 ). However, when ThT assays were repeated using Aβ 1-42 , modest, but statistically significant, profibrillogenic activity was detected in both apolipoprotein E ε3- and apolipoprotein E ε4-containing media and was similar in magnitude for the two isoforms (Table 1 ). The observation of a profibrillogenic effect of apolipoprotein E specifically for Aβ 1-42 has been noted [ 5 ] and is of particular interest in light of biophysical and molecular neuropathological evidence suggesting that "long" Aβ peptides ending at positions N-42 or N-43 are apparently crucial for the initiation ("seeding") of Aβ deposition [ 13 ]. Table 1 Effects of native apolipoprotein E isoforms on fibrillogenesis of Aβ 1-40 and Aβ 1-42 . Fold-effects represent means ± SEM of the quotients of ThT fluorescence values obtained for each Aβ peptide in the presence of apolipoprotein E-isoform-containing conditioned medium divided by ThT values obtained in the presence of conditioned medium lacking apolipoprotein E, derived from empty vector-transfected cells (n = 5–6). Aβ 1-40 1 day co-incubation CHO apolipoprotein E ε3 1.0 ± 0.l-fold N.S. CHO apolipoprotein E ε4 1.0 ± 0.l-fold N.S. 7 day co-incubation CHO apolipoprotein E ε3 1.0 ± 0.l-fold N.S. CHO apolipoprotein E ε4 1.2 ± 0.l-fold N.S. Aβ 1-42 4 day co-incubation CHO apolipoprotein E ε3 1.7 ± 0.27-fold p < 0.01 CHO apolipoprotein E ε4 1.6 ± 0.18-fold p < 0.005 7 day co-incubation CHO apolipoprotein E ε3 1.7 ± 0.22-fold p < 0.005 CHO apolipoprotein E ε4 1.8 ± 0.19-fold p < 0.0005 These data demonstrate that native apolipoprotein E possesses "pathological chaperone"-type activity for Aβ: in other words, the data indicate that a chaperone-like misfolding reaction can occur between native apolipoprotein E and Aβ, at least at the concentrations and proportions evaluated herein. However, the equipotent activities of the apolipoprotein E ε3 and ε4 isoforms suggests the possibility that either extended co-incubation of apolipoprotein E and Aβ, or, perhaps, the inclusion in the reaction of other fibrillogenesis-modulation co-factors (such as metal ions, or inflammatory mediators such as reactive oxygen species, α 1 -antichymotrypsin, heparin sulfate-protcoglycan, non-Aβ component, apolipoprotein J, complement, etc.) may be required for modeling in vitro the apolipoprotein E-isoform-specific-regulation of extracellular Aβ accumulation that occurs in vivo . Alternatively, other events, such as differential apolipoprotein E-isoform-mediated clearance of Aβ or of apolipoprotein E/Aβ complexes [ 8 , 9 , 14 ] may contribute to apolipoprotein E-isoform-dependent Aβ accumulation. Differential anti-inflammatory activity might also play a role. Further investigation will be required in order to elucidate the precise mechanism(s) which specify how apolipoprotein E ε4 promotes Aβ accumulation in human brain and cerebral vessels in vivo . List of abbreviations ThT, thioflavine T; Aβ, amyloid-β peptide; PBS, phosphate-buffered saline; CHO, Chinese hamster ovary cells. Competing interests None declared. Authors' contributions DS performed all assays, including the ThT assay, which was originated by HL. HL also oversaw the transfer of the assay from his lab to ours. RM prepared standard conditioned media from transfected cells provided by JDS. SG oversaw the project, supported the project as noted below, and wrote the manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516045.xml

529262

Differential diagnosis of tuberculous meningitis from partially-treated pyogenic meningitis by cell ELISA

Background Tuberculous meningitis (TBM) is a major global health problem, and it is sometimes difficult to perform a differential diagnosis of this disease from other diseases, particularly partially-treated pyogenic meningitis (PTPM). In an earlier study, we demonstrated the presence of a 30-kD protein antigen in cerebrospinal fluid (CSF) of TBM patients. We have also shown that lymphocytes from CSF of TBM patients respond differently to this antigen than do those from PTPM patients. The purpose of this study was to develop an assay that can discriminate between TBM and PTPM. Methods We developed a cell enzyme-linked immunosorbant assay (Cell ELISA) to quantitatively measure production of antibodies against the 30-kD protein in B cells from CSF of TBM and PTPM patients. Results The cell ELISA yielded 92% (11/12) sensitivity and 92% (11/12) specificity for the differential diagnosis of TBM from PTPM. Conclusion When induced with the 30-kD protein antigen, B cells derived from CSF of TBM patients respond to IgG production within 24 h while those derived from PTPM patients do not respond.

Background Tuberculous meningitis (TBM) is an infection of the central nervous system (CNS) that is prevalent in both under-developed and developing countries. An increased incidence of TBM has occurred in recent years due to the growing number of people infected with human immunodeficiency virus (HIV). Diagnosis of TBM remains problematic despite many new, advanced diagnostic methods [ 1 , 2 ]. Previous clinical studies have clearly demonstrated that the timing of TBM treatment is the most critical factor in determining the ultimate outcome, which underscores the importance of early diagnosis [ 3 ]. The laboratory confirmation for the diagnosis of TBM is based on the detection of acid-fast bacilli (AFB) in the cerebrospinal fluid (CSF) and by culturing CSF for Mycobacterium tuberculosis bacilli (MTB) [ 4 ]. However, the sensitivity of direct AFB smears from CSF ranges from 5–10% and culturing techniques take 4–6 weeks. It has been recently reported that the staining efficiency of the AFB smear test can be increased to detect up to 50% of TBM cases, but this technique requires a very large amount of CSF [ 5 ]. Clinical as well as CSF features are helpful for diagnosing TBM, but they cannot be used to differentiate TBM from other infectious and non-infectious disorders [ 6 , 7 ]. In particular, clinicians often encounter difficulty when performing a differential diagnosis of TBM from partially-treated pyogenic meningitis (PTPM) cases. Both the results from biochemical and pathological analysis of CSF and the clinical presentation of TBM are often similar to those of PTPM, which results in frequent misdiagnosis. In an earlier study, we reported the presence of a diagnostic 30-kD protein antigen in CSF of confirmed and suspected TBM patients [ 8 ]. Immunological methods such as antibody-capture enzyme-linked immunosorbant assay (ELISA) have been previously used for diagnosing TBM [ 9 ]. The cell ELISA method allows further confirmation of the results obtained by antibody-capture ELISA. Cellular immune function is characterized by the existence of various types of lymphoid cells. As lymphocytes participate in the production of humoral immunity, they may respond to the 30-kD protein antigen in TBM and PTPM patients. We have developed a cell ELISA to study the response of B cells derived from CSF of TBM and PTPM cases following challenge with the 30-kD protein antigen. The purpose of the present study was to evaluate the antibody response to the 30-kD protein antigen in CSF of TBM and PTPM patients by cell ELISA and to determine whether this method may be used in differential diagnosis of TBM from PTPM. Methods Patients and sample collection The Central India Institute of Medical Sciences (CIIMS), Nagpur, is a tertiary referral center. CSF was collected from patients who were suspected of having TBM or other infections before they received any treatment. For patients undergoing cranial surgery, analysis of CSF was performed if they were suspected of having meningitis. These patients were already on broad-spectrum antibiotics, such as third-generation cephalosporins and aminoglycosides. To establish a diagnosis of meningitis, 2–5 ml CSF was withdrawn from patients using a lumbar puncture. CSF was then subjected to routine biochemical analysis and pathological analysis including Gram staining, India ink staining, and AFB staining and culturing. One milliliter of CSF was used for the cell ELISA study, and 1 ml was used for detection of the 30-kD protein band by SDS-PAGE analysis in 12 randomly selected TBM and PTPM patients. Diagnosis of TBM and PTPM was based on the criteria described below. Diagnostic criteria 1. Tuberculous Meningitis (TBM) Presence of Mycobacterium tuberculosis in CSF by staining and/or culture, OR Clinical meningitis with the following observations: A. Sub-acute or chronic fever with features of meningeal irritation such as headache, neck stiffness, and vomiting with or without other features of CNS involvement B. CSF findings showing increased proteins, decreased glucose (CSF:blood glucose ratio <0.5), and/or pleocytosis with lymphocytic predominance C. Presence of the 30-kD protein band in CSF on SDS-PAGE analysis D. Good clinical response to antituberculous drugs None of the 12 TBM patients had positive AFB staining. 2. Partially-treated pyogenic meningitis (PTPM) Presence of pathogenic bacteria in CSF by staining and/or culture, OR Clinical meningitis with the following observations: A. Fever and/or signs of meningeal irritation (patients who have undergone cranial surgery to treat tumor(s), stroke, or head injury and who have received antibiotics), OR High fever and/or signs of meningeal irritation with or without CNS manifestations (patients who received broad-spectrum antibiotics) B. CSF findings showing increased proteins, decreased glucose (CSF:blood glucose ratio <0.2), and/or pleocytosis with a predominance of polymorphonuclear cells; CSF may resemble that of chronic meningitis patients C. Absence of the 30-kD protein band in CSF on SDS-PAGE analysis D. Good clinical response to broad-spectrum antibiotics 3. Control group Peripheral blood samples from six healthy volunteers were also analyzed and included as negative controls. Laboratory studies Sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE CSF samples obtained from confirmed and suspected TBM cases were subjected to SDS-PAGE. SDS-PAGE was performed with a vertical slab gel electrophoresis system (Broviga, India) using the standard Laemmali method (10). A 4% stacking gel and 10% running gel were used. Electrophoresis was carried out at 250 volts/50 mAmps. Gels were developed by staining with Coomassie brilliant blue GR-250 and the protein profiles were then studied. Band size (i.e., molecular weight) was estimated using molecular weight markers (Genei, Bangalore, India) in a parallel lane. Antigen (30-kD) preparation Following separation of proteins from CSF of confirmed TBM patients (AFB-positive) by SDS-PAGE, the 30-kD protein band was sliced out of the gel and pre-equilibrated in elution buffer (0.15 M phosphate-buffered saline [PBS], pH 7.4) and then electro-eluted in a whole gel eluter system (Biotech, India) for 90 min at 30 volts (11). The sample was then harvested from the unit and dialyzed against PBS and the protein content was measured using a Bio Lab KIT. Protein purity was checked using native PAGE and was then used to evaluate the antibody response of B cells derived from CSF of TBM and PTPM patients. Preparation of CSF Cells One milliliter of CSF collected from TBM and PTPM patients was centrifuged at 400 rpm for approximately 20 min. The supernatant was then discarded and the cell pellet was washed two times with PBS and then diluted in RPMI 11640 tissue culture medium containing 10% fetal calf serum. Preparation of Blood Cells Heparinized blood samples were obtained from six healthy volunteers. Peripheral blood mononuclear cells (PBMC) were isolated from heparinized blood by standard Ficoll-Hypaque gradient centrifugation. PBMCs were dissolved in PBS and centrifuged at 400 rpm for approximately 15–20 min, and the PBMCs were then diluted in RPMI 11640 tissue culture medium containing 10% fetal calf serum. Cell ELISA Flat-bottomed, 96-well ELISA plates were coated with 10 μg 30-kD antigen/ml diluted in PBS (pH 7.2). Following overnight incubation, the plates were washed with PBS and then coated with 5% BSA-PBS for 4 h. The plates were then washed five times with PBS. Two-hundred μl of the cell preparation derived from CSF of patients with TBM or PTPM were then added to the wells and coated. Each sample was prepared in duplicate. Plates were maintained overnight at 37°C in 5% CO 2 in a carbon dioxide incubator. The following day, the plates were washed with PBS and horseradish peroxidase (HRP)-conjugated rabbit anti-human IgG (1:10,000) was then added to the plates. After a 2-hr incubation at 37°C, the plates were washed again with PBS and 100 μl tetramethylbenzidine (TMB)/H 2 O 2 were added. The TMB/H 2 O 2 served as a substrate for HRP. After a 15-min incubation, 100 μl stop solution (2.5 N sulphuric acid) were added and the plates were then read with an ELISA reader at 450 nm (12). Results Detailed clinical data for TBM and PTPM patients are presented in Table 1 . Out of the 12 PTPM patients, two cases harbored microorganisms, which were cultured (gram-positive cocci in one case and gram-negative bacilli in the other case). Among the 12 patients who fulfilled the criteria for TBM (shown in Table 1 ), CSF of all these patients was positive for the 30-kD protein antigen and was negative for AFB. None of the patients had a previous history of extra-CNS tuberculosis. In addition to the patients described in Table 1 , we also tested an additional 700 CSF samples, including 150 from TBM patients. The 30-kD protein antigen was observed in >90% of these TBM patients (data not shown). Figure 1 shows the presence of the 30-kD protein band in the CSF of suspected TBM cases. This band was markedly absent from the CSF of PTPM patients. Table 1 Clinical and CSF Findings for TBM and PTPM Patients CSF Analysis Case No. Age (years)/Sex TLC P % L % Protein (mg/dl) Sugar (mg/dl) CSF:blood sugar ratio Neck Stiffness Duration of fever (weeks) Headache TBM 1* 16/f 30 - 100 83 27 0.32 Present 16 Present 2 8/m 90 5 95 105 30 0.31 Absent 8 Present 3 54/f 25 2 98 47 67 0.54 Absent 3 Present 4 16/f 450 18 82 535 21 0.24 Absent 12 Present 5* 26/m 120 60 38 143 43 0.66 Present 12 Present 6 58/m 14 - 100 68 37 0.44 Present - Absent 7 31/f 180 12 88 203 86 0.53 Absent 1 Present 8 55/f 112 10 90 231 22 0.38 Present 4 Present 9 65/m 150 1 99 131 23 0.33 Present 4 Present 10 56/m 121 12 88 217 41 0.47 Present 2 Present 11 7/f 32 35 65 68 31 0.32 Present 4 Present 12 43/f 60 - 100 97 39 0.48 Present 2 Present PTPM 1 # 38/m 220 90 10 96 26 0.20 Present 1 Present 2 # 63/f 1600 61 38 401 15 0.14 Present 1 Present 3**# 23/m 61 95 5 270 33 0.20 Absent 6 Present 4 +π 25/m 180 88 12 203 86 0.53 Present 1 Present 5 # 48/m 450 77 32 868 32 0.21 Present 8 Present 6 π 14/f 150 84 11 61 38 0.31 Absent 1 Present 7 ψ 52/m 140 92 8 471 12 0.14 Absent - Absent 8 # 56/m 430 80 20 518 17 0.22 Absent - Present 9 ++ψ 6/m 36 82 9 61 107 0.33 Absent - Absent 10 # 62/f 40 73 27 142 25 0.25 Absent - Present 11 # 4/m 50 90 5 71 24 0.19 Absent - Absent 12 π 27/m 200 78 22 131 21 0.18 Absent 2 Present %P- Polymorphs, % L- Lymphocytes. *Pulmonary tuberculosis (on chest x-ray) **Consolidation (on chest x-ray) # Cranial surgery post operative. ψ Post Head injury, π Presented to CIIMS as PTPM. + gram-negative bacilli observed, ++ gram-positive& non-capsulated cocci pairs observed. Figure 1 SDS-PAGE electrophoretogram of CSF from control (lanes B, C, E) and suspected (lane D) TBM subjects. Molecular weight marker is shown in lane A. The arrow indicates the 30-kD band, which represents the 30-kD protein antigen The ELISA absorbance values of IgG to the 30-kD protein antigen in CSF from TBM and PTPM patients are presented in Figure 2 . The cut-off value (OD at 450 nm) for positivity to the 30-kD protein antigen IgG in the control CSF is 0.6. High-titer values for IgG antibody production against the 30-kD protein antigen were observed in 11 out of 12 TBM patients. However, the titer in PTPM patients was much lower than that observed in TBM patients. IgG antibody production (expressed as ELISA absorbance value) ranged from 0.7 to 2.0 for cells derived from CSF of TBM patients, with the exception of case no. 5 (ELISA absorbance value, 0.59), and from 0.05 to 0.38 for cells derived from CSF of pyogenic meningitis cases, with the exception of case no. 4 (ELISA absorbance value, 0.79). The sensitivity of the cell ELISA was 92% and the specificity was 92% for differential diagnosis of TBM from PTPM. No IgG antibodies to the 30-kD protein antigen were produced by PBMCs from six healthy individuals within 48 h of exposure to the 30-kD protein antigen. Figure 2 B Cell response (IgG reactivity) to the 30-kD protein antigen in CSF cells derived from tuberculous meningitis (TBM) and partially-treated pyogenic meningitis (PTPM) patients and peripheral blood cells from control subjects Discussion During the past decade, several conventional immunoassays including ELISA, dot immunobinding assays, immunoblot assays, and various molecular methods such as the polymerase chain reaction (PCR) have been reported as adjuncts in the diagnosis of TBM [ 4 , 13 - 15 ]. However, difficulties have been encountered when using many of the aforementioned techniques to differentiate TBM from PTPM. CSF TLC, DLC (total and differential leukocyte count), protein, and glucose estimation are helpful parameters for establishing a TBM diagnosis and for differentiating other infectious and non-infectious neurological disorders, but these tests are non-specific and often cannot differentiate TBM from PTPM in patients in whom organisms are not observed. Delays in diagnosis and treatment are regarded as major contributing factors to the high mortality and morbidity of TBM, and any delay in starting appropriate medication for TBM and PTPM worsens the outcome. We previously used SDS-PAGE to demonstrate the presence of a 30-kD protein antigen in the CSF of TBM patients that is specific to M. tuberculosis and may be considered to be a diagnostic marker for TBM. In this study, we used this 30-kD protein antigen to evaluate the IgG antibody response of B cells derived from CSF of TBM and PTPM patients and from peripheral blood samples from six healthy volunteers. A cell ELISA was developed for the quantitative measurement of antibody production against the 30-kD protein antigen by these cells. Higher titers of IgG antibody production were observed in TBM patients compared to PTPM patients. The cells obtained from CSF of TBM patients gave an early response, presumably because they were already sensitized against the TBM antigen. However, when challenged with the 30-kD protein antigen, the cells obtained from PTPM patients and healthy volunteers gave a delayed response since they are not sensitized against this antigen. Therefore, an early response on this time scale is indicative of TBM. We have thus shown that cell ELISA is a sensitive technique for the differential diagnosis of TBM from PTPM. This method involves the demonstration of active antibody production by cells, particularly those derived from the affected site [ 16 ]. Previously, we standardized cell ELISA methodology in our laboratory using standard culture filtrate protein of M. tuberculosis (H37Rv strain) received from Colorado State University, Fort Collins USA (data not shown). The only limitation of this method is the time period (24–30 h) involved. However, the sensitivity of the test overcomes this drawback since it the only reported method that can discriminate TBM from PTPM. The sensitivity and specificity of IgG antibody in differential diagnosis of TBM from PTPM using the 30-kD protein antigen by cell ELISA was found to be 92% (11/12). We have also demonstrated that antibody production against the 30-kD protein antigen is higher in cells derived from CSF of TBM patients compared to PTPM patients. Various methods have been developed in our laboratory that yield a high specificity and sensitivity for diagnosis of TBM, but a small number of false positive results have been observed in pyogenic meningitis cases, particularly PTPM cases [ 17 , 18 ]. The cell ELISA method developed in our laboratory using the 30-kD protein antigen marker can potentially provide additional information to the treating physician that may enable a differential diagnosis of TBM from PTPM. The cell ELISA method for diagnosing TBM is based on the assumption that local synthesis of humoral antibodies against MTB antigen occurs. Various researchers have shown that CSF-derived cells have a significantly higher proliferation response to purified protein derivative (PPD) in patients with TBM, which is suggestive of an intrathecal immune response [ 11 , 19 ]. Our data can be summarized by the following observations: first, cell ELISA is a useful method for differentiating TBM from PTPM using the 30-kD protein antigen; second, the method of challenging B cells from CSF of suspected TBM patients with the 30-kD protein antigen can be helpful in confirming a TBM diagnosis; and third, the cell ELISA allows several samples to be analyzed simultaneously. Hence, the cell ELISA should be a very useful tool for the differential diagnosis of TBM from PTPM. Conclusion The presence of a 30-kD protein antigen in CSF of TBM patients indicates that this protein carries the candidate marker antigen which is specific to M. tuberculosis . We have demonstrated that by using cell ELISA, we can differentiate TBM patients from PTPM patients, which should be helpful for diagnosing TBM. Additionally our results suggest that lymphocytes from CSF of TBM patients when challenged with 30 kD protein give a quick response by producing IgG antibodies when compared with that of PTPM and healthy volunteers. This may be because lymphocytes from TBM patients have already been exposed to 30 kD MTB antigens. Competing interests The authors declare that they have no competing interests. Authors' contributions RSK carried out the study design, data collection, statistical analysis, data interpretation, literature search, and manuscript preparation; NPA, RPK, and RMS assisted in data analysis collection; NHC assisted in data collection, statistical analysis, and data interpretation; HJP participated in the preparation of the manuscript, data interpretation, and study design; GMT provided assistance in preparation of the manuscript, data interpretation, study design, and funds collection; and HFD supervised the study design, statistical analysis, data interpretation, manuscript preparation, and literature search. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529262.xml

526186

Radiotherapy fractionation for the palliation of uncomplicated painful bone metastases – an evidence-based practice guideline

Background This practice guideline was developed to provide recommendations to clinicians in Ontario on the preferred standard radiotherapy fractionation schedule for the treatment of painful bone metastases. Methods A systematic review and meta-analysis was performed and published elsewhere. The Supportive Care Guidelines Group, a multidisciplinary guideline development panel, formulated clinical recommendations based on their interpretation of the evidence. In addition to evidence from clinical trials, the panel also considered patient convenience and ease of administration of palliative radiotherapy. External review of the draft report by Ontario practitioners was obtained through a mailed survey, and final approval was obtained from the Practice Guidelines Coordinating Committee. Results Meta-analysis did not detect a significant difference in complete or overall pain relief between single treatment and multifraction palliative radiotherapy for bone metastases. Fifty-nine Ontario practitioners responded to the mailed survey (return rate 62%). Forty-two percent also returned written comments. Eighty-three percent of respondents agreed with the interpretation of the evidence and 75% agreed that the report should be approved as a practice guideline. Minor revisions were made based on feedback from the external reviewers and the Practice Guidelines Coordinating Committee. The Practice Guidelines Coordinating Committee approved the final practice guideline report. Conclusion For adult patients with single or multiple radiographically confirmed bone metastases of any histology corresponding to painful areas in previously non-irradiated areas without pathologic fractures or spinal cord/cauda equine compression, we conclude that: • Where the treatment objective is pain relief, a single 8 Gy treatment, prescribed to the appropriate target volume, is recommended as the standard dose-fractionation schedule for the treatment of symptomatic and uncomplicated bone metastases. Several factors frequently considered in clinical practice when applying this evidence such as the effect of primary histology, anatomical site of treatment, risk of pathological fracture, soft tissue disease and cord compression, use of antiemetics, and the role of retreatment are discussed as qualifying statements. Our systematic review and meta-analysis provided high quality evidence for the key recommendation in this clinical practice guideline. Qualifying statements addressing factors that should be considered when applying this recommendation in clinical practice facilitate its clinical application. The rigorous development and approval process result in a final document that is strongly endorsed by practitioners as a practice guideline.

Background Radiotherapy is a well-recognized, effective modality in the palliative treatment of painful bone metastases. Bone metastases are a common manifestation of distant relapse from many types of malignant tumours, especially from cancers of the lung, breast, and prostate. With the advent of effective systemic therapies and improvements in supportive care, cancer patients are expected to live longer and may suffer from metastatic disease for a considerable length of time. Many patients with bone metastases suffer from compromised mobility and performance status. The optimal dose-fractionation schedule for the treatment of bone metastases is unclear. Two surveys of Canadian patterns of practice found that various fractionation schedules are employed by radiation oncologists, ranging from a single large-dose fraction (e.g., 8 Gy) to a more prolonged course of 30 Gy/10 fractions over 2 weeks [ 1 , 2 ]. It has been suggested that the choice of fractionation is influenced not only by patient-related factors but also by physician education and attitudes, treatment toxicity, resource utilization, and departmental policy [ 3 - 7 ]. While different clinicians may associate " optimal " with different treatment goals, one could recommend that a " preferred " dose-fractionation is one that provides pain relief without undue toxicity and is least onerous to the patient. During the past decade, significant clinical trial efforts have been devoted to comparing single large-dose radiation (8 Gy to 10 Gy) with multifraction regimens (five to ten fractions) [ 8 - 14 ]. The two largest trials were published in 1999 by the Bone Pain Trial Working Party [ 10 ] and the Dutch Bone Metastasis Study group [ 11 ]. Results of a Canadian study were presented at the Canadian Association of Radiation Oncologists (CARO) meeting in 2000 and reported in an abstract for the 2000 meeting of the American Society for Therapeutic Radiology and Oncology (ASTRO) [ 9 ]. Those randomized trials should provide substantial evidence to address the question of a "preferred" fractionation for the majority of patients with bone metastases. This provincial guideline was initiated to summarize the evidence and to provide recommendations on the preferred standard radiotherapy fractionation schedule for the treatment of painful bone metastases. Clinical practice guideline development This practice guideline was developed by Cancer Care Ontario's Practice Guidelines Initiative (PGI), using the methods of the Practice Guidelines Development Cycle [ 15 ]. The practice guideline report is a convenient and up-to-date source of the best available evidence on the preferred dose-fractionation of radiotherapy for the treatment of uncomplicated painful bone metastases, developed through systematic reviews, evidence synthesis, and input from practitioners in Ontario. The report is intended to promote evidence-based practice. The PGI is editorially independent of Cancer Care Ontario and the Ontario Ministry of Health and Long-term Care. The PGI has a formal standardized process to ensure the currency of each guideline report. This process consists of the periodic review and evaluation of the scientific literature and, where appropriate, integration of this literature with the original guideline information. Evidence was selected and summarized by four members of the Supportive Care Guidelines Group (SCGG) and methodologists. Members of the SCGG disclosed potential conflict of interest information, reviewed the analysis of the evidence, and prepared draft recommendations. The SCGG includes palliative care physicians, medical and radiation oncologists, psychiatrists, nurses, psychologists, a chaplain, an anesthetist, a surgeon, methodologists, and administrators. After reviewing the evidence and considering issues of patient convenience and resource utilization, the SCGG reached consensus on draft recommendations. The systematic review and meta-analysis, conducted as the initial step in formulating this practice guideline, has been described elsewhere [ 16 ]. External review by Ontario practitioners was obtained through a mailed survey consisting of items that address the quality of the draft practice guideline report and recommendations and whether the recommendations should serve as a practice guideline. The efficacy of the practitioner feedback survey process has been previously described [ 17 ]. Final approval of the original guideline report was obtained from the Practice Guidelines Coordinating Committee (PGCC). Methods External review – Ontario practitioner feedback Practitioner feedback was obtained through a mailed survey of 95 radiation oncologists across Ontario. The survey consisted of items evaluating the methods, results, and interpretive summary used to inform the draft recommendation and whether the draft recommendation should be approved as a practice guideline. Written comments were invited. Follow-up reminders were sent at two weeks (post card) and four weeks (complete package mailed again). The SCGG reviewed the results of the survey. Results of practitioner feedback Fifty-nine of the 95 surveys were returned (62% return rate). Key results of the practitioner feedback survey are summarized in Table 1 . The survey results indicated that 83% of respondents agreed with the interpretation of the evidence and 74% agreed with the draft recommendations as stated. Seventy-five percent of respondents agreed that the report should be approved as a practice guideline. Twenty-one respondents (42%) also provided written comments. The final recommendation was revised to reflect feedback from practitioners and currently applies to patients for whom the treatment objective is pain relief. Practice guidelines coordinating committee approval process The practice guideline report was circulated to PGCC members for review and approval. Eleven of the fourteen PGCC members completed and returned ballots. Ten PGCC members approved the practice guideline report as written, and one member approved the guideline and provided suggestions for consideration by the SCGG. Suggestions made were to reword the Target Population section and to clarify the second qualifying statement. The SCGG agreed with the suggestions and modified the guideline accordingly. Discussion The preferred radiotherapy dose-fractionation schedule for the palliation of uncomplicated painful bone metastases has been a controversial subject [ 18 - 21 ]. The goal of our systematic review was to enable guideline developers and practitioners to determine whether the available evidence supports the notion of a "standard" dose-fractionation. " Standard " refers to what is applicable to the majority of patients, with a preference for patient convenience and ease of administration without compromising treatment efficacy or morbidity. Our meta-analysis of all published randomized trials found no difference in pain relief between single fraction and multifraction treatments [ 16 ]. Based on this information, the authors of this practice guideline conclude that a single fraction at 8 Gy is the preferred standard dose-fractionation for patients with uncomplicated painful bone metastases. In applying this evidence into practice, however, the following clinical factors merit consideration: 1. How durable is the pain relief? Available evidence does not support the notion that a more durable response can be achieved with higher dose-fractionation [ 11 ]. 2. Is the recommendation appropriate when preventing pathological response is an important consideration? There is a lack of firm evidence relating the fractionation schedule to the prevention of pathologic fracture because no study evaluated the risk of pathologic fracture prior to treatment. Although the pathologic fracture rate was significantly higher after single fraction radiotherapy than after multifraction in the Dutch study [ 11 ], the absolute difference was only 2%. The RTOG study, on the other hand, showed a higher fracture rate following high dose-fractionation (40 Gy) than low-dose treatment (20 Gy) in patients with a solitary metastasis [ 22 ]. Until CT (computed tomography)-based bone density measurements [ 12 ] are correlated with pathologic fractures, no evidence-based recommendation can be given. Patients at risk of pathologic fractures in long or weight-bearing bones should be assessed by an orthopaedic surgeon. Where radiotherapy is considered for tumour downsizing prior to an orthopaedic procedure or for such patients who are not surgical candidates, fractionated treatment (e.g., 20 Gy/5 fractions, 30 Gy/10 fractions) would be considered appropriate by many clinicians. A discussion of fracture risk assessment is beyond the scope of this review but has been published elsewhere [ 23 - 25 ]. 3. Does the recommendation apply to all pathologies? It should be noted that the published studies included a heterogeneous group of patients differing in histologies, performance status, severity of pain, extent of disease, and so forth. The fact that breast, prostate, and lung cancer patients constituted the majority of trial patients implies a greater confidence in reproducing treatment results for these patients in practice. However, the evidence does not provide sufficient materials to allow a recommendation based on treatment outcomes among subgroups of different primary tumours or other patient- and tumour-related factors. 4. Do treatment field size and anatomical location affect application of the recommendation? The evidence reviewed does not specifically address the results of large-volume (i.e. wide-field, hemibody irradiation), single fraction treatment. Although average treatment volumes were not reported in any of the single fraction trials, a significant proportion of patients did receive treatments to the lumbar spine and pelvis [ 10 , 11 , 13 ][Kirkbride P: Personal communications. 2001]. Since treatment volume was not an inclusion or exclusion criterion among those studies, it is reasonable to assume that study patients represent the majority of treatment volumes delivered in an average department. No difference in nausea and vomiting was seen in the subgroup of 133 patients from the Bone Pain Trial Working Party study, who were asked to self-assess nausea/vomiting experience in the first 14 days following treatments [ 10 ]. Therefore, the evidence does not support the choice of fractionated treatment based on volume consideration. However, the use of prophylactic ondansetron was shown to significantly reduce vomiting episodes in the single fraction arm compared with the 20 Gy arm (no prophylactic ondansetron) in the Canadian Bone Mets study [Kirkbride P: Personal communications. 2001]. For treatment over the epigastrium or lumbar spine, or with larger treatment volumes in the pelvis, it is reasonable to use a prophylactic antiemetic, as one would for hemibody irradiation. Patients may also be instructed to use anti-diarrheal agents if enteritis is experienced. 5. Do age and life expectancy affect application of the recommendation? The underlying concern for this group of patients is whether single large-dose radiation compromises subsequent tolerance to re-irradiation. Although no untoward late effects were reported by the single fraction studies with follow-up of one year or more [ 10 , 11 ], clinicians may be uneasy about the long-term effects of repeated radiation. Given the lack of evidence to the contrary, single fraction radiotherapy remains an appropriate treatment option in this subgroup. 6. Does the presence of soft tissue disease around bone metastases affect application of the recommendation? With CT/MRI (magnetic resonance imaging) diagnostic investigations becoming more routinely available, and the introduction of the CT-simulator into many departments, the extent of metastatic disease is likely to be better evaluated than in the past. In cases where lytic disease is associated with a large soft-tissue mass (e.g., in the acetabulum and adjacent pelvic bone), the desired palliative endpoint may be tumour shrinkage as well as pain control. No evidence-based recommendation can be given for this scenario. 7. When should re-irradiation be considered? Re-irradiation may be considered in three scenarios: 1) no pain relief or pain progression after initial radiotherapy, 2) partial response with initial radiotherapy and the hope of achieving further pain reduction with more radiotherapy, and 3) partial or complete response with initial radiotherapy but subsequent recurrence of pain. The response after re-irradiation may be different for each of these scenarios. Two published studies reported response rates to re-irradiation [ 26 , 27 ] with doses ranging from 4 Gy as a single dose to 30 Gy in 10 fractions over two weeks. The Dutch Bone Metastases Study [ 11 ] was re-analyzed to separate the response to initial treatment from the response to re-treatment. Van der Linden presented the results at the 2003 ASTRO meeting [ 28 ]. The majority of the 173 patients re-irradiated received single fraction 8 Gy as the initial treatment. Overall response rate to re-treatment was 63%. At present no clear guideline can be given regarding dose-fractionation of re-irradiation. A new intergroup randomized trial supported by the National Cancer Institute of Canada (NCIC) of single versus multiple fractions for re-irradiation opened in January 2004 and is expected to accrue patients from Canada, the United Kingdom, the Netherlands, and Australia [ 29 ]. 8. How is radiation-induced emesis best managed when delivering spinal irradiation? No increase in acute gastrointestinal morbidity was observed with single fraction treatment compared to multiple fractions. The Canadian study showed significantly fewer vomiting episodes with single fraction treatment after prophylactic ondansetron was used in cases with treatment fields over the abdomen or pelvis [Kirkbride P: Personal communications. 2001]. Antiemetic agents should be considered as a prophylaxis, given that 30% or more of patients experienced vomiting following single or multifraction treatment in the two studies that specifically collected patient-assessed nausea/vomiting data [ 9 , 10 ]. Conclusions For patients where the treatment objective is pain relief, a single 8 Gy treatment, prescribed to the appropriate target volume, is recommended as the standard dose-fractionation schedule for the treatment of symptomatic and uncomplicated bone metastases. This recommendation applies to adult patients with single or multiple radiographically confirmed bone metastases of any histology corresponding to painful areas in previously non-irradiated areas without pathologic fractures or spinal cord/cauda equina compression. It does not apply to the management of malignant primary bone tumour. The following qualifying statements are provided to support the application of the recommendation in clinical practice: • " Standard " refers to what is applicable to the majority of patients, with a preference for patient convenience and ease of administration and without compromising treatment efficacy or morbidity. • The recommendation does not apply to lesions previously irradiated, or lesions causing cord compression or pathologic fractures, because such patients were mostly excluded from clinical trials examining fractionation schedules. • Prophylactic antiemetic agents should be considered when a significant proportion of the gastrointestinal tract is in the irradiated volume. • Patients and referring physicians should be advised that repeat irradiation to the treated area may be possible. • There is insufficient evidence at this time to make a dose-fractionation recommendation for other treatment indications, such as long-term disease control for patients with solitary bone metastasis, prevention/treatment of cord compression, prevention/treatment of pathologic fractures, and treatment of soft tissue masses associated with bony disease. This practice guideline incorporates recommendations based on a systematic review, comprehensive consideration of how the evidence may be applied to clinical practice, feedback from Ontario practitioners, and input from the Practice Guidelines Coordinating Committee prior to final approval. It is strongly endorsed by practitioners for whom it was developed. List of Abbreviations Used ASTRO, American Society for Therapeutic Radiology and Oncology; CARO, Canadian Association of Radiation Oncologists; CT, computed tomography; Gy, gray(s); met, metastasis(es); MRI, magnetic resonance imaging; NCIC, National Cancer Institute of Canada; PGCC, Practice Guidelines Coordinating Committee; PGI, Practice Guidelines Initiative; RTOG, Radiation Therapy Oncology Group; SCGG, Supportive Care Guidelines Group. Competing interests The authors declare that they have no competing interests. Authors' contributions JW was the lead author responsible for designing and conducting the systematic review of the literature and the meta-analysis that informed the practice guideline, and for drafting and modifying the practice guideline report. JW was a member of the Supportive Care Guidelines Group and a Radiation Oncologist at the Juravinski Cancer Centre during the development of this guideline. RW reviewed all drafts of the guideline report and made major contributions to performing the meta-analysis that informed the practice guideline, and provided extensive input to the guideline as a radiation oncologist and methodologist. RW is co-Chair of the Supportive Care Guidelines Group. NL and MJ coordinated input from members of the SCGG, conducted literature searches, and drafted and edited the guideline report. MJ conducted duplicate data extraction and meta-analysis. NL updated the literature search, incorporated new data, conducted the practitioner feedback survey, and coordinated approval of the guideline by the Practice Guidelines Coordinating Committee. As members of the Supportive Care Guidelines Group, AB and TW provided substantial feedback on the guideline report at several points during its development, from both a radiation oncology and methodology perspective. Members of the SCGG provided feedback on all draft guideline reports. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526186.xml

544930

Loss of Sight and Enhanced Hearing: A Neural Picture

Stevie Wonder and Ray Charles are often cited as evidence that blindness confers superior musical ability. Wonder lost his sight after an incubator-related oxygen overdose during infancy; Charles lost his as a boy to glaucoma. It's impossible to know whether sight would have compromised their success, but many gifted musicians, from Jose Feliciano to Rahsaan Roland Kirk, lost their sight at an early age. A number of human studies show that blind persons perform nonvisual tasks better than those with sight. Neuroimaging studies of blind persons performing nonvisual tasks, including hearing, show activity in brain areas normally associated with vision. But much remains to be learned about the nature and extent of this phenomenon: how these “visual areas” are used, the mechanisms that generate individual differences (not all blind persons can localize sounds better than the sighted, for example), and the neural processes that underlie it. The task of localizing sound—which requires integrating information available to one ear only (monaural sounds available, for example, when one ear is plugged) or information derived from comparing sounds binaurally—is particularly suited to investigating the neural remapping that seems to follow vision loss. In a previous study, Franco Lepore and colleagues showed that people who lost their sight at an early age could localize sound, particularly from monaural cues, better than those who could see. These findings suggested that areas of the brain normally dedicated to processing visual stimuli (the visual cortex, located at the back of the brain in the occipital lobe) might play a role in processing sound in these individuals. In a new report, Lepore and colleagues use functional imaging studies to investigate the functional relationship between neural activity and enhanced hearing abilities in the blind, and find a strong correlation between superior sound localization skills and increased activity in the brain's visual center. The authors hypothesized that if visual cortex recruitment bolstered auditory function in some individuals, then visual cortex activity would correlate with individual differences in performance, and the degree of activity should predict such differences. Nineteen people—seven sighted and twelve who lost their sight at an early age—were placed in an echo-free chamber and asked to indicate where a sound was coming from, using either one (monaural) or both (binaural) ears. The participants then performed the same tasks within a positron emission tomography (PET) machine, which measures brain activity through changes in cerebral blood flow (CBF). Five of the blind participants could accurately localize sounds monaurally; most of the sighted could not. (All 19 participants had no trouble localizing binaural sounds.) Only the blind individuals with superior localization skills showed increased CBF in the visual cortex while performing monaural localization tasks. Interestingly, during binaural localization, the sighted participants showed decreased CBF in visual cortical areas. This decrease comports with previous studies showing that engaging one brain center—say, the temporal lobe, which processes sound—inhibits activation of others—such as the occipital lobe, which processes visual cues. These inhibitions appear to be absent in blind persons, though it's not clear why. It could be that blind persons don't need such inhibitions, the authors speculate, or maybe unrestricted access to the visual center serves to compensate for vision loss by boosting nonvisual senses. Whether the enhanced auditory performance reported here simply reflects increased efficiency of auditory processing or indicates “supranormal” powers, Lepore and colleagues argue that their results show that the visual cortex is “specifically recruited to process subtle monaural cues more effectively.” It will be interesting to learn whether blind persons can recruit visual centers for other auditory tasks or to help them navigate the world without sight. Such studies would be vital for tailoring sensory support to suit individual needs and maybe even suggest ways to facilitate the neural cross talk that enhances auditory performance. But don't expect such innovations to recreate the likes of Rahsaan Kirk or Ray Charles anytime soon.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544930.xml

176546

DNA Analysis Indicates That Asian Elephants Are Native to Borneo and Are Therefore a High Priority for Conservation

The origin of Borneo's elephants is controversial. Two competing hypotheses argue that they are either indigenous, tracing back to the Pleistocene, or were introduced, descending from elephants imported in the 16th–18th centuries. Taxonomically, they have either been classified as a unique subspecies or placed under the Indian or Sumatran subspecies. If shown to be a unique indigenous population, this would extend the natural species range of the Asian elephant by 1300 km, and therefore Borneo elephants would have much greater conservation importance than if they were a feral population. We compared DNA of Borneo elephants to that of elephants from across the range of the Asian elephant, using a fragment of mitochondrial DNA, including part of the hypervariable d-loop, and five autosomal microsatellite loci. We find that Borneo's elephants are genetically distinct, with molecular divergence indicative of a Pleistocene colonisation of Borneo and subsequent isolation. We reject the hypothesis that Borneo's elephants were introduced. The genetic divergence of Borneo elephants warrants their recognition as a separate evolutionary significant unit. Thus, interbreeding Borneo elephants with those from other populations would be contraindicated in ex situ conservation, and their genetic distinctiveness makes them one of the highest priority populations for Asian elephant conservation.

Introduction Elephants have a very limited distribution in Borneo, being restricted to approximately 5% of the island in the extreme northeast ( Figure 1 ). There are no historical records of elephants outside of this range. Fossil evidence for the prehistoric presence of elephants on Borneo is limited to a single specimen of a tooth from a cave in Brunei ( Hooijer 1972 ). Figure 1 Asian Elephant Range and Sampling Locations in Borneo Solid lines demarcate country borders and the dotted line the boundary between the Malaysian states of Sabah and Sarawak. Black dots indicate areas of sample collection. Popular belief holds that elephants presented to the Sultan of Sulu in 1750 by the East India Trading Company and subsequently transported to Borneo founded the current population ( Harrisson and Harrisson 1971 ; Medway 1977 ). These animals presumably originated in India ( Shoshani and Eisenberg 1982 ), where company operations and trade in domesticated elephants were centred. Alternatively, considering the geographic proximity to Borneo, the elephant trade that flourished in Sumatra and peninsular Malaysia during the 16th–18th centuries ( Andaya 1979 ; Marsden 1986 [1811]) may have been the source. Thus, if elephants were introduced to Borneo, the source population could have been India, Sumatra, or peninsular Malaysia, and as a feral population, Borneo's elephants would have low conservation importance. Conversely, if elephants occurred naturally on Borneo, they would have colonised the island during Pleistocene glaciations, when much of the Sunda shelf was exposed ( Figure 2 ) and the western Indo-Malayan archipelago formed a single landmass designated as Sundaland ( MacKinnon et al. 1996 ). Thus, the isolation of Borneo's elephants from other conspecific populations would minimally date from the last glacial maximum, 18,000 years ago, when land bridges last linked the Sunda Islands and the mainland ( MacKinnon et al. 1996 ). If Borneo's elephants are of indigenous origin, this would push the natural range of Asian elephants 1300 km to the east, and as a unique population at an extreme of the species' range, Borneo elephants' in situ conservation would be a priority and ex situ cross-breeding with other populations would be contraindicated. Figure 2 Asian Elephant Range and Sampling Locations Central sampling locations denote the countries sampled and represent a number of actual sampling locations within each country. 1. Sri Lanka, 2. India, 3. Bhutan, 4. Bangladesh, 5. Thailand, 6. Laos, 7. Vietnam, 8. Cambodia, 9. Peninsular Malaysia, 10. Sumatra (Indonesia) 11. Borneo (Sabah–Malaysia). Initially, Borneo elephants were classified as a unique subspecies ( Elephas maximus borneensis ) based on morphological differences from other populations ( Deraniyagala 1950 , 1955). Subsequently, they were subsumed under the Indian Elephas maximus indicus ( Shoshani and Eisenberg 1982 ) or the Sumatran Elephas maximus sumatrensis ( Medway 1977 ) subspecies, based on an assumption of their introduction to the region or on the reasoning that morphological divergence was insufficient to warrant separate status. While unique subspecific status would highlight their conservation importance, evaluation of their status in terms of evolutionary significant units (ESUs) and management units (MUs) ( Ryder 1986 ; Moritz 1994 ) would be more relevant to conservation management. Results We PCR-amplified and sequenced a 630 bp fragment of mitochondrial DNA (mtDNA), including the hypervariable left domain of the d-loop ( Fernando et al. 2000 ), from 20 Borneo elephants and compared them with 317 sequences we generated for elephants across ten of the 13 Asian elephant range states ( Figure 2 ). Asian elephant haplotypes segregated into two distinct clades, α and β ( Fernando et al. 2000 ). All ‘Sundaland’ (peninsular Malaysia, Sumatra, and Borneo) haplotypes fell in clade β, while α and β clades were observed in Sri Lanka and mainland populations ( Figures 3 and 4 ). The Borneo population was fixed for the unique β-haplotype BD. Similar tree topologies were obtained by maximum parsimony, neighbour joining, and maximum-likelihood methods of phylogenetic analyses, with some minor rearrangements of the terminal branches. In all trees, Bornean and other haplotypes unique to ‘Sundaland' (Borneo: BD; peninsular Malaysia: BQ, BV; Sumatra: BS, BU, BT, BR) occupied basal positions in the β-clade phylogeny ( Figure 3 ) and were derived from internal nodes in a parsimony network of haplotypes ( Figure 4 ). Uncorrected p distances between the Borneo haplotype and other β-haplotypes ranged from 0.012 (haplotypes BQ, BP, BO, BS, BU) to 0.020 (haplotype BE), with a mean of 0.014. Assuming a nucleotide substitution rate of 3.5% per million years for the elephant mtDNA d-loop ( Fleischer et al. 2001 ), the observed genetic distance indicates divergence of the Borneo haplotype BD and its closest relative from a common ancestor approximately 300,000 years ago. Owing to stochastic coalescent processes, the use of a single gene to infer population parameters is prone to error. Despite any such error, the magnitude of the genetic difference between Borneo and other Asian elephant haplotypes is such that it indisputably excludes divergence since introduction; the observed divergence is so great that even if there was some error it would not have any influence on the conclusion that places the Borneo haplotype in a timeframe supporting a Pleistocene colonisation rather than introduction by humans. Figure 3 A Neighbour-Joining Phylogram of Asian Elephant Haplotypes Rooted with an African Elephant Out-Group Sunda Region haplotypes are in bold. Figure 4 Network of Asian Elephant Haplotypes Based on Statistical Parsimony Grey circles with letters denote haplotypes unique to the Sunda region (BD: Borneo; BQ, BV: peninsular Malaysia; BR, BS, BT, BU: Sumatra). White circles with letters denote haplotypes found in mainland Asia (excluding peninsular Malaysia) and Sri Lanka. The small open circles denote hypothetical haplotypes. Haplotypes beginning with the letters A and B belong to the two clades α and β, respectively. We also genotyped 15 Borneo elephants for five polymorphic autosomal microsatellite loci ( Nyakaana and Arctander 1998 ; Fernando et al. 2001 ) and compared them to 136 five-locus genotypes we generated for Asian elephants from nine range states. Tests of Hardy–Weinberg equilibrium and linkage disequilibrium in all populations indicated simple Mendelian inheritance of five unlinked, selectively neutral loci. The total number of alleles per locus across populations in the Asian elephant ranged from 2.0 ( EMX-2 ) to 11.0 ( LafMS03 ) ( x¯ , SE = 4.60, 1.51); the average number of alleles across loci, per population (excluding Borneo), from 2.0 (Sumatra) to 3.6 (Sri Lanka) ( x¯ , SE = 2.93, 0.155); the observed heterozygosity H 0 across all populations (excluding Borneo) from 0.38 ( EMX-4 ) to 0.63 ( LafMS03 ) ( x¯ , SE = 0.44, 0.041); and gene diversity from 0.39 ( EMX-4 ) to 0.69 ( LafMS03 ) ( x¯ , SE = 0.47, 0.050). Comparatively, all indices demonstrated very low genetic diversity in the Borneo population: proportion of polymorphic loci, 0.4; number of alleles per locus, 1–2 ( x¯ , SE = 1.40, 0.219); gene diversity, 0–0.13 ( x¯ , SE = 0.04, 0.024); heterozygosity H 0 = 0–0.07 ( x¯ , SE = 0.01, 0.013). The number of alleles, observed heterozygosity, and gene diversity, averaged across Asian elephant populations, were all higher than those in Borneo, at all loci ( Table 1 ). Similarly, in all populations, the number of alleles and observed heterozygosity, averaged across loci, were higher than in Borneo ( Table 2 ). Five unique genotypes were identified in the 15 Borneo elephants sampled. In tests of population subdivision, all pairwise comparisons between Borneo and other populations demonstrated highly significant differentiation, F ST 0.32–0.63 ( x¯ , SE = 0.44, 0.034) ( Table 3 ). In tests of a recent bottleneck, no heterozygote excess ( Maruyama and Fuerst 1985 ) or mode-shift distortion of allele frequency distributions ( Luikart et al. 1998a ), characteristic of a recent bottleneck, was observed in the Borneo population. In assignment tests indicating the distinctness of a population's genotypes, all five Borneo genotypes were assigned with maximum likelihood to Borneo (likelihoods ranging from 0.004 to 0.80, x¯ , SE = 0.51, 0.175), and maximum-likelihood ratios of the most-likely (Borneo) to the next-most-likely population ranged from 2.97 to 48.20 ( x¯ , SE = 25.02, 8.795). Borneo was significantly more likely to be the source than any other population for all five genotypes, since each of the assignment likelihoods to Borneo fell outside the upper end of the corresponding distribution of assignment likelihoods to the other populations. Assignment likelihoods to the putative Indian, Sumatran, and peninsular Malaysian source populations were very small (India: 0–0.0004, x¯ , SE = 0.000126, 0.000065; Sumatra: 0–0.0355, x¯ , SE = 0.007146, 0.006336; peninsular Malaysia: 0.0003–0.1195, x¯ , SE = 0.0301, 0.0201), indicating that Borneo's genotypes were highly unlikely to have originated from any of these populations. Table 1 Comparison of Measures of Genetic Variation at Individual Loci in Borneo with Those of the Other Populations Table 2 Measures of Genetic Variation Using Five Loci, in Asian Elephant Populations from across the Range Table 3 F ST Values in Pairwise Comparison of Borneo with Other Populations Discussion mtDNA evidence supports an indigenous hypothesis in three ways. First, this hypothesis assumes an ancient, independent evolution of Borneo's elephants, resulting in the unique, divergent Borneo haplotype(s), as we observed. Conversely, the introduction hypothesis assumes an introduction at 500 years ago or less, which approximates zero time on a scale of mtDNA d-loop evolution, and hence requires Borneo and source population haplotypes to be identical. This was not observed. Second, the estimated divergence time between the Borneo haplotype and other Asian elephant haplotypes is concordant with a mid- to late-Pleistocene isolation of elephants on Borneo and the vicariant history of the island ( MacKinnon et al. 1996 ). Third, all observed ‘Sundaland' haplotypes, including Borneo's, were of the β clade, had basal relationships to that clade in a phylogenetic tree, and were independently derived from internal nodes in a haplotype network, suggesting an ancient isolation of these lineages on Borneo, Sumatra, and peninsular Malaysia. Thus, the Borneo haplotype fits a pattern of distribution and relatedness to other ‘Sundaland' haplotypes that is congruent with an ancient colonisation of the Sunda region by β clade and subsequent allopatric divergence of populations on its larger landmasses. Microsatellite data also support the indigenous hypothesis. If the Borneo population originated from animals introduced in the 16th–18th centuries, it would have reached its mid-20th-century size of approximately 2,000 individuals ( deSilva 1968 ) in fewer than 30 generations, assuming an Asian elephant generation time of 15–20 years ( Sukumar 1989 ). Thus, the Borneo population would have experienced a rapid demographic expansion after the ‘recent’ bottleneck caused by the founder-event of introduction. We did not observe a heterozygote excess or a mode-shift distortion in allele frequency distribution in the Borneo population, suggesting that the population did not undergo a recent bottleneck and hence did not arise from a few introduced animals. However, this result by itself is not conclusive, since with a sample size of 15 and five loci, the test for heterozygosity excess has low power and bottlenecks may not be detected ( Luikart et al. 1998b ). We observed extremely low genetic diversity at Borneo elephant microsatellite loci, including fixation at three of the five loci. Sequential founder-events or persistent small population size, as would be expected in a small population isolated since the Pleistocene, would lead to substantial loss of genetic variation ( Nei et al. 1975 ) and hence is consistent with the data. Successful founding of a population by a very few individuals from a single introduction could also result in a severe bottleneck. However, given the adversities faced by translocated elephants ( Fernando 1997 ) and the importance of social structure in the reproduction and survival of elephants ( Fernando and Lande 2000 ; McComb et al. 2001 ), such an explanation is unlikely. In the assignment tests, all five Borneo genotypes, which included free-ranging as well as captive animals, were assigned to Borneo with significantly higher likelihoods than to other populations and with extremely low likelihoods to the putative source populations. An introduced population may be highly divergent from the source population in terms of F statistics ( Williams et al. 2002 ) due to allelic loss from founder-events. However, the probability of loss for a particular allele is inversely proportional to its frequency in the founder and hence the source population. Thus, genotypes in an introduced population would retain a high likelihood of assignment to the source population, enabling its identification from among a number of candidate populations. Therefore, the assignment tests strongly suggest that the Borneo elephants were not derived from another population in the recent past. Thus, microsatellite data strongly suggest a Pleistocene colonisation, independent evolution through a long period of isolation, and long-term small population size for the Borneo population. It strongly rejects a recent origin from any of the putative source populations. Mitochondrial and microsatellite analyses indicate that Borneo's elephants are indigenous to Borneo, have undergone independent evolution since a Pleistocene colonisation, and are not descended from animals introduced by humans. The evolutionary history of Borneo's elephants warrants their recognition as a separate ESU ( Moritz 1994 ). Thus, they should not be cross-bred with other Asian elephants in ex situ management. The genetic distinctiveness and evolutionary history of Borneo elephants support their recognition as a unique subspecies. However, one of the reasons E. maximus borneensis was subsumed under E. m. indicus and E. m. sumatrensis was the inadequacy of the original description of E. m. borneensis in terms of the morphological characters assessed and sample size. Therefore, we suggest that a formal reinstatement of the E. m. borneensis taxa await a detailed morphological analysis of Borneo elephants and their comparison with other populations. While Borneo's elephants appear to be genetically depauperate, through a long history of isolation and inbreeding, they may have purged deleterious recessive alleles from their genome and decreased their genetic load, thus becoming less susceptible to inbreeding depression. We recommend research on reproductive rates, juvenile survival, and other indicators of detrimental effects of inbreeding such as sperm deformities, sperm mobility, and genetic diversity at MHC loci. While increasing genetic diversity by introducing a small number of elephants from other populations ( Whitehouse and Harley 2001 ) may have to be considered if deleterious inbreeding effects are evident, in the absence of such findings Borneo's elephants should be managed separately from other Asian elephants. Materials and Methods Samples. Samples consisted of dung from free-ranging and dung or blood from captive elephants. Sample collection, storage, and DNA extraction followed published protocols ( Fernando et al. 2000 , 2003 ). For mitochondrial and microsatellite analysis, respectively, 20 and 15 samples from Borneo (nine blood samples from elephants captured for management purposes—eight from the Kretam area and one individual originating from around Lahad Datu—and the rest from dung samples from free-ranging elephants collected during a survey of the Kinabatangan watershed) were compared with 317 and 136 samples from across the current Asian elephant range, Sri Lanka (n = 81, 20), India (n = 81, 20), Bhutan (n = 13, 13), Bangladesh (n = 30, 20), Thailand (n = 8, 8), Cambodia (n = 30, 20), Vietnam (n = 5, 0), Laos (n = 20, 6), Indonesia (Sumatra) (n = 40, 20), and peninsular Malaysia (n = 9, 9). Vietnam was excluded from the microsatellite analysis owing to nonamplification of a number of samples. mtDNA amplification and sequencing. Approximately 630 bp of mtDNA, including the left domain of the d-loop, were amplified using published primers ( Fernando et al. 2000 ). PCR products were sequenced in both directions, using internal sequencing primers MDLseq-1 (CCTACAYCATTATYGGCCAAA) and MDLseq-2 (AGAAGAGGGACACGAAGATGG), and resolved in 4% polyacrylamide gels in an ABI 377 automated sequencer (Perkin-Elmer, Wellesley, Massachusetts, United States). mtDNA phylogenetic analysis. We used 600 bp of the amplified segment in the analysis. Sequences were aligned and edited using SEQUENCHER version 3.1.1 (GeneCodes Corporation, Ann Arbor, Michigan, United States). Sequences were deposited in GenBank (accession numbers AY245538 and AY245802 to AY245827). Phylogenetic analyses were conducted using PAUP* version 4.0 ( Swofford 1998 ). Three African elephant ( Loxodonta africana ) sequences from zoo animals in the United States were used as an out-group. Genetic distances among sequences were calculated using uncorrected p distances. Maximum-parsimony analysis was conducted using a heuristic search with random stepwise addition of taxa, tree bisection/reconnection branch swapping, and equal weighting; neighbour joining, with Kimura two-parameter distances; and maximum likelihood, using empirical base frequencies and estimated values for the shape parameter for among-site rate variation and transition/transversion ratios. A network of haplotypes was created using statistical parsimony in the software TCS version 1.13 ( Clement et al. 2001 ). Microsatellite amplification. Samples were screened with five published microsatellite loci, EMX-1 to EMX-4 ( Fernando et al. 2001 ) and LafMS03 ( Nyakaana and Arctander 1998 ). Forward primers were fluorescent labelled (FAM, HEX, or TET), samples were amplified in 12.5 μl volumes with relevant cycling profiles ( Fernando et al. 2001 ), and 1 μl of PCR product was mixed with 0.2 μl of loading-dye and 0.5 μl of Tamra 500 size standard (Applied Biosystems, Foster City, California, United States) and was resolved in 4% polyacrylamide gels in an ABI 377 automated sequencer. Alleles were scored using GENESCAN software (Applied Biosystems) and published guidelines ( Fernando et al. 2003 ). Microsatellite data analysis. Deviations from Hardy–Weinberg equilibrium for each locus and population were tested using the exact Hardy–Weinberg test as implemented in GENEPOP 3.2 ( Raymond and Rousset 1995 ), with the complete enumeration method ( Louis and Dempster 1987 ) for loci with fewer than four alleles and with the Markov chain method ( Guo and Thompson 1992 ) (dememorization: 1000; batches: 100; iterations per batch: 1000) for loci with more than four alleles. GENEPOP was also used to test for linkage disequilibrium between loci, using the Markov chain method. Population differentiation was tested with estimates of Wright's fixation index ( Weir and Cockerham 1984 ), F ST , using the program Arlequin version 2 ( Schneider et al. 2000 ). Evidence for a recent bottleneck in the Borneo population in terms of a heterozygote excess ( Cornuet and Luikart 1996 ) or a mode-shift distortion in allele frequencies ( Luikart et al. 1998a ) was conducted using the program BOTTLENECK version 1.2.02 ( Piry et al. 1997 ) and a graphical method ( Luikart et al. 1998a ). Assignment tests were performed using WHICHRUN version 4.1 ( Banks and Eichert 2000 ). Assuming Hardy–Weinberg equilibrium in each baseline population and linkage equilibrium between loci, the likelihood that an individual originates from a particular population is the Hardy–Weinberg frequency of the individual's genotype at that locus, in that population. This likelihood was multiplied across loci to obtain a multilocus assignment likelihood of the test individual to each population, and the population with the highest value was identified as the ‘most-likely’ source population. To test for statistical significance of the most-likely source population, this assignment likelihood was compared with the distribution of assignment likelihoods of the other populations. Maximum-likelihood ratios were calculated as the ratio between the likelihood of assignment to the most-likely population to that for a particular population. Supporting Information Accession Numbers The GenBank accession numbers for the sequences reported in this paper are AY245538 and AY245802 to AY245827.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC176546.xml

548677

Infiltrative microgliosis: activation and long-distance migration of subependymal microglia following periventricular insults

Background Subventricular microglia (SVMs) are positioned at the interface of the cerebrospinal fluid and brain parenchyma and may play a role in periventricular inflammatory reactions. However, SVMs have not been previously investigated in detail due to the lack of a specific methodology for their study exclusive of deeper parenchymal microglia. Methods We have developed and characterized a novel model for the investigation of subventricular microglial reactions in mice using intracerebroventricular (ICV) injection of high-dose rhodamine dyes. Dynamic studies using timelapse confocal microscopy in situ complemented the histopathological analysis. Results We demonstrate that high-dose ICV rhodamine dye injection resulted in selective uptake by the ependyma and ependymal death within hours. Phagocytosis of ependymal debris by activated SVMs was evident by 1d as demonstrated by the appearance of rhodamine-positive SVMs. In the absence of further manipulation, labelled SVMs remained in the subventricular space. However, these cells exhibited the ability to migrate several hundred microns into the parenchyma towards a deafferentation injury of the hippocampus. This "infiltrative microgliosis" was verified in situ using timelapse confocal microscopy. Finally, supporting the disease relevance of this event, the triad of ependymal cell death, SVM activation, and infiltrative microgliosis was recapitulated by a single ICV injection of HIV-1 tat protein. Conclusions Subependymal microglia exhibit robust activation and migration in periventricular inflammatory responses. Further study of this population of microglia may provide insight into neurological diseases with tendencies to involve the ventricular system and periventricular tissues.

Background It has become increasingly evident that the central nervous system is an immunocompetent organ [ 1 ]. Microglia are the primary immune effector cells of the brain parenchyma and functionally resemble tissue macrophages elsewhere in the body [ 1 , 2 ]. The brain ventricles are also under immune surveillance by intraventricular macrophages which patrol the cerebrospinal fluid (CSF), choroid plexus, and supraependymal surface [ 3 ]. At the interface of the CSF and brain proper is the ciliated neuroepithelial ependymal cell which lines the ventricular system of the brain and spinal canal. The ependyma not only functions as a physical barrier preventing foreign proteins and organisms from entering the brain from the CSF, but also displays immunological effector ability such as phagocytosis of fluorescent beads injected into the CSF [ 4 ] and upregulation of MHC-II in response to interferon gamma challenge in vivo [ 5 ]. These diverse cell types may work in concert establishing the basis for the innate immune system of the CNS. Importantly, a population of resident subventricular microglia (SVMs) are found in the subependymal zone [ 6 , 7 ] suggesting the ependyma and microglia may cooperate to prevent invasion of the CNS from the ventricular system. Juxtaventricular microglia have been shown to react to both direct periventricular/ependymal damage as well as the mere presence of cytokines in the CSF. For instance, the intracerebroventricular (ICV) injection of lentiviral tat protein in low nanomolar quantities is sufficient to kill ependymal cells and cause a periventricular inflammatory reaction including the characteristic microglial nodules of human HIV-1 encephalopathy (HIVE) [ 8 ]. Alternatively, Kong et al [ 9 ], noting the high CSF levels of inflammatory cytokines in multiple sclerosis (MS), demonstrated a vigorous periventricular activation of microglia with ICV injection of IFN-γ alone or in combination with endotoxin or TNF-α. In these cases, microglia were activated in the absence of primary tissue damage, but were thought to contribute secondarily to immune-mediated periventricular damage via potentiation of cytokine release and bystander effect. Of note, both HIVE and MS often present with enigmatic periventricular inflammatory lesions in humans [ 10 - 13 ]. The reaction of SVMs to periventricular damage has not been described in detail. Specifically, the functional repertoire displayed by activated SVMs including phagocytosis and long-distance migration have not been investigated. Here, we directly examine SVM function by exploiting a novel methodology which selectively activates and labels SVMs in vivo combined with confocal timelapse techniques for dynamic analyses in adult mouse brain tissue. We hypothesized that SVM activation is a general consequence of periventricular insults that can be caused by diverse circulating substances in the CSF known to damage the ependyma. Our work indicates that activated, phagocytic SVMs are capable of infiltrating deep within the parenchyma. Methods Animals Adult male C57bl/6 mice (6–8 wks) were obtained commercially from Harlan (Indianapolis, IN) and cared for in accordance with Public Health Service and University of Virginia guidelines. Surgical procedures To damage the ependyma 2–3 μl of 0.2% Sp-DiI (D-7777, Molecular Probes) in DMSO; 1:20 rhodamine-conjugated latex microspheres (Lumafluor, Naples, FL) in sterile PBS; 0.25 U Neuraminidase (Sigma); or 2.0 nM recombinant HIV-1 tat protein in 100 mg/ml BSA, 0.1 mM DTT in PBS were stereotaxically injected into the left lateral ventricle at the following coordinates: L, 1.5 mm; P, -0.5 mm; D, 2.0 mm. GFP-expressing adenovirus (10 9 plaque forming units); 100 mg/ml BSA, 0.1 mM DTT in PBS, or deactivated tat [ 14 ] were used as volume-matched controls. Forebrain stab lesion for deafferenting lesion of the contralateral hippocampus was performed as previously described [ 15 ]. Briefly, mice were anesthetized with a Xylazine/Ketamine mixture and placed in a stereotaxic head holder (Benchmark, myneurolab.com). Temperature was maintained with a ventral heat pad. A right parietal craniectomy extending 3–4 mm from midline and spanning Lambda to Bregma was created with a microdrill. Beginning at the level of Bregma, a 3 mm lesion was created in the sagittal plane 1.5 mm from midline at a depth of 3.5 mm with a sterile no. 11 scalpel blade held in the stereotaxic device. After achieving hemostasis, the bone was replaced and sealed with carboxylate cement (Durelon, Fisher Sci). This right sided lesion results in deafferentation injury to primarily stratum oriens of the left hippocampus. Static analyses Brains were collected and processed for histopathology as previously described [ 15 ]. All antibodies and staining procedures have been described previously [ 15 ] except for rat anti-cd11b/MAC-1 (1:50), mouse anti-foxj1 (1:1000), mouse anti-nestin/rat-401 (1:100), and rat anti-F4/80 (1:10). Histochemistry for biotinylated Griffonia simplicifolia lectin IB 4 was performed 1:100 in PBS overnight at 4°C and visualized with either 1:200 FITC- or Alexa Fluor 350-conjugated streptavidin (Sigma and Molecular Probes, respectively). Dynamic analyses 200–400 μm live slices were prepared from adult C57BL/6 mice as described previously [ 16 , 17 ]. Briefly, mice were acutely anesthetized in a chamber with halothane and decapitated. The brain was rapidly removed, blocked, and covered with 10% agar at 37°C in a specimen mold (Tissue-Tek 4566, Fisher). Live slices were obtained with a vibratome and placed individually on Millicell-CM inserts (Millipore PICM03050, Fisher). Culture medium consisted of CCM1 (Hyclone, Logan, UT) with 20% heat-inactivated normal horse serum. Vital labelling of microglia in tat experiments was performed with Alexa 488 or 568 IB 4 (Molecular Probes) [ 17 ]. Laser-scanning confocal images were acquired on a Nikon IX-70 inverted microscope with Fluoview 300 software (Olympus). A z-series stack covering 40 μm of slice thickness was taken every 1.5–4 minutes, creating a three-dimensional timelapse data set. To create timelapse movies from the data set, 4 to 6 z-plane images were collapsed as 2D projections using ImageJ 1.31 u and compiled into quicktime movies with Quicktime Pro 6.3. Movies were analyzed for migration speed and distance as described [ 18 ]. Values represent the mean ± SEM. Statistical analysis was performed with ANOVA or student's t -test. Pairwise post-hoc analysis was performed with a t -test and the Bonferroni correction factor. A p < 0.05 was considered statistically significant. Rose plots were created in Kaleidagraph 3.0. Results Selective ependymal death is induced by high-dose rhodamine dyes Rhodamine dyes such as SP-DiI and rhodamine latex microbeads (RhoB) have been classically used for tract tracing studies in vivo and in fixed tissues [ 19 , 20 ]. Recently, intracerebroventricular (ICV) injection of these dyes into the CSF has also been shown to selectively label ependymal cells [ 4 , 21 ]. In pilot studies for other projects, we discovered doses of these dyes that, in addition to labelling, result in selective death and denudation of the ependyma (not shown). To investigate the timecourse of ependymal damage in response to rhodamine dyes we injected a toxic bolus of SP-DiI (0.2% in DMSO) or rhodamine latex microbeads (1:20 in PBS) into the left lateral ventricle of mice. Overt ependymal damage was appreciated beginning by 12 h (Fig. 1a , left panel) post-injection and progressed rapidly by 24 h (Fig. 1a , middle panel) where the ependyma appeared swollen and ragged with frequent pyknotic profiles (Fig. 1b ). At these doses, damage was largely restricted to the lateral ventricle ipsilateral to the injection and the third ventricle (not shown) while sparing the contralateral lateral ventricle (Fig. 1b, c ) suggesting a dose- and diffusion-dependent toxicity. Near complete ependymal cell loss occured in regions of the lateral ventricle proximal to the injection site within 3 days with both dyes (Fig. 1A , right panel & 1D ). Mild subependymal astrogliosis as revealed by nestin immunohistochemistry was also evident by 3d (Fig. 1E ). The loss of ependyma was further confirmed by chronic loss of immunoreactivity for the ciliated cell-specific transcription factor foxj1 at 1 month after injection (Fig. 1f ). Animals injected with equal volumes of GFP-reporter adenovirus (Fig. 1g ) or low-dose rhodamine microbeads in PBS (1:50, not shown) demonstrated no ependymal loss or activation of subventricular microglia (SVMs, not shown). Thus, rhodamine dyes rapidly and selectively damage ependymal cells at high doses. Figure 1 Ependymal damage with rhodamine dyes. (A) Timecourse of ependymal death in the lateral ventricle after rhodamine dye injection demonstrated with digital subtraction. Damage to the ependyma was evident at 12 h and rapidly progressed by 24 h. (B) Histology at 24 h demonstrates swollen ependyma with numerous pyknotic profiles in injected, but not the contralateral, hemisphere. e, ependyma; lv, lateral ventricle; p, parenchyma. RHO fluorescence overlaid on brightfield hematoxylin images. (C) Low-power view of lateral ventricles 3 d after injection demonstrates halo of rhodamine-positive cells around injected ventricle (white arrow). The contralateral ventricle demonstrates labelled ependyma in the absence of damage. (D) By 3 d, near-complete loss of the ependyma was evident. This coincided with the appearance of dye-laden SVMs, black arrowheads. The ependyma remained intact in the contralateral hemisphere (right panels). e, ependyma; lv, lateral ventricle; p, parenchyma; RhoB, rhodamine beads. RHO fluorescence overlaid on brightfield hematoxylin image (RhoB) and photoconverted DiI counterstained with hematoxylin. (E) Periventricular reactive astrocytes (black arrows) visualized with nestin immunohistochemistry (IHC) at 3d post-injection at wall of injected ventricle (left), but not in the contralateral hemisphere (right). lv, lateral ventricle; sp, septum. (F) IHC for ciliated cell-specific foxj1 28d after dye injection demonstrates persistent loss of ependyma in injected hemispere (left). cc, corpus callosum, cp, caudate/putamen; sp, septum. (G) Equivalent volume control injection of GFP-reporter adenovirus demonstrates no ependymal damage 3 weeks after injection. e, ependyma; lv, lateral ventricle; p, parenchyma. GFP fluorescence overlaid on brightfield hematoxylin image. SVMs phagocytose ependymal debris Loss of the ependyma in the above areas coincided with the appearance of dye-laden periventricular cells resembling macrophages (Fig. 1d , arrowheads). At low power, affected ventricles were surrounded by a halo of these rhodamine-positive (RHO+) cells (Fig. 1a , last panel; 1c ). To determine the identity of the RHO+ cells we performed transmission electron microscopy (Fig. 2a ), immunohistochemistry for microglial/macrophage markers F4/80 (Fig. 2b ) and MAC-1/cd11b (not shown), and histochemistry for IB 4 lectin from Griffonia simplicifolia (Fig. 2b ). These techniques demonstrated periventricular RHO+ cells to be microglia. Figure 2 Selective labelling of SVMs with rhodamine dyes. (A) RHO+ cells are microglia. Transmission electron microscopy demonstrates dye-laden inclusions (white arrows) in a SVM. n, nucleus. (B) Immunohistochemistry for F4/80 (top) and histochemistry for lectin IB 4 (bottom) demonstrate double-labelled periventricular cells, white arrows. (C) Time-lapse confocal microscopy in live brain slices demonstrates SVM (white arrow) extending (time 0' and 9') and retracting (time 4.5' and 13.5') a process toward ependymal debris (yellow star) highly suggestive of phagocytosis. See also Video 1. lv, lateral ventricle; p, parenchyma. (D) Neuraminidase injection following sublethal ependymal labelling similarly results in RHO+ SVMs (black arrows). e, ependyma; lv, lateral ventricle; p, parenchyma. Left panels, hematoxylin; Right panels, RHO fluorescence overlaid on hematoxylin. SVMs may become RHO+ as a result of phagocytosis of the dye-labeled ependymal debris. To provide direct evidence of this hypothesis we performed confocal time-lapse microscopy in living slices from adult mice given dye injection 24 h prior to sacrifice. Grossly, we observed a dramatic increase in RHO+ periventricular cells over several hours suggesting active clearance of labelled debris (not shown). Further, SVMs displayed dynamic behavior consistent with phagocytosis of ependymal debris (Fig. 2c , Video 1(Additional file 1 )). Therefore, SVMs became rhodamine positive after high-dose dye injection due to phagocytosis of labelled ependymal debris. To determine if SVM activation is a general response to periventricular damage we injected animals with a sublethal dose of RhoB to label ependymal cells without causing damage. 24 h later we injected 0.25 U neuraminidase to damage the ependyma [ 22 ] via an alternate mechanism. Histological examination of sections at 7 d revealed loss of the ependyma and the presence of RHO+ SVMs (Fig. 2d , left panels). Control injection of PBS vehicle alone resulted in no ependymal damage or SVM labelling (Fig. 2d , right panels). Thus, ependymal cell damage of diverse etiologies incites a reactive response by SVMs including phagocytosis of debris. Long-distance infiltration by SVMs after parenchymal injury Unpurturbed, SVMs remained in the immediate periventricular vicinity with no deeper parenchymal migration (Fig. 1c ). Indeed, his population was stable in animals sacrificed up to 30d following dye injection (not shown). Periventricular lesions in HIVE and MS often extend deep into the parenchyma suggesting long-distance infiltration of reactive cells. Microglia have been shown to migrate in vitro in response to many chemokines and growth factors present in brain lesions and plaques [ 23 - 26 ]. In order to investigate whether activated SVMs can migrate towards parenchymal brain damage in vivo we gave mice a deafferenting lesion (FSL) of the hippocampus 24 hours following rhodamine dye injection and allowed survival for up to 28 days. Invasion of the parenchyma by RHO+ cells occurred in the stratum oriens of the denervated hippocampus (cSO) in 21/21 animals but not in sham animals (0/4) (Fig. 3a ). Infiltrating cells were found an average of 849 ± 34 μm from the lateral ventricle after FSL compared to 210 ± 16 μm in uninjured mice (p < 0.01, Fig. 3b ). Based on population distribution histograms, greater than 75% of RHO+ cells in sham animals were found within 300 μm of the ventricles (maximum: 860 μm) whereas greater than 75% were found beyond 400 μm (maximum: 2377 μm) in injured mice. Temporal quantification of RHO+ cell infiltration in the cSO demonstrates that this event commences between 1 and 3 days post-injury (PI) and peaks at 5 days PI (Fig. 3c ). This timecourse mirrors that of the appearance of degeneration debris (Fig. 3d , GSD), activated resident hippocampal microglia (Fig. 3d , IB4), and reactive gliosis in the cSO (Fig. 3d , pERK [ 15 ]) supporting migration of RHO+ cells towards injury cues [ 24 , 25 ]. 94.6% of all RHO+ cells in the cSO were immunoreactive for F4/80 confirming the infiltrating cells are microglia. Finally, BrdU-positive/RHO+ cells were observed in the cSO maximally at the 3 day timepoint suggesting mitosis occurred primarily after the SVMs had migrated to the hippocampus (not shown). Therefore, activated SVMs are capable of infiltrating deep into the parenchyma in response to brain injury. Figure 3 Infiltration of parenchyma by SVMs after injury. (A) SVMs infiltrated the stratum oriens of the hippocampus in injured mice (right panel) but not in sham animals (left panel). cSO, contralateral stratum oriens of hippocampus; ffx, fimbria/fornix; lv, lateral ventricle; th, thalamus. (B) SVMs migrate significantly farther into parenchyma of injured animals compared to sham injury (*p < 0.01). (C) Infiltration of hippocampus begins days after injury and cells remain for weeks (*p < 0.05 compared to sham). (D) Temporal pattern of infiltration corresponds to neuropil degeneration (black punctate staining, bottom left) activation of resident microglia (shown by increased IB4 staining, bottom middle) and glial activation (indicated by phospho-ERK immunoreactivity, bottom right). GSD, Gallyas silver degeneration stain; pERK, phospho-extracellular signal-related kinase. To provide direct evidence for the migration of SVMs into the parenchyma and characterize their general migratory behavior we prepared live brain slices from dye-injected/lesioned mice and rendered confocal time-lapse movies in the cSO (Fig. 4a ; Video 2 (Additional file 2 )). RHO+ cells migrated in a directed fashion from the periventricular region into the cSO (Fig. 4b ) and demonstrated an average speed of 80 ± 6 μm/h. Migrating cells displayed polarized morphologies with a prominent leading protrusion demonstrating numerous side branches (Fig. 4c ). We conclude that activated SVMs are able to migrate long distances into the brain parenchyma towards damaged regions in vivo and in situ . We have named this event "infiltrative microgliosis" (IMG). Figure 4 Dynamics of infiltrative microgliosis. (A) 2D projections of confocal images demonstrate three migratory cells (large and small white arrows) migrating into the cSO. white arrowhead, non-migratory cell for reference. e, ependyma; cSO, stratum oriens . See also Video 2. (B) Migration was highly directed from ventricle to hippocampus, five representative cells from a single experiment. lv, lateral ventricle; black stars, cell origin (C) Highly polarized, migratory morphologies of RHO+ cells as demonstrated by confocal 3D reconstruction. cb, cell body; lpr, leading process; tpr, trailing process. ICV injection of HIV-1 tat protein causes IMG Lentiviral tat protein has been shown to be neurotoxic [ 8 ], stimulate microglial migration in vitro possibly by mimicking, and inducing expression of, chemokines [ 27 , 28 ], and soluble tat protein is released from HIV-infected cells [ 29 ]. Further, ependymal lesions were found in 16% of AIDS patients at autopsy [ 30 ] and HIV-1 tat has been shown to damage the ependymal layer of mice in low nanomolar concentrations [ 8 ]. To establish IMG as an event relevant to neurologic disease we tested the idea that ependymal damage caused by an ICV injection of 2.0 nM recombinant HIV-1 tat protein in mice would cause activation, and possibly intraparenchymal migration, of SVMs. 24 hours post-injection mice demonstrated ependymal cell damage (Fig. 5a , top left) and extensive activation of SVMs (Fig. 5a , bottom left). No damage or SVM activation was seen after injection of deactivated tat (Fig. 5a , right panels). Interestingly, activated microglia could often be found several hundred microns from ventricular surfaces as demonstrated by IB 4 histochemistry (not shown). To determine whether this was due to migration of SVMs into the parenchyma or spreading activation of stationary cells we performed timelapse confocal analysis of live brain slices taken from animals 24 h after ICV tat injection. We found that nearly all activated periventricular microglia were motile and many were locomotory after tat injection (Video 3 (Additional file 3 ). Injection of deactivated tat did not result in migration (Video 4 (Additional file 4 )). Velocities of HIV-1 tat activated microglia averaged approximately 500 μm/h. Further, we observed many microglia which migrated deep into the parenchyma from the periventricular zone (Fig. 5b , Video 5 (Additional file 5 )). Intense microglial activity at the ependyma suggestive of phagocytosis was also observed (Video 5). We conclude that nanomolar concentrations of ICV-injected HIV-1 tat protein alone is sufficient to cause ependymal damage, SVM activation, and diffuse IMG. Figure 5 HIV-1 tat injection activates SVMs and incites IMG. (A) Ependymal loss (top left) and subventricular microgliosis (bottom left) 24 h following injection of 2.0 nmol tat protein but not in animals injected with deactivated tat (right panels). (B) To determine if tat-activated SVMs migrated in situ we rendered timelapse confocal movies 24 h post-injection. Colored arrowheads demonstrate three SVMs which migrate from the region near the ventricle (green line) deep into the parenchyma (colored dashed lines). Field measures ~200 μm horizontally. See also Video 5. Discussion We have shown that SVMs represent a pool of microglial cells which are highly reactive to periventricular damage and are responsible for clearance of resulting cellular debris. Further, activated SVMs are capable of migrating away from the ventricle towards injury cues from damaged regions in the parenchyma several hundred microns away. We confirmed these findings dynamically in acute slice preparations from adult mice. Both the juxtaventricular origin and the extensive migratory capacity of activated SVMs have important implications for neurobiology and disease. Periventricular/subependymal microglia have been noted by histologists since microglia were first identified as a distinct cell type (reviewed in [ 31 ]). Little attention has been paid to these cells in the literature until recently due to their intimate arrangement among the subventricular neural progenitors [ 6 , 7 ]. Possible phenotypic differences between SVMs and parenchymal microglia have not been investigated, however, the location of SVMs among stem cells and within close proximity to ependymal cells and ventricular CSF is unique. A few neurological diseases demonstrate altered CSF constituents and often pathological involvement of the periventricular tissues [ 8 - 13 ]. Therefore, the specific function of microglia in these specialized regions of the brain under normal and pathological conditions deserve further investigation. While evaluating rhodamine dyes for selective labeling of ependymal cells [ 21 ] for other studies we discovered that the ependyma of the injected hemisphere became rapidly damaged after dye uptake. Upon death of the ependyma, SVMs phagocytosed the ependymal debris, and thereby became rhodamine-positive. This serendipitous finding results in rapid and selective labelling of SVMs allowing study of this specific cell population in vivo . For instance, in this study we were able to demonstrate phagocytosis, mitosis, and migration of activated SVMs using histopathological techniques alone. We have shown these cellular activities can be confirmed in situ with timelapse confocal microscopy further validating this versatile protocol. Mechanistic investigations are possible by combining our in vivo and in situ protocols with genetic or pharmacological techniques. We found SVMs only infiltrated the brain after selective ependymal damage with rhodamine dyes if a distant lesion was also present, likely providing a gradient of chemoattractive cues. CC chemokines are known to be upregulated rapidly after deafferenting injury of the hippocampus [ 26 ]. The extensive migration of SVMs in response to HIV-tat injection, on the other hand, may be due to a direct effect of tat on microglia or possibly an indirect effect due to upregulation of chemokines by neurons and glia [ 27 , 28 ]. Further, that ICV injection of recombinant HIV-1 tat protein alone is sufficient to damage the ependyma, activate SVMs, and incite infiltrative microgliosis supports the "cytokine dysregulation hypothesis" [ 8 , 22 ] of damage in HIV-1 encephalitis whereby overactivation of microglia/monocytes may be more critical than actual CNS viral load [ 33 , 34 ]. Conclusions In summary, we have shown that SVMs are a highly reactive pool of cells which, when activated, can infiltrate the parenchyma in response to injury cues from damaged brain regions or exposure to HIV-1 tat. These findings provide new in vivo and in situ models for the study of SVM function, further insight into microglial dynamics after brain injury, and novel hypotheses for the role of microglia in periventricular reactions in neurological diseases. List of abbreviations BrdU, 5-bromo 2-deoxyuridine; BSA, bovine serum albumin; CSF, cerebrospinal fluid; cSO, stratum oriens of the hippocampus contralateral to stab lesion; DMSO, dimethyl sulfoxide; DTT, dithiothreitol; GFP, green fluorescent protein; GSD, modified Gallyas silver degeneration stain; HIVE, human immune deficiency virus 1 encephalitis; ICV, intracerebroventricular; IFN-γ, interferon gamma; IMG, infiltrative microgliosis; MS, multiple sclerosis; PBS, phosphate-buffered saline; pERK, phosphorylated/activated extracellular signal-regulated kinase; PI, post-injury; RHO+, rhodamine-positive; RhoB, rhodamine latex microbeads; SEM, standard error of the mean; Sp-DiI, 1,1'-dioctadecyl-6,6'-di(4-sulfophenyl)-3,3,3',3'-tetramethylindocarbocyanine; SVM, subventricular microglia; TNF-α, tumor necrosis factor alpha. Competing interests The authors declare that they have no competing interests. Authors' contributions WSC conceived of and designed the study, carried out all experiments, performed data analysis, and drafted the manuscript. S-IM participated in study design especially with regards to the timelapse experiments. AFH participated in study design and coordination and provided the confocal facilities, equipment, and expertise for timelapse experiments. JWM participated in study design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Supplementary Material Additional File 1 SVM phagocytosis of ependymal debris. Activity of SVMs suggestive of phagocytosis of dye-labeled ependymal cell debris 24 h following injection. SVMs can be seen extending processes towards debris. Examples of ependymal debris are pseudocolored yellow. Each frame is a 2D projection representing a stack of 6 images 8 μm apart. Original magnification, 40×. Click here for file Additional File 2 Dynamics of infiltrative microgliosis. Infiltrative microgliosis of SVMs into the hippocampal stratum oriens from the subependymal region of the posterior lateral ventricle. Note highly directed migration into the hippocampus. Each frame is a 2D projection representing a stack of 4 images 10 μm apart taken every 3 minutes. Original magnification, 20×. Click here for file Additional File 3 SVM dynamics in response to HIV-1 tat protein. Extensive migratory activation of periventricular microglia in response to 2.0 nM ICV HIV-1 tat protein. This migratory reaction extends several hundred microns into the parenchyma. v3v, ventral third ventricle. Each frame is a 2D projection representing a stack of 6 images 8 μm apart taken every 90 seconds. Original magnification, 20×. Field measures 700 × 700 μm. Click here for file Additional File 4 Control video for HIV-1 tat protein. Lack of activation of SVMs and migration with ICV injection of deactivated HIV-1 tat protein (compare to Video 3, similar field). The paucity of IB4 labeling indicates the limited microglial activation. Note blood vessel endothelial cell labeling and gradual photobleaching. Each frame is a 2D projection representing a stack of 6 images 8 μm apart taken every 90 seconds. Original magnification, 20×. Field measures 700 × 700 μm. Click here for file Additional File 5 HIV-1 tat incites infiltrative microgliosis. HIV-1 tat activated subventricular microglia infiltrate the parenchyma. Three highlighted cells correspond to those in Figure 5b . Note also intense activity of SVMs at ventricle (red line) suggestive of phagocytosis of ependymal cell debris. Each frame is a 2D projection representing a stack of 6 images 8 μm apart taken every 90 seconds. Original magnification, 40×. Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548677.xml

535893

Male Reproductive Health: A village based study of camp attenders in rural India

Background A paucity of information about male reproductive health and a perceived interest in involvement among local men provided the impetus for carrying out a village based male reproductive health camp. The aim was to investigate men's willingness to participate in such camps, and to describe reproductive health problems in men. Methods Structured interviews were carried out with 120 men attending a reproductive health check-up in a village in rural West Bengal, India. General information, details of family planning methods used and data on reproductive health complaints were collected. Clinical examinations were also carried out. Socio-demographic characteristics were compared for men with and without reproductive health and urinary complaints. Results Three quarters of the married men were using contraception, but the majority stated that their wives were responsible for it. The most common reproductive health complaint was urinary problems; 28% had burning on urination, and 22% reported frequent and/or difficult urination. There were few social or demographic differences between men with and without problems. Seventeen percent of the men had clinically diagnosed reproductive health problems, the most common being urethral discharge. None of the men with diagnosed problems were using condoms. Conclusions This study highlights the interest of men in their reproductive health, but also highlights the high proportion of men with problems. In addition, a number of men with clinically diagnosed problems had not reported them in the interviews, illustrating either the reticence to report or the lack of knowledge about symptoms of reproductive health problems. Recommendations for future programmes and research in this field are given.

Background Reproductive health is a major world priority, with particular problems in developing countries. However, as Ndong [ 1 ] states, "reproductive health generally has been synonymous with women's health", and reproductive health of men has received little attention [ 2 ]. Researchers and health planners have pointed out that better outcomes for reproductive health programmes would be expected if men were involved [ 2 - 6 ], and there are a number of mechanisms by which this might occur. Hawkes [ 7 ] indicated that the treatment of male reproductive health problems might actually encourage more women to seek treatment, and therefore improve the overall level of reproductive health. Other studies have highlighted that women often need the support of their husbands, including financial support, to attend reproductive health services [ 4 , 8 , 9 ] and that the health status of couples, in particular the reproductive health status, is strongly linked to the knowledge, attitudes and behaviour of men [ 6 , 10 ]. The documentation of the prevalence of male reproductive health problems has been identified as a research priority by the World Health Organization (WHO) [ 5 ]. Until now limited information on reproductive health problems and health care needs in men from developing countries, particularly from rural areas of India, is available [ 11 , 12 ]. Verma [ 12 ] studied male sexual problems in a slum population in Mumbai (India) and found that around half of the men studied could identify symptoms such as itching, burning on urination and white discharge, although the prevalence of these symptoms among these men was not reported. One study from Bangladesh reported that over 10% of men had symptoms that were possibly indicative of a sexually transmitted infection, with the main complaints being urethral discharge and pain passing urine [ 2 ]. Information about male reproductive health problems may help in the planning of future services, the needs for screening, prevention and treatment. In order to collect this information, a village based reproductive health camp for men was carried out. Health camps are used to provide services in places where the perceived need and demand is high but access to treatment is poor due to lack of information about available services, distance from health facilities, or low economic status of the population. It is one of the preferred methods of service provision in rural villages. The usual model is a temporary clinic set up by a team of health personnel at a community-owned location (e.g. school or community hall) to provide services and medicines usually free of cost based on the health needs of the community. The aim of this study was to investigate men's willingness to participate in such camps, and to describe reproductive health problems in this group of men. This paper reports on the findings from this health camp. Methods Study participants Men aged 15–60 years, both married and unmarried and living in one rural village were invited to attend. The village is relatively small (approximately 4000 residents), mostly consisting of low-income, Muslim population, and had been highlighted by local health workers as having particularly poor knowledge about reproductive health issues. Camp procedure The camp was held in a village school on one day in February 2000. The community was consulted about the camp through a variety of routes, including men's groups, Panchayats (local self government), schoolteachers and specially convened public meetings; all of these groups agreed to the holding of the camp in the village. The camp was publicised through peer educators, health workers and group meetings, and was described as a reproductive health check-up, with free treatment if necessary. A male team was used as feedback from health workers and peer educators in the village indicated that men would be more willing to attend if male doctors carried out the examinations. Health workers carried out a confidential interview on arrival to obtain demographic information (age, marital status, number of children, employment and income), information about personal hygiene, details about family planning issues and about general health complaints (including fever, dizziness, backache, headache, skin problems) by using a structured interview schedule. Male doctors then asked about reproductive health complaints (including urinary problems, genital itching, penile sores/swelling, urethral discharge) and current sexual activity (sex outside marriage or with a prostitute), and carried out clinical examinations (recording of urethral discharge, genital ulcer, inguinal lymph nodes, penile lesions, scrotal swelling, genital warts or scabies). Blood samples were taken for Venereal Disease Research Laboratory (VDRL) test from all attendees. Those men detected as having sexually transmitted infections and urinary problems were given the appropriate medication, provided with condoms and referred to hospital when necessary. Advice and instructions for condom use were given and patients were encouraged to bring their partners for treatment. Setting The Child in Need Institute (CINI) is a non-governmental organisation (NGO) working in West Bengal, and is one of the four national NGOs in India. The organisation has, for the last 30 years, focused on the health and nutrition of women and children, with the additional involvement of adolescents and men. CINI started working on reproductive health with men in rural areas in 1997. To date, this involvement has entailed orientation and education with men's groups, intensive training of voluntary peer educators and facilitators and individual interaction with health workers. There are at present no specific clinical services for men provided by CINI: men who are identified as having a problem are given individual information and counselling and are referred to other clinics for treatment. Analysis The information collected during the interviews, clinical examination and pathology reports was entered into the Epi-Info data entry and analysis package [ 13 ]. Statistical analyses were carried out using the SPSS statistical package [ 14 ]. The relationship between reproductive health complaints, clinical findings and other health and socio-demographic factors was investigated. P-values for the differences between means were calculated using the independent samples t-test. P-values for the differences between proportions were calculated using the chi-squared test or Fisher's exact test. Results One hundred and twenty men attended the camp; all men responded freely to the interview, and all but two agreed to a clinical examination. This sample represents about 10% of the local male population aged 15–60 years. Demographic information The mean age in this group of men was 29.6 years (range 16–60). Fifty-nine percent of the men were married (n = 71); the mean age at marriage was 22.6 years (range 14–34). Fifty-two percent of the attendees had children (comprising 90% of the married men); among the men with children, the mean number was 2.8 (range 1–7). The literacy rate was 71% among the attendees, although more than half of these had only studied up to class V or less. Eighty four percent of the attendees (n = 85, data missing for 19 attendees) had a monthly income of less than Rs. 2000 (~$45); 40% (n = 48) were daily wage earners, and a further 30% (n = 36) were in service (usually unskilled office workers), 15% (n = 18) were either unemployed or students. One third of the men worked away from home for at least one night each month, half of these for 1–5 days, and half for more than 5 days per month. Only 11 attendees indicated that they consumed alcohol. Twenty percent (n = 24) reported good personal hygiene (washing genital areas and changing underwear daily), however 28% (n = 33) had very poor hygiene practices (washing genital area and/or changing underwear less than twice a week). General health Four fifths of men (82%) reported at least one problem with their general health, the most commonly reported problems were coughs and colds (45%, n = 54), and weakness (28%, n = 34), other problems included indigestion, backache and headache. Two men were identified as having hypertension. Family planning Forty-three percent of all men (n = 51) reported that they or their wives were using some kind of family planning, this represented 73% of the married men; only one unmarried man was using contraception (condoms). Over half of the men stated that their wives are responsible for family planning, they are using the oral contraceptive pill, have an intrauterine device or had sterilisation. Eight men were using condoms, and one man had had a vasectomy. The main reason given for not using family planning was being unmarried. Two men reported a lack of knowledge of family planning, two men had religious objections and five men said that they wanted children. Sexual behaviour Six married men admitted to having extramarital sex, three of them with prostitutes. Seven unmarried men (14%) reported having sexual relations; three of them with prostitutes. None of the men visiting prostitutes reported using condoms, however one married man having extra marital sex reported using condoms. It did not appear that the men visiting prostitutes or having extra-marital sex were any different from the other men in terms of monthly income, education or number of nights working away from home. Reproductive health complaints Half of the attendees reported at least one reproductive health complaint (n = 60) (table 1 ). Only three men with a reproductive health complaint reported using condoms (table 2 ). There were no statistically significant differences in age, monthly income, days of absence from home or marital status between men with and without reporting of sexual health problems. However, men who have complaints have on average fewer years of education (3.3) compared to men who do not have complaints (4.9). Similar proportions of men were married in those with and without reproductive health complaints (table 2 ). The most common complaints were urinary problems, which were reported by 43% of attendees (n = 51); 28% (n = 33), of men complained of burning on urination and 23 % (n = 27) reported frequent and/or difficult urination. Table 1 Self reported complaints of reproductive health problems Complaint No of attendees % (n = 120) Burning on urination 33 28% Frequent urination 20 17% Difficulty urinating 7 6% Strong urine 4 3% Any urinary complaint 51 43% Premature ejaculation 6 5% Urethral discharge 6 5% Genital itching 5 4% Impotence 2 2% Primary infertility 2 2% Any reproductive health complaint 60 50% Table 2 Socio-demographic characteristics in men with and without reproductive health & urinary complaints Demographic variable Any reproductive health complaint Burning on urination Frequent and/or difficult urination Yes n = 60 No n = 60 p-value Yes n = 87 No n = 33 p-value Yes n = 94 No n = 26 p-value Age (mean years) 29.8 29.1 0.761 † 28.3 29.9 0.498 † 31.7 28.8 0.249 † Monthly income (mean rupees) 1447 1411 0.905 † 1283 1493 0.528 † 1146 1515 0.307 † Education (mean years) 3.3 4.9 0.009 † 3.7 4.2 0.476 † 3.2 4.3 0.126 † Time away from home (mean days per month) 2.8 1.9 0.330 † 2.5 2.3 0.825 † 3.6 2.0 0.141 † Marital status: no., (%) married 35 (58%) 34 (57%) 0.853 ‡ 17 (52%) 52 (60%) 0.414 ‡ 16 (62%) 53 (56%) 0.638 ‡ Condom use: no. (%) using condoms 3 (5%) 5 (8%) 0.717* 3 (5%) 5 (8%) 0.683* 1 (2%) 7 (12%) 1.000* Sex with prostitute: no. (%) of men 2 (3%) 4 (7%) 0.679* 1 (2%) 5 (8%) 1.000* 0 6 (10%) 0.338* Good personal hygiene: no. (%) of men** 13 (26%) 11 (19%) 0.492* 10 (32%) 14 (18%) 0.129* 4 (20%) 20 (23%) 1.000* † p-value for the difference between means using independent samples t-test; ‡ p-value using chi-squared test * p-value using Fisher's exact test; ** Reporting washing genital areas and changing underwear daily Clinical reproductive health findings On clinical examination, 20 men (17%) had findings of reproductive health problems, (table 3 ), fourteen of these men had reported reproductive health complaints in the interview. None of the men with clinical findings reported using condoms, and nearly half (n = 9) had poor self-reported hygiene. Two of the men with positive findings reported visiting prostitutes, and five reported having extramarital sex. The most common clinical finding was urethral discharge, which was found in 13 men, this included all 6 men who had reported urethral discharge in the interview (table 4 ); two of these men reported having sex with a prostitute. Painful genital lesions were found in two men, this could be an indicator of the herpes virus; these men were both married and one was visiting a prostitute, neither used condoms and neither reported penile sores in the interview, although both reported genital itching (table 4 ). Only one man had a positive VDRL result, he had reported genital itching, the clinical examination found a genital ulcer and swollen and painful lymph nodes. He was married, and reported having extra marital sex with a prostitute, and said he did not use condoms. Table 3 Clinical findings of reproductive health problems Clinical sign No of attendees % (n = 118) Urethral discharge 13 11% Genital ulcer 2 2% Swollen lymph nodes 5 4% Painful genital lesions 2 2% Scrotal swelling 0 - Genital warts 0 - Scabies 2 2% Fungal infection 2 2% Any clinical sign 20 17% Table 4 Comparison of self reported complaints and clinical findings of reproductive health problems (no. of attendees) Clinical finding Self-reported complaint Urethral discharge (n = 13) Genital ulcer (n = 2) Swollen lymph nodes (n = 5) Painful genital lesions (n = 2) Scabies (n = 2) Fungal infection (n = 2) Any clinical sign (n = 20) Burning on urination (n = 33) 4 0 2 1 2 0 7 Frequent urination (n = 20) 4 0 1 0 0 0 4 Difficulty urinating (n = 7) 1 0 2 1 1 0 4 Strong urine (n = 4) 2 0 1 1 2 0 3 Any urinary complaint (n = 51) 8 0 3 1 2 0 11 Premature ejaculation (n = 6) 0 0 1 1 1 1 2 Urethral discharge (n = 6) 6 0 1 0 1 0 6 Genital itching (n = 5) 1 1 2 2 1 0 3 Any reproductive health complaint (n = 60) 10 1 4 2 2 1 15 There was a trend that men having extramarital sex and men having sex with prostitutes were more likely to have positive clinical findings than men not having extramarital sex (5/13 vs. 15/105; odds ratio 3.8, 95% confidence interval 1.1 to 13.0 and 3/6 vs. 17/112; OR 5.6, 95% CI 1.0 to 30.0, respectively). None of the eight men using condoms had positive clinical findings compared with 18% (n = 20) of the men not using condoms. Discussion This study indicates that there is a high level of reproductive health problems among those men attending a reproductive health camp in a rural area of India. More than one in ten men had urethral discharge and over one third reported urinary problems. The high number of men reporting urinary symptoms is similar to that of an unpublished study in Uttar Pradesh, India [ 12 ]. However, a survey in Bangladesh reported pain passing urine in only 8% of men questioned [ 2 ]. In our study, 14% of unmarried men admitted to having sexual relations and 8% of married men reported extra-marital sex in our study, which lies within the expected range from other studies in Southern Asia [ 11 , 12 ]. The enthusiasm from men for involvement in reproductive health programmes has been reported elsewhere in Southern Asia [ 2 , 15 ]. Factors perceived to have maximised the attendance rate may have included the use of purely male staff at the camp, the promotion of a 'check-up' rather than just treatment, the consultation with local people and the placement of the camp in a local school (based on feedback from the health workers and peer educators). The attendees appear to be similar to the local population in terms of demographic factors such as marital status, education and income; however, the attendees may not be representative of all the men in the area. It is possible that men who thought they had a problem might be more likely to attend a reproductive health check-up; conversely, men who perceived that they had problems might be less likely to attend through fear. Studies maximising the generalisability of the sample would improve our understanding of the problems in this section of the population. Wang [ 5 ] identified the level of responsibility that men hold for the consequences of their sexual behaviour as a priority area, and other researchers have put this emphasis primarily on condom use [ 3 ]. This study has shown that none of the men with clinical findings of reproductive health problems, nor any of the men visiting prostitutes, reported using condoms. Men who did not use a condom, who had sex with a prostitute or who had extramarital sex also appeared to be more likely to have clinical findings of reproductive health problems. The uptake of condom use was equally low compared to other reports, making the promotion of condom use in this population even more important [ 11 ]. The high proportion of men reporting urinary problems has also highlighted the need testing of urinary samples. One study in Bangladesh indicated that among 47 men reporting pain passing urine, only one was infected with Chlamydia trachomatis [ 2 ]; we are not aware of similar data for Indian samples. A significant number of men did not complain of reproductive health problems (such as urethral discharge) during the interview, but had positive clinical findings (table 4 ). This incoherence has been reported in other studies [ 2 , 11 ], and could reflect a reticence about admitting to such problems, or that the men were unaware of their symptoms which may explain the low proportion of men seeking treatment for their problems [ 8 ]. Therefore, the dissemination of information about symptoms of reproductive health problems highlights another important area for further investigation. There are some limitations to this study. First of all the numbers were relatively small and a larger sample size might have given further insight. Information obtained from urethral smears of all attendees may have provided important information. However, this was not possible in our setting due to technical problems. Further, not all areas of relevance to reproductive health were addressed in this study. Psychosexual problems have been commonly reported elsewhere [ 2 ], but were not asked about as part of this study. Some priority areas for male reproductive health, namely HIV/AIDS and male infertility [ 5 ], have not been addressed here. These areas were not addressed as CINI did not have sufficient resources, and there were no referral linkages to appropriate facilities at the time. Inclusion of these problems in future research is vital in improving overall reproductive health. Conclusions This study provides important information about male reproductive health problems in a sample of men in rural West Bengal, India. The response of the men and the fact that nearly all attenders were willing to undergo a clinical examination is encouraging, and emphasises the possibilities for future research. The high level of reproductive health problems identified on clinical examination, but not reported in interviews, plus the low levels of condom use illustrate the need for reproductive health interventions with such groups. Future research could build on the findings of this exploratory study. List of abbreviations CINI = Child in Need Institute VDRL = Venereal Disease Research Laboratory Competing interests The author(s) declare that they have no competing interests. Authors' contributions KMD participated in the design of the study, performed the statistical analysis and drafted the manuscript. SD conceived the study, participated in its design, co-ordination, data entry and analysis and contributed to the manuscript. RD conceived the study, and participated in the design, co-ordination and analysis. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535893.xml

549582

Teaching Health Workers Malaria Diagnosis

In most parts of the world, microscopy is still the gold standard for diagnosing malaria. An online tool could help to improve your diagnostic skills

Malaria kills over one million people in Africa each year and contributes 10% of the continent's burden of disease [ 1 ]. One of the factors that affects the morbidity and mortality rate is incorrect diagnosis [ 2 ]. In this article, we describe a freely available online training tool for health professionals to learn malaria diagnosis. We describe why we launched it and discuss how it is being used. The Burden of Disease Malaria affects at least 200–300 million people every year and causes from 1–2 million deaths—mostly children under five and pregnant women in sub-Saharan Africa [ 1 ]. These deaths largely occur in remote rural areas with poor access to health services. In non-pregnant adults, although mortality rates are lower, the debilitating disease affects quality of life. The economic burden is also extremely high, accounting for a reduction of 1.3% in the annual economic growth rate of countries where malaria is endemic [ 3 ]. Malaria costs Africa more than US$12 billion every year in lost GDP, even though it could be controlled for a fraction of that sum [ 1 ]. Malaria is not limited to Africa—40% of the world's population is at risk of acquiring the disease [ 4 , 5 ]—and it is crucial for all physicians, medical scientists, and other healthcare professionals to be alert to the diagnosis. In addition, each year, 25–30 million people from non-tropical countries visit areas where malaria is endemic [ 6 ], and between 10,000 and 30,000 contract malaria [ 7 ]. Some 90% of infected travellers do not become ill until they return home; this “imported malaria” is easily treated, but only if it is diagnosed promptly [ 2 , 8 ]. The Importance of Correct Diagnosis Despite the efforts of a global campaign to roll back the disease, the number of deaths from malaria is increasing in Africa [ 9 ]. This statistic highlights the importance of local capacity to diagnose and treat malaria in order to prevent illness and death [ 10 ]. In many parts of the world, education about malaria diagnosis and treatment is limited—and the incorrect diagnosis of malaria by clinical, laboratory staff, and other healthcare workers can contribute to morbidity and mortality [ 2 , 11 ]. There is another important reason why correct diagnosis matters. Because of rising drug resistance in Africa, the conventional drugs for treating malaria—chloroquine and sulfadoxine-pyrimethamine—are now failing in up to 80% of cases [ 12 ]. There is a highly effective alternative to these drugs, which is artemisinin-based combination therapy (ACT) [ 13 ], but ACT is expensive (an adult dose of chloroquine costs around 10 cents, whereas a dose of ACT costs at least ten times that amount) [ 14 ]. Scaling up the use of ACT is now a core strategy in the global campaign to control malaria. But in order to control costs, and to prevent the emergence of resistance to ACT, its use should be targeted to real (rather than presumptive) cases of malaria [ 14 ]. Currently, however, a comparison of the number of parasitologically confirmed cases of malaria with those that are presumptively diagnosed shows high rates of overdiagnosis outside of hospitals at the community level, where self treatment is routine [ 14 ]. Microscopy as the Gold Standard for Diagnosis We still regard microscopy as the gold standard for the diagnosis and characterisation of malaria infection ( Figure 1 ). Modern “dipstick” technology and molecular techniques have emerged as an aid to diagnosis [ 13 , 14 ]. Some of these are useful particularly for Plasmodium falciparum [ 15 ]—although they are less effective for other species or for mixed infections [ 16 , 17 ]—but these techniques are often only available in larger hospitals, which are more likely to be able to afford them. Figure 1 Thin Film Micrograph Showing a Red Blood Cell Containing Two Ring-Form P. vivax Parasites P. vivax rings have a large quantity of cytoplasm and a large chromatin dot, as well as occasional pseudopods. The red blood cells are normal, to 1.5× normal, sized, round, and contain fine Schüffner's dots, and they quite often contain multiple parasites. (Photograph: CDC/Dr Mae Melvin) Some authors have suggested that these new tests should not be considered a complete substitute for direct microscopic examination of blood smears [ 16 ]. We agree. While microscopy has limitations—morphology can be misleading if a patient has received partial treatment or incomplete prophylaxis, and in some locations a microscope (let alone a microscopist) can be rare—the new dipstick tests also have their flaws. For example, they must be stored correctly, used correctly, and interpreted correctly. There are concerns about the variability in the different tests' false positive and false negative rates and about their cost–benefit ratio. There are many supporters of the new dipstick technology, and as it improves and becomes more robust and reliable, it may well replace the microscope in practice. But for much of the world, the front-line diagnosis of malaria still remains in the hands of the reasonably trained microscopist. Analysis of DNA by the polymerase chain reaction (PCR) may be a useful tool for diagnosis of malaria when the results of conventional techniques are negative, especially since PCR allows accurate species identification [ 18 ]. And when compared with the “gold standard” of microscopy, PCR has a sensitivity and specificity of 100%, with a detection limit of just one P. falciparum or three P. vivax parasites per microlitre of blood [ 19 ]. However, in most areas with malaria transmission, limited financial resources, persistent subclinical parasitaemia, and inadequate laboratory infrastructure preclude PCR as a routine diagnostic method [ 20 ]. Even in affluent, nonendemic countries, PCR is not a suitable method for routine use. Capital investment and ongoing running costs are prohibitive for many laboratories, and certainly for the foreseeable future this technology will remain a tool for the more specialised services in the more affluent societies. At this stage, it is not considered a “routine” assay. Teaching Microscopy Online In the hope of improving health professionals' understanding of malaria diagnosis, treatment, and prophylaxis, we launched an online malaria education tool that went live in July 1998 ( http://www.rph.wa.gov.au/labs/haem/malaria/index.html ). The materials are now available in English, French, and Spanish. One of the most important features is a “Test and Teach” section to allow microscopists to develop and sharpen their skills in malaria diagnosis ( Figure 2 ). Figure 2 A Clinical Case from the “Test and Teach” Section Because of regional variability in Internet access, we also made the tool freely available as a CD-ROM. The first CD-ROMs we produced, which were in English, reached institutions and centres in 41 countries during the first few months after the launch. By October 1999, we produced a new CD-ROM that incorporated a French translation of key areas of the project; a Spanish version followed. By the end of 1999, the CD-ROM version had reached over 70 countries and the website had received nearly 30,000 visitors. To date, the website has received more than 500,000 visitors. Around 83 million pages of text were downloaded from the website between July 2001 and September 2003. The CD-ROM version (over 8,000 copies) has been sent on request to institutions and centres in 149 countries, a number that does not include unauthorised (pirated) copies of the CD-ROM. The number of hits to the site (the number of pages or images of the site that are accessed) has risen dramatically, from 1,500 to over 100,000 per week. Feedback from Users Through a circulated questionnaire with a 63% response rate, we have received feedback from users in 15 countries. Respondents said they valued the content, presentation, and usefulness of the online information. Who is using the material? A wide range of people—from students through to experienced personnel, and from clinicians to scientific, research, and technical staff. Many tropical medicine institutions have requested copies of the CD-ROM to supplement their own training programs. We have discussed the project at scientific conferences around the world and in scientific journals, advertising the fact that this educational material is freely available. Institutions wishing to obtain a free copy of the CD-ROM should contact E-mail: Graham.Icke@health.wa.gov.au . We also became aware that unauthorised copies of the CD-ROM were being produced. Although there is a real need to guarantee the integrity of the information on the CD-ROMs and ensure proper credit for the authors and patrons, our main aim is to spread the “good word”, and such unauthorised copies are a powerful aid to distribution. Given that there is no need to recoup profit from this project, our only concern is its usefulness. Consequently, we have welcomed the production of these pirated copies, on condition that they were of good quality and were made freely available at no cost. Conclusion The Internet and associated technologies make it possible for educators with a desire to teach to contact those with a need to learn, regardless of the geographical distances involved. By enlisting assistance from international colleagues, language barriers can be overcome. In addition to the English, French, and Spanish versions, we are aware that our project has been translated into German, Thai, and Vietnamese and has been distributed through small regional health group networks. There can be little doubt that these new technologies have the potential to revolutionise information dissemination, with particularly significant implications for healthcare professionals in developing countries. The malaria educational program could be a model on which future health education programs are based. With rapidly increasing access to these new information technologies, health care professionals from anywhere in the world can now join the global health community.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549582.xml

526780

Palmistry

A review of "On Intelligence", a book exploring brain function by entrepreneur Jeff Hawkins and science writer Sandra Blakeslee

Is Michael Moore liberal America's Rush Limbaugh? If so, is he filling a much needed, or a much lamented, gap in turning issues that are really cast in pastel shades into Day-Glo relief? In this hale monograph, Jeff Hawkins (rendered by Sandra Blakeslee) plays exactly this role for theoretical neuroscience. As a pastel practitioner myself, but furtively sharing many of Hawkins' prejudices and hunches about computational modelling in neuroscience, I am caught between commendation and consternation. Hawkins is an engineer, entrepreneur, and scientist who founded and led the companies Palm and then Handspring. He created, against what must have been considerable obstacles, the first widely successful PDA, and continued the development of this platform. He has thus amply earned a bully pulpit. The autobiographical segments of this book detail that, throughout his career, he has been interested in understanding how the brain works, using his substantial knowledge and intuition about the architecture and design of conventional computers as a counterpoint. More recently, Hawkins has generously put his money where his ideas about mentation dictate, founding the Redwood Neuroscience Institute and also funding various conferences and workshops. The institute is dedicated to ‘studying and promoting biologically accurate mathematical models of memory and cognition.’Despite its youth, the Institute already has attracted notable attention as a centre for theoretical neuroscience. Hawkins' quest, and-depending on which statements of the book you read-its endpoint (‘… a comprehensive theory of how the brain works … describ[ing] what intelligence is and how your brain creates it’) or just its tipping point (‘join me, along with others who take up the challenge’), are the subject here. There are really three books jostling inside the covers. One is the (highly abbreviated) autobiography. The history of modern computing is very brief and (at least judging by the sales) very glorious, and this story is most entertaining. Don't miss the wonderfully faux naive letter from Hawkins to Gordon Moore asking, in 1980, to set up a research group within Intel devoted to the brain. That Hawkins prospered in clear opposition to accepted wisdom is perhaps one of the key subtexts of the book. The second, and rather less satisfying, book is about the philosophy of mind and the history of artificial intelligence and neural network approaches to understanding the brain and replicating cognition. With respect to the fields of artificial intelligence and neural nets, the text seems rather to be fighting yesterday's battles. The importance of learning, flexibility in representation and inference, and even decentralisation of control has been more than amply recognised in the inexorable rise of probabilistic approaches in both fields. With respect to the philosophy of mind, there seems to be something of an enthusiast's disdain for the niceties of philosophical pettifogging, even arguing by assertion. The discussions at the end on creativity and consciousness all seem a bit gossamer. The book is somewhat careless about functionalism, a key doctrine for computational theorists about how brains give rise to minds. According to this doctrine, at least roughly, it is the functional roles of, and functional interactions among, the physical elements of brain that matter, and not their precise physical nature. If you can capture those functional aspects correctly, for instance, in a computer program, then you can (re-)create what's important about mental states. Functionalism licenses a form of inquiry into the computational jobs played by structures in the brain. However, although formally agreeing that ‘there's nothing inherently special or magical about the brain that allows it to be intelligent,’the book slips into statements such as ‘brains and computers do fundamentally different things,’which are, at best, unfortunate shorthand. The book is a little apt to sneak plausible, but misleading, claims under the radar. Just to give one instance, it compellingly compares a six year old hopping from rock to rock in a streambed with a lumbering robot failing to do the same task. However, this is a bit unfair. One of Hawkins' self-denying ordinances is to consider the cortex pretty much by itself. As aficionados of the cerebellum (an evolutionarily ancient brain region with a special role in the organisation of smooth, precise, well-timed, and task-sensitive motor output) would be quick to point out, the singular role for the cortex in such graceful behaviour is rather questionable. The third book is what I think is intended to be the real contribution. This contains a (not wholly convincing) attempt to conceptualise the definition of intelligence in terms of prediction rather than behaviour, and then to describe its possible instantiation in the anatomy (and mostly only the anatomy) of the cortex. Unsupervised Learning To situate Hawkins' suggestions, it is instructive to consider current models of how the cerebral cortex represents, and learns to represent, information about the world without being explicitly taught. Being a popular account, the book fairly breezes by these so-called unsupervised learning models (see Hinton and Ghahramani 1997 ; Rao et al. 2002 ), in which the neocortex is treated as a general device for finding relationships or structure in its input. The algorithms are called unsupervised since they have to work without detailed information from a teacher or a supervisor about the actual structure in each input. Rather, they must rely on general, statistical characteristics. First, where does the structure in the inputs come from? For the sake of concreteness, think of the input as being something like movies on a television screen. Movies don't look like white noise, or ‘snow’, because of their statistical structure. For instance, in movies, pixel activities tend to change rather slowly over time, and pixels that are close to each other on the screen tend to have relatively similar activities at any given time. Neither of these is true of white noise. More technically, movies constitute only a tiny fraction of the space of all possible activations of all the pixels on your screen. They (and indeed real visual scenes) have a particular statistical structure that the cortex is supposed to extract. What is the cortex supposed to do with this structure? The idea is that the cortex learns to model, or ‘parameterize’, it. Then, the activities of cortical cells over time for a particular input, for example, a particular face in a movie, indicate the values of the parameters associated with that face. Thereby the cortical activities represent the input. The parameters for a face might include one set for its physical structure (e.g., the separation between the eyes and whether it is more round or more square), another set for the expression, and yet others, too. Cortical representations are thus intended to reflect directly the statistical structure in the input. Importantly, for inputs such as movies, this structure is thought to be hierarchical and, concomitantly, to provide an account of the observed hierarchical structure of sensory cortical areas. One source of hierarchical structure in movies is the simple fact that objects (such as the faces) have parts (such as eyes and cheeks) whose form and changes in form over time are interdependent. Another source of hierarchical structure is that the same face can appear in many different poses, under many different forms of illumination, and so on. Pattern theory ( Grenander 1995 ), one of the parent disciplines of the field, calls these dimensions of variation deformations. Loosely, the deformations are independent of the objects themselves, and we might expect this independence to be reflected in the cortical representations. Indeed, there is neurophysiological evidence for just such invariant neural responses to deformations of a stimulus. How does the cortex do all this? Of course, some fraction of this structure was built in over evolution. However, the unsupervised learning tradition concentrates on ontogenic adaptation, based on multiple presented input movies. An additional facet of the lack of supervision is that this adaptation is taken as not depending on any particular behavioural task. Finally, what does this process allow the cortex to do? The whole representational structure is intended to support inference. Crudely, this involves turning partial or noisy inputs into the completed, cleaned-up patterns they imply, using connections between areas in the cortical hierarchy. Construed this way, probabilistic inference actually instantiates a very general form of computation. Crucially, over the course of the development of unsupervised learning methods, it has been realised that the best way to approach the extraction of input structure, and inference with it, is through the language and tools of probability theory and statistics. The same realisation has driven substantial developments in artificial intelligence, machine learning, computer vision, and a host of other disciplines. Predictive Auto-Association We can now return to the book. Hawkins compactly sums up his thesis in the following way. ‘To make predictions of future events, your neocortex has to store sequences of patterns. To recall appropriate memories, it has to retrieve patterns by their similarity to past patterns (auto-associative recall). And finally, memories have to be stored in an invariant form so that the knowledge of past events can be applied to new situations that are similar but not identical to the past.’In fact, to take the latter points first, the sort of auto-associative storage and recall to which Hawkins refers is a theoretically and practically hobbled version of unsupervised learning's probabilistic inference. Invariance is closely related to the deformations we described above in the context of pattern theory. Unsupervised learning has certainly paid substantial attention to sequences of inputs and prediction, and to some good effect. For instance, (artificial) speech recognition programs are based on a probabilistic device called a hidden Markov model, which is a key element in a wealth of unsupervised learning approaches to prediction. However, despite heroic efforts, these modelling methods are incapable of capturing the sort of complex structure seen in inputs such as natural languages. They fail on phenomena like long-distance dependencies, for example, the agreement between the cases of subjects and verbs, which are rife. This does tend to offer a vaccine against Hawkins' otherwise infectious optimism. Once place in which Hawkins goes beyond existing unsupervised learning models is in an extension to actions and control, and in an ascription of parts of the model to cortical anatomy. The hierarchical conception of cortex here goes all the way down to primary motor cortex (the neocortical area most directly associated with motor output). This allows auto-associative recall of sequences of past inputs and outputs to be used to specify actions that have formerly been successful. The discussion of this possibility is, unfortunately, rather brief. Central issues are omitted, such as the way that planning over multiple actions might happen. Also, the way that value is assigned to outcomes to determine success or failure is not discussed. The latter is widely believed to involve the neuromodulatory systems that lie below the cortex and that the book's cortical chauvinism leads it cheerfully to ignore. By contrast, the book has a rather detailed description of how the model should map onto the anatomy of the cerebral cortex. Like many unsupervised learning modellers, Hawkins is a self-confessed ‘lumper’. He ignores huge swathes of complexity and specificity in cortical structure and connections in favour of a scheme of crystalline regularity. Though this will doubtless irk many readers (as will the lack of citations to some influential prior proponents such as Douglas and Martin [1991] ), some (though not necessarily this) strong form of abstraction and omission is necessary to get to clear functional ideas. This part has interesting suggestions, such as a neat solution for a persistent dilemma for proponents of hierarchical models. The battle comes between cases in which information in a higher cortical area, acting as prior information, boosts activities in a lower cortical area, and cases of predictive coding, in which the higher cortical area informs the lower cortical area about what it already knows and therefore suppresses the information that the lower area would otherwise just repeat up the hierarchy. The proposed solution involves the invention (or rather prediction) of two different sorts of neurons in a particular layer of cortex. Unsupervised learning models of cortex are without doubt very elegant. However, if pushed, purveyors of this approach will often admit to being kept awake at night by a number of critical concerns even apart from the difficulty of getting the models to work in interestingly rich sensory domains. Does the book provide computational Halcyon? First, the representations acquired by unsupervised learning are intended to be used for something-such as accomplishing more specific learning tasks, for example, making predictions of reward. However, most aspects of the statistical structure of inputs are irrelevant. This might be called the ‘carpet’problem: there is a wealth of statistical structure in the visual texture of carpets; however, this structure is irrelevant for almost any task. Capturing it might therefore (a) constitute a terrible waste of cortical representational power, or, worse, (b) interfere with, or warp, the parameterization of the aspects of the input that are important, making it harder to extract critical distinctions. The book does not address this issue, relying on there being enough predictive power to capture any and all predictions, including predictive characterisation of motor control. Second, although our subjective sense is that we build a sophisticated predictive model of the entire sensory input, experiments into such phenomena as change blindness ( Rensink 2002 ) show this probably isn't true. A classic example involves alternating the presentation of two pictures, which differ in some significant way (e.g., the colour of the trousers of one of the main protagonists). Subjects have great difficulty in identifying the difference between the pictures, even though (a) they are explicitly told to look for it, (b) they have the subjective sense that they have represented all the information in each picture, and (c) if the location of the change is pointed out, they see it as blindingly obvious. This, and other attentional phenomena, suggests that substantially less is actually represented than we might naively think. In fact, elaborate computations go into selecting aspects of the input to which the models might be applied, and sophisticated models of these computations, such as Li's salience circuit (2002) , involve aspects of cortical anatomy and physiology ignored in the book. As a final example of a spur to insomnia, unsupervised learners worry that Damasio (1994) might be somewhat right. That is, cool logic and hot emotion may be tightly coupled in a way that a model such as this that is rigidly confined to cortical processing, ignoring key subcortical contributions to practical decision making, will find hard to capture. To sum up, in terms of the adage that genius is 1% inspiration and 99% perspiration, the book's enthymematic nature suggests that not quite enough sweat has been broken. Were it 1% inspiration and 99% aspiration, though, then the appealing call to arms for a new generation of modellers should more than suffice.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526780.xml

523834

The Future of Surgical Research

Surgeons seem to love publishing case series, which are of limited usefulness. How can we encourage them to do randomized clinical trials?

In 1996, Richard Horton, editor of the Lancet, chastised much of current surgical research and, in particular, questioned the usefulness of the case series as a predominant form of communication among surgeons [ 1 ]. He asked a poignant question: “Does surgical research have a future?” Nearly a decade later, it is important for surgeons and non-surgeons alike to revisit Horton's challenge. Why Surgeons Favor Case Series Randomized controlled trials (RCTs) have become the pillar of clinical research. Such trials attempt to obtain an unbiased randomization of patients with respect to known and unknown baseline conditions and to assess the effects of an intervention. However, only a minority of surgical studies involve a valid randomization scheme. The case series remains a favored method of clinical investigation in surgery. Case series are easy to perform, require less resources in terms of personnel and funds, can be performed at a single center, and, for many surgeons, represent a means to illustrate their surgical method and skills. In many instances, case series also serve as valuable intellectual background for future clinical or scientific work. For example, consider Dennis Burkitt's report on jaw tumors in African children, Alfred Blalock's initial efforts in cardiac surgery, or, more recently, Starzl and colleagues' observations, in a small collection of patients, of donor leukocyte chimerism, whereby recipients acquire tolerance to foreign donor cells. In all three cases, the authors' work led to powerful shifts in our understanding of the biology and treatment of disease [ 2 , 3 , 4 ]. All were case reports or case series—but under the current paradigm adopted by most journals and evidence-based databases, they would not be valued [ 5 , 6 , 7 ]. Surgical research needs to move from case series to RCTs (Photo by Linda Bartlett, National Cancer Institute) Barriers to Surgical RCTs There are many reasons why RCTs in surgical patients may be more difficult to perform than those in non-surgical patients. One of the most important—though least understood—is that the complexities of human disease in surgical patients makes them a more difficult group to study. Surgical patients are often heterogeneous in many more ways than non-surgical patients. So it would be inherently easier, for example, to study a new medication for generally healthy young adults with essential hypertension than a surgical technique for older patients with hepatic failure needing transplantation. In addition, while there may be value in studying patients from multiple centers, there may be important differences in the skill levels of different surgeons, either between centers or across the country. For example, the skill levels of surgeons in trials of carotid endarterectomy may be greater than those across the surgical community as a whole. This makes the applicability of some surgical RCTs to the wider community less certain than trials of medical therapies. So when it comes to surgical research, for both researchers and funding agencies, it is easier to grapple with a difficult, but ultimately soluble, basic science question than to face the uncertainty of clinical research. Investigators understand these implicit issues and trim their sails accordingly. Improving the Rigor of Research Nonetheless, too much surgical work is conducted in the less rigorous format of the case series. What can and should be done to improve the rigor of surgical investigation? It would seem that improvements are required from within and beyond the surgical world. First, as Horner observed, and several eminent surgeons have since agreed, reforms must begin within the field itself [ 1 , 5 , 6 , 7 ]. Both during surgical training and in the early years of faculty development, surgeons must obtain a thorough grounding in the principles of basic research and proper clinical investigation. Second, surgeons must establish firm and friendly relations with biostatisticians so that the latter may play a strong role in helping to develop adequately powered studies that can answer critical questions raised by new therapies and techniques. This is an especially acute need in an accelerating age of targeted therapies and disease biomarkers. Third, surgeons must re-engage in the clinical research enterprise and resume leadership roles in local and national clinical trials that involve surgical patients. In the United States, for example, an important step in this regard has been the establishment of the American College of Surgeons Oncology Group, which invites surgeons from all sectors, including private practice, to become active participants in well-designed, multi-institutional trials [ 5 ]. Similar efforts are needed on a global level. Finally, similar to the pressures faced by their colleagues elsewhere in academia, surgeon clinician-investigators must be nurtured, protected, and valued by their colleagues and medical administrators. The financial health of academic medical centers relies heavily on the generation of clinical revenue, which in many centers falls disproportionately on the shoulders of surgeons. New paradigms for revenue generation and funding of clinical research are needed. Funding for Surgical Research Beyond the walls of the academic medical center, there also needs to be greater recognition of the value of scientifically sound surgical research and clinical investigation. However, the National Institutes of Health (NIH), the major source of biomedical funding in the United States, continues to convey a less welcoming attitude toward surgical research than toward other types of clinical or basic science[ 8 , 9 ]. At the NIH, the principal instrument for performing peer review and making grant funding decisions is the study section, composed of about 10–20 members with expertise in a given field. There are few study sections devoted to surgically oriented clinical research and only two study sections (from among more than 100) in which surgeons make up even a reasonable minority of the committee members [ 8 ]. In comparison to those in other clinical departments, surgical grant proposals are less likely to be funded, and awards, when funded, are smaller [ 8 ]. Funding agencies need to recognize the importance of the surgical endeavor to modern medicine. Surgical research is also impeded by processes affecting other types of research as well. The number of researchers under 35 years of age receiving a first RO1 grant, the main NIH mechanism for external funding, in any field, is below 4%. The average age of initial funding for US physicians is about 44 years, and shows a trend toward advancing age that has progressed significantly in the past two decades. Thus, the NIH appears to reward experience and proven results very heavily, which may stifle innovation and likely serves as an innate barrier for younger physician-investigators contemplating research careers [ 9 ]. To help correct for this worrisome trend, the NIH created the “K” award system—career development grants designed to help starting researchers gain the experience needed to compete for RO1 grants. However, nearly 40% of the clinicians who receive KO8 awards never apply for RO1 funding [ 10 ], which suggests that the overall support—both explicit and implicit—for clinical research at the institutional and funding levels is inadequate. Finally, outside the US, surgeons face similar, if not greater problems. This bodes poorly for countries where the cost of evaluating new therapies and technologies may be an unaffordable luxury. These challenges to the surgical research enterprise are therefore global issues and should merit the attention of surgeons, medical institutions, and funding agencies in all countries. The Future What can be done? On the national and international level, funding agencies need to recognize the importance of the surgical endeavor to modern medicine. Recently, in the US the NIH unveiled a “roadmap” ( http://nihroadmap.nih.gov ) designed to provide “new pathways to discovery.” Clear, careful, scientific surgical investigation must be part of this roadmap, although it is not specifically mentioned. Outreach efforts to include surgeons in a variety of study sections should be made to ensure that important insights into the pathophysiology and treatment of disease, with which surgeons are concerned on a daily basis, are not overlooked. Additional efforts are needed to improve funding for clinical research, both for individuals at early stages of their careers and for multi-disciplinary clinical research and clinical trials. Locally, and individually, surgeons must join efforts to improve the clinical research enterprise by including training in clinical investigation at an early stage in medical school and during surgical residency training, fostering the careers of young surgeon-investigators through committed, protected time, participating in local and national clinical research groups, and recognizing that development as a clinical researcher takes time—many years in fact. These efforts may help ensure that surgical research is a vital part of the future of medicine and that it leads to the kind of high-quality work that shapes and remodels the face of medicine. To foster these efforts, surgeons must change and adapt to the currents of modern medical research. If this is successful, the case series will become the occasional rather than the common form of surgical communication. And surgeons, other clinicians, and, most importantly, basic scientists will be better able to take advantage of the new avenues of biomedical science opening before us. But the case series will always represent one important tool for early studies or uncommon conditions. It remains true that while the method one uses influences the answer one receives, it can be just as important to ask the right questions, which can be asked even in a series of one patient [ 11 ]. And surely that is the place one must begin.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC523834.xml

514652

A single amino acid determines preference between phospholipids and reveals length restriction for activation ofthe S1P4 receptor

Background Sphingosine-1-phosphate and lysophosphatidic acid (LPA) are ligands for two related families of G protein-coupled receptors, the S1P and LPA receptors, respectively. The lysophospholipid ligands of these receptors are structurally similar, however recognition of these lipids by these receptors is highly selective. A single residue present within the third transmembrane domain (TM) of S1P receptors is thought to determine ligand selectivity; replacement of the naturally occurring glutamic acid with glutamine (present at this position in the LPA receptors) has previously been shown to be sufficient to change the specificity of S1P 1 from S1P to 18:1 LPA. Results We tested whether mutation of this "ligand selectivity" residue to glutamine could confer LPA-responsiveness to the related S1P receptor, S1P 4 . This mutation severely affected the response of S1P 4 to S1P in a [ 35 S]GTPγS binding assay, and imparted sensitivity to LPA species in the order 14:0 LPA > 16:0 LPA > 18:1 LPA. These results indicate a length restriction for activation of this receptor and demonstrate the utility of using LPA-responsive S1P receptor mutants to probe binding pocket length using readily available LPA species. Computational modelling of the interactions between these ligands and both wild type and mutant S1P 4 receptors showed excellent agreement with experimental data, therefore confirming the fundamental role of this residue in ligand recognition by S1P receptors. Conclusions Glutamic acid in the third transmembrane domain of the S1P receptors is a general selectivity switch regulating response to S1P over the closely related phospholipids, LPA. Mutation of this residue to glutamine confers LPA responsiveness with preference for short-chain species. The preference for short-chain LPA species indicates a length restriction different from the closely related S1P 1 receptor.

Background Sphingosine-1-phosphate (S1P) and lysophosphatidic acid (LPA) are phospholipid growth factors which are present in normal serum and plasma. These lipids elicit diverse responses from a wide range of cell types, including enhanced cell survival, cell proliferation, induction of cytoskeletal changes and chemotaxis (reviewed in [ 1 - 4 ]. Some of these responses reflect activation of G protein-coupled receptors of the endothelial differentiation gene (Edg) family. The Edg receptor family is classified into two clusters based on ligand selectivity: S1P 1/2/3/4/5 (formerly Edg1/5/3/6/8) specifically respond to S1P whilst LPA 1/2/3 (formerly Edg2/4/7) respond to LPA [ 5 ]. Members of the S1P receptor family display higher sequence similarity to each other (approximately 40% identity) than to members of the LPA receptor family (approximately 30% identity). These homologies, coupled with observed differences in the structure of S1P and LPA receptor genes, suggest that these receptor families evolved from distinct ancestral genes. The S1P receptors contain a conserved glutamic acid residue present within the third TM that corresponds to glutamine in the LPA receptors. Interaction between distinct functional groups present on S1P and LPA with this residue was shown for the S1P 1 and LPA 1 receptors using computational modelling techniques [ 6 , 7 ] and was demonstrated as the basis for the ligand preference displayed by the receptors. Experimental characterisation confirmed that replacement of glutamic acid with glutamine in S1P 1 changed ligand specificity from S1P to LPA, and the reciprocal mutation in LPA 1 resulted in recognition of both LPA and S1P [ 7 ]. In the present study, the role of this residue in determining ligand selectivity for the S1P 4 receptor was examined. Phylogenetic analysis of the Edg family of receptors indicates that S1P 4 is more closely related to other S1P receptors than receptors which respond to LPA. However, S1P 4 lies on the edge of the S1P family cluster and has been shown to bind S1P with lower affinity than other S1P receptors and hence it has been suggested that S1P is not the true endogenous agonist of this receptor [ 8 ]. We therefore decided to investigate whether replacement of this residue (E 3.29(122) ) with glutamine conferred LPA-responsiveness to the S1P 4 receptor and hence determine the role of this residue in this lower-affinity S1P receptor. To achieve this, we expressed wild type and E 3.29(122) Q mutant S1P 4 receptors in CHO-K1 cells and studied responses to lysophospholipids using a [ 35 S]GTPγS binding assay. Since CHO-K1 cells respond to LPA, we utilised fusion proteins constructed between the S1P 4 receptor and a pertussis toxin-insensitive Gα i1 (C 351 I) G protein. Expression of these proteins in CHO-K1 cells followed by treatment with pertussis toxin prior to harvest allowed elimination of any signal due to stimulation of endogenous LPA receptors. Within this study, we also examined how the length of the LPA acyl chain affected potency at the mutant S1P 4 receptor, using a panel of naturally occurring LPA analogues. Computational models of complexes between the wild type or mutant S1P 4 receptor and S1P and LPA species were used to provide a molecular interpretation of the experimental findings. Results Human HA-S1P 4 was mutated at position 122 to replace the naturally occurring glutamic acid with glutamine. The mutant and wild type receptors were stably expressed in CHO-K1 cells as in-frame GPCR-G protein fusions with pertussis toxin-insensitive Gα i1 (C 351 I). Western blotting was used to detect expression of these fusion proteins. Membranes from HA-S1P 4 -Gα i1 (C 351 I)- or HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I)-transfected cells contained a polypeptide with an apparent molecular mass of approximately 110 kDa, which reacted with anti-HA and anti-SG1 antibodies (Figure 1A and 1B ) and was consistent with expression of the GPCR-G protein fusion. Confirmation of comparable cell-surface expression of these proteins was obtained via FACS analysis using an anti-HA antibody directly conjugated with fluorescein (Figure 1 Panel III). Figure 1 Expression of HA-S1P 4 -Gα i1 and HA-S1P 4 (E 122 Q)-Gα i1 (C 351 I) in CHO-K1 cells. Membranes from untransfected CHO-K1 cells (lane 2) and CHO-K1 cells stably expressing HA-S1P 4 -Gα i1 (C 351 I) (A) or HA-S1P 4 (E 122 Q)-Gα i1 (C 351 I) (B) were analysed by Western blotting using anti-HA (panel I) or anti-Gα i1 (panel II) antibodies. Visualisation of immunoreactive proteins was achieved using chemiluminescence after incubation of the blot with appropriate HRP-conjugated secondary antibodies. The position of each HA-S1P 4 fusion protein is indicated by an arrow. Cell-surface expressed HA-S1P 4 receptor was detected by FACS analysis (panel III) using a Fluorescein conjugate of the anti-HA antibody (blue line). Cells were also stained with an isotype matched control antibody (red line). No staining of untransfected CHO-K1 cells was observed using the Fluorescein conjugate of the anti-HA antibody (not shown). Data are presented as overlay histograms and are representative of at least five independent experiments. The response of HA-S1P 4 -Gα i1 (C 351 I) and HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) to S1P was assessed using membranes from cells transfected to express these proteins and treated with pertussis toxin prior to harvest. S1P promoted dose-dependent increase in [ 35 S]GTPγS binding to membranes containing HA-S1P 4 -Gα i1 (C 351 I) with an EC 50 of 355 nM ± 155 nM (n = 3); in contrast, membranes from HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I)-expressing cells demonstrated severely impaired response to S1P (Figure 2A ). The EC 50 for S1P stimulation of the HA-S1P 4 -Gα i1 fusion protein (355 ± 155 nM) was not statistically significantly different from that obtained using the unfused HA-S1P 4 receptor (439 ± 187 nM, Figure 2B ) and compared favourably with published values for this receptor in HEK293T cells from two different research groups of 270 nM [ 9 ] and 790 nM [ 10 ]. Figure 2 Sensitivity of HA-S1P 4 -Gα i1 (C 351 I) and HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) to S1P in [ 35 S]GTPγS binding assay. A. Membranes from CHO-K1 cells transfected with HA-S1P 4 -Gα i1 (C 351 I) ( filled squares ) or HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) ( open circles ) which had been cultured in the presence of 100 ng/mL pertussis toxin for 24 hours prior to harvest were stimulated with varying concentrations of S1P for 30 minutes at 30°C in the [ 35 S]GTPγS binding assay. Gα i G proteins were immunoprecipitated after solubilisation and preclearance with non-immune serum. Data are the mean of three determinations ± SEM from a single experiment, and are representative of three such experiments performed. B. Dose-dependent stimulation of wild-type HA-S1P 4 ( filled squares ) and HA-S1P 4 -Gα i1 fusion protein ( filled triangles ) by S1P measured as described in panel A. These data are representative of three such experiments performed and analysis of mean EC 50 values obtained for each protein showed them to be not statistically different. Structure activity relationships were determined by the [ 35 S]GTPγS assay for S1P 4 or its E 3.29(122) Q mutant using S1P and LPA species with 14:0, 16:0 and 18:1 acyl chains at a single (10 μM) concentration. Of the lysophospholipids tested, only S1P induced a strong response over basal levels (approximately 48% ± 5%) in membranes containing HA-S1P 4 -Gα i1 (Figure 3A ), whilst 18:1 LPA did not stimulate a statistically significant response; weak stimulation of [ 35 S]GTPγS binding was observed with 14:0 and 16:0 LPA (approximately 11% ± 3% above basal in each case). In contrast, membranes containing HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) showed at least weak response to each ligand (Figure 3A ). The weakest agonist at the E 3.29(122) Q mutant S1P 4 receptor was 18:1 LPA, which produced only 14% ± 2% stimulation over basal. S1P and 16:0 LPA gave approximately 21% ± 4% and 19% ± 4% stimulation over basal response, respectively. The best agonist for the E 3.29(122) Q mutant was 14:0 LPA, which gave a 40% ± 2% enhancement over basal levels. The stimulation promoted by 14:0 LPA was statistically different from that produced by 18:1 LPA (p < 0.01). Sensitivity of the receptor to stimulation by each form of LPA, and particularly 14:0 LPA, was markedly increased after introduction of the E 3.29(122) Q mutation and indicated that this position was important in influencing HA-S1P 4 ligand preference. Figure 3 Ligand preference of HA-S1P 4 (E 122 Q)-Gα i1 (C 351 I) in [ 35 S]GTPγS binding assay. Membranes were stimulated for 30 minutes at 30°C with lysophospholipid ligands in the [ 35 S]GTPγS binding assay and Gα i G proteins immunoprecipitated after solubilisation and preclearance with non-immune serum. (A) Membranes from CHO-K1 cells transfected to express HA-S1P 4 -Gα i1 (C 351 I) or the mutant HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) and incubated with 100 ng/mL pertussis toxin for 24 hours prior to harvest, were left untreated (basal), or treated with vehicle or 10 μM concentrations of 18:1 LPA, 16:0 LPA, 14:0 LPA or S1P. Data are the mean of three determinations ± SEM from a single experiment and are representative of three such experiments performed. Statistical significance from the basal responses of each set of membranes tested is denoted by * (P < 0.05) or ** (P < 0.01); ## denotes statistical significance from the response to 18:1 LPA (P < 0.01) for the HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I)-transfected membranes. (B) Membranes from CHO-K1 cells transfected to express HA-S1P 4 (E 122 Q)-Gα i1 (C 351 I) and incubated with 100 ng/mL pertussis toxin for 24 hours prior to harvest, were stimulated with various concentrations of S1P ( crosses ), 18:1 LPA ( open circles ) or 14:0 LPA ( filled triangles ). Data are the mean of three determinations ± SEM and are representative of three such determinations performed. These results indicate that introduction of the E 3.29(122) Q mutation in the S1P 4 receptor confers LPA-responsiveness, and that a short form of LPA was a more effective agonist than the intermediate and longer forms, when tested at this single concentration. Dose response curves were constructed for ligand-induced activation of the E 3.29(122) Q S1P 4 mutant by the 14:0 and 18:1 forms of LPA as well as S1P (Figure 3B ). An EC 50 could only be determined for the 14:0 form of LPA as S1P and 18:1 LPA caused minimal stimulation at only the highest concentration tested. The EC 50 value for activation of HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) was calculated to be 3.8 ± 1.4 μM. However, since a plateau of maximal stimulation was not achieved, interpretation of this EC 50 value needs caution. This result clearly showed that 14:0 LPA was a weak agonist of HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) and hence confirmed the involvement of residue 122 in S1P 4 ligand preference. Similar results were obtained using a second CHO-K1 clone expressing this fusion protein (not shown). Computational modelling of the S1P complex with the wild type S1P 4 receptor identifies the best S1P binding site within the TM with the phosphate group at the extracellular end (Figure 4A ). Ion pairs appear between the phosphate group of S1P and two cationic amino acids, R 3.28(121) and K 5.38(202) . An additional ion pair occurs between the cationic ammonium of S1P and E 3.29(122) . Hydrophobic residues from TM2, TM3, TM5 and TM6 line the binding pocket and surround the alkyl chain of S1P. Figure 4 Computational models of wild type S1P 4 and its E3.29(122)Q mutant with S1P and LPA species. Computational models of the complexes between the wild type S1P 4 or its E 3.29(122) Q mutant with S1P or various LPA species generated by Autodock 3.0 and minimised using the MMFF94 forcefield in the MOE program. Complexes in each panel are shown from the same viewpoint with the extracellular end of the receptors oriented to the top of the figure. Standard element color codes are used with grey, white red, blue and magenta representing carbon, hydrogen, oxygen, nitrogen and phosphorous. Ribbons are shaded from red at the amino-terminus to blue at the carboxy-terminus. (A) Model of the complex between S1P (spacefilling) and the wild type S1P 4 receptor. Residues in the receptor involved in ion pairs with S1P are shown as stick models and labelled. (B) Superimposition of the wild type S1P 4 complex with S1P (orange) and the E 3.29(122) Q S1P 4 mutant complex with 14:0 LPA (green). For clarity, the only position at which the modelled amino acid position is shown for both receptor models is 3.29(122). Other residues had very similar optimised positions in the two model structures. (C) Superimposition of wild type S1P 4 complexes with 18:1 LPA (cyan), 16:0 LPA (yellow) and 14:0 LPA (green) on E 3.29(122) Q mutant complexes with 18:1 LPA (blue-green), 16:0 LPA (gold) and S1P (orange). For clarity, the only position at which modelled amino acid position is shown for both the wild type and mutant receptor models is 3.29(122). Other residues had very similar optimised positions in all model structures. (D) Space-filling models which represent the minimised extended conformation of each structure were constructed using SYBYL 6.9 software (Tripos Inc., St. Louis, MO., U.S.A.). The distance between phosphorus and terminal carbon atoms was predicted for each structure listed from top to bottom: 18:1 LPA, 27.0 Å; 16:0 LPA, 26.7 Å; 14:0 LPA, 24.2 Å; S1P, 24.0 Å. The best complex of 14:0 LPA in the E 3.29(122) Q S1P 4 mutant receptor model has striking similarity to the best complex of S1P in the wild type S1P 4 receptor model (Figure 4B ). Both models demonstrate ion pairing between the phosphate group and two cationic amino acids, R 3.28(121) and K 5.38(202) . Each ligand interacts with the amino acid at position 3.29(122), S1P by an ion pair with the carboxylate of the wild type glutamate and 14:0 LPA by a hydrogen bond with the mutated glutamine. Multiple hydrophobic residues surround the nonpolar tails of the lipid ligands. The superimposition of the two complexes (Figure 4B ) also demonstrates that the ligands occupy almost identical volumes. Common interactions and overlap volumes are qualitatively consistent with the experimental findings that these ligands give similar 48% and 40% maximal stimulation over basal for S1P at the wild type and 14:0 LPA at the mutant receptor, respectively. In contrast to the complexes of 14:0 LPA with E 3.29(122) Q S1P 4 and S1P with wild type S1P 4 , the remaining complexes show much less common volume (Figure 4C ). Most complexes exhibit the phosphate interactions described for 14:0 LPA with E 3.29(122) Q S1P 4 and S1P with wild type S1P 4 . Of particular interest is the observation that the best complexes generated by Autodock for the 18:1 LPA species with wild type S1P 4 has a very high positive van der Waals interaction energy, > 3000 kcal/mol, compared to values well under 200 kcal/mol for every other complex studied. In the best complexes found for 16:0 LPA and 18:1 LPA in both constructs, the terminal six to eight carbons of the hydrophobic tails fold into L-shaped conformations quite different from the extended conformations observed in the S1P complex with wild type S1P 4 or the 14:0 LPA complex with the E 3.29(122) Q S1P 4 mutant. The terminal carbons in several complexes curl out of the receptor between TM5 and TM6 (Figure 4C ) due to the restricted length of the binding pocket. These results suggest that the complete lack of S1P 4 activation in response to 18:1 LPA is likely due to failure to form a complex. The strongest complexes formed, S1P with wild type S1P 4 and 14:0 LPA with the E 3.29(122) Q S1P 4 mutant, have complementary interactions with the residue at position 3.29(122). These strong complexes give the most robust activation. Weak complexes are formed for other combinations due to mismatched interactions with position 3.29(122) or excessive length of the hydrophobic tail. The presence of hydrophobic tails of 16:0 or 18:1 LPA between transmembrane domains may additionally impair the conformational change necessary for full agonist responses. Discussion Parental CHO-K1 cells respond to LPA in functional assays, reflecting expression of endogenous LPA 1 (G. Holdsworth, et al ., manuscript in preparation). For this reason, fusion proteins between wild type or mutant HA-S1P 4 and the pertussis toxin-insensitive Gα i1 (C 351 I) G protein were used in these studies. Expression of these proteins in CHO-K1 cells followed by treatment with pertussis toxin prior to harvest allowed elimination of any signal due to stimulation of endogenous LPA receptors. McAllister et al . [ 11 ] (.(adopted a similar approach for studies of the LPA 1 receptor. We examined the role of residue E 3.29(122) in controlling S1P 4 ligand selectivity using functional and computational methods. This residue, which is conserved throughout the S1P receptors, has been shown to control ligand specificity for the related S1P 1 receptor [ 7 ]. Introduction of the E 3.29(122) Q mutation severely affected the response of S1P 4 to S1P: in dose-response experiments S1P caused minimal stimulation at only the highest concentration of ligand used. This is in agreement with published observations for activation of the equivalent S1P 1 mutant [ 7 ]. 14:0 LPA was able to induce dose-dependent stimulation of S1P 4 (E 3.29(122) Q) with an EC 50 of approximately 3.8 μM but only promoted minimal stimulation of the wild type S1P 4 receptor. The modelled complexes of 14:0 LPA with E 3.29(122) Q S1P 4 and S1P with wild type S1P 4 demonstrate nearly identical volumes occupied by the two ligands and very similar interactions between these ligands and their respective receptors. Of particular importance are amino acid residues at positions 3.28(121), 3.29(122) and 5.38(202), which either ion pair with the phosphate or interact with the 2-amino or 2-hydroxyl group in S1P and 14:0 LPA, respectively. The importance of interactions with amino acids at positions 3.28 and 3.29 has been previously noted for the S1P 1 [ 6 , 7 ] and LPA 1,2,3 [ 7 , 12 ] receptors. The S1P 4 receptor exhibits marked constitutive (agonist-independent) activity (Figure 5 ) which was unaffected by the introduction of the E 3.29(122) Q mutation (data not shown). This indicates that the mutation perturbs S1P recognition without affecting the ability of the receptor to spontaneously adopt an active conformation. Similar observations have been reported for the β 2 AR, where a mutation in the sixth transmembrane domain abolished agonist activation but not constitutive activity [ 13 ]. Figure 5 Constitutive activity of HA-S1P 4 and HA-S1P 4 -Gα i1 (C 351 I). Comparison of basal, vehicle and stimulated (10 μM S1P) GTPγS binding for membranes prepared from CHO cells transfected to express HA-S1P 4 or HA-S1P 4 fused to PTx insensitive Gα i1 . Where indicated, cells were treated with 100ng/ml PTx for 24 hours prior to harvesting. PTx treatment of HA-S1P 4 -transfected membranes prevented activation by S1P and also caused a dramatic reduction in basal signalling, indicative of constitutive activity. In contrast, when the HA-S1P 4 -Gα i1 fusion-expressing membranes were treated with PTx, there was only a slight reduction in basal signalling and the receptor still responded to exogenous S1P, indicating that the receptor signalled via the tethered, PTx-insensitive G protein. Basal signalling in PTx-treated HA-S1P 4 -Gα i1 -expressing membranes exceeded that seen with membranes from untransfected CHO cells. Unlike S1P, which exists as a single species in vivo , the term LPA actually refers to a family of molecules that take the general form 1- o -acyl-2-hydroxy- sn -glyceryl-3-phosphate. Naturally occurring forms of LPA contain acyl chains of differing lengths, with differing degrees of saturation. Investigations into the effect of the length and degree of saturation of the acyl chain of LPA have been undertaken for the LPA receptors [ 14 , 15 ], but limited SAR information is available for S1P receptors (22). The LPA-responsive E 3.29(122) Q S1P 4 mutant facilitates structure activity relationship (SAR) studies due to the greater availability of LPA analogs relative to S1P analogs. Comparison of space-filling models of the structures of S1P and three analogues of LPA (Figure 4D ) revealed that 14:0 LPA most closely resembled S1P in terms of apparent length. [ 35 S]GTPγS binding assays demonstrated greater agonist activity of 14:0 LPA at the mutant receptor relative to 18:1 or 16:0 LPA. This SAR indicates a length restriction for the S1P 4 agonist binding site. Model complexes of 16:0 and 18:1 LPA contained alkyl chains that fold at the bottom of the binding pocket, defined by a cluster of hydrophobic amino acids. Three of these differ either in position of sidechain branching or size relative to LPA receptors and the other S1P receptors. Position 2.46, I88 in S1P 4 , is leucine in LPA 1–3 and other S1P receptors. Residue I6.40(256) is larger than the valine found in the other four S1P receptors, LPA 1 and LPA 3 . Finally, I7.51(305) corresponds to the smaller valine in S1P 2 and S1P 3 and the much smaller alanine in LPA 2 . These findings provide a molecular explanation for a similar SAR observed using para-alkyl amide analogs of S1P [ 9 ]. SAR obtained with the S1P 4 mutant are in contrast to that shown by LPA receptors, which exhibit the general trend of 18:1 ≥ 16:0 > 14:0 for potency and maximal stimulation [ 15 ]. Since mutation of residue 122 in the S1P 4 receptor from the naturally occurring glutamic acid to glutamine conferred responsiveness to 14:0 LPA and severely affected responses to S1P, our observations support the hypothesis that this conserved residue in the third transmembrane domain of the S1P receptors is involved in ligand recognition. This is in contrast to a recent paper describing models of several GPCRs, including S1P 4 , which had been generated using novel first principle methods [ 16 ]. In this model of S1P 4 , interactions between S1P and residues T 7.34(127) and W 7.37(291) and E 7.30(284) were observed. Interaction of E 7.30(284) with the ammonium group of S1P appeared to control ligand selectivity since the other residues appeared to interact with the phosphate group, which is present on both LPA and S1P. It is therefore surprising that none of these residues are conserved throughout the S1P or LPA receptor families. The data presented here support the assertion that glutamic acid residue 3.29 present in the third transmembrane domain of the S1P receptors controls ligand selectivity and suggest that the S1P 4 model described by Vaidehi et al . [ 16 ] is inaccurate. The current study provides new information for the development of more selective S1P receptor agonists. In particular, an S1P analog with its hydrophobic chain extended by either 2 or 4 carbons would be a very poor agonist of the S1P 4 receptor. On the other hand, the activation of the S1P 1 -E 3.29(121) Q mutant by 18:1 LPA [ 7 ] indicates that a chain-extended S1P analog should retain agonist activity at the S1P 1 receptor. S1P receptor agonists with differing selectivity profiles will be useful tools to more completely map the physiological and pathophysiological roles of these receptors. Conclusions These studies confirm that glutamic acid residue 3.29, present in the third transmembrane domain of the S1P receptors is important for the selective recognition of S1P, versus the closely related lipid, LPA. Mutation of E3.29 to glutamine diminished response to S1P and allowed structure activity studies using the diverse available LPA species. The mutant S1P 4 receptor is stimulated most strongly by LPA 14:0 and is not activated by the longer LPA 18:1, in contrast with a previous report on the analogous S1P 1 receptor mutant that responded to LPA 18:1. Thus the S1P 4 receptor ligand binding pocket is shorter in length than the S1P 1 ligand binding pocket. Methods Residue nomenclature Amino acids within the TM of S1P 4 can be assigned index positions to facilitate comparison between receptors with different numbers of amino acids, as described by Weinstein and coworkers [ 17 ]. An index position is in the format x.xx. The first number denotes the TM in which the residue appears. The second number indicates the position of that residue relative to the most highly conserved residue in that TM which is arbitrarily assigned position 50. E3.29, then, indicates the relative position of glutamate 122 in TM3 relative to the highly conserved arginine 143 in the E(D)RY motif which is assigned index position 3.50 [ 17 ]. Materials Materials for tissue culture were supplied by Invitrogen Ltd. (Paisley, Scotland, U.K.). Foetal bovine serum was obtained from Helena Biosciences Ltd., (Sunderland, U.K.) or PAA Labs GmbH., (Linz, Austria). Pertussis toxin was purchased from CN Biosciences Ltd., (Nottingham, U.K.). Lysophosphatidic acid (18:1, 16:0 and 14:0) and S1P were from Avanti Polar Lipids Inc., (Alabaster, AL., U.S.A.). The SG1 antiserum was produced previously [ 18 ]. All other chemicals were from Sigma Aldrich Company Ltd., (Gillingham, Dorset, U.K.) or BDH Ltd., (Poole, Dorset, U.K.) unless stated otherwise. Construction of receptor expression plasmids The S1P 4 coding sequence was cloned from a human PBMC cDNA library using the sense primer 5'-GAGAGA GCGGCCGC CACCATG TATCCATATGATGTTCCAGATTATGCT AACGCCACGGGGACCCCGGTG-3', which contains a Not I restriction site (bold) and the haemagglutinin HA epitope tag (YPYDVPVYA, underlined) immediately after the initiator methionine, and the antisense primer 5'-GAGAGA GAATTC GGC GATGCTCCGCACGCTGGAGATG-3', which contains an Eco RI restriction site (bold) and changes the S1P 4 stop codon to alanine (underlined). A C 351 I mutant of the Gα i1 G protein (previously produced, [ 19 ]) was amplified using PCR with the sense cloning primer 5'-GAGAGA GAATTC GCCA CCATGGGCTGCACACTGAGCG-3', which contains the Eco RI restriction site (bold), and the antisense cloning primer 5'-GAGAGA GGATCC TTAGAAGAGACCGATGTCTTTTA G-3', which contains a Bam HI restriction site (bold). After digestion of each PCR product with the appropriate restriction enzymes, fragments were ligated into the pIRESpuro mammalian expression vector (Invitrogen Ltd.) to generate an in-frame fusion between HA-S1P 4 and Gα i1 (C 351 I). The E 3.29(122) Q mutation was introduced into the S1P 4 sequence in parallel PCR reactions. Complementary oligonucleotides were designed across the residue which was to be mutated such that each primer contained the necessary base change to mutate residue 122 to glutamine (underlined in each primer): sense mutational primer: 5'-CAGTGGTTCCTACGG CAG GGCCTGCTCTTCAC-3'; antisense mutational primer: 5'-GTGAAGAGCAGGCC CTG CCGTAGGAACCACTG-3'. Mutational sense or antisense primers were used in parallel PCR reactions with the appropriate antisense or sense cloning primer, with HA-S1P 4 plasmid DNA as template. Equimolar amounts of each purified PCR product were mixed and amplified in a further reaction, using the cloning primers described above. The resultant product was digested with the appropriate restriction enzymes and ligated with the Gα i1 sequence in the pIRESpuro expression vector to generate an in-frame fusion between HA-S1P 4 (E 3.29(122) Q) and Gα i1 (C 351 I). Cell culture and transfection CHO-K1 cells were maintained at 37°C with 5% CO 2 in Dulbecco's modified Eagle's medium (DMEM), supplemented with 10% foetal bovine serum (FBS), 2 mM L-glutamine and non-essential amino acids. Sub-confluent cell monolayers were stably transfected to express either HA-S1P 4 -Gα i1 (C 351 I) or HA-S1P 4 (E 3.29(122) Q)-Gα i1 (C 351 I) fusion proteins using Lipofectamine reagent (Invitrogen). 72 hours post-transfection, cells were seeded in media supplemented with 7.5 μg/mL puromycin and the resultant clones examined for expression of cell surface receptor using FACS analysis. Clonal cell lines were expanded in complete DMEM containing 7.5 μg/mL puromycin and were transferred to serum free DMEM approximately 24 hours prior to harvesting. Where indicated, 100 ng/mL pertussis toxin was included in the serum free medium. It should be noted that we initially expressed S1P 4 in RH7777 cells, which are unresponsive to S1P and LPA and have been commonly used for studies of Edg family receptors [ 20 ]. Unfortunately, our attempts to detect activation of S1P 4 expressed in these cells using a variety of functional assays were unsuccessful. Therefore, we used CHO-K1 cells as an alternative host in these studies; expression of functional S1P 4 in CHO-K1 cells has also been reported by Mandala et al. [ 21 ]. FACS analysis The amino-terminal HA-epitope tag was detected using a fluorescein conjugate of the anti-HA antibody, clone 3F10 (Roche Molecular Biochemicals Ltd., Lewes, U.K.). Cells were harvested non-enzymatically and washed with FACS buffer (PBS containing 3% FBS and 0.1% NaN 3 ) then stained with the 3F10 antibody (or an isotype matched control) for 40 minutes at 4°C in the dark. After washing with FACS buffer, cells were analysed using a FACScalibur flow cytometer (BD Biosciences, Oxon., U.K.). Preparation of cell membranes Cells were harvested non-enzymatically, washed with PBS and resuspended in "assay buffer" (20 mM Hepes, pH 7.4, 3 mM MgCl 2 , 100 mM NaCl), supplemented with "complete" protease inhibitors (Roche Molecular Biochemicals Ltd.). Cells were homogenised in a nitrogen cavitation chamber (500 psi for 15 minutes). Unbroken cells and nuclei were pelleted by centrifugation (500 × g , 10 minutes, 4°C) and the supernatant fraction was centrifuged at 45,000 × g for 45 minutes at 4°C. Membrane pellets were resuspended in assay buffer, titurated through a fine gauge needle and stored at -80°C until required. Immunoblot analysis Samples were resolved by SDS-Page on 4–20% Tris-Glycine gels (Invitrogen) and were transferred to Immobilon-P membrane (Millipore Ltd., Herts., U.K.). The membrane was blocked using 2.5% Marvel in PBS before incubating with primary antibodies which had been diluted in PBS/0.1% Tween-20 containing 1% Marvel. The high affinity rat anti-HA antibody was diluted 1 in 500; the anti-G αi1 antibody (Autogen Bioclear Ltd., Wilts., U.K.) was diluted 1 in 1000. Immunoreactivity was detected using an appropriate horseradish peroxidase-conjugated secondary antibody, diluted 1 in 10,000 in PBS/0.1% Tween-20 containing 1% Marvel, followed by detection using SuperSignal reagents (Perbio Science Ltd., Cheshire, U.K.). [ 35 S]GTPγS binding assay [ 35 S]GTPγS binding experiments were performed essentially as described previously [ 22 ]. Briefly, membranes were incubated with or without the indicated ligand for 30 minutes at 30°C in assay buffer containing [ 35 S]GTPγS (100 nCi/point), saponin (20 μg/point) and 0.1 μM GDP. 18:1 LPA was prepared as a 2 mM DMSO stock whilst 16:0 and 14:0 LPA were prepared as 2 mM stock solutions in 1:1 ethanol:water per supplier recommendation due to their poor solubility in DMSO. S1P had previously been dispensed as thin film aliquots (dissolved in MeOH and the solvent evaporated under nitrogen) in brown glass vials and stored at -70°C prior to use. Lipids (S1P or LPA forms) were diluted in assay buffer containing 1% fatty acid free BSA, such that the final concentration of BSA in the assay was 0.1%. Following incubation, membrane protein was solubilised with 1.25% NP-40 and 0.4% SDS and after pre-clearance using non-immune serum, Gα i1/2 subunits were immunoprecipitated with SG1 antiserum, used at a dilution of 1 in 200. Non-specific binding was determined by the addition of 100 μM GTPγS. Bound radioactivity was measured using liquid scintillation counting. Experimental data analysis Numerical data are expressed as means ± standard error, shown as error bars in the appropriate figures. Statistical comparisons were made using one-way ANOVA with Dunnett's multiple comparison post test. Receptor model development A model of human S1P 4 (GenBank™ accession number AAP84350) was developed by homology to the experimentally-validated model of S1P 1 [ 23 ]. Alignment of the S1P receptor sequences was performed using the MOE software package (version 2003. 01 ed. Chemical Computing Group, Montreal, Canada). The alignment was optimised by the manual removal of gaps within the TM, and alignment in the region of TM5 was shifted one position to correctly orient K5.38(202) toward the interior of the helical bundle (Pham, et al ., unpublished data). A preliminary model was generated by homology modelling using default parameters and subsequently manually refined to optimise interhelical hydrogen bonding. Cis-amide bonds present in the loop regions were converted to the trans conformation by manual rotation followed by the minimisation of two residues on either side of the amide linkage to a root mean square (RMS) gradient of 0.1 kcal/mol·Å using the MMFF94 forcefield [ 24 ]. After these manual refinements, the receptor model was optimised using the MMFF94 forcefield to an RMS gradient of 0.1 kcal/mol·Å. A model of S1P 4 with the E 3.29(122) Q mutation was developed by performing the appropriate mutation in MOE, and saturating the residue with hydrogen atoms. To allow the sidechains of the other residues in the binding pocket to adapt to the presence of the new moiety, the backbone atoms of the receptor were fixed and the receptor was optimised to an RMS gradient of 0.1 kcal/mol·Å using the MMFF94 forcefield [ 24 ]. Ligand model development Computational models of the naturally-occurring stereoisomers of 14:0 LPA, 16:0 LPA, 18:1 LPA, and S1P were built using the MOE software package. The -1 ionization state for the phosphate functionality was chosen for all ligands, and the +1 ionization state was chosen for the amine moiety of S1P. Previous docking studies using the -2 ionization state for phosphate in related systems yield essentially identical geometries as studies using the -1 ionization state. These ligands were geometry optimised using the MMFF94 force field [ 24 ]. Docking Using the AUTODOCK 3.0 software package [ 25 ], 14:0 LPA, 16:0 LPA, 18:1 LPA, and S1P were docked into the S1P 4 wild type and S1P 4 E 3.29(122) Q mutant receptor models. Each docking box was centered near F 3.33(126) with dimensions of 30.75 × 23.25 × 23.25 or 32.25 × 23.25 × 23.25 Å for shorter (S1P and 14:0 LPA) or longer (16:0 and 18:1 LPA) ligands, respectively. At least 20 putative complexes were generated for each receptor:ligand pair using docking parameters at default values with the exception of the number of energy evaluations (2.5 × 10 8 ), generations (10000) and maximum iterations (3000). Resultant complexes were evaluated based on final docked energy, Van der Waals interaction energies from the MMFF94 forcefield as well as visual analysis. The complexes with the lowest final docked energies and others of interest were geometry optimised using the MMFF94 force field [ 24 ], and the lowest energy complex after minimisation was chosen as the final complex structure. Abbreviations CHO, Chinese hamster ovary; Edg, endothelial differentiation gene; ERK, extracellularly regulated kinase; FACS, fluorescence activated cell sorter; G protein, guanine nucleotide-binding protein; GPCR, G protein-coupled receptor; HA, haemagglutinin; LPA, lysophosphatidic acid; MAP, mitogen-activated protein kinase; PBMC, peripheral blood mononuclear cell; PTx, pertussis toxin; SIP, sphingosine-1-phosphate; TM, transmembrane domain Authors' contributions G Holdsworth performed and interpreted all studies with experimental S1P 4 fusion proteins and drafted the manuscript. D Osborne performed and interpreted docking studies to generate all mutant complexes with all LPA species and S1P and all wild type complexes with LPA species. TC Pham generated the homology model of the human S1P 4 receptor. J Fells performed docking studies of S1P with the wild type S1P 4 receptor. G Hutchinson and G Milligan participated in the design and coordination of the experimental studies with S1P 4 fusion proteins. A Parrill participated in the design and coordination of the modelling studies and edited the manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514652.xml

515376

Spotting Signs of Natural Selection

Milk, cheese, and yogurt are so ingrained in the diets of Europeans that it's easy to forget that their ancestors ever ate differently. But about 9,000 years ago, before the domestication of cows, sheep, and goats, milk was a staple only for babies. Back then—just as in most Asian and African cultures today—individuals lost their ability to digest lactose, a sugar found in milk, as they grew up. But with the domestication of animals, milk became abundant. Among herders, individuals had an advantage if they had versions of genes, also known as alleles, that allowed them to digest lactose into adulthood. They would tend to be healthier and reproduce more than those who could not digest lactose. Thus, by natural selection within herding groups, over generations those who could drink milk into adulthood became more common. Researchers have found the allele that allows adults to digest lactose, and it's one of the clearest signs of natural selection in humans. In groups with a history of herding, the vast majority of people have the allele, whereas in non-herding groups, most people lack it. Researchers can find such footprints of selection by comparing groups of people that have lived in different environments, for example, or have eaten different diets. In this issue of PLoS Biology , Joshua Akey and colleagues report new-found signs of natural selection in several human genes—including a chunk of Chromosome 7 encompassing four genes, the largest footprint of selection found yet. The research group analyzed the complete sequences of 132 genes in a set of 23 European-Americans and 24 African-Americans. All these genes are involved in inflammation, blood clotting, or blood pressure regulation and were studied as part of a larger project looking for alleles that contribute to disease. In general, solid evidence of natural selection acting on genes is hard to find. The history of selection can be obscured by a variety of processes. For one, genes can undergo “neutral changes,” in which some of its base pairs change, but without altering the sequence or function of the protein the gene codes for. Also, idiosyncrasies in the history of a population can leave marks on the gene pool. A lineage can go through a “bottleneck,” for example, if a small group splinters off from a larger population and then later multiplies. In general, the splinter group won't perfectly represent the larger population, so the frequencies of alleles for many genes will be skewed in the splinter group's lineage. Having first ruled out irrelevant changes in genes and population history effects, Akey and colleagues found strong signs of natural selection only in the European-Americans, suggesting this group went through significant changes in climate, diet, or culture more recently than the African-American group. This fits with the well-accepted idea that European populations came from small groups that split off from the larger African population. The researchers find evidence for such an event in the European population about 40,000 years ago. They also estimate that the region of Chromosome 7 was subjected to strong selection around 10,000 years ago, roughly when European herders began drinking milk. Interestingly, two of these genes, TRPV5 and TRPV6 , limit the rate of calcium uptake, so selection on one or both of these genes in Europeans could have originated with herding. Recent studies also found TRPV6 to be more active in prostate cancer cells. In addition, African-Americans suffer higher rates of prostate cancer, and Akey and colleagues found that European-Americans have alleles of TRPV6 different from those of African-Americans. Given this evidence, the researchers suggest that this gene may be involved in susceptibility to prostate cancer. This research could therefore shed light on the evolution of complex diseases such as cancer and why different populations suffer different rates of disease.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515376.xml

524034

Regulation of FeLV-945 by c-Myb binding and CBP recruitment to the LTR

Background Feline leukemia virus (FeLV) induces degenerative, proliferative and malignant hematologic disorders in its natural host, the domestic cat. FeLV-945 is a viral variant identified as predominant in a cohort of naturally infected animals. FeLV-945 contains a unique sequence motif in the long terminal repeat (LTR) comprised of a single copy of transcriptional enhancer followed by a 21-bp sequence triplicated in tandem. The LTR is precisely conserved among independent cases of multicentric lymphoma, myeloproliferative disease and anemia in animals from the cohort. The 21-bp triplication was previously shown to act as a transcriptional enhancer preferentially in hematopoietic cells and to confer a replicative advantage. The objective of the present study was to examine the molecular mechanism by which the 21-bp triplication exerts its influence and the selective advantage responsible for its precise conservation. Results Potential binding sites for the transcription factor, c-Myb, were identified across the repeat junctions of the 21-bp triplication. Such sites would not occur in the absence of the repeat; thus, a requirement for c-Myb binding to the repeat junctions of the triplication would exert a selective pressure to conserve its sequence precisely. Electrophoretic mobility shift assays demonstrated specific binding of c-Myb to the 21-bp triplication. Reporter gene assays showed that the triplication-containing LTR is responsive to c-Myb, and that responsiveness requires the presence of both c-Myb binding sites. Results further indicated that c-Myb in complex with the 21-bp triplication recruits the transcriptional co-activator, CBP, a regulator of normal hematopoiesis. FeLV-945 replication was shown to be positively regulated by CBP in a manner dependent on the presence of the 21-bp triplication. Conclusion Binding sites for c-Myb across the repeat junctions of the 21-bp triplication may account for its precise conservation in the FeLV-945 LTR. c-Myb binding and CBP recruitment to the LTR positively regulated virus production, and thus may be responsible for the replicative advantage conferred by the 21-bp triplication. Considering that CBP is present in hematopoietic cells in limiting amounts, we hypothesize that FeLV-945 replication in bone marrow may influence CBP availability and thereby alter the regulation of CBP-responsive genes, thus contributing to altered hematopoiesis and consequent hematologic disease.

Background Feline leukemia virus (FeLV) is a simple gammaretrovirus that induces degenerative, proliferative and malignant hematologic disorders in its natural host, the domestic cat. Like other natural retroviruses, FeLV is not a single genomic species but is a genetically complex family of closely related viruses subject to selective pressures in the host. Variant genomes are generated during virus replication as a result of both error-prone reverse transcription and recombination. The consequence of this variation is a diverse population that is continuously shaped in vivo and from which variants with selective advantages arise as predominant species. The variable clinical outcome of FeLV infection is thought to reflect this genetic diversity [ 1 , 2 ]. FeLV-945, a natural FeLV variant, was originally identified as the predominant species in a temporal and geographic cohort of infected cats. FeLV-945 was originally derived from a multicentric lymphoma of unknown phenotype and subsequently identified in degenerative and proliferative diseases of myeloid and erythroid origin from the cohort. FeLV-945 contains a unique sequence motif in the long terminal repeat (LTR) comprised of a single copy of transcriptional enhancer followed 25-bp downstream by a 21-bp sequence triplicated in tandem. The sequence and position of the 21-bp triplication in the FeLV-945 LTR was observed to be precisely conserved among eight independent multicentric lymphomas and in cases of myeloproliferative disease and anemia in animals from the cohort [[ 3 , 4 ], Chandhasin et al ., manuscript submitted]. The 21-bp triplication was previously shown to provide transcriptional enhancer function to the LTR that contains it, and to function preferentially in primitive hematopoietic cells [ 5 ]. In K-562 cells, a human leukemia cell line considered to be primitive and multipotential [ 6 , 7 ], the FeLV-945 LTR was 12-fold more active than other naturally occurring FeLV LTRs examined. Further, the FeLV-945 LTR was preferentially active in K-562 cells, 4.2-fold more active than in FEA feline embryo fibroblasts [ 5 ]. Interestingly, when the U3 region of the LTR containing the 21-bp triplication was placed downstream of a heterologous promoter, the preferential activity in K-562 cells was lost. These findings suggest that the ability of the 21-bp triplication to enhance transcription preferentially in hematopoietic cells depends on the presence of the adjacent LTR binding sites in their natural array, a possibility examined further in the present study. Previous studies also showed that the 21-bp triplication in the FeLV-945 LTR confers a replicative advantage to the virus that contains it, preferentially in hematopoietic cells [ 8 ]. This growth advantage may account for the induction of tumors of the type in which FeLV-945 was identified, and may represent a selective advantage that contributes to precise conservation of the unusual LTR sequence. Regarding the molecular mechanism by which the 21-bp triplication functions in the context of the LTR, at least two possibilities have been considered. One possibility is that the 21-bp triplication functions to maintain the appropriate spacing in the LTR between the enhancer and the promoter. A spacer function might be particularly relevant in an LTR like FeLV-945 in which the enhancer is not tandemly repeated. Substitution of the 21-bp repeat element with unrelated sequence of the same length, however, was observed to ablate the replicative advantage, thus indicating that the 21-bp triplication does not perform solely a spacer function [ 8 ]. An alternative mechanism may be that the 21-bp triplication contributes genuine enhancer function, perhaps via the binding of nuclear transcription factors. Indeed, electrophoretic mobility shift assay demonstrated that the 21-bp triplication contains binding sites for specific nuclear proteins. These observations suggested that preserving the protein binding sites may confer a selective advantage that accounts for the precise sequence conservation of the 21-bp triplication in this natural FeLV isolate [ 8 ]. The present study examined this possibility further. Binding sites were identified for the transcription factor, c-Myb, that crossed the repeat junctions of the triplication. Further, once c-Myb was bound to the triplication, the transcriptional co-activator CBP was recruited and was shown to positively regulate virus production. Considering that CBP is present in hematopoietic cells in limiting amounts, these observations suggest that FeLV-945 replication in bone marrow may influence CBP availability and thereby alter the regulation of CBP-responsive genes, thus contributing to altered hematopoiesis and consequent hematologic disease. Results As described above, previous studies suggested that preserving the protein binding sites may confer a selective advantage that accounts for the precise sequence conservation of the 21-bp triplication in the FeLV-945 LTR [ 8 ]. In the present study, the sequence of the 21-bp triplication was compared to a transcription factor binding site database (TFSEARCH, based on TRANSFAC; [ 9 ]) in order to identify potential binding proteins. This analysis identified two putative binding sites for the transcription factor c-Myb formed across the repeat junctions of the triplication (Figure 1 ). The sequence of those sites, 5'-AAACTG, closely matched the consensus c-Myb binding sequence, YAACG/TG (Y = pyrimidine; [ 10 , 11 ]). Mismatch between the putative binding site and the consensus sequence was observed at position 1, a position whose change from T to A is known to have little effect on binding affinity [ 12 ]. To determine whether c-Myb binds to the FeLV-945 21-bp triplication, EMSA was performed by reacting a radiolabeled triplication-containing probe with nuclear extracts from K-562 cells in the presence of increasing amounts of a known high-affinity c-Myb binding site as competitor. K-562 cells were chosen because c-Myb is known to be expressed and is thought to be a regulator of their differentiation along multiple hematopoietic lineages [ 13 ]. The results demonstrated a significant reduction in complex formation in the presence of the c-Myb site competitor, especially at amounts in ≥ 100-fold molar excess. In contrast, 250-fold molar excess of the unrelated CREB binding site had no effect (Figure 2 ). To confirm the presence of c-Myb in the specific protein-DNA complex formed on the 21-bp triplication, supershift EMSA was performed using nuclear extracts from K-562 cells in the presence of a monoclonal anti-c-Myb antibody. The results clearly showed decreased mobility of the specific complex in the presence of the c-Myb antibody (Figure 3A ), but not in the presence of an isotype control antibody (Figure 3B ). As a control to confirm that c-Myb binding required repetition of the 21-bp element, EMSA was repeated with a homologous probe derived from FeLV-A/61E, a natural isolate that contains only a single copy of the 21-bp sequence in the LTR. The results demonstrated no specific complex formation on the FeLV-A/61E-derived probe (Figure 3C ), confirming that the specific complex formed on the FeLV-945-derived probe is attributable to the 21-bp triplication. Figure 1 Diagram of the U3 region of the FeLV-945 LTR, indicating the transcriptional enhancer (hatched box), 21-bp triplication (open boxes) and transcriptional promoter (Pro). Below the diagram is shown the sequence of the 21-bp triplication, indicating putative binding sites for the c-Myb transcription factor formed across the repeat junctions. The c-Myb binding site consensus occurs in the negative strand. Figure 2 Electrophoretic mobility shift assays (EMSA) performed using a radiolabeled probe representing the 21-bp triplication from the FeLV-945 LTR. Nuclear extracts (3.5 μg) from K-562 cells were incubated with the radiolabeled probe (1 ng). Double-stranded competitor oligonucleotides were omitted from the reaction (lanes 0), or were included in increasing amounts from 10-fold to 250-fold molar excess (10-, 25-, 50-, 100- and 250-fold excess shown). The competitors used contained a c-Myb consensus binding site (5'-TACAGGCA TAACGG TTCCGTAGTGA) or a CREB consensus binding site (5'-AGAGATTGCC TGACGTCA GAGAGCTAG). Also indicated is the migration of the radiolabeled probe without the addition of nuclear extract (lanes C). Figure 3 Supershift EMSA in the presence of a c-Myb-specific antibody. (A) . Nuclear extracts (5 μg) from K-562 cells were incubated with the radiolabeled GS945 probe (2.4 ng) representing the 21-bp triplication from the FeLV-945 LTR. Shown are probe only (lane 1), complex formation in the presence of nuclear extract (lane 2), and complex formation in the presence of 200-fold molar excess of non-specific (lane 3) or specific competitor (lane 4). Reaction performed in the presence of monoclonal antibody to c-Myb (4 μg) resulted in supershift of the specific complex (lane 5) which was not observed in the presence of 200-fold molar excess of specific competitor (lane 6). (B) . Lanes 1, 2 and 3 represent repetitions of lanes 1, 2 and 5 of (A). Reaction with a isotype control antibody (lane 4) did not result in supershift. Indicated are the specific complex (solid arrow), non-specific complexes (open arrows), and the supershifted complex (asterisk). (C) . EMSA performed using the radiolabeled GS61E probe, which contains only a single copy of the 21-bp element. Shown are probe only (lane 1), reaction performed in the presence of K-562 nuclear extract (5 μg; lane 2), and reaction performed in the presence of 100-fold molar excess of unlabeled GS945 (lane 3), GS61E (lane 4) or non-specific competitor (lane 5). The absence of complex formation using the GS61E probe demonstrates the requirement for the 21-bp triplication. To evaluate whether c-Myb binding to the 21-bp triplication regulates LTR function, reporter plasmids were constructed in which expression of the firefly luciferase gene was driven by the U3 region of an FeLV LTR containing one, two or three copies of the 21-bp element. Reporter gene constructs were introduced by lipid-mediated transfection into feline embryonic fibroblasts (FEA) along with increasing amounts of a c-Myb expression vector. Fibroblasts were selected because the level of endogenous c-Myb expression in those cells is low or absent [ 14 , 15 ]. The results (Figure 4 ) demonstrated that the FeLV-945 LTR (3 × 21) responds to increasing levels of c-Myb expression to an extent statistically indistinguishable from the positive control, i.e., a reporter plasmid containing five tandem Myb-responsive elements (5X MRE). In contrast, an FeLV LTR containing only a single 21-bp element was unresponsive to c-Myb. The responsiveness of an LTR containing two 21-bp elements was also examined, since the 21-bp duplication would be predicted to encode one c-Myb binding site across the repeat junction. Interestingly, this LTR responded to increasing c-Myb expression to a low but statistically significant extent (p < 0.05 as compared to 1 × 21; Figure 4 ). These data show that the triplication-containing LTR is responsive to c-Myb in a dose-dependent manner and suggest that full responsiveness requires the presence of both c-Myb binding sites. To confirm the latter finding, a point mutation previously shown to ablate c-Myb binding [ 16 ] was introduced alternately into each of the sites (Figure 5A ). Synthetic oligonucleotides containing the respective mutations were substituted into the LTR, and luciferase reporter gene constructs containing the mutant LTRs were introduced into FEA cells along with increasing concentrations of a c-Myb expression vector. LTRs in which either c-Myb binding site was ablated were observed to respond only weakly to increasing levels of c-Myb, and to significantly lower levels than the wild type LTR containing both binding sites (p < 0.05; Figure 5B ). Figure 4 Response to exogenous c-Myb expression of FeLV LTRs containing variable numbers of the 21-bp element. Recombinant FeLV LTRs were constructed that contained 1, 2 or 3 copies of the 21-bp element and were cloned into a firefly luciferase reporter plasmid. LTR reporter plasmids or a 5X MRE positive control plasmid (500 ng) were introduced by lipid-mediated transfection in triplicate into feline embryonic fibroblasts (FEA) together with the Renilla luciferase reporter plasmid pRL-SV40 (5 ng) and a c-Myb expression plasmid in increasing concentrations (0 – 500 ng). Cell lysates were harvested 24 hours later and luciferase activity was quantified. Data are reported as a ratio of firefly to Renilla luciferase activity. Shown are data from a representative experiment repeated three times independently. Figure 5 Response to exogenous c-Myb expression of FeLV LTRs containing c-Myb binding site mutations. (A) . Diagram of the 21-bp triplication as contained in the FeLV-945 LTR, indicating the sequence of c-Myb binding sites across the repeat junctions of the triplication ( +/+ ). LTRs were constructed in which the first ( -/+ ) or second ( +/- ) binding site was mutated. (B) . Firefly luciferase reporter gene plasmids containing the FeLV LTR with wild type or mutant c-Myb binding sites (500 ng) were introduced by lipid-mediated transfection in triplicate into feline embryonic fibroblasts (FEA) together with the Renilla luciferase reporter plasmid pRL-SV40 (5 ng) and a c-Myb expression plasmid in increasing concentrations (0 – 500 ng). Cell lysates were harvested 24 hours later and luciferase activity was quantified. Data are reported as a ratio of firefly to Renilla luciferase activity. Shown are data from a representative experiment repeated three times independently. Previous studies had shown that the 21-bp triplication contributes enhancer function to the LTR in a cell type-specific manner, and that it is significantly more active in K-562 cells as compared to a fibroblast line [ 5 ]. In hindsight, these results may be explained by the relatively high levels of c-Myb expression in K-562 cells and its relative absence in fibroblasts [ 14 , 15 ]. When the U3 region of the FeLV-945 LTR was placed downstream of a heterologous promoter, however, the cell type-specific preference for enhancer function was lost [ 5 ]. One explanation for these findings is that c-Myb bound to the 21-bp triplication may function through interactions with other proteins bound to the LTR, and that such interactions require the LTR binding sites to be present in their natural array. Indeed, c-Myb is known to function in a combinatorial manner with other transcription factors and co-activators to activate target gene expression [ 14 ]. Studies were performed in the present study to evaluate this possibility further. First, an oligonucleotide containing only the 21-bp triplication was cloned into a luciferase reporter plasmid upstream of a heterologous SV40 promoter. This construct, when introduced into FEA cells, was observed to be unresponsive to increasing levels of c-Myb expression (data not shown). Thus, the presentation of c-Myb binding sites through the 21-bp triplication is apparently insufficient to regulate transcription in the absence of the normally adjacent LTR enhancer and promoter. These findings are consistent with the possibility that c-Myb binding to the 21-bp triplication functions to activate transcription by interacting with proteins bound to adjacent sites on the LTR. c-Myb is known to interact directly with a number of different proteins, including the transcriptional co-activator CREB-binding protein (CBP) [ 14 , 15 ]. Indeed, CBP is thought to act as a bridge that physically connects c-Myb to the promoter-bound basal transcription machinery, thus stabilizing the transcription-preinitiation complex [ 14 , 15 , 17 ]. Experiments were therefore performed in the present study to examine the possibility that c-Myb bound to the 21-bp triplication interacts with CBP. Supershift EMSA was used to evaluate whether c-Myb and CBP might be present in the specific complex that forms on the 21-bp triplication. Specific complex formation was observed using a radiolabeled probe representing the 21-bp triplication in the presence of nuclear extracts from either K-562 cells or feline 3201 T-cells. In both cases, mobility of the complex was reduced (supershifted) when reacted with antibody to c-Myb. When the same reaction was performed in the presence of an antibody to CBP, an identical supershift was observed (Figure 6A,6B ). No complex formation was observed when the probe was reacted with nuclear extracts from FEA cells (Figure 6C ), consistent with the lack of c-Myb expression in fibroblasts [ 14 , 15 ]. CBP is ubiquitously expressed [ 14 ]; therefore, this observation indicates that CBP does not participate in complex formation on the 21-bp triplication in the absence of c-Myb. When c-Myb was expressed exogenously in FEA cells, specific complex formation and supershift were observed in the presence of antibody to either c-Myb or CBP (Figure 6D ). Finally, analysis of FeLV-945 replication indicated a regulatory role for c-Myb binding and recruitment of CBP to the 21-bp triplication. K-562 cells were infected with recombinant FeLV [ 8 ] containing the LTR of either FeLV-945 or FeLV-A/61E, the latter having only a single copy of the 21-bp element. A CBP expression vector was introduced into cells chronically infected with either virus, and virus production was measured three days later by quantifying reverse transcriptase activity in the culture supernatants. The results showed significantly increased levels of production of virus containing the FeLV-945 LTR. In contrast, virus containing the FeLV-A/61E LTR was unaffected (Figure 7 ). These findings indicate that the 21-bp triplication in the LTR renders the virus responsive to the amount of available CBP. Figure 6 Supershift EMSA in the presence of antibody specific for c-Myb or CBP. (A – C) . Nuclear extracts (5 μg) from K-562, 3201 or FEA cells were incubated with the radiolabeled GS945 probe (2.4 ng) representing the 21-bp triplication from the FeLV-945 LTR. Shown in each panel is specific complex formation in the presence of nuclear extract (closed circle), and with the addition of monoclonal antibody to c-Myb or CBP (4 μg). Reduced mobility of the complex (supershift) is indicated (asterisk). (D) shows the same reactions performed with nuclear extracts from FEA cells in which c-Myb was exogenously overexpressed. Figure 7 Regulation of FeLV replication in response to exogenous overexpression of CBP. K-562 cells were chronically infected with recombinant FeLV containing the LTR of FeLV-945 or FeLV-A/61E. A CBP expression plasmid was then introduced by lipid-mediated transfection. Culture supernatants were collected 3 days later and reverse transcription activity was quantified as a measure of virus production. Results are reported as cpm/ml of 3 H-TTP incorporated. The data shown were pooled from two independent experiments each performed in triplicate. Discussion The natural FeLV isolate, FeLV-945, was originally identified from lymphoid and other hematopoietic disorders in a geographic and temporal cohort. A unique 21-bp repeat motif in the FeLV-945 LTR was observed to be precisely conserved among animals in the cohort that exhibited malignant, proliferative or degenerative hematopoietic diseases of non-T-cell origin [[ 3 , 4 ], Chandhasin et al ., manuscript submitted]. The 21-bp triplication was shown to enhance transcription from the FeLV LTR and to confer a replicative advantage to the virus, at least in part through the specific binding of unidentified nuclear proteins to the repeat motif [ 5 , 8 ]. In the present study, sequence analysis revealed two potential c-Myb binding sites formed across the repeat junctions of the 21-bp triplication (Figure 1 ). While the sequence of the potential binding sites (AAACTG) did not match the consensus c-Myb binding site precisely (YAACG/TG; Y = pyrimidine; [ 10 , 11 ]), the sequence was observed to be as closely related to the consensus binding site as are several sites in the HTLV-I LTR that are known to bind c-Myb [ 18 , 19 ]. Indeed, electrophoretic mobility shift assays indicated the specific binding of c-Myb to the 21-bp triplication by showing that a known high-affinity c-Myb binding site competed for DNA-protein complex formation but an unrelated site did not. It was noteworthy in these assays that significant competition for complex formation occurred only when the competitor was present at relatively high amounts (≥ 100-fold molar excess; Figure 2 ). By comparison, c-Myb binding to a consensus sequence in the bcl -2 promoter was shown to be effectively competed by 50-fold molar excess of a cold oligonucleotide carrying a high affinity c-Myb binding site [ 20 ]. A possible explanation for this difference may be that, while c-Myb can recognize a single consensus binding site such as that found in the competitor oligonucleotide we used, the natural recognition sites are generally found in multiple, closely aligned copies as in the 21-bp triplication. Thus, the affinity of binding to the triplication may be higher than to the competitor. It is further known that sequences flanking the consensus binding site may also be important in determining c-Myb binding affinity [ 21 ]. Electrophoretic mobility shift assays performed in the presence of an antibody to c-Myb confirmed the presence of c-Myb in the specific DNA-protein complex (Figure 3A,3B ), and confirmed that repeat of the 21-bp element was required for complex formation (Figure 3C ). The c-Myb transcription factor is a critical regulator of gene expression, proliferation and differentiation in early hematopoietic progenitors [ 14 , 15 , 22 ] and has been exploited as a transcriptional regulator by many viruses that infect bone marrow cells [ 19 , 23 - 25 ]]. Considering that FeLV is known to replicate in the bone marrow [ 26 , 27 ], and that FeLV-945 infection was associated with various diseases of hematopoietic origin [Chandhasin et al ., manuscript submitted], it is likely that the tropism of FeLV-945 in vivo included the hematopoietic progenitors in which c-Myb is expressed. Considering this possibility, we hypothesized that c-Myb may act as a transcriptional regulator of FeLV-945. In support of this hypothesis, reporter gene assays showed that an LTR containing the triplication was responsive to c-Myb in a dose-dependent manner (Figure 4 ), and that optimal responsiveness required the presence of both c-Myb binding sites (Figure 5 ). The identification of c-Myb binding sites that spanned the repeat junctions of the 21-bp triplication was particularly noteworthy because such sites would not occur in the absence of the repeat. Thus, a requirement for c-Myb binding to the repeat junctions of the triplication would exert a selective pressure to conserve its sequence precisely. Results indicated further that when c-Myb binds to the 21-bp triplication, it interacts with the transcriptional co-activator CBP, a critical regulator of normal hematopoiesis [ 17 ]. Identical electrophoretic mobility supershifts were observed when protein-DNA complexes were formed in the presence of antibody either to c-Myb or CBP, consistent with the hypothesis that both proteins are present in the same complex (Figure 6A,6B ). The data further indicated that c-Myb recruits CBP to the 21-bp triplication, since no CBP-containing complex formation could be demonstrated unless c-Myb was also expressed (Figure 6C,6D ). Finally, virus production was shown to be positively regulated by CBP in a manner dependent on the presence of the 21-bp triplication (Figure 7 ). These results indicated that the interaction between c-Myb and CBP is functional, and suggest that the c-Myb-mediated recruitment of CBP to the FeLV-945 LTR could be responsible for the previously reported replicative advantage conferred by the 21-bp triplication [ 8 ]. CBP and c-Myb are thought to activate target genes in hematopoietic progenitors through various mechanisms of interaction. One of those mechanisms involves a bridging function in which CBP links c-Myb with components of the basal transcription machinery, thereby establishing and/or stabilizing the transcription complex [ 14 , 15 , 17 ]. While the mechanism of interaction was not investigated in the present study, the bridging function is an intriguing possibility because it might explain the observed requirement of the 21-bp triplication for an intact LTR enhancer and promoter. Specifically, when the isolated 21-bp triplication was positioned upstream of a heterologous promoter, it did not confer responsiveness to exogenously supplied c-Myb (data not shown). Previous studies had similarly shown that the 21-bp triplication could not exert its influence when placed downstream of a heterologous promoter [ 5 ]. These observations indicated that transcriptional activation of the FeLV-945 LTR through c-Myb/CBP interaction requires that the LTR binding sites be present in their natural array. The possibility that CBP exerts its influence on the FeLV-945 LTR through a bridging function is significant because it implies that CBP acts stoichiometrically. CBP is known to be present in bone marrow cells in limiting amounts, playing a major role in hematopoiesis through competitive utilization on target promoters [ 17 , 28 - 31 ]]. Considering the competitive utilization of limiting amounts of CBP in hematopoiesis, its stoichiometric recruitment to the FeLV-945 LTR might interfere with CBP availability and thereby alter the regulation of CBP-responsive genes. Such alteration might then contribute to altered hematopoiesis and consequent hematologic disease. Conclusions FeLV-945 contains a unique 21-bp triplication in the LTR, conserved among animals in a geographic cohort with multicentric lymphoma, myeloproiferative disease or anemia. Binding sites for the c-Myb transcription factor were identified across the repeat junctions of the 21-bp triplication. Optimal responsiveness of the FeLV-945 LTR to c-Myb was shown to require the presence of both c-Myb binding sites. Since the binding sites would not occur in the absence of the repeat, a requirement for c-Myb binding would be predicted to exert a selective pressure for conserving the 21-bp triplication precisely. c-Myb binding to the 21-bp triplication was shown to recruit CBP, a transcriptional co-activator essential for hematopoiesis and known to be present in limiting amounts. Interaction of c-Myb and CBP with the 21-bp triplication was shown to positively regulate virus production, and thus may be responsible for the replicative advantage conferred by the repeat sequence. Considering that CBP is present in hematopoietic cells in limiting amounts, we hypothesize that FeLV-945 replication in bone marrow may influence CBP availability and thereby alter the regulation of CBP-responsive genes, thus contributing to altered hematopoiesis and consequent hematologic disease. Methods Cell lines and viruses K-562, a malignant multipotential human hematopoietic cell line, was obtained from the American Type Culture Collection (CCL-243) and was maintained in RPMI 1640 medium with 10% FBS. The FEA cell line, a continuous line of feline embryonic fibroblasts, was obtained from Dr. Jennifer Rojko and was grown in Eagle minimal essential culture medium supplemented with 10% fetal bovine serum (FBS), 0.1 mM non essential amino acids and 50 μg/ml gentamicin reagent solution (Invitrogen, Carlsbad, CA,). 3201 is an FeLV-negative thymic lymphoma cell line of feline origin [ 32 ] and was maintained in 50% Leibovitz L-15 medium/50% RPMI 1640 supplemented with 15% FBS. Infectious recombinant FeLVs GA-945L and GA-61EL were constructed from an infectious molecular clone of FeLV-B/Gardner-Arnstein into which was substituted the LTR of FeLV-945 or of FeLV-A/61E, respectively, between EcoRV and Hinc II restriction enzyme sites [ 5 , 8 ]. The FeLV-A/61E LTR was selected because it represents a naturally occurring isolate of FeLV typical of those horizontally transmitted among cats in nature, and it contains only a single copy of the 21-bp element [ 33 ]. Electrophoretic mobility shift assays A double-stranded oligonucleotide probe containing the 21-bp triplication and 40 bp of flanking sequence from the FeLV-945 LTR was radiolabeled using the synthetic oligonucleotide GS945 as template (5'- GCTGAAACAGCAGAAGTTTCAAGGCCACTGCCAGCAGTTTCAAGGCCACTGCCAGCAGTTTCAAGGCCACTGCCAGCAGTCTCCAGGCTCCCCAGTTGAC -3'), the filling primer (5'-CTGGTCAACTGGGGAGCCT-3') and the Klenow fragment of DNA polymerase (Invitrogen, Carlsbad, CA) to complete the duplex. A homologous probe containing only a single copy of the 21-bp element was similarly synthesized using the oligonucleotide GS61E as template (5'-GCTGAAACAGCAGAAGTTTCAAGGCCACTGCCAGCAGTCTCCAGG CTCCCCAGTTGAC-3'). Nuclear extract from K-562 cells was obtained from Active Motif (Carlsbad, CA). Nuclear extracts from FEA and 3201 cells were prepared using the Nuclear Extract Kit from Active Motif (Carlsbad, CA) according to manufacturer specifications. Nuclear extracts were also prepared from FEA cells following the lipid-mediated transfection (Lipofectamine Plus reagent; Invitrogen, Carlsbad, CA) of FL-Myb, a c-Myb expression vector in which full length murine c-Myb cDNA was inserted into the multiple cloning site of pcDNA3.1 (a gift of Dr. Linda Wolff, National Cancer Institute). DNA-protein binding reactions included 5 μg of nuclear extract and 2.4 ng of radiolabeled probe in a 20 μl reaction containing 1 mM Tris pH 7.5, 7.5 mM NaCl, 1 mM EDTA, 0.7% glycerol, 0.1 mM DTT and 2 μg poly(dI-dC). Reactions containing nuclear extracts from K-562 cells were incubated at 4°C for 30 minutes. Reactions containing nuclear extracts from 3201 or FEA cells were incubated at 30°C for 30 minutes. In some reactions, unlabeled probe was added to the reaction as a specific competitor, or HindIII/HaeIII-digested bacteriophage lambda DNA was included as non-specific competitor. Some reactions included as competitor a double-stranded oligonucleotide containing a known high-affinity c-Myb consensus binding site (5'-TACAGGCA TAACGG TTCCGTAGTGA) or a CREB consensus binding site (5'-AGAGATTGCC TGACGTCA GAGAGCTAG). Protein-DNA complexes were resolved by 6% polyacrylamide gel electrophoresis in 0.25X TBE buffer (1X TBE buffer is 89 mM Tris base, 89 mM Boric acid and 2 mM EDTA). Gels were then dried at 80°C and exposed to radiographic film for varying periods of time. In some reactions, monoclonal antibody (4 μg) to either c-Myb or CBP, or isotype control antibody, was added after the 30-minute incubation period and incubated overnight at 4°C. Complexes were then resolved by 6% polyacrylamide gel electrophoresis as described above. The mouse monoclonal IgG1 antibody to c-Myb was raised against a recombinant protein corresponding to amino acids 500–640 of the human protein (Santa Cruz Biotechnology, Santa Cruz, CA). The mouse monoclonal IgG1 antibody to CBP was raised against a peptide corresponding to amino acids 2422–2441 of CBP of human origin (Santa Cruz Biotechnology, Santa Cruz, CA). Reporter gene constructs and luciferase expression assays Luciferase reporter plasmids were constructed to contain the U3 region of an FeLV LTR containing one, two or three copies of the 21-bp element. The U3 region of the FeLV-A/61E LTR, containing one copy of the 21-bp element, was cloned into the firefly luciferase reporter plasmid pGL2-Basic (Promega Corp., Madison, WI). The LTR was then substituted between Pst I and Hinc II restriction sites with homologous sequences from a naturally occurring LTR containing two 21-bp elements [Chandhasin et al ., manuscript submitted] or from FeLV-945, which contains three 21-bp elements. Luciferase reporter plasmids were also constructed in which point mutations were introduced into either the first or second c-Myb binding site in the 21-bp triplication of the FeLV-945 LTR. Binding site mutants were constructed by designing synthetic oligonucleotides -/+ (5'-GCTGAAACAGCAGAAGTTTCAAGGCCACTGCCAGCAG A TTCAAGGCCACTGCCAGCAGTTTCAAGGCCACTGCCAGCAGTCTCCAGGCTCCCCAGTTGAC-3') and +/- (5'-GCTGAAACAGCAGAAGTTTCAAGGCCACTGCCAGCAGTTTCAAGGCCACTGCCAGCAG A TTCAAGGCCACTGCCAGCAGTCTCCAGGCTCCCCAGTTGAC-3') that contained a point mutation in the first or second binding site, respectively (mutated base indicated by boldface and underline). The indicated point mutation had previously been shown to ablate c-Myb binding [ 16 ]. A double-stranded form of each sequence was generated using the filling primer (5'-GAACTCTGGTCAACTGGGGAGCCTGGAGACTGCTG-3') and the Klenow fragment of DNA polymerase. The resulting double stranded oligonucleotides were digested with AluI/HincII and substituted into the LTR of FeLV-A/61E. The KpnI/PstI fragment of the resulting recombinant LTR was then excised and cloned into the pGL2-Basic luciferase reporter plasmid. Finally, a luciferase reporter plasmid was developed that contained the isolated 21-bp triplication cloned upstream of the SV40 promoter in pGL2-Promoter (Promega Corp., Madison, WI). A double-stranded DNA fragment containing the 21-bp triplication was generated by PCR amplification using the oligonucleotide GS945 (described above) as template and primers fw945-kpn1 (5'- GCTCGGTACCAGCTGAAACAGCAGAAGTTTC) and rv945-sac1 (5'- ATGCTGAGCTCAACTGGGGAGCCTGGAGACT). The resulting amplification product was digested with KpnI/SstI and inserted into the multiple cloning site upstream of the SV40 promoter in the reporter plasmid. For reporter gene assays, 2 × 10 5 cells were seeded in triplicate into 6-well tissue culture plates. The next day, reporter plasmids (500 ng) were introduced into cultured cells by lipid-mediated transfection (Lipofectamine Plus; Invitrogen, Carlsbad, CA) in the presence of pRL-TK (5 ng) in a 100:1 ratio. pRL-TK encodes Renilla luciferase and was used as an internal control for transfection efficiency. Firefly and Renilla luciferase activities were quantified 24 hours later using the Dual-Luciferase Reporter Assay System (Promega Corp., Madison, WI). Data from triplicate wells were analyzed statistically using one-way ANOVA and Bonferroni post test. Statistical significance was considered as p < 0.05. In some assays, the c-Myb expression vector FL-Myb (described above) was added to the transfection in increasing amounts (50 ng – 500 ng). The 5XMRE plasmid was used in reporter gene assays as a positive control. This plasmid contains five tandem c-Myb binding sites cloned upstream of a luciferase gene (a gift of Dr. Linda Wolff, National Cancer Institute). Virus Replication Assay 5 × 10 5 K-562 cells, uninfected or chronically infected with recombinant FeLVs GA-945L and GA-61EL (described above) were seeded in triplicate into 24-well tissue culture plates. A full-length CBP expression vector (4 μg; a gift from Dr. Matthew Burow, Tulane University Medical School) was introduced into the cells by lipid mediated transfection using Lipofectamine 2000 reagent (Invitrogen, Carlsbad, CA). Culture supernatants were collected three days later and reverse transcriptase activity was quantified as previously described [ 8 ]. Data from triplicate wells were analyzed statistically using one-way ANOVA and Bonferroni post test. Statistical significance was considered as p < 0.05. Competing interests None declared. Authors' contributions SLF developed reporter gene constructs and performed binding assays, gene expression and virus replication assays. SP identified and initially demonstrated c-Myb binding sites. KRR developed the reporter gene assays. LSL directed the experimental design, implementation and interpretation of data. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524034.xml

524183

Emphysematous cystitis: An unusual disease of the Genito-Urinary system suspected on imaging

Emphysematous cystitis is a rare disease entity caused by gas fermenting bacterial and fungal pathogens. Clinical symptoms are nonspecific and diagnostic clues often arise from the unanticipated imaging findings. We report a case of 52-year-old male who presented with fever, dysuria and gross hematuria who was found to have emphysematous cystitis.

Introduction Emphysematous cystitis is an uncommon, but severe manifestation of infection of the urinary bladder produced by gas forming organisms. The presentation may be atypical and contrary to the degree of inflammation, patients may present with subtle clinical findings. A high index of suspicion, especially in susceptible populations, is needed. We report a case of middle-aged, poorly controlled, diabetic male who presented with dysuria, fever and hematuria and was found to have Escherichia coli emphysematous cystitis that resolved with antibiotic treatment. Case Report A 52-year-old male presented with fever, chills and blood in the urine. Four days prior to admission, he noted increased urinary frequency, urgency, occasional incontinence and burning sensation during micturition. His past medical history is significant for coronary artery disease, hyperlipidemia, hypertension and insulin dependent diabetes mellitus (most recent HgbA1c of 11.7 two weeks prior). He had 66-pack year history of smoking. Medications on admission included, aspirin, atenolol, cyclobenzaprine, gabapentin, gemfibrozil, glipizide, NPH insulin, hydrochlorthiazide/triamterene, nifedipine, nitroglycerin patch, omeprazole, and sertraline. On physical examination, his temperature was 100.9°F, pulse 88, BP 123/71 mm/hg. Cardiac and pulmonary examinations were unremarkable. His abdomen was soft, nontender with positive bowel sounds and no organomegaly detected. The rectal examination was normal with a non-tender, smooth prostate. Neurological exam was exam was unremarkable. Laboratory evaluation revealed white cell count 17,400 cells/mm 3 , hemoglobin 15.1 gm/dl, sodium 128 mmol/l, potassium 4.1 mmol/l, chloride 91 mmol/l, bicarbonate 24 mmol/L, glucose 273 mg/dl and an anion gap of 13. Urine analysis revealed red, cloudy urine with pH 5.0, nitrite positive and numerous white and red blood cells on microscopy. A pelvic X-ray (figure 1 ) showed circumferential air in the bladder wall. CT scan of the pelvis (figure 2 ) revealed gas in the bladder lumen and wall extending to the right ureter. He was empirically treated with gentamicin and piperacillin/tazobactam. He responded with defervescence and a decline in his white cell count. Urine cultures grew Escherichia coli and based on the sensitivities he was discharged home on ciprofloxacin. Figure 1 X-ray of pelvis showing gas in the urinary bladder wall (Arrow) Figure 2 CT scan of the pelvis revealing gas in the bladder and the bladder wall. Discussion Emphysematous cystitis is a rare entity characterized by pockets of gas in and around the bladder wall produced by bacterial or fungal fermentation [ 1 , 2 ]. Patients may complain of irritative symptoms, abdominal discomfort or pneumaturia. A history of pneumaturia is highly suggestive, but is rarely offered by the patient. As occurred in our case and in a number of cases in the literature, the clinical features were inconclusive or actually unhelpful [ 3 - 6 ]. The radiographic findings provided the first and only diagnostic clue. The disease is often associated with female sex, immunocompromised state, diabetes mellitus, previous recurrent urinary tract infections, urinary stasis, neurogenic bladder and in transplant recipients [ 7 ]. Therefore, in susceptible patients, with the above risk factors along with signs and symptoms of urinary tract infection, the index of suspicion for this entity should be high. The most common organism is E. coli [ 2 ], but other organisms reported to produce emphysematous cystitis include Enterobacter aerogenes , Klebsiella pneumonia , Proteus mirabilis , Staphylococcus aureus , streptococci , Clostridium perfringens [ 8 ], and Candida albicans [ 9 ]. The mechanism by which gas appears in the wall of the bladder may involve either transluminal dissection of gas or true infection of the bladder wall with pathogens. Diagnostic entities associated with gas in the genitourinary tract include emphysematous pyelonephritis, emphysematous pyelitis, and gas-forming renal abcess. Patients with emphysematous cystitis are not as acutely ill as those with pyelonephritis or pyelitis. Abdomino-pelvic CT scan can further delineate the extent of disease. It is important to differentiate emphysematous cystitis from emphysematous pyelonephritis, in which gas involves the renal parenchyma, since the latter has an increased mortality and generally requires nephrectomy. In contrast surgical intervention is rarely needed in emphysematous cystitis except when an anatomical abnormality like an obstruction or stone is present [ 10 ]. The source of this gas within the urinary tract is from infection, trauma, vesico-enteric fistulas from radiation therapy, rectal carcinoma, diverticular disease or Crohn's disease and iatrogenic causes, such as diagnostic or surgical instrumentation. History, physical exam and imaging are the best modalities to differentiate the above etiologic causes. Fistulous tracts, abscess, can be excluded on CT scan. Emphysematous cystitis requires aggressive treatment with parenteral antibiotics and bladder drainage [ 11 ]. Delayed diagnosis may lead to unfavorable outcomes including overwhelming infection, extension to ureters and renal parenchyma, bladder rupture and death. Improved outcomes may be achieved by early recognition of the infection, by clinical and radiological assessment, and by appropriate antibiotic therapy. Conclusion Emphysematous cystitis most often is not diagnosed by routine or systematic approach. It is a rare entity, detected on imaging, and the physician should be cautious, tailor the diagnostic approach to individual patients based on the suspicion, available clinical and radiological data, and consider emphysematous cystitis in the differential diagnosis of hematuria in a patient with known risk factors.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524183.xml

535930

Reexamining the effects of gestational age, fetal growth, and maternal smoking on neonatal mortality

Background Low birth weight (<2,500 g) is a strong predictor of infant mortality. Yet low birth weight, in isolation, is uninformative since it is comprised of two intertwined components: preterm delivery and reduced fetal growth. Through nonparametric logistic regression models, we examine the effects of gestational age, fetal growth, and maternal smoking on neonatal mortality. Methods We derived data on over 10 million singleton live births delivered at ≥ 24 weeks from the 1998–2000 U.S. natality data files. Nonparametric multivariable logistic regression based on generalized additive models was used to examine neonatal mortality (deaths within the first 28 days) in relation to fetal growth (gestational age-specific standardized birth weight), gestational age, and number of cigarettes smoked per day. All analyses were further adjusted for the confounding effects due to maternal age and gravidity. Results The relationship between standardized birth weight and neonatal mortality is nonlinear; mortality is high at low z-score birth weights, drops precipitously with increasing z-score birth weight, and begins to flatten for heavier infants. Gestational age is also strongly associated with mortality, with patterns similar to those of z-score birth weight. Although the direct effect of smoking on neonatal mortality is weak, its effects (on mortality) appear to be largely mediated through reduced fetal growth and, to a lesser extent, through shortened gestation. In fact, the association between smoking and reduced fetal growth gets stronger as pregnancies approach term. Conclusions Our study provides important insights regarding the combined effects of fetal growth, gestational age, and smoking on neonatal mortality. The findings suggest that the effect of maternal smoking on neonatal mortality is largely mediated through reduced fetal growth.

Background Birth weight is arguably one of the strongest predictors of infant survival, yet its role as a causal predictor of mortality is poorly understood [ 1 ]. This is at least partly because low birth weight (<2,500 g) is a construct of two intricately intertwined components: preterm delivery and reduced fetal growth, or both. Our lack of understanding of the complex relationship among birth weight, gestational age and perinatal mortality stems from mixing etiologically distinct pathways to mortality, namely effects chiefly due to fetal maturity ( i.e ., gestational age) versus those related to fetal growth. Disentangling the intricate pathways of gestational age and fetal growth to neonatal mortality gets even more complicated by the consideration of a third factor – maternal smoking during pregnancy. Smoking has been clearly associated with poor reproductive outcomes, including increased risk of preterm birth, stillbirth, and a range of other outcomes [ 2 - 6 ]. Recent studies suggest a more direct and stronger association between maternal smoking and "fetal growth" (birth weight-for-gestational age) than with preterm delivery [ 7 ], suggesting that the effect of smoking on mortality may be largely mediated through restricted fetal growth rather than preterm delivery. To better understand the relationship among these indices of "fetal wellbeing", we examined neonatal mortality in relation to standardized birth weight ( i.e ., z-score birth weight), gestational age, and smoking during pregnancy. We applied nonparametric logistic regression based on generalized additive models to examine neonatal mortality in relation to 3 factors. Methods Cohort composition of United States live births Data for this study were derived from the 1998–2000 United States vital statistics data files (live births linked to infant deaths), assembled by the National Center for Health Statistics of the Centers for Disease Control and Prevention [ 8 ]. The analysis was restricted to singleton live births, with neonatal mortality defined as deaths within the first 28 days. Gestational age assignment in these data are predominantly based on self-reported last menstrual period, with a small fraction (<5%) based on the clinical estimate [ 9 ]. Further, the National Center for Health Statistics imputed missing gestational ages in these data files prior to release of the data [ 10 ]. Information on smoking during pregnancy was available in two forms on the vital statistics data: one as an indicator variable (yes or no), and the other as a continuous variable denoting the number of cigarettes smoked per day during pregnancy. Both of these smoking measures were based on maternal self-report. Information on smoking patterns across different trimesters in pregnancy was not available on the vital records. Fetal growth was defined as birth weight-for-gestational age, and was expressed as gestational age-specific birth weight z-score. This z-score construct is interpreted as units of standard deviations from the population-specific mean birth weight at each gestational age. The z-score or standardized birth weight follows a Gaussian distribution with mean 0 and variance 1. In addition to the full analysis, we also examined in a sub-analysis the impact of implausible birth weight/gestational age combinations on overall results. These implausible birth weight/gestational ages were identified if infants' birth weights were outside the gestational age-specific birth weight cutoffs [ 11 ]. This was done to examine the impact of largely apparent gestational age errors ( e.g ., infant delivered at 26 weeks with a birth weight of 4,000 g) on neonatal mortality. Data exclusions There were 11,677,103 singleton live births from which we excluded infants with missing birth weight or gestational age (n = 237,433), and birth weight <500 g or gestational age <24 weeks (n = 28,732). Since smoking data was not reported on vital statistics in California, Indiana, New York state, and South Dakota [ 8 ], births from these states were also excluded (n = 1,326,841). After all exclusions, 6,117,808 singleton live births remained for analysis. Statistical analysis We examined the distributions of z-score birth weight, gestational age, and number of cigarettes smoked per day, and compared these distributions between the two groups of neonatal mortality. Neonatal mortality was then modeled using nonparametric logistic regression based on generalized additive models [ 12 ]. GAM is one modeling approach that makes no assumptions about the functional form of the exposure-disease relationship except for smoothness, i.e ., continuity of the dose-response function and its low-order derivatives [ 13 ]. When combined with more traditional modeling approaches, GAMs are powerful graphical tools that can provide interesting insights about complex relationships. While polynomial models [ 14 ] could be used to the same end as GAM-based approaches, such models result in restricted shapes, especially at the tail of the distribution, and may not be as statistically efficient as nonparametric models. Therefore, these models were not considered. All regression models were adjusted for the confounding effects due to maternal age and gravidity ( i.e ., number of pregnancies). We examined the associations between neonatal mortality and each of the 3 factors z-score birth weight, gestational age, and number of cigarettes smoked per day separately. We then fit a full model for mortality after forcing all 3 predictors (in addition to the confounders) as described in the Appendix [see additional file 1 ]. The independent effect of each of these 3 factors on neonatal mortality was assessed by comparing the residual deviances [ 12 ] between nested models ( i.e ., comparing the residual deviances from a full model to a model without the predictor). Under the large sample assumption, the deviance has an approximate chi-square distribution, with degrees-of-freedom for the test being the difference in the degrees of freedom between the nested models being compared. We also examined the distribution of partial residuals [ 12 ] from fitting the model to assess departures from adequate fit. In addition, we tested for all possible two-factor interactions between the predictors. Although all interactions were statistically significant (owing to the large study size), none provided any additional insights that were different from a model that contained no interaction terms. Therefore, we did not consider assessing two-way interactions in the analysis. All statistical analyses were performed in S-Plus (Insightful Corporation, Seattle WA) version 6.2 on the UNIX (Sun Microsystems, Inc: Palo Alto, CA). Nonparametric logistic regression models were fit using the gam( ) function based on the it loess scatterplot smoother [ 14 ], using the default span of 50%. Given the large size of the study, small changes in the span resulted in statistically significant improvement in the fit, while offering very little clinical insight. Thus, we resorted to the default span. Results The overall neonatal mortality rate was 2.4 per 1,000 live births. The distribution of z-score birth weight among infants that died during the neonatal period was shifted more towards lower standardized birth weights than among those that survived the neonatal period (Fig 1 , left panel). Infants who died during the neonatal period were delivered earlier than those that survived the neonatal period (Fig 1 , right panel). Surviving infants weighed, on average, 1,582 g more compared with those who died during the neonatal period (P < 0.0001; Table 1 ). Likewise, infants who died during the neonatal period were delivered, on average, 7 weeks earlier than those who survived the neonatal period (P < 0.0001). The proportion of mothers that smoked during their pregnancy was higher among infants that died during the neonatal period (19.1%) compared with those that survived the neonatal period (16.5%; P < 0.0001). Figure 1 Distributions of z-score birth weight (left panel) and gestational age (right panel) among neonatal deaths (thick line) and neonatal survivors (thin line). Table 1 Distributions of birth weight, gestational age, and maternal smoking in relation to neonatal survival status Neonatal survivors Neonatal deaths Total events 10,084,106 27,355 Maternal age (years)† 27.0 (6.2) 26.4 (6.7) Primigravida 33.2% 34.1%¶ Birth weight (grams)† 3,347 (572) 1,765 (1,145) Birth weight <2,500 grams 6.1% 69.6% Birth weight <1,500 grams 1.1% 62.3% z-score birth weight† 0.00 (1.00) -0.62 (1.07) Gestational age (weeks)† 38.9 (2.3) 31.3 (6.6) Delivered <37 weeks 10.3% 67.9% Delivered <34 weeks 3.0% 63.6% Delivered <32 weeks 1.7% 61.2% Smoking during pregnancy Smokers 16.5% 19.1% Cigarettes smoked/day‡ 11 (1–40) 15 (1–40) † Data expressed as mean (standard deviation). ‡ Data expressed as median (range) among all smokers. ¶ P-value <0.01. For all other comparisons, P < 0.0001. We first separately examined the effect of each of the 3 covariates standardized birth weight, gestational age, and number of cigarettes smoked per day, on neonatal mortality. This was done by fitting nonparametric logistic regression models (GAM). The univariable GAM strongly suggests that the unadjusted association between standardized z-score birth weight and neonatal mortality is nonlinear (not shown). The association between gestational age and neonatal mortality was also nonlinear, whereas the association between number of cigarettes smoked per day mortality was virtually flat. The adjusted smooth curves for these 3 covariates, along with their corresponding 95% point-wise confidence bands are displayed in Figure 2 . These curves were adjusted for the other two factors in addition to maternal age and gravida. It is interesting to note that the relationship between standardized birth weight and neonatal mortality (adjusted for gestational age and smoking and confounders) was virtually flat at increased birth weight z-scores ( i.e ., at z-scores ≥ 4.0). Figure 2 Adjusted log-odds of neonatal mortality (thick curve) with 95 percent point-wise confidence bands (shaded area) for z-score birth weight (left panel), gestational age (middle panel), and number of cigarettes smoked per day (right panel). Each factor was adjusted for the other two factors as well as for maternal age and gravidity. Since smoking was weakly associated with neonatal mortality, we examined if the effect of smoking on mortality was mediated through either standardized birth weight or gestational age (or both). We therefore modeled neonatal mortality in relation to these 2 covariates (and adjusted for confounders) within broad categories of smokers and nonsmokers (Fig 3 ). Compared with nonsmokers, neonatal mortality among women that smoked during their pregnancy was higher among infants that were between -5 and -1, and between 1 and 5 standard deviation units of the birth weight distribution among smokers. Infants with birth weight z-scores between -1 and 1 had mortality rates that were similar regardless of maternal smoking status. When neonatal mortality rates were examined by gestational age, the mortality curve was consistently higher at every gestational age among smokers than among nonsmokers (P < 0.001). In order to better understand whether smoking affects fetal growth, we examined the distributions of gestational age-specific standardized birth weight z-scores between the two groups of smokers (Fig 4 ). The results indicate that the adjusted mean z-score birth weight among nonsmokers is fairly constant across gestational age. However, among women that smoked during pregnancy, the adjusted mean z-score is higher that those of nonsmokers between 22 and 28 weeks, and begins to drop precipitously with increasing gestational age. This pattern indicates that smoking results in more growth restricted infants, and that the effect of smoking on reduced fetal growth appears to get stronger at gestational ages 32 weeks and beyond. Figure 3 Adjusted log-odds of neonatal mortality based on z-score birth weight (left panel) and gestational age (right panel) among smokers (thick curve) and nonsmokers (thin curve). Each factor was adjusted for the other factor as well as for maternal age and gravidity. Figure 4 Distribution of gestational age-specific mean z-score birth weight among smokers (thick curve) and nonsmokers (thin curve). The curves were adjusted for maternal age and gravidity. The logistic regression models discussed thus far are based on the implicit assumption that the combined effects of standardized birth weight and gestational age are multiplicative on a logistic scale. We examined the sensitivity of this assumption by modelling neonatal mortality by allowing an interaction term between these two factors based on nonparametric smooth fit. The joint effect of standardized birth weight and gestational age on neonatal mortality reveals that both reduced fetal growth and early delivery result in increasing mortality risk, with the mortality plane progressively diminishing with increasing standardized birth weight and gestational age (Fig 5 ). Figure 5 Adjusted smoothed surface of risk of neonatal mortality in relation to z-score birth weight and gestational age. The curve was adjusted for smoking, maternal age, and gravidity. Discussion For decades, several researchers have focused on trying to understand the complex biological relationship among pregnancy duration, infant size, and neonatal mortality. Not only are gestational age and birth weight highly correlated, but both are powerful predictors of neonatal mortality [ 14 - 16 ]. The chief findings from our study include (i) z-score birth weight and preterm delivery (independent of birth weight) exert strong influences on neonatal mortality; (ii) the effect of maternal smoking is mediated largely through reduced fetal growth and, to a smaller extent, through shortened gestation; and (iii) mortality among babies born to smoking mothers is virtually higher at every z-score birth weight (independent of gestational age) than those born to nonsmoking mothers. The inverted "J"-shaped relationship between birth weight and mortality essentially holds for analyses relating to gestational age and mortality. While birth weight is considered a marker for fetal size, gestational age is thought of as an indicator of fetal maturity. Almost 3 decades ago, Susser and colleagues [ 15 ] proposed that gestational age is causally precedent to birth weight (implying that birth weight is in the causal pathway of the gestational age-mortality relationship). Wilcox and Skjaerven [ 16 ] examined close to 400,000 singleton births from Norway in an effort to separate the influences of birth weight and gestational age on neonatal mortality. They showed that, comparisons using the "relative birth weight" scale, there were two strong and separable factors related to mortality: gestational age independent of birth weight, and relative birth weight at any given gestational age. On these similar lines, Herman and Hastie [ 17 ] examined neonatal mortality in relation to (absolute) birth weight and gestational age. They initially speculated that among preterm (<37 weeks) babies, maturity would serve as a strong predictor of mortality, while among term babies, the increased mortality was probably due to growth restriction. However, their analysis showed that mortality was associated only with birth weight and not with gestational age. Their approach to analysis may have suffered from collinearity (between birth weight and gestational age), perhaps leading to the attenuated gestational age-mortality relationship [ 17 ]. Coory [ 18 ] analyzed neonatal mortality in relation to birth weight and gestational age. He concluded that both birth weight and gestational age have independent effects on mortality, and that both are fundamental risk-adjusting variables. However, he was cautious in not interpreting the effects of gestational age, but focused his interpretations almost entirely on birth weight. Our construction of standardized birth weight z-score was developed conditional on gestational age. Thus, this birth weight z-score (independent of gestational age) enabled us to assess the effects of shortened gestation and fetal growth restriction on mortality. It is widely acknowledged that smoking mothers give birth to infants that are lighter compared with those born to nonsmoking mothers. This reduction in birth weight is thought mainly to result in fetal growth restriction, as well as to shortened gestation [ 19 , 20 ]. Although the precise mechanism by which smoking during pregnancy affects the fetus is unclear, two possible pathways have been proposed. Smoking results in increased capillary fragility and vasoconstriction of arterial walls, leading to reduced blood flow to the uterus and eventually to the placenta [ 21 ]. The second is the "fetal hypoxia" hypothesis, whereby smoking leads to a villous shrinkage due to an alteration in the thickness of the villous membrane, thereby reducing oxygen transfer to the fetus [ 22 ]. Both mechanisms are likely to increase the risk of uteroplacental bleeding in pregnancy [ 23 ], which, in turn, increases the risk of not only neonatal deaths [ 20 , 24 ], but also preterm delivery and growth restriction [ 23 ]. Our study provides circumstantial evidence that after the general effects of (shortened) gestational age and (reduced) fetal growth are accounted for, smoking has little direct impact on neonatal mortality. Our study has some limitations. First, errors in the estimation of gestational age [ 25 , 26 ] are likely to affect our results to some extent. Our study was based on gestational age largely determined from the date of last menstrual period as opposed to one based on early ultrasound. Sonographically estimated gestational age is likely to shift the overall gestational age distribution to lower gestational ages [ 26 ] sometimes by as much as a full menstrual cycle [ 25 ], possibly due to delayed ovulation or amenorrhea. Second, the impact of congenital malformations and chromosomal abnormalities on the risk of neonatal death could have been partly responsible for the findings noted here. Although data on malformations are contained on the vital statistics files, they are recorded poorly. Third, although we adjusted all the analysis for maternal age and gravidity, the study does not take into account other known or suspected risk factors for neonatal mortality. These risk factors may account for a part of the associations noted here, but is unlikely that these factors could explain the powerful effects of fetal growth restriction and preterm delivery on neonatal mortality. Finally, non-differential misclassification of smoking data on vital records is likely [ 27 ] and may have attenuated the smoking-mortality association to some extent. Application of generalized additive regression models to examine neonatal mortality appears useful towards understanding the complex biological relationship amongst the predictors. However, we make no claim that GAMs serve as adjuncts to other modeling approaches; on the contrary, we believe that GAMs can provide the first step toward modeling complex exposure-disease relationships. Conclusions Our study provides important insights about the combined effects of gestational age, fetal growth, and smoking during pregnancy on neonatal mortality. Both standardized z-score birth weight and preterm delivery are strongly associated with neonatal mortality, and the effect of maternal smoking appears largely mediated largely through reduced fetal growth and, to a smaller extent, through shortened gestation. Competing interests The author(s) declare that there are no competing interests. Authors' contributions CVA and RWP conceived the idea for the study. CVA assembled the data and performed all statistical analyses. CVA drafted the manuscript with essential contributions from RWP. Both authors critically reviewed the manuscript, edited it for content, and revised it to the final form. Both authors have read and approve the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Description of generalized additive models. Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535930.xml

387275

Semi-Supervised Methods to Predict Patient Survival from Gene Expression Data

An important goal of DNA microarray research is to develop tools to diagnose cancer more accurately based on the genetic profile of a tumor. There are several existing techniques in the literature for performing this type of diagnosis. Unfortunately, most of these techniques assume that different subtypes of cancer are already known to exist. Their utility is limited when such subtypes have not been previously identified. Although methods for identifying such subtypes exist, these methods do not work well for all datasets. It would be desirable to develop a procedure to find such subtypes that is applicable in a wide variety of circumstances. Even if no information is known about possible subtypes of a certain form of cancer, clinical information about the patients, such as their survival time, is often available. In this study, we develop some procedures that utilize both the gene expression data and the clinical data to identify subtypes of cancer and use this knowledge to diagnose future patients. These procedures were successfully applied to several publicly available datasets. We present diagnostic procedures that accurately predict the survival of future patients based on the gene expression profile and survival times of previous patients. This has the potential to be a powerful tool for diagnosing and treating cancer.

Introduction Predicting Patient Survival When a patient is diagnosed with cancer, various clinical parameters are used to assess the patient's risk profile. However, patients with a similar prognosis frequently respond very differently to the same treatment. This may occur because two apparently similar tumors are actually completely different diseases at the molecular level ( Alizadeh et al. 2000 ; Sorlie et al. 2001 ; van de Vijver et al. 2002 ; van't Veer et al. 2002 ; Bullinger et al. 2004 ; Lapointe et al. 2004 ). The main example discussed in this paper concerns diffuse large B-cell lymphoma (DLBCL). This is the most common type of lymphoma in adults, and it can be treated by chemotherapy in only approximately 40% of patients ( NHLCP 1997 ; Vose 1998 ; Coiffier 2001 ). Several recent studies used DNA microarrays to study the gene expression profiles of patients with DLBCL. They found that it is possible to identify subgroups of patients with different survival rates based on gene expression data ( Alizadeh et al. 2000 ; Rosenwald et al. 2002 ; Shipp et al. 2002 ). If different subtypes of cancer are known to exist, there are a variety of existing techniques that can be used to identify which subtype is present in a given patient ( Golub et al. 1999 ; Hastie et al. 2001a ; Hedenfalk et al. 2001 ; Khan et al. 2001 ; Ramaswamy et al. 2001 ; Nguyen and Rocke 2002a , 2002b ; Shipp et al. 2002 ; Tibshirani et al. 2002 ; van de Vijver et al. 2002 ; van't Veer et al. 2002 ; Nutt et al. 2003 ). However, most of these techniques are only applicable when the tumor subtypes are known in advance. The question of how to identify such subtypes, however, is still largely unanswered. There are two main approaches in the literature to identify such subtypes. One approach uses unsupervised learning techniques, such as hierarchical clustering, to identify patient subgroups. This type of procedure is called “unsupervised” since it does not use any of the clinical information about the patient. The subgroups are identified using only the gene expression data. (In contrast, “supervised learning” would use the clinical data to build the model.) For an overview of unsupervised learning techniques, see Gordon (1999 ) or Hastie et al. (2001b ). Hierarchical clustering ( Eisen et al. 1998 ) has successfully identified clinically relevant cancer subtypes in several different studies ( Alizadeh et al. 2000 ; Bhattacharjee et al. 2001 ; Sorlie et al. 2001 ; Beer et al. 2002 ; Lapointe et al. 2004 ). However, one drawback to unsupervised learning procedures is that they may identify cancer subtypes that are unrelated to patient survival. Although several different subtypes of a given cancer may exist, if the prognosis for all patients is the same regardless of which subtype they have, then the utility of this information is limited. Since unsupervised learning procedures by definition do not use the clinical data to identify subtypes, there is no guarantee that the subtypes they identify will be correlated with the clinical outcome. The second approach to identifying subtypes of cancer is based exclusively on the clinical data. For example, patients can be assigned to a “low-risk” or a “high-risk” subgroup based on whether they were still alive or whether their tumor had metastasized after a certain amount of time. This approach has also been used successfully to develop procedures to diagnose patients ( Shipp et al. 2002 ; van de Vijver et al. 2002 ; van't Veer et al. 2002 ). However, by dividing the patients into subgroups based on their survival times, the resulting subgroups may not be biologically meaningful. Suppose, for example, that there are two tumor cell types. Suppose further that patients with cell type 2 live slightly longer than patients with cell type 1 but that there is considerable overlap between the two groups ( Figure 1 ). Assume also that the underlying cell types of each patient are unknown. If we were to assign patients to “low-risk” and “high-risk” groups based on their survival times, many patients would be assigned to the wrong group, and any future predictions based on this model would be suspect. We can obtain more accurate predictions by identifying these underlying subtypes and building a model that can determine which subtype is present in future patients. Figure 1 Two Patient Subgroups with Overlapping Survival Times Proposed Semi-Supervised Methods To overcome these difficulties, we propose a novel procedure that combines both the gene expression data and the clinical data to identify cancer subtypes. The crux of the idea is to use the clinical data to identify a list of genes that correlate with the clinical variable of interest and then apply unsupervised clustering techniques to this subset of the genes. For instance, in many studies, the survival times of the patients are known even though no tumor subtypes have been identified ( Alizadeh et al. 2000 ; Bhattacharjee et al. 2001 ; Sorlie et al. 2001 ; Beer et al. 2002 ; Rosenwald et al. 2002 ; Shipp et al. 2002 ; van de Vijver et al. 2002 ; van't Veer et al. 2002 ; Nutt et al. 2003 ; Bullinger et al. 2004 ). We can calculate the Cox score for each gene in the expression data—the Cox score measures the correlation between the gene's expression level and patient survival—and consider only the genes with a Cox score that exceeds a certain threshold. Once such a list of significant genes is compiled, there are several methods we can use to identify clinical subgroups. We can apply clustering techniques to identify subgroups of patients with similar expression profiles. Once such subgroups are identified, we can apply existing supervised learning techniques to classify future patients into the appropriate subgroup. In this study, we will use the “nearest shrunken centroids” procedure of Tibshirani et al. (2002 ), which is implemented in the package PAM ( Tibshirani et al. 2003 ). For a brief description of the procedure, see “ Materials and Methods .” Sometimes, however, a continuous predictor of survival is desired. We also describe a supervised version of principal components analysis that can be used to calculate a continuous risk score for a given patient and identify subtypes of cancer. The resulting predictor performs very well when applied to several published datasets. These two methods will produce satisfactory results in most datasets. However, we will describe some variations of these methods that can sometimes improve their performance. When we cluster a dataset using only a subset of the genes, it is important that we choose the correct subset of genes. Choosing the genes with the largest Cox scores is generally a good strategy, but this procedure sometimes selects some spurious genes. We will show that one can use partial least squares (PLS) to compute a “corrected” Cox score. Selecting the genes with the largest “corrected” Cox scores can produce better clusters than selecting genes with largest raw Cox scores. Additionally, we will describe two other continuous predictors of survival that we will call β˜ and γ^. For some problems, they are better predictors than the continuous predictor based on supervised principal components ( Figures S1–S3 ). These methods are described in Protocol S1 . Related Methods in the Literature Ben-Dor et al. (2001 ) and von Heydebreck et al. (2001 ) attempt to identify biologically meaningful tumor subtypes from gene expression data by clustering on a subset of the genes. The important distinction between these methods and our semi-supervised clustering method is that our method uses the available clinical data to choose the subset of the genes that is used to perform the clustering. The methods of von Heydebreck et al. (2001 ) and Ben-Dor et al. (2001 ) do not use this clinical information. We will show that utilizing the available clinical data can improve the quality of the clustering. There are also related methods for predicting the survival of cancer patients using gene expression data. Nguyen and Rocke (2002a ) use a form of PLS to predict survival. Li and Luan (2003 ) use support vector machines (SVMs). However, a drawback of these methods is the fact that they use a combination of all of the genes to predict survival. Since the vast majority of the genes in a given dataset are unrelated to survival, the result is that many of the inputs to the model are superfluous, which reduces the predictive accuracy of the model. We will show that our semi-supervised methods, which use only a subset of the genes, generally perform better than these methods. Moreover, in many applications, we would like to identify which genes are the best predictors of survival. These genes could be analyzed in the laboratory to attempt to discover how they influence survival. They could also be used to develop a diagnostic test based on immunostaining or reverse transcriptase PCR. For these applications, it is important to have a predictor of survival that is based on a small subset of the genes. This is another important advantage of our methods over existing methods. Beer et al. (2002 ) utilized an ad hoc method that fit a series of univariate Cox proportional hazards models and took a linear combination of the resulting coefficients. A brief description of their method is given in Protocol S1 . This method is similar to our methods in that it selects a relevant subset of genes by choosing the genes with the largest Cox scores. However, this method is a purely supervised procedure. It does not apply any unsupervised methods (such as clustering or principal components analysis) to this subset of genes to identify additional patterns in the data. We will show that our semi-supervised procedures generally perform better than this method. Summary Our goal is to identify subtypes of cancer that are both clinically relevant and biologically meaningful. Suppose that we have 𝓃 patients, and we measure the expression level of p genes for each patient. (Note that 𝓃 ≫ p .) We assume that there are several different types (classes) of cancer, each of which responds differently to treatment, and each of which is distinct at the molecular level. Therefore, given a set of 𝓃 patients with different classes of cancer, we wish to train a classifier that can diagnose which type of cancer a future patient has, given the expression levels of the patient's p genes. We will show that it is possible to identify such subgroups using the semi-supervised learning techniques described in the previous paragraph, and that identification of such subgroups can enable us to predict the clinical outcome of cancer more accurately. Results Fully Unsupervised Clustering As noted in the Introduction, we needed to assign each patient to a subgroup before we could apply nearest shrunken centroids. First, we applied an unsupervised 2-means clustering procedure to the DLBCL data of Rosenwald et al. (2002 ). This dataset consisted of measurements on 7399 genes from 240 patients. Of these 240 patients, 160 were used for training the model and 80 were reserved for validating the model. A survival time was given for each patient, which ranged between 0 and 21.8 y. We compared the survival times of the two subgroups using a log-rank test. The log-rank test statistic was 0.7, with a corresponding p- value of 0.416. Thus, conventional clustering techniques failed to identify subgroups that differed with respect to their survival times. Subgroups identified using hierarchical clustering also did not differ with respect to survival (data not shown). Using Clinical Data Alone to Generate Class Labels We assigned each patient in the training data to either a “low-risk” or “high-risk” subgroup based on their survival time (see “ Materials and Methods ” for details.) After applying nearest shrunken centroids with crossvalidation, we selected a model that used 249 genes. We then used this model to assign each patient in the independent test data to the “low-risk” or “high-risk” group. The plots of the two survival curves associated with the two classes generated by the model are shown in Figure 2 . The p -value of the log-rank test was 0.03. Figure 2 Comparison of the Survival Curves of the “Low-Risk” and “High-Risk” Groups These were obtained by applying nearest shrunken centroids to the DLBCL test data. Patients in the training data were assigned to either the “low-risk” or “high-risk” group depending on whether or not their survival time was greater than the median survival time of all the patients. Supervised Clustering In order to identify tumor subclasses that were both biologically meaningful and clinically relevant, we applied a novel, supervised clustering procedure to the DLBCL data. We ranked all of the genes based on their univariate Cox proportional hazards scores, and performed clustering using only the “most significant” genes. Recall that when we performed 2-means clustering on the patients in the test data using all 7,399 genes and used a log-rank test to compare the survival times of the patients in the two resulting clusters, the result was not significant. To test our new clustering method, we calculated the Cox scores of all 7,399 genes based on the 160 training observations and ranked the genes from largest to smallest based on their absolute Cox scores. We then clustered the 80 test observations using only the 25 top-scoring genes. This time, the log-rank statistic comparing the survival times of the two clusters was highly significant ( p = 0.0001). A plot of the two resulting survival curves is shown in Figure 3 . A plot of the survival curves that we obtained by applying 2-means clustering to all of the genes is also shown for comparison. Figure 3 Comparison of the Survival Curves Resulting from Applying Two Different Clustering Methods to the DLBCL Data Other Clustering Methods Both Ben-Dor et al. (2001 ) and von Heydebreck et al. (2001 ) used a subset of the genes to try to cluster a microarray dataset in a biologically meaningful manner. They observed that clustering using a subset of the genes can produce better results than using all of the genes. However, these methods were still fully unsupervised since they used only the gene expression data to perform the clustering. They did not use the clinical data to identify subgroups. Although these methods do a better job of identifying biologically meaningful clusters than clustering based on all of the genes, there is no guarantee that the clusters thus identified are associated with the clinical outcome of interest. Indeed, both Ben-Dor et al. (2001 ) and von Heydebreck et al. (2001 ) applied their procedures to a small DLBCL dataset of 40 patients ( Alizadeh et al. 2000 ). The clusters they identified did not (with a few exceptions) differ significantly from one another with respect to survival. We applied the clustering procedure of von Heydebreck et al. (2001 ) to the larger DLBCL dataset of Rosenwald et al. (2002 ). Figure 4 shows the survival curves of the two clusters generated using this method. The survival curves generated by clustering on the genes with the largest Cox scores are included for comparison. Note that the two subgroups identified using the clustering procedure of von Heydebreck et al. (2001 ) do not differ significantly with respect to survival. Figure 4 Comparison of the Survival Curves Resulting from Applying Two Different Clustering Methods to the DLBCL Data Survival Diagnosis We showed that the cancer subgroups identified using this supervised clustering method can be used to predict survival in future patients. The idea is straightforward. First, we identified subgroups of patients using supervised clustering. Then we trained a nearest shrunken centroid classifier to predict the subgroup to which each patient belonged. Details are given in “ Material and Methods .” We tested this procedure on the DLBCL data. A clustering based on 343 genes produced the smallest crossvalidation error rate, so we used a classifier based on this clustering to assign each of the 80 test patients to one of the two subgroups. The survival curves of the two predicted subgroups are shown in Figure 5 ; the p -value of the log-rank test comparing the two survival curves is 0.008. Figure 5 Survival Curves for Clusters Derived from the DLBCL Data Supervised Principal Components We used a form of the principal components of the expression matrix to predict survival. Principal components analysis is an unsupervised learning technique that is used to reduce the dimensionality of a dataset by calculating a series of “principal components.” The hope is that the first few principal components will summarize a large percentage of the variability in the entire dataset. See Hastie et al. (2001b ) for a description of principal components analysis. Unfortunately, principal components analysis suffers from the same limitations as purely unsupervised clustering. If we perform principal components analysis using all of the genes in a dataset, there is no guarantee that the resulting principal components will be associated with survival. Thus, we propose a semi-supervised form of principal components analysis that we call “supervised principal components.” Rather than using all of the genes when we perform principal components analysis, we use only a subset of the genes that are correlated with survival. Using the 160 training observations, we computed the Cox scores for each gene. We kept the 17 genes with Cox scores of 2.39 or greater. We calculated the principal components of the training data using only these 17 genes. Then we approximated the principal components of the test data using equation (11) (see “ Materials and Methods ” for details.) Figure 6 shows that there does appear to be a correlation between the value of the first principal component, υ^I 1 , and patient survival. To confirm this observation, we fit a Cox proportional hazards model to a linear combination of υ^I 1 and υ^I 2 , the estimated first and second principal components of the test data, respectively. (See “ Materials and Methods ” for a description of how this linear combination was obtained.) The resulting sum was a significant predictor of survival ( R 2 = 0.113, likelihood ratio test statistic = 9.58, 1 d.f., p = 0.00197). This predictor is stronger than the discrete predictor shown in Figure 5 ( R 2 = 0.08, likelihood ratio test statistic = 6.7, 1 d.f., p = 0.00966). Figure 6 Plot of Survival Versus the Predictor υ^I for the DLBCL Data A Breast Cancer Example Thus far, all of our examples have been based on the DLBCL data of Rosenwald et al. (2002 ). We now apply our methodology to a set of breast cancer microarray data. In a recent study, van't Veer et al. (2002) built a model to predict the time to metastasis of breast cancer in patients based on microarray data from 78 patients. They showed that this model could be used to predict the times to metastasis of 20 independent test patients. Later, in a separate study, this same model was applied to a much larger set of 292 patients ( van de Vijver et al. 2002 ). Unfortunately, the expression levels of only 70 genes were available for the 292 patient dataset, making it difficult to test our methodology. However, we were able to apply our supervised principal components method. The expression levels of approximately 25,000 genes were available for the earlier study (consisting of 78 patients). After applying crossvalidation, we selected a model consisting of eight genes, five of which were included among the 70 genes in the larger dataset. Thus, we fit a supervised principal components model using these five genes and applied it to the dataset of 292 patients. The results are shown in Table 1 . (To compare the predictive power of the various models, we fit a Cox proportional hazards model to each predictor and computed the R 2 statistic for each model. R 2 measures the percentage of the variation in survival time that is explained by the model. Thus, when comparing models, one would prefer the model with the larger R 2 statistic.) We see that our supervised principal components method produced a stronger predictor of metastasis than the procedure described in van't Veer et al. (2002) . Furthermore, our method used only five genes, whereas the predictor of van't Veer et al. (2002) used 70 genes. These results held even though we did not have the expression data for three genes that we would like to have included in our model. Table 1 Supervised Principal Components Applied to Breast Cancer Data Comparison of the values of the R 2 statistic of the Cox proportional hazards model (and the p- value of the associated log-rank statistic) obtained by fitting the times to metastasis to our supervised principal components method and the discrete predictor described in van't Veer et al. (2002) (Of the 78 patients used to build the model in the original study, 61 were included in the larger dataset of 292 patients. Thus, the values of R 2 calculated using all 292 patients are inflated, since part of the dataset used to validate the model was also used to train the model. We include these results merely to demonstrate the greater predictive power of our methodology. Moreover, we repeated these calculations using only the 234 patients that were not included in the earlier study to ensure that our results were still valid.) Comparison With Related Methods in the Literature We compared each of our proposed methods to several previously published methods for predicting survival based on microarray data. In particular, we examined three previously published procedures: a method based on SVMs ( Li and Luan 2003 ), a method based on PLS ( Nguyen and Rocke 2002a ), and an ad hoc procedure that calculated a “risk index” for each patient by taking an appropriate linear combination of a subset of the genes ( Beer et al. 2002 ). Finally, we considered a naive procedure that split the training data into two groups by finding the bipartition that minimized the p -value of the resulting log-rank statistic. A brief description of each of these procedures is given in Protocol S1 ; for a full description of these procedures, see the original papers. We compared these methods on four different datasets (See Datasets S1-S13 ). First, we examined the DLBCL dataset ( Rosenwald et al. 2002 ) that we used in the earlier examples. Recall that there were 7,399 genes, 160 training patients, and 80 test patients. Second, we considered a breast cancer dataset ( van't Veer et al. 2002 ). There were 4,751 genes and 97 patients in this dataset. We partitioned this dataset into a training set of 44 patients and a test set of 53 patients. Third, we examined a lung cancer dataset ( Beer et al. 2002 ). There were 7,129 genes and 86 patients, which we partitioned into a training set of 43 patients and a test set of 43 patients. Finally, we considered a dataset of acute myeloid leukemia patients ( Bullinger et al. 2004 ). It consisted of 6,283 genes and 116 patients. This dataset was partitioned into a training set of 59 patients and a test set of 53 patients. The results are shown in Table 2 . Table 2 Comparison of the Different Methods on Four Datasets Comparison of the different methods applied to the DLBCL data of Rosenwald et al. (2002 ), the breast cancer data of van't Veer et al. (2002) , the lung cancer data of Beer et al (2002), and the acute myeloid leukemia (AML) data of Bullinger et al. (2004 ). The methods are (1) assigning samples to a “low-risk” or “high-risk” group based on their median survival time; (2) using 2-means clustering based on the genes with the largest Cox scores; (3) using the supervised principal components method; (4) using 2-means clustering based on the genes with the largest PLS-corrected Cox scores; (5) using the continuous predictor ; (6) using 2-means clustering to identify two subgroups; (7) partitioning the training data into “low-risk” and “high-risk” subgroups by choosing the split that minimizes the p- value of the log-rank test when applied to the two resulting groups; (8) using SVMs, similar to the method of Li and Luan (2003 ); (9) using a discretized version of (8); (10) Using partial least squares regression, similar to the method of Nguyen and Rocke (2002a ); (11) using a discretized version of (11); (12) using the method of Beer et al. (2002 ) A Simulation Study We compared each of the methods we proposed above on two simulated datasets. (See Data S1-S4 .) The first simulated dataset X had 5,000 genes and 100 samples. All expression values were generated as standard normal random numbers with a few exceptions. Genes 1–50 in samples 1–50 had a mean of 1.0. We randomly selected 40% of the samples to have a mean of 2.0 in genes 51–100, 50% of the samples to have a mean of 1.0 in genes 101–200, and 70% of the samples to have a mean of 0.5 in genes 201–300. We then generated survival times. The survival times of samples 1–50 were generated as normal random numbers with a mean of 10.0 and a standard deviation of 2.0, and the survival times of samples 51–100 were generated as normal random numbers with a mean of 8.0 and a standard deviation of 3.0. For each sample, a censoring time was generated as a normal random number with a mean of 10.0 and a standard deviation of 3.0. If the censoring time turned out to be less than the survival time, the observation was considered to be censored. Finally, we generated another 5000 × 100 matrix of test data X˜I, which was generated the same way X was generated. Survival times for X˜I were also generated in an identical manner. We defined samples 1–50 as belonging to “tumor type 1” and samples 51–100 as belonging to “tumor type 2.” Thus, a successful subgroup discovery procedure should assign samples 1–50 to one subgroup, and samples 51–100 to the other subgroup. We applied the methods discussed above to identify these subgroups (and predict survival) for the simulated dataset. This simulation was repeated ten times. The results are shown in Table 3 . The first column of the table shows how many samples were misclassified when the dataset was originally divided into two subgroups. The second column shows the number of crossvalidation errors that occurred when the nearest shrunken centroids model was applied to these putative class labels. The third column shows the number of incorrectly labeled samples when the optimal nearest shrunken centroids model was used to assign labels to the samples in the test data X˜I. The final column is the value of R 2 obtained by fitting a Cox proportional hazards model to the predicted class labels for the test data (or by fitting a Cox model to γ^ in the case of methods 4 and 6). Table 3 Comparison of the Different Methods on Our Simulated Data The methods are (1) assigning samples to a “low-risk” or “high-risk” group based on their median survival time; (2) using 2-means clustering to identify two subgroups; (3) using 2-means clustering based on the genes with the largest Cox scores; (4) using the supervised principal components method; (5) using 2-means clustering based on the genes with the largest PLS-corrected Cox scores; (6) using the continuous predictor . Each entry in the table represents the mean over 10 simulations; the standard error is given in parentheses In the first simulation, we found that the fully supervised and the fully unsupervised methods produced much worse results than the semi-supervised methods. (For each iteration of the “median cut” method, the crossvalidation error was minimized when all of the observations were assigned to the same class. Hence, each such model had no predictive power, and the value of R 2 was zero for each iteration. If we had chosen a smaller value of the tuning parameter Δ, the procedure would have performed better, although not significantly better.) The continuous predictor based on supervised principal components performed nearly as well as the methods based on semi-supervised clustering. Next, we performed a second simulation. The second simulated dataset X had 1000 genes and 100 samples. All expression values were generated as Gaussian random variables with a mean of zero and a variance of 1.5, although again there were a few exceptions. Genes 1–50 had a mean of 0.5 in samples 1–20, a mean of 0.6 in samples 21–40, a mean of 0.7 in samples 41–60, a mean of 0.8 in samples 61–80, and a mean of 0.9 in samples 81–100. And again, we randomly selected 40% of the samples to have a mean of 2.0 in genes 51–100, 50% of the samples to have a mean of 1.0 in genes 101–200, and 70% of the samples to have a mean of 0.5 in genes 201–300. To generate the survival time of each “patient,” we took the sum of the expression levels of the first 50 genes and added a Gaussian noise term with variance 0.01. There was no censoring mechanism for the second simulation. We also generated another 1000 × 100 matrix of test data using an analogous procedure. Under this model, there are actually five “tumor subgroups.” However, we still used 2-means clustering on this simulated dataset in order to evaluate the performance of our methods when the number of clusters is chosen incorrectly. Thus, in this simulation, it does not make sense to talk about the number of “misclassification errors;” we can only compare the methods on the basis of their predictive ability. We applied the six different methods to this new simulated dataset and repeated this simulation ten times; the results are shown in Table 4 . The supervised principal component method is the clear winner in the second simulation study. The semi-supervised methods performed poorly because there were a large number of subgroups and there was a considerable overlap between subgroups. This example demonstrates that the supervised principal component method performs well regardless of the number of tumor subclasses and that it seems to perform especially well when survival is an additive function of the expression level of certain genes. Table 4 Comparison of the Different Methods on Our Simulated Data The methods are (1) assigning samples to a “low-risk” or “high-risk” group based on their median survival time; (2) using 2-means clustering to identify two subgroups; (3) using 2-means clustering based on the genes with the largest Cox scores; (4) using the supervised principal components method; (5) using 2-means clustering based on the genes with the largest PLS-corrected Cox scores; (6) using the continuous predictor . Each entry in the table represents the mean over 10 simulations; the standard error is given in parentheses Discussion One important goal of microarray research is to develop more powerful diagnostic tools for cancer and other diseases. Consider a hypothetical cancer that has two subtypes. One subtype is known to spread much more rapidly than the other subtype, and hence must be treated much more aggressively. We would like to be able to diagnose which type of cancer patients have and give them the appropriate treatment. If it is known that two such subtypes of a certain cancer exist, and if we have a training set where it is known which patients have which subtype, then we can use nearest shrunken centroids or other classification methods to build a model to diagnose this cancer in future patients. However, in many cases, we do not know how many subtypes are present, nor do we know which patients belong to which subgroup. Thus, it is important to develop methods to identify such subgroups. Unsupervised methods, such as hierarchical clustering, are popular techniques for identifying such subgroups. However, there is no guarantee that subgroups discovered using unsupervised methods will have clinical significance. An alternative is to generate class labels using clinical data. The simplicity of the approach of dividing the patients into two subclasses based on their survival time is attractive, and there is evidence that this procedure can successfully predict survival. Indeed, this procedure produced a significant predictor of survival in four different datasets, suggesting that this approach has some utility. However, as noted in the Introduction, subgroups identified in this manner may not be biologically meaningful. When we applied this model to the DLBCL data described earlier, the misclassification error rate for the shrunken centroids model was very high (around 40%), so a diagnosis based on this procedure is likely to be inaccurate. Supervised clustering methods can overcome these problems. We have seen that if we selected significant genes prior to clustering the data, we could identify clusters that were clinically relevant. We have also seen how knowledge of these clusters could be used to diagnose future patients. This supervised clustering methodology is a useful prognostic tool. It is also easy to interpret. However, it has certain shortcomings as well. Recall our conceptual model shown in Figure 1 . Patients with tumor type 2 live longer than patients with tumor type 1 on average, but there is still significant variability within each tumor type. Even if we can diagnose a patient with the correct tumor type 100% of the time, the prognosis of the patient may be inaccurate if the variability in survival time within each tumor type is large. Thus, it would be desirable to find a continuous predictor of survival that accounts for this within-group variability. One possible such predictor is our supervised principal components procedure. This procedure used the principal components of a subset of the expression matrix X as a predictor of patient survival. The chosen subset contained the genes with the largest Cox scores. This method could also be used to detect cancer subtypes, since the principal components will presumably capture the variation that exists between subtypes. It is also capable of identifying variation within these subtypes, which, as discussed above, cannot be identified using supervised clustering. We showed that this procedure could produce a stronger predictor of survival than the discrete predictor based on supervised clustering. We compared our methods to several previously published methods for predicting survival based on microarray data. In general, our methods performed significantly better than these existing methods. In particular, our supervised principal components method gave the best results on three of the four datasets. (It performed slightly worse than our γ^ method on the DLBCL data, but it still outperformed almost all of the other methods.) Furthermore, each of our proposed methods was a significant predictor of survival (at p = 0.05) for all four datasets, which was not true for any of the other methods. Finally, if we consider only discrete predictors of survival, our semi-supervised clustering methods performed better than the other models on at least three of the four datasets. Another important advantage of our methods is that they select a subset of the genes to use as predictors. The methods of Nguyen and Rocke (2002a ) and Li and Luan (2003 ), by contrast, require the use of all (or a large number) of the genes. If we can identify a small subset of genes that predict the survival of cancer patients, it may be possible to develop a diagnostic test using immunostaining or reverse transcriptase PCR. However, such a test would not be feasible if hundreds or thousands of genes were necessary to make the diagnosis. Throughout this study, we have used survival data to help us identify genes of interest. However, other clinical variables could also be used, such as the stage of the tumor, or whether or not it has metastasized. Rather than ranking genes based on their Cox scores, one would use a different metric to measure the association between a given gene and the clinical variable of interest. For example, suppose we wished to identify a subgroup of cancer that was associated with a high risk of metastasis. For each gene, we could compute a t-statistic comparing the expression levels in the patients whose cancer metastasized to those in the patients with no metastasis. Tusher et al. (2001 ) described methods for generating such “significant gene lists” for a variety of possible clinical variables. Many of these methods are implemented in the significance analysis of microarrays software package ( Chu et al. 2002 ). Information about the risk of metastasis (and death) for a given patient is essential to treat cancer successfully. If the risk of metastasis is high, the cancer must be treated aggressively; if the risk is low, milder forms of treatment can be used. Using DNA microarrays, researchers have successfully identified subtypes of cancer that can be used to assess a patient's risk profile. Our results show that semi-supervised learning methods can identify these subtypes of cancer and predict patient survival better than existing methods. Thus, we believe they can be a powerful tool for diagnosing and treating cancer and other genetic diseases. Materials and Methods Overview of nearest shrunken centroids The nearest shrunken centroids procedure calculates the mean expression of each gene within each class. Then it shrinks these centroids toward the overall mean for that gene by a fixed quantity, Δ. Diagonal linear discriminant analysis (LDA) is then applied to the genes that survive the thresholding. Details are given in Tibshirani et al. (2002 ). It has successfully classified tumors based on gene expression data in previous studies. In one experiment, there were a total of 88 patients, each of which had one of four different types of small round blue cell tumors . Nearest shrunken centroids classified 63 training samples and 25 test samples without a single misclassification error ( Tibshirani et al. 2002 ). Generation of “median cut” class labels We created two classes by cutting the survival times at the median survival time (2.8 y). Any patient who lived longer than 2.8 y was considered to be a “low-risk” patient, and any patient that lived less than 2.8 y was considered to be a “high-risk” patient. In this manner, we assigned a class label to each observation in the training data. Unfortunately, many of the patients' survival times were censored, meaning that the individual left the study before the study was completed. When this occurs, we do not know how long the patient survived; we only know how long the patient remained in the study prior to being lost to follow-up. If an observation is censored, we may not know to which class it belongs. For example, suppose that the median survival time is 2.8 y, but that a patient left the study after 1.7 y. If the patient died in the interval between 1.7 y and 2.8 y, then the patient should be assigned to the “high-risk” group. Otherwise, the patient should be assigned to the “low-risk” group. However, there is no way to determine which possibility is correct. Based on the Kaplan-Meier survival curve for all the patients, we can estimate the probability that a censored case survives a specified length of time ( Cox and Oakes 1984 ; Therneau and Grambsch 2000 ). For example, suppose that the median survival time is 50 months and a patient left the study after 20 months. Let T denote the survival time of this patient. Then, using the Kaplan-Meier curve, we can estimate p ( T >50) and p ( T >20). Then we can estimate p ( T >50| T >20) as follows: and, of course, In this manner, we can estimate the probability that each censored observation belongs to the “low-risk” and “high-risk” classes, respectively. However, it is still unclear how we would train our classifier based on this information. Nearest shrunken centroids is a modified version of LDA. It is described in detail in Hastie et al. (2001b ). Like most classification techniques, LDA assumes that the class labels of the training observations are known with complete certainty. The version of LDA described in Hastie et al. (2001b ) and most other books cannot handle probabilistic class labels, where there is a certain probability that a training observation belongs to one class, and a certain probability that it belongs to a different class. We will now describe a simple modification of LDA that can be trained based on this type of data. It is similar to a technique described in McLachlan (1992 ) for training an LDA classifier when some of the training observations are missing. Let { x i } denote the set of input variables, and let { y i } represent the corresponding response variables. Also, let g represent the number of discrete classes to which the y i s may belong. (If we are dividing the training data into “low-risk” and “high-risk” patients, then g = 2.) When we perform LDA when all of the y i s are known, the problem is to fit the mixture model (Generally, each 𝒻 i is a Gaussian density function, and the θ i s correspond to the mean of the observations in each class. The π i s correspond to “prior” probabilities that an observation belongs to class 𝒾.) In this case, we must fit this model on the basis of the classified (uncensored) training data, which we denote by t , and the unclassified (censored) feature vectors x 𝒿 ( 𝒿 = 𝓃+1, …,𝓃+𝓂), which we denote by t 𝓊 . (Also, note that Φ = (π′,θ′)′ denotes the vector of all unknown parameters.) We define the latent variables 𝓏 ij to be equal to one if the 𝒿th observation belongs to the 𝒾th class, and zero otherwise. Then the complete-data log likelihood is The EM algorithm is applied to this model by treating z j (𝒿 = 𝓃+1,…, 𝓃+𝓂) as missing data. It turns out to be very simple in the case of LDA. The E-step is effected here simply by replacing each unobserved indicator variable 𝓏 ij by its expectation conditional on x j . That is, 𝓏 ij is replaced by the estimate of the posterior probability that the 𝒿th entity with feature vector x j belongs to G i (𝒾 = 1, …, G , 𝒿 = n + 1, …, n + m ) ( McLachlan 1992 ). We take the initial estimates of 𝓏 ij to be the earlier estimate that the 𝒾th censored observation belongs to class 𝒿 based on the Kaplan-Meier curve. The estimates of π i and μ i in the M-step are equally simple: and where In these expressions, τ i ( x ;Φ) is the posterior probability that the 𝒿th entity with feature vector x j belongs to G i , or, in other words, We continue these imputations until the algorithm converges. In practice, one imputation seems to be sufficient for most problems, since each imputation is computationally intensive, and additional imputations did not seem to change the results significantly. Diagnosing patient survival via supervised clustering We calculated the Cox scores of each gene based on the 160 training observations, and obtained a list of the most significant genes. Then we performed 2-means clustering on these 160 observations using the genes with the largest absolute Cox scores and obtained two subgroups. We repeated this procedure multiple times with different numbers of genes. For each such clustering, we trained a nearest shrunken centroid classifier to assign future patients to one subgroup or the other and examined the crossvalidation error rate. The problem of choosing the number of genes on which to perform the clustering is more complicated than it appears. The obvious way to choose the optimal number of genes on which to cluster is to simply minimize the crossvalidation error rate of the nearest shrunken centroids model based on the clustering. This works up to a certain point. It is possible that the clustering procedure will identify a cluster that is unrelated to survival. (Since we are clustering on the genes with the highest Cox scores, this is unlikely to occur. However, it is still possible, especially if the number of genes on which we are clustering is large.) Thus, we needed to build a safeguard against this possibility into our procedure. After performing clustering based on a given set of high-scoring genes, we performed a log-rank test to determine if the resulting clusters differed with respect to survival. If they did not, the clustering was discarded without further analysis. An outline of the procedure follows: (1) Choose a set G of possible values of Γ. (2) Let p min = 1 and e min = 1. (3) For each Γ in G , do the following: (4) Perform k -means clustering using only those genes with absolute Cox scores greater than Γ. (5) Perform a log-rank test to test the hypothesis that the k clusters have different survival rates. Call the p -value of this test p . (6) If p ≥ p min , then return to step 3. (7) Fit a nearest shrunken centroids model based on the clusters obtained in step 3. Calculate the minimum crossvalidation error rate across all values of the shrinkage parameter, and call it e . (8) If e < e min , then let Γ best = Γ, and return to step 3. Otherwise return to step 3 without changing the value of Γ best . The optimal value of Γ is taken to be the value of Γ best when this procedure terminates. Several comments about this procedure are in order. First, note that we did not recalculate the Cox scores at each fold of the crossvalidation procedure. We calculated them only once, using all of the patients in the dataset. There are several reasons for doing this. Recalculating the Cox scores at each fold would be extremely expensive computationally. Moreover, we found that the Cox score of a given gene varied depending on the number of patients (and which patients) we included in the model. Thus, if a given value of Γ produced a low crossvalidation error rate, there was no guarantee that a model based on the full dataset using this value of Γ would produce good results, since the model based on the full dataset may use a different list of genes. Other studies have found that using the entire dataset to produce a “significant gene list” prior to performing crossvalidation can produce more accurate predictions ( van't Veer et al. 2002 ). Also, the set G was left unspecified in the procedure. The choice of which (and how many) possible values of Γ to include in G depends on the problem at hand, as well as the computational power available. As a default, we recommend trying 100 evenly spaced values of Γ between the 90th percentile of the Cox scores and the maximum of the Cox scores. However, the optimal Γ best varies greatly from dataset to dataset, so we recommend trying several different forms of G if adequate computing power exists. Furthermore, note that when we calculated the p -value of the log-rank test after performing the original clustering, we insisted not only that the p -value be significant, but also that it be lower than the best p -value obtained thus far. The reasons for this are twofold. First, experience suggests that if a given set of genes produces a good clustering on the training data (“good” defined as having a low p -value from a log-rank test), then it is likely to produce a good clustering on the test data. (We offer no theoretical or biological justification for this statement; it simply represents our experience. However, we have observed this result a sufficient number of times to convince us that it is not coincidental.) Moreover, this speeds up the algorithm substantially. Calculating the nearest shrunken centroids crossvalidation error rate for a given clustering is the slowest part of the procedure; the time required to perform the clustering and calculate the log-rank statistic is insignificant in comparison. Thus, by only considering clusterings which produce a log-rank statistic with a small p -value, we allow the set G to be much larger than would be feasible otherwise. Finally, the number of clusters k was unspecified in the procedure. We have experimented with some algorithms to choose the value of k automatically, but without success. If possible, we recommend that the value of k be chosen based on prior biological knowledge. (Perhaps one could first perform hierarchical clustering, examine a dendogram of the data, and visually search for major subgroups.) If this is not possible, we recommend trying several different small values of k and choosing the one that gives the best results. (Our experience suggests that choosing k = 2 will give good results for almost all datasets.) Supervised principal components As above, let X be the p × n matrix of expression values, for p genes and n patients. Let x ij denote the expression level of the 𝒾th gene in the 𝒿th patient. Assume that each patient has one of two possible underlying tumor types. Without loss of generality, assume that patients 1, …, m have tumor type 1, and that patients m + 1,…, n have tumor type 2. Then assume that the genetic profiles of the two tumor types are distinct from one another, which is equivalent to assuming that the joint distribution of ( x 1 j , …, x pj ) is different for 1 ≤ 𝒿 ≤ 𝓂 than it is for 𝓂 + 1 ≤ 𝒿 ≤ 𝓃. Thus, if we choose constants { a i } p i =1 , the distribution of ∑ p j =1 a j x ij will be different for 1 ≤ 𝒿 ≤ 𝓂 than it is for 𝓂 + 1 ≤ 𝒿 ≤ 𝓃. (Obviously, this is not true for all values of { a i } p i =1 . For example, if we let a i = 0 for all 𝒾, then this statement will not hold. However, it will generally be true unless we deliberately choose a pathological set { a i } p i =1 .) In particular, consider the singular value decomposition of X : where U is a p × 𝓃 orthogonal matrix, D is an 𝓃 × 𝓃 diagonal matrix, and V is an 𝓃 × 𝓃 orthogonal matrix ( Horn and Johnson 1985 ). Then the matrix V can be written as In other words, for a given column of V , each row of V is a linear combination of the expression values in the corresponding column of X . Thus, by the argument in the preceding paragraph, rows 1 through 𝓂 should have a different distribution than rows 𝓂 + 1 through 𝓃. Hence, we propose that the first few columns of V be used as continuous predictors of survival for each patient. Formally, Moreover, suppose that we have an independent test set X˜I. Then let where U and D are the same as in equation (11) (i.e., derived from the singular value decomposition of the training data). In this case, the first few columns of V^I can be used to estimate survival for the patients in the independent test set. The reason for choosing the first few columns of V is because the matrix U was chosen so that X T u 1 has the largest sample variance amongst all normalized linear combinations of the rows of X ( Hastie et al. 2001b ). (Here, u 1 represents the first column of U .) Hence, assuming that variation in gene expression accounts for variation in survival, we would expect that X T u 1 captures a large percentage of the variation in survival. (Indeed, in some simple models, it can be proven that equation [11] is the best possible predictor of survival; see Protocol S1 .) In theory, we could calculate V using the entire dataset X , and the rows of V would have different distributions depending on the tumor type of the corresponding patient. In practice, however, many of the genes in X are unrelated to survival, and if we use the entire dataset X to compute V , the quality of the resulting predictor is poor. We can overcome this difficulty by using only the genes with the largest Cox scores. Formally, we construct a matrix X ′ consisting of only those genes whose Cox scores are greater than some threshold Γ, and take the singular value decomposition of X ′. To choose the optimal value of Γ, we employ the following procedure: (1) Choose a set G of possible values of Γ. (2) For each Γ in G , split the training data into k random partitions (i.e., perform k -fold crossvalidations). For most problems (and for the rest of this discussion), we can let k = 10. (3) For each crossvalidation fold, take a singular value decomposition of X , leaving out one of the 10 partitions for validation purposes. Use only those genes with absolute Cox scores greater than Γ. (4) Calculate υ^ for the 10% of the data that was withheld, as described above. (5) Fit a Cox proportional hazards model to υ^, and calculate the chi-square statistic for the log-rank test associated with this model. Denote the chi-square statistic for the 𝒾th crossvalidation fold by 𝓌 i . (6) Average the 𝓌 i s over the 10 crossvalidation folds. Call this average 𝓌 Γ . (7) If 𝓌 Γ is greater than the value of 𝓌 Γ∗ , then let Γ∗ = Γ and 𝓌 Γ∗ = 𝓌 Γ . (8) Return to step 2. The set G is left unspecified in the procedure. As a default, we recommend trying 30 evenly spaced values of Γ between the 90th percentile of the Cox scores and the maximum of the Cox scores, although this recommendation is somewhat arbitrary. In some cases, we can improve the predictive power of our model by taking a linear combination of several columns of V (rather than simply taking the first column of V ). Suppose we wish to find a predictor based on the first k columns of V . We can perform the following procedure: (1) Let X denote the training data. Take the singular value decomposition of X = UDV T as described above (after selecting an appropriate subset of the genes). (2) Fit a Cox proportional hazards model using the first k columns of V as predictors. (3) Calculate the matrix V^I for the test data using equation (12) above. (4) Take a linear combination of the first k columns of V^I using the Cox regression coefficients obtained in step 2. Use the resulting sum as a continuous predictor of survival. Software and computational details All computations in this study were performed using the R statistical package, which is available on the Internet at http://cran.r-project.org/ . R source code for the procedures described in this paper are available from the authors upon request (see also Data S1–Data S4 ). These methods will also be implemented in a future version of the PAM microarray analysis package ( Tibshirani et al. 2003 ). (The “median cut” method has been implemented in version 1.20, which is now available.) Supporting Information Data S1 Documentation of Our R Functions This file contains a brief description of the functions contained in the semi-super.R file. (1 KB TXT). Click here for additional data file. Data S2 R Source Code This file contains R functions for implementing the procedures we have described in our study. (6 KB TXT). Click here for additional data file. Data S3 Source Code for Simulation Study 1 This file contains the R source code that we used to perform the first simulation study in our paper. (31 KB TXT). Click here for additional data file. Data S4 Source Code for Simulation Study 2 This file contains the R source code that we used to perform the second simulation study in our paper. (39 KB TXT). Click here for additional data file. Dataset S1 Breast Cancer Expression Data: van't Veer et al. (2002) Study The gene expression data for the breast cancer study of van't Veer et al. (2002) . We include only the expression levels of 4,751 genes identified in the study whose expression varied (2.9 MB CSV). Click here for additional data file. Dataset S2 Breast Cancer Gene Names: van't Veer et al. (2002) Study The names of each of the 4,751 genes in the study of van't Veer et al. (2002) . (74 KB CSV). Click here for additional data file. Dataset S3 Breast Cancer Survival Data: van't Veer et al. (2002) Study The clinical data for the study of van't Veer et al. (2002) . The first column represents the time until metastasis (or the time until the patient left the study); the second column is 1 if the tumor metastasized and 0 if it did not. (1 KB CSV). Click here for additional data file. Dataset S4 Breast Cancer Expression Data: van de Vijver et al. (2002) Study The gene expression data for the 70 genes in the breast cancer study of van de Vijver et al. (2002) . (141 KB CSV). Click here for additional data file. Dataset S5 Breast Cancer Gene Names: van de Vijver et al. (2002) Study The names of the 70 genes in the study of van de Vijver et al. (2002) . (1 KB CSV). Click here for additional data file. Dataset S6 Repeated Breast Cancer Samples A single column that is 1 if the patient was included in the earlier study (that of van't Veer et al. [2002 ]), and 0 if the patient was not included in the earlier study. (1 KB CSV). Click here for additional data file. Dataset S7 Breast Cancer Survival Data: van de Vijver et al. (2002) Study The clinical data for the study of van de Vijver et al. (2002) . The format is the same as the format of the earlier file of clinical data. (5 KB CSV). Click here for additional data file. Dataset S8 DLBCL Expression Data The gene expression data for the DLBCL study of Rosenwald et al. (2002 ). (24.38 MB CSV). Click here for additional data file. Dataset S9 DLBCL Survival Data The clinical data for the study of Rosenwald et al. (2002 ). The format is the same as the format of the clinical data above. (2 KB CSV). Click here for additional data file. Dataset S10 Lung Cancer Gene Expression Data This is the gene expression data for the lung cancer dataset of Beer et al. (2002 ). (5.39 MB TXT). Click here for additional data file. Dataset S11 Lung Cancer Survival Data This is the clinical data for the lung cancer dataset of Beer et al. (2002 ). The format is the same as in the clinical data above. (1 KB TXT). Click here for additional data file. Dataset S12 AML Gene Expression Data This is the gene expression data for the AML dataset of Bullinger et al. (2004 ). (9.96 MB TXT). Click here for additional data file. Dataset S13 AML Survival Data This is the clinical data for the AML dataset of Bullinger et al. (2004 ). It has the same format as the clinical data above. (1 KB TXT). Click here for additional data file. Figure S1 Results of Using PLS-Derived Cox Scores in the Supervised Clustering Procedure (8.26 MB TIFF). Click here for additional data file. Figure S2 Plot of Survival Versus the Least Squares Estimate of β˜ for the DLBCL Data (8.33 MB TIFF). Click here for additional data file. Figure S3 Plot of Survival Versus the Least Squares Estimate of γ˜ for the DLBCL Data (8.42 MB TIFF). Click here for additional data file. Protocol S1 Additional Models and Methods (28 KB TEX). Click here for additional data file.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC387275.xml

526179

Use of a Dense Single Nucleotide Polymorphism Map for In Silico Mapping in the Mouse

Rapid expansion of available data, both phenotypic and genotypic, for multiple strains of mice has enabled the development of new methods to interrogate the mouse genome for functional genetic perturbations. In silico mapping provides an expedient way to associate the natural diversity of phenotypic traits with ancestrally inherited polymorphisms for the purpose of dissecting genetic traits. In mouse, the current single nucleotide polymorphism (SNP) data have lacked the density across the genome and coverage of enough strains to properly achieve this goal. To remedy this, 470,407 allele calls were produced for 10,990 evenly spaced SNP loci across 48 inbred mouse strains. Use of the SNP set with statistical models that considered unique patterns within blocks of three SNPs as an inferred haplotype could successfully map known single gene traits and a cloned quantitative trait gene. Application of this method to high-density lipoprotein and gallstone phenotypes reproduced previously characterized quantitative trait loci (QTL). The inferred haplotype data also facilitates the refinement of QTL regions such that candidate genes can be more easily identified and characterized as shown for adenylate cyclase 7.

Introduction The combined efforts of the public and private mouse genome sequencing consortiums have yielded important advances in understanding the structure and content of the genome ( Mural et al. 2002 ; Waterston et al. 2002 ). Identification of new genes from the sequence data and placement of all genes, along with genetic markers, on a physical assembly has greatly aided in the search for phenotypically important genes in both quantitative trait loci (QTL) and mutagenesis-based mapping. The sequencing of four different strains of laboratory mice for the initial genome assemblies also produced a catalog of natural sequence variations that are present between these commonly used strains. Other smaller scale resequencing efforts have increased the breadth of this information by including additional strains ( Lindblad-Toh et al. 2000 ; Grupe et al. 2001 ; Wade et al. 2002 ; Wiltshire et al. 2003 ). The utility of this sequence-variation data is 2-fold. First, the single nucleotide polymorphisms (SNPs) identified by these sequencing projects provide denser coverage marker sets that are well suited for high-throughput genotyping systems. Currently, these benefits have only been available for crosses between the relatively few strains for which substantial polymorphism discovery has been undertaken. Second, the distribution of SNPs between any two strains, or more precisely, the lack of SNPs between two mouse strains, indicates regions of their genomes that were inherited from a common ancestor. Phenotypic differences that are traditionally mapped in QTL studies are almost exclusively due to loci inherited from different ancestral progenitors rather than new mutations ( Frazer et al. 2004 ). Thus, a detailed knowledge of where common ancestral regions lie between strain pairs would speed QTL mapping by elimination of shared regions from consideration as candidate loci ( Wade et al. 2002 ). Additionally, it has been proposed that the actual haplotype structures marking these ancestral relationships can be determined and that the relationship of haplotype distribution among mouse strains and phenotypic variation could be used to directly map the genetic controls for the phenotypes ( Grupe et al. 2001 ). However, three major factors have seriously curtailed the implementation of in silico mapping methods: a lack of the necessary SNP density and distribution along the genome for more than just a few strains; incomplete phenotype data for multiple strains, and lastly, the appropriate analysis tools for making genotype–phenotype associations. More recently uncertainties have also been expressed concerning the level of data that will be required to make in silico mapping a viable method. This is in part due to the emerging complexity of the haplotype structure in mouse and also to such issues as how many strains need to be phenotyped to be able to gain statistical power for in silico mapping ( Darvasi 2001 ; Frazer et al. 2004 ; Yalcin et al. 2004 ). To overcome the barriers to in silico mapping, 10,990 SNP assays have been typed against 48 mouse strains in this study. These assays provide an extensive polymorphic marker set enabling expansion of traditional mapping efforts to other strains. Wide-ranging phenotyping projects that have been coordinated by The Jackson Laboratory ( http://www.jax.org/phenome ) have collated multistrain phenotype data. We demonstrate that when using these datasets, in combination with new analysis methods, statistically significant associations between discrete genomic regions and biologically important phenotypes can be identified. Confirmation of these associations was obtained by comparison to data from traditional QTL mapping methods. Results SNP assays were designed based on sequence data from the Celera Mouse SNP Database and typed, in duplicate, against the genomic DNA of 48 strains of mice, including all 40 Mouse Phenome Project priority strains (see list of strains in Tables S1 and S2 ) ( Bogue 2003 ). Twelve wild-derived inbred strains were included in the set. Two strains, SPRET/EiJ and SEG/Pas (Mus spretus), represent a different species of mouse from the other lines tested. Not surprisingly, fewer genotypes were obtained from these two because of the divergent genomes (Tables S1 and S2 ) of the distinct species, which led to a higher failure rate in the genotyping reactions. For the 36 non-wild-derived strains typed, 8,349 SNP assays produced at least 90% of the possible allelic data. Previously, sufficient polymorphic markers have not been available for many strain–pair combinations. The development of this SNP panel provides a resource of polymorphic markers to enable traditional mapping projects between almost any strain–pair combination of the 48 strains. In mapping a phenotype, introduction of modifier genes can be a confounding influence, and selection of more closely related mapping partners can alleviate this problem. The SNP density in this set is sufficient that mapping can now be accomplished between strains that had previously been too closely related for sufficient markers to be found. For example, for C58L/J by C57BL/6J and C57BL/6J by C57BL/10J comparisons, over 2,000 and 400 polymorphic markers, respectively, are available. Although large gaps do exist in the coverage of these strain–pair combinations, markers are present on all chromosomes to allow for initial candidate region identification ( Figure 1 ). Details of all allele calls and SNP assays are available in Dataset S1 . Figure 1 Visualization of the SNP Sets Allows for Mapping in Crosses That Minimize the Number of Potential Modifiers When the distribution of the SNPs is plotted out genome-wide, the expected irregular clustering of SNPs mark regions where heterozygosity was continuing to segregate during the inbreeding of the C57 family. Likewise, there are regions that were successfully homozygosed before the split of C58/J from the rest of the family members. In all five strain comparisons, no SNPs were found in the distal 25 Mb of MMU19. Just as in humans, a spectrum of phenotypic values can be observed among the inbred strains of mice. SNPs that occur between these strains may produce a specific functional change in a gene leading to this phenotypic variation but are more often simply markers for an ancestral haplotype. The goal of in silico mapping is to identify which haplotype patterns (genetic measure) track with the phenotypic outcome with the idea that these haplotypes contain causative mutations. For in silico mapping to be successful, a requirement is that the SNP data accurately represent this ancestral relationship of the mice at the genomic level. At a gross level, this was examined by comparing branches of the phylogenic tree generated from this SNP dataset with the known breeding histories of the strains used in this study ( Beck et al. 2000 ). An inspection of the C57-related family of mice, derived from a tree built from the SNP data of all 48 strains ( Figure S1 ), recapitulates the family's lineage in the phylogenic tree ( Figure 2 A). Figure 2 Genome-Wide SNP Data Accurately Represent the Known Ancestries of the Genotyped Strains (A) A tree, adapted from Beck et al. (2000) , tracing the lineage of the C57 family of mice (upper tree) shows almost perfect correlation with a phylogenic tree based solely on SNP data (lower tree). The only difference in the two trees is the location of C57BLKS/J, which splits from C57BL/6J sooner in the phylogenic tree because of the genomic contributions of the non-C57 strain, DBA/2J. The maximum parsimony phylogenic tree of the strain relatedness was built using the pseudoalignment of the 10,990 SNP alleles for 48 strains with the Phylip package, version 3.6b. (B) The DBA/2J contribution to C57BLKS/J can be visualized in its allelic patterns. The region from 104 Mb to 109 Mb on MMU9 shows the same SNP alleles for both C57BLKS/J and its other parental strain, C57BL/6J (a period represents identity with the C57BLKS/J allele). At 110 Mb, the pattern switches and every C57BLKS/J allele matches the DBA/2J content through 120 Mb. SNP marker names are positioned above the alleles with the first number representing the chromosome the marker is located on, the second number being the Mb position on the chromosome, and the third number being an approximate location within the Mb. At a more detailed level the specific genomic contributions of mouse strains derived as hybrids of other common laboratory strains can be estimated. For example, DBA/2J is considered to have contributed approximately 16% of the genomic content to the C57BLKS/J mouse ( Naggert et al. 1995 ). Comparisons of C57BL/6J, the other founder strain of C57BLKS/J, and DBA/2J-specific alleles to the SNP content of C57BLKS/J clearly defines these large regions of DBA/2J contribution as shown for Mus musculus Chromosome 9 (MMU9) ( Figure 2 B). Based on the SNP data, it can be estimated that 20% of the C57BLKS/J genome came from a DBA/2J origin, including almost all of MMUX. This type of analysis also indicates that at least one additional strain, possibly 129-like, contributed genomic content to C57BLKS/J in regions where the allelic pattern matches neither DBA nor C57BL/6J. A lack of sufficient underlying SNP data to this point have prevented the thorough development and testing of an algorithm to carry out in silico mapping ( Chesler et al. 2001 ; Darvasi 2001 ). Previously published methods were severely limited by the lack of SNP density and strain coverage leading to a method that utilized generalized genetic distances and resulted in lack of resolution in the analysis ( Grupe et al. 2001 ; Smith et al. 2003 ). Based on the data above, this SNP set provides sufficient spacing and resolution to distinguish discrete ancestral patterns, allowing for subsequent in silico analyses to treat the genetic measure used in these calculations as categorical. Although the number of SNPs used here still does not allow the precise definition of haplotype blocks, the relatively even spacing of the SNPs every 300 kb does allow for an inference of ancestral relationships across 1-megabase (Mb) regions. For this reason, a sliding window of three SNPs is used to infer haplotypes at each locus. Strains showing the same pattern are grouped in the same inferred haplotype, as a single category, and any variations are considered to form distinct inferred haplotypes. All of the strain-distribution patterns created by this definition of inferred haplotype were compared across the genome to determine their uniqueness. Replication of the same strain-distribution pattern at multiple locations across the genome, or “mirror loci,” would result in regions that are all equally associated with the phenotype and produce false positives. No occurrences of mirror loci were found outside of a 5-Mb region of any three-SNP block. With this in mind a logistic regression model followed by analysis of deviance was used to determine the association between a sliding window of three SNPs and phenotype scores of 1 or 0 for the presence or absence of three Mendelian traits: coat color traits of nonagouti and albino and retinal degeneration. All of the phenotypes were determined from phenotypic descriptions in The Jackson Laboratory mouse database ( http://jaxmice.jax.org/jaxmice-cgi/jaxmicedb.cgi ). Albino mice were excluded from the mapping of nonagouti because the nature of the phenotype prevents the ascertainment of agouti or nonagouti coat colors. In each case, the appropriate locus for the gene responsible for the particular trait was identified from this SNP collection with the most significant p -value ( Figure 3 A; Kwon et al. 1987 ; Bowes et al. 1990 ; Bultman et al. 1992 ). Interestingly, for in silico mapping of the nonagouti locus, a highly significant score was also obtained for a locus on MMU7 at 29.9 Mb. This is also the approximate location of the dark locus, an unidentified gene that also influences coat color ( Silvers 1979 ). Figure 3 In Silico Mapping Method Correctly Identifies Coat Color, Retinal Degeneration, and Sweet Preference Loci from SNP Data (A) Presence or absence of the retinal degeneration, albino, or agouti phenotypes was given a numerical value of 1 or 0 for use in the mapping algorithm. In each case, the most significant p -value (indicated by an arrow) was obtained for the region that contains the gene known to produce these phenotypes. A closer inspection of the retinal degeneration mapping shows that the maximum linkage region indicated by the algorithm covered a 0.4-Mb region from 102.4 Mb to 102.8 Mb on MMU5. (B) Tas1r3 is known to be a major control gene for the complex trait of preference for sweet tastes. Values for the sweet preference of 23 strains of mice produced a highly significant association with the one Mb region of MMU4 that contains Tas1r3. The ability to properly identify causative genes for monogenic traits is a minimum requirement for a viable in silico mapping method, but to serve its intended purpose it must be able to point to controlling loci when multiple genes act in concert to contribute to a phenotype. To examine a quantitative trait, data from a two-bottle saccharin preference test for 23 strains of mice were analyzed with the Fisher permutation-based analysis of variance (ANOVA) statistical model. Briefly, at each three-SNP window, a modified F-statistic based on the true genotype–phenotype pairings is calculated (detailed in Materials and Methods ). The significance of this test statistic is estimated by comparing to a distribution of 1 million random bootstrap samples of phenotypic values. A three-SNP window beginning with marker 04.155.136 obtained the lowest p -value of the genome scan ( Figure 3 B). This locus corresponds to the position of the gene, Tas1r3, identified by traditional QTL methods as a primary contributor to the variability of the sweet preference quantitative trait ( Bachmanov et al. 2001 ; Max et al. 2001 ) Three other saccharin preference QTL were also found to overlap significant associations from this mapping ( Table 1 ). Table 1 Comparison of In Silico QTL with Experimentally Derived QTL B6, C57BL/6J. For definitions of other abbreviations, see Abbreviations section in text a References indicate source of QTL data b The in silico mapping algorithm was run twice for each phenotype, and only SNP blocks that obtained a log p -value above the cutoff in both runs are shown here. When more than one marker within the same genomic region obtained a log p -value above the cutoff, only the marker with the most significant p -value is shown c The median p -value from each algorithm run was averaged d QTL regions were defined as the experimentally determined 95% CIs for the particular strain-cross referenced unless otherwise noted e The 95% CIs were not available, so the QTL regions were defined as ± 10 cM from the position with the highest likelihood-of-odds score (peak), as it represents the approximate size of the available 95% CIs for this dataset and has been previously published in X. Wang and Paigen (2002) as the definition of HDL QTL size. One marker 10 cM away from the peak on either side was chosen, and their physical positions were retrieved from Celera Mouse Database f Additional crosses support this QTL region g MMUX has not been studied in mouse crosses used to detect HDL QTL h QTL for gallbladder mucin content, an early step in gallstone formation After validation with both monogenic traits and a quantitative trait, the same strategy was applied to map quantitative trait loci for the control of plasma high-density lipoprotein cholesterol (HDL) and gallstone development. The average HDL values from 10-wk-old mice fed on a normal chow diet were taken from the Mouse Phenome Database ( Paigen et al. 2002 ; Bogue 2003 ). Because of the complexity of these traits, a conservative approach was used for strain selection. Data used for only 23 of the most related laboratory strains and two of the M. musculus domesticus strains because if a strain is from a unique lineage and contains unique haplotypes, it will not add any power to the analysis and risks increasing the level of noise (see Materials and Methods for a list of the 25 strains). Using a three-SNP window to analyze the 25 strains, there were no mirror loci present, and on average, 3.8 distinct inferred haplotypes were found at each locus. Where multiple loci may be expected to be found, as is the case for multigenic traits, a significance threshold was defined. To determine the false positive rate of each p -value, a recently described method by Dudoit et al. (2004) was used. The generalized family-wise error rate (gFWER) method uses a bootstrap estimation of the null distribution to assign a significance cutoff. In the case of the HDL phenotype, a significance threshold associated with a false positive rate of less than 0.005 ( p -value = 0.000506; −log [p] = 3.2958) was used ( Figure S2 ). Nineteen three-SNP windows were identified as having significant association with the HDL phenotype, which collapsed into 11 distinct loci ( Figure S3 ). To gauge the reliability of the in silico predictions, the results were compared to previously described QTL regions. Nine of these 11 loci fell within one of the regions identified by traditional two-strain crosses ( Table 1 ). Of the two that were not found to match previously identified QTL, the in silico MMUX QTL would not be expected to be matched because MMUX has been excluded from consideration in prior HDL QTL work. This same type of analysis was repeated for a phenotype that scored the formation of gallstones in 25 strains of male mice ( Paigen et al. 2000 ). Eleven regions were produced that exceeded the gFWER false positive cutoff ( p -value = 0.000398; −log [p] = 3.400117), and seven of these regions fell within the range of traditionally identified QTL for gallstone formation or mucin accumulation, which is considered a precursor to gallstone formation ( Table 1 ; D. Q. Wang et al. 1997 ). As well as identifying QTL, the inferred haplotype data from this SNP set also can be used to assist the narrowing of candidate regions, aiding in the selection of candidate genes. An association for a region overlapping an HDL QTL previously identified on MMU8 did not meet the stringent statistical cutoff set for the in silico method ( X. Wang and B. Paigen 2002 ). The most significant p -values obtained for the MMU8 QTL region were consistently found between 89–94 Mb. Sample sequencing of the region confirmed, at a slightly higher resolution, the SNP pattern that generated the association. This sequencing also replicated an inferred haplotype break point in the BTBR strain that narrowed the region to 88.52–90.88 Mb. A candidate gene within this 2-Mb region, adenylate cyclase 7 (Adcy7), located at 89.55 Mb, is expressed in the liver and adipose tissue ( http://symatlas.gnf.org ) and functions by producing cyclic adenosine monophosphate ( Watson et al. 1994 ). Cyclic adenosine monophosphate is known to be an important signaling component in the pathway to lipolysis ( Cammisotto and Bukowiecki 2002 ). Homologous regions containing the rat and human ortholog of Adcy7 have also been identified as containing an HDL QTL ( Bottger et al. 1996 ; Mahaney et al. 2003 ; Pajukanta et al. 2003 ). Adcy7 was sequenced in strains representing the three inferred haplotypes identified for this locus in the SNP dataset. Twenty-eight SNPs were identified in the gene, three of which produced amino acid changes. Nineteen of these SNPs, including the three nonsynonymous changes, were typed against all 48 strains of mice ( Figure 4 A). One of the haplotypes showed a higher average HDL level than all the others (77.5 mg/dl + 20.3 versus 67.2 mg/dl + 24.3 and 57.9 mg/dl + 16.7). This haplotype also contained a SNP causing a C717Y change in exon 20. Among the 48 strains, the members of this haplotype are the only ones with a replacement of this cysteine, which is conserved in the rat, cow, and human versions of the gene ( Figure 4 B), making it a good candidate for being a gene that contributes to the variability of HDL levels in the blood ( Abiola et al. 2003 ). Figure 4 Analysis of Adcy7 Haplotypes Reveals Amino Acid Change Associated with HDL Phenotype (A) Sequencing of Adcy7 in multiple strains revealed 28 SNPs distinguishing three distinct haplotype patterns. All strains were typed with markers selected to represent the three haplotypes. The strain distribution pattern predicted by the SNP data and the sample sequencing for this region was confirmed with NZB/BlNJ and BTBR T+ tf/J, I/LnJ and MA/MyJ, and C3H/HeJ, C57BL6/J, and C57L/J, each separating into unique haplotypes. (B) The SNP represented by marker 08.089.597 resulted in a change from a cysteine to a tyrosine in the resulting protein (asterisk). This cysteine is conserved in orthologs of the gene in human, rat, and cow. It is also found at the beginning of a stretch of ten amino acids (indicated by black line) predicted to be one of the protein's ten transmembrane domains. Identical amino acids are black and conserved amino acid changes are gray. Discussion The SNP data here provide new resources for traditional mapping projects and enable development of inbred strain haplotype methods for QTL detection. The analyses presented here indicate that the inferred haplotype structures derived from this dataset provide sufficient estimation of genetic diversity/similarity to map Mendelian traits to within 1-Mb intervals. QTL can also be defined as inferred haplotype loci of several megabases in size. The analysis for QTL provides a rank order of significant phenotype/genotype associations, and using the gFWER method of controlling for multiple-testing error, the loci reported as statistically significant are very likely to be biologically relevant. This point is borne out by the high concordance between the in silico QTL and the traditionally determined QTL ( Table 1 ). The traditionally determined HDL QTL identified in the mouse covers 42% of the genome and are in concordance with nine of ten in silico QTL—a significant result ( p < 0.0025). This excludes the MMUX in silico QTL since they cannot be verified from current traditional QTL data. The false positive cutoff employed here is very restrictive and could be relaxed to find additional real associations, but the chances of including false positives would then increase. For the gallstones phenotype the concordance is not demonstrated to the same level; however, the top-ranked loci still show overlap with previously defined QTL. What about the loci that do not show overlap; are they still real? From a statistical analysis it is unlikely they are false positives. In these results, 25 strains are simultaneously combined, unlike standard QTL mapping using two-strain comparisons, and some phenotype–genotype associations may occur that have not been observed by classical methods. Contributions from diverse strains that have not normally been used in F2 crosses or available in RI lines can now be incorporated. Even showing that this method does find significant associations, the question arises about its general utility and applicability. The methods of in silico mapping as described here should be viewed as a complement to, and not a substitute for, traditional methods for mapping QTL. Although we have demonstrated a robust approach to in silico mapping, it would certainly not be expected to find all QTL for a given phenotype. Major contributors to phenotypic variation will show up, but weaker contributors would be expected to be lost because of the limited power of 25 strains. It would also be expected to miss QTL resulting from recent strain-specific mutations or low-frequency haplotypes. Traditional QTL methods will still be required to identify the more subtle interactions, including those involving epistasis and modifying genes. However, these methods would provide a useful starting point for a new phenotype that is being investigated, where often the first step for any QTL analysis is a strain survey to quantify the range of the phenotype. Additionally, if these analyses are overlaid with the results from a traditional two-strain QTL mapping, one of the major advantages to be gained from this approach is that associative loci are defined in terms of a few megabases instead of tens of centiMorgans. The number and selection of strains and appropriate phenotype are also important considerations. Here we have limited our analysis of the complex traits, HDL and gallstones, to 25 strains—those that are best interrogated by this SNP set. While it is true that more strains have the potential to add greater statistical power to resolve QTL, this potential is limited by our ability to accurately represent the ancestral relationship of those additional strains. If we add more strains, but cannot accurately infer haplotype structures in those strains, we only add more noise to the analysis. The ability to detect all possible haplotypes in the utilized strains from the SNP data suffers from the availability of sequence data, currently from only four strains of mice. Because the source SNPs come from the sequencing of only four closely related strains, this current set is biased toward interrogating ancestry of M. m. domesticus. To be successful, phenotypes must have a low intrastrain variation but sufficient variance within the strain set selected. This however, is not a requirement restricted to in silico mapping. The overall power of this method will only improve as the biases and limitations of the SNP panel are addressed and additional strains are genotyped and phenotyped. Unique strains would become more useful if all possible SNPs are known and the mapping is then done directly with the causative polymorphism or at least with a large unbiased set of SNPs. As resequencing of other mouse genomes progresses, the ability to correctly infer the complete number and structure of haplotypes will improve, and the number of QTL regions reaching statistically significant levels will increase. Recently, two similar studies of haplotype structures across 5-Mb regions were published, although they produced differing conclusions on how their findings might affect in silico mapping efforts ( Frazer et al. 2004 ; Yalcin et al. 2004 ). Yalcin et al. has suggested that the complex nature of mouse haplotype structure and the small size of many haplotypes in inbred strains will make in silico mapping methods untenable and will preclude the mapping of any meaningful genotype–phenotype association short of whole genome resequencing ( Yalcin et al. 2004 ). This assessment would presumably hold true even for the well-defined Mendelian traits. The inferred haplotypes from a three-SNP window spanning on average 900 kb would not be able to reflect ancestral relationships, so the appropriate genotype–phenotype association could not be made no matter the strength of the allele in determining phenotype. Yet, clearly they can. The Frazer et al. (2004) study, which utilized more strains and produced significantly greater coverage of their 5-Mb region, estimates that the average ancestral segment length among classical inbred strains is in the order of 1.5 Mb in size, within the resolution of this work. In fact, the Yalcin et al. (2004) data show similar megabase-long ancestral relationships between strain pairs (for example, 5 Mb of near identity between A/J and C3H/J). This in silico approach concurs with the conclusions of Frazer et al. (2004) . Despite the complexities of haplotype structures, the use of a large enough set of strains with a dense SNP map does allow for significant and real associations to be found. This is not to suggest that fragmented small haplotypes are not common in the genome of the inbred mouse. This clearly does mean that there will be regions of the genome that will not be interrogated well by an in silico method. This approach is still limited by the density of this SNP map and can only be expected to visualize inferred haplotype patterns of approximately 1 Mb in size, and therefore smaller haplotype structures are hidden and potential phenotype–genotype associations will be missed. Here future, larger SNP sets that will allow more SNPs to infer haplotype will become important. However, this is the best resolved whole genome view of the diversity of the commonly used inbred strains to date. The algorithms employed here provide a starting point for further development of in silico mapping. We have shown that they can be used to identify Mendelian traits and replicate classical QTL associations. Clearly, the next goals are to validate some of the previously unreported associations, and this work is ongoing. Materials and Methods SNP selection and detection SNPs for use in genotyping were selected on a weighted basis from the Celera Mouse SNP Database containing data from the DBA/2J, A/J, C57BL/6J, 129S1/SvImJ, and 129X1/SvJ strains. Sufficient SNPs were selected for coverage of at least one SNP per 300 kb on average. The 129S1/SvImJ and 129X1/SvJ strains were considered as the same strain when their alleles agreed; preference was given first to SNPs where each allele of the SNP was represented by two strains. This was done to favor selection of SNPs representing ancestral inheritance, not recent strain-specific mutations, and to favor real SNPs as opposed to errors in sequence annotation. Additional selective value was incorporated based on whether the SNP was in a gene, how many sequencing runs supported the presence of the SNP, and the proximity of the SNP to previously selected SNPs. Additional SNPs used to characterize the Tas1r3 locus were gathered from sequence from multiple strains available in GenBank ( http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=nucleotide&cmd=search&term=tas1r3 ). All physical positions presented in the paper are from the Celera Mouse Genome Assembly R13. Primers for PCR and single-base extension were designed by using the SpectroDESIGNER software package (Sequenom, San Diego, California, United States). Assay designs are available as Supporting Information. All SNP assays were named for their position in the genome in the following format: the chromosomal location, the Mb position on the chromosome, and the kb position with a period separating each number. For SNP genotyping, genomic DNA from pedigreed mice (Mouse DNA Resources, The Jackson Laboratory, Bar Harbor, Maine, United States) was diluted to 10 ng/μl, and 1 μl of DNA was combined with 2.45 μl of water, 0.1μl of 25 mM dNTPs (Invitrogen, Carlsbad, California, United States), 0.03μl of 5 units/μl HotStar Taq (Qiagen, Valencia, California, United States), 0.625 μl of 10X HotStar PCR buffer containing 15 mM MgCl 2 , 0.5μl PCR primers mixed together at a concentration of 1.25 μM for multiplexed reactions, and 0.325 μl of 25 mM MgCl 2 . Reactions were heated at 95 °C for 15 min followed by 45 cycles at 95 °C for 20 s, 56 °C for 30 s, and 72 °C for 1 min and a final incubation at 72 °C for 3 min. After PCR amplification, remaining dNTPs were dephosphorylated by adding 1.5 μl of water, 0.17 μl of homogeneous mass extend reaction buffer (Sequenom), 0.3 units of shrimp alkaline phosphatase (Sequenom), and 0.03 μl of 10 units/μl exonuclease (USB Corporation, Cleveland, Ohio, United States). The reaction was placed at 37 °C for 20 min, and the enzyme was deactivated by incubating at 85 °C for 15 min. After shrimp alkaline phosphatase treatment, the genotyping reaction was combined with 0.76 μl of water, 0.2 μl of 10X Termination mix (Sequenom), 0.04 μl of 0.063 units/μl Thermosequenase (Sequenom), and 1μl of 10 mM extension primer. The MassEXTEND reaction was carried out at 94 °C for 2 min and then 99 cycles of 94 °C for 5 s, 52 °C for 5 s, and 72 °C for 5 s The reaction mix was desalted by adding 3 mg of a cationic resin, SpectroCLEAN (Sequenom), and resuspended in 30 μl of water. Completed genotyping reactions were spotted in nanoliter volumes onto a matrix arrayed into 384 elements on a silicon chip (Sequenom SpectroCHIP), and the allele-specific mass of the extension product was determined by matrix-assisted laser desorption ionization time-of-flight MS. Analysis of data was by automated allele calling from the SpectroTYPER software. All SNP data are available at NCBI dbSNP ( http://www.ncbi.nih.gov/entrez/query.fcgi?db=snp ) and The Jackson Laboratory Mouse Phenome Database ( http://www.jax.org/phenome ). Placement of the SNP data across the genome and major and minor allele distributions can be visualized using SNPview ( http://snp.gnf.org ). Statistical modeling for in silico mapping The use of a single marker is restrictive in the sense that it only allows a representation of the genome as diallelic. The use of windows of multiple markers enables the visualization of more complex genomic relationships between multiple strains. This more accurately models actual haplotype patterns than does a binary approach. In determining the size of the SNP window to use as a definition of inferred haplotype for purposes of the algorithm, sizes of two, three, four, and five SNPs were examined. A window of only two SNPs was still found to be too limiting. Windows of three, four, and five SNPs produced similar results, but as window size is increased biologically meaningful patterns become fragmented, creating more single-strain inferred haplotypes, resulting in an increase in noise. Singly represented haplotypes can never be informative in this analysis because the commonality of haplotypes is required to achieve significant association with a phenotype. Three SNP windows were also analyzed across the whole genome to identify mirror loci. This would be a locus that has exactly the same strain distribution pattern across all 25 strains used in an in silico run. There were no mirror loci, or 1-off, or 2-off mirror loci (with one or two strains not grouped identically) that occurred outside of a 5-Mb interval. Defining the genetic measure as a categorical unit necessitated the use of an ANOVA-based model. The type of ANOVA to use was determined by the characteristics of the phenotypic values. The phenotypes studied here fell into two categories: binary or continuous. The coat color phenotype was considered as binary, where phenotypic values were set to 1 and 0. The HDL phenotype is an example of a continuous phenotype since the phenotypic values are measured on a continuous scale. Two different statistical methods were employed based on this distinction. When phenotypic values are binary, the appropriate statistical approach involves first fitting a binomial generalized linear model to the data. An analysis of deviance table is then computed for the fitted model. The R language function glm with the parameter family set to binomial was used. This was followed by an application of anova.glm with the parameter test set to Chisq. For continuous phenotypic values, a log transformation was applied to reduce the effects of outliers in the phenotypic data. Next, an F-statistic weighted for the genotypic diversity of the strains within an inferred haplotype group was used. The weighted F-statistic had the following form: where and where n g is the number of strains in a given inferred haplotype, μ g is the mean of phenotypic values in a given inferred haplotype , μ T is the mean of all phenotypic values, k is the number of inferred haplotypes, N is the total number of data values, and w g is the weight representing the genetic diversity of the inferred haplotype. The genetic diversity ratio ( w g ) between two strains is the number of SNPs genome-wide in which both strains have genetic information and they disagree, divided by the total number of SNPs in which they both have genetic information. The genetic diversity coefficient for an inferred haplotype in the weighted F-statistic is the average w g between all strain pairs contained in the inferred haplotype. The weighted F-statistic calculated at each SNP window determines if at least one of the inferred haplotypes has an average phenotypic value significantly different from the other inferred haplotypes. To assess the significance of the computed value, the null distribution of the weighted F-statistic was simulated at each SNP window by taking a million bootstrap samples of the phenotypic values. As in the algorithm used for binary phenotypes, inferred haplotype patterns present in only one strain were not included in the calculation because they are not informative in elucidating shared ancestral blocks. From this distribution of a million random F-statistics, 200 bootstrap samples of size 1 million were computed. For each bootstrap sample, a p -value was computed by dividing the number of random F-statistics larger than the true F-statistic by the total number of random F-statistics (million). In this way 200 p -values were collected. The vertical heights reported in the bar graphs (see Figure S2 ) are the −log( p ) transform of the median of these 200 p -values. A 95% confidence interval (CI) for the p -value at this window was also calculated from this bootstrap distribution. To estimate the overall false positive rate for this type of calculation, calculating a significance threshold based on the family-wise error rate (FWER) has been proposed ( Churchill and Doerge 1994 ). Others have noted that the traditional FWER calculation is too strict in the context of multiple testing and leads to a significant loss of power ( Lander and Kruglyak 1995 ). Therefore, we employed a recently developed method of bootstrap estimation of common cutoffs based on the gFWER ( Dudoit et al. 2004 ). Whereas the FWER method reports significance, using the most conservative criterion of only one false positive, the gFWER method controls for multiple testing while allowing for an acceptable false positive rate (in our case, α < 0.005). The gFWER method to control for false positives as applied to in silico mapping is briefly described as follows. A null reference distribution was constructed using random bootstrap tests to determine a significance cutoff. Ten thousand bootstrap samples of phenotype values were randomly assigned to the true haplotype structure. For each random bootstrap sample, the nonparametric ANOVA approach outlined above was performed at each three-SNP window, with one difference. Whereas the initial true calculation reports the median of 200 bootstrap p -values, the gFWER method requires an estimate of the “supremum” (least upper bound) of expected values reported at each locus, so the most significant value is reported from the 200 bootstrap p -values (following Procedure 3 in Dudoit et al. 2004 ), ensuring a conservative false positive estimate. For each bootstrap sample, the genome-wide −log( p -value) corresponding to the (1 − α) percentile was added to the null distribution (as described in Procedure 5, Dudoit et al. 2004 ). Finally, after the 10,000 bootstraps are complete, the significance threshold is set as the (1 −α) percentile in the entire null reference distribution (computed from our 10,000 randomly bootstrapped iterations). While this threshold still represents a conservative estimate of the desired false positive rate, the gFWER has significantly more power than the traditional FWER calculation. Using this method for calculation of false positives, it is not necessary to specify the marginal distribution of the test statistic at each window of SNPs. Estimations of false positives or power that assume some parametric form of test statistic's distribution are not reliable in this context. This distribution can alter radically at each SNP window. In this context, the statistical problem of calculating quantities like discovery power (that is, ultimately the type I and type II error) is further complicated. Nearly 11,000 hypothesis tests (one at each three-SNP window) are conducted in a single run of the algorithm. Therefore, equations that currently exist for the estimation of power for QTL mapping by traditional methods cannot be applied here because they assume that the test statistic has some previously defined parametric form. Code for the above described algorithms is available upon request. For calculation of the significance of the number of in silico QTL that overlapped with previously identified QTL for the HDL phenotype, a binomial distribution was used given p = probability of success of 0.42 (overlap with HDL QTL in previous literature). Therefore q = probability of failure; the 0.0025 result is the probability of at least nine successes in ten trials. Only ten loci could be assessed for this result as no information is available for traditional HDL QTL present on the X chromosome. For the mapping of the retinal degeneration traits, 37 strains were used. This represented all of the strains for which information existed in The Jackson Laboratory database minus the most divergent wild-derived strains for which inference of haplotype would be expected to be most inaccurate. These strains were A/J, AKR/J, BALB/cByJ, BUB/BnJ, C3H/HeJ, C57BL/10J, C57BL/6J, C57BLKS/J, C57BR/cdJ, C57 l/J, C58/J, CBA/J, CE/J, DBA/1J, DBA/2J, FVB/NJ, I/LnJ, KK/HlJ, LG/J, LP/J, MA/MyJ, NOD/LtJ, NON/LtJ, NZB/BlNJ, NZW/LacJ, PERA/EiJ, PL/J, RIIIS/J, SEA/GnJ, SJL/J, SM/J, ST/bJ, SWR/J, WSB/EiJ, ZALENDE/EiJ, 129S1/SvImJ, and 129X1/SvJ. Because of the added complexity of the coat color traits, mapping was restricted to the 25 most related strains for which coat color phenotype could clearly be determined. For the albino analysis, 129S1/SvImJ, A/J, AKR/J, BALB/cByJ, C3H/HeJ, C57BL/10J, C57BL/6J, C57BLKS/J, C57BR/cdJ, C57 l/J, C58/J, CBA/J, DBA/1J, DBA/2J, I/LnJ, LP/J, MA/MyJ, NZB/BlNJ, NZW/LacJ, PERA/EiJ, PL/J, SEA/GnJ, SM/J, WSB/EiJ, and ZALENDE/EiJ strains were used. For the nonagouti mapping, the same strain set as the albino mapping was used except for the mice presenting the albino phenotype. The strains were 129S1/SvImJ, C3H/HeJ, C57BL/10J, C57BL/6J, C57BLKS/J, C57BR/cdJ, C57 l/J, C58/J, CBA/J, DBA/1J, DBA/2J, I/LnJ, LP/J, NZB/BlNJ, PERA/EiJ, SEA/GnJ, SM/J, WSB/EiJ, and ZALENDE/EiJ. Any mouse showing an agouti coat color was considered to be agouti for this analysis regardless of genotype at the agouti locus. Only limited phenotype data were available for saccharin preference, so again all strains with available data except the most divergent wild-derived strains for which inference of haplotype would be expected to be most inaccurate were used. These strains were A/J, AKR/J, BALB/cByJ, BUB/BnJ, C3H/HeJ, C57BL/6J, C57 L/J, CBA/J, CE/J, DBA/2J, FVB/NJ, I/LnJ, KK/HlJ, LP/J, NOD/LtJ, NZB/BlNJ, PL/J, RIIIS/J, SEA/GnJ, SJL/J, SM/J, ST/bJ, and SWR/J. For the mapping of the other complex traits, only the 25 strains with the closest ancestral relationship were used. These strains were 129S1/SvImJ, A/J, AKR/J, BALB/cByJ, BTBR T+ tf/J, C3H/HeJ, C57BL/10J, C57BL/6J, C57BLKS/J, C57BR/cdJ, C57 l/J, C58/J, CBA/J, DBA/1J, DBA/2J, I/LnJ, LP/J, MA/MyJ, NZB/BlNJ, NZW/LacJ, PERA/EiJ, PL/J, SEA/GnJ, SM/J, and WSB/EiJ. Supporting Information Dataset S1 Complete Allele Call and Assay List (16.1 MB XLS). Click here for additional data file. Figure S1 Phylogenic Tree of 48 Strains Generated from SNP Dataset Ancestral relationships between strains can be seen within clusters of the tree such as the fact that BALB/cByJ is a progenitor strain to SEA/GnJ. The bias of the SNP set can also be viewed by the exaggerated distance between the C57 and 129 clusters and the DBA and A/J cluster. The wild-derived strains make up the outermost cluster, but the three M. m. domesticus strains show a much closer relationship than the other wild-derived strains to the common laboratory strains. (2.2 MB EPS). Click here for additional data file. Figure S2 Duplicate In Silico Genome Scans for the HDL Phenotype The log p -value at each three-SNP window was calculated and plotted along the x-axis. Because any log p -value below 3 will not reach significance, calculations are halted at any locus once obtaining a log p -value of 3 becomes impossible in order to increase computational throughput. As such all log p -values below 3 are reported at 3. The false positive cutoff established by the gFWER calculation is indicated by a horizontal red line. Every quantitative trait was run twice through the algorithm to ensure consistency of results. (5.8 MB EPS). Click here for additional data file. Figure S3 Distribution of log p- Values from gFWER Calculation of Significance for HDL In Silico Analysis To estimate an appropriate false positive cutoff, 10,000 genome scans are conducted on randomized datasets and the 99.5 percentile log p -value is reported from each run. The significance cutoff is indicated by the vertical red line. (3.1 MB EPS). Click here for additional data file. Table S1 Frequency of Polymorphic Alleles between Strain Pairs (44 KB XLS). Click here for additional data file. Table S2 Total Number of SNP Alleles between Strain Pairs (34 KB XLS). Click here for additional data file. Accession Numbers The Mouse Phenome Database ( http://www.jax.org/phenome ) accession numbers for the phenomes discussed in this paper are MPD:29 and MPD:99.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526179.xml

535924

PASBio: predicate-argument structures for event extraction in molecular biology

Background The exploitation of information extraction (IE), a technology aiming to provide instances of structured representations from free-form text, has been rapidly growing within the molecular biology (MB) research community to keep track of the latest results reported in literature. IE systems have traditionally used shallow syntactic patterns for matching facts in sentences but such approaches appear inadequate to achieve high accuracy in MB event extraction due to complex sentence structure. A consensus in the IE community is emerging on the necessity for exploiting deeper knowledge structures such as through the relations between a verb and its arguments shown by predicate-argument structure (PAS). PAS is of interest as structures typically correspond to events of interest and their participating entities. For this to be realized within IE a key knowledge component is the definition of PAS frames. PAS frames for non-technical domains such as newswire are already being constructed in several projects such as PropBank, VerbNet, and FrameNet. Knowledge from PAS should enable more accurate applications in several areas where sentence understanding is required like machine translation and text summarization. In this article, we explore the need to adapt PAS for the MB domain and specify PAS frames to support IE, as well as outlining the major issues that require consideration in their construction. Results We introduce PASBio by extending a model based on PropBank to the MB domain. The hypothesis we explore is that PAS holds the key for understanding relationships describing the roles of genes and gene products in mediating their biological functions. We chose predicates describing gene expression, molecular interactions and signal transduction events with the aim of covering a number of research areas in MB. Analysis was performed on sentences containing a set of verbal predicates from MEDLINE and full text journals. Results confirm the necessity to analyze PAS specifically for MB domain. Conclusions At present PASBio contains the analyzed PAS of over 30 verbs, publicly available on the Internet for use in advanced applications. In the future we aim to expand the knowledge base to cover more verbs and the nominal form of each predicate.

Background We are now in an era where full genomes, data from high throughput experimental methods (e.g. micro-arrays) and electronic versions of scientific literature are easily available to every researcher over the Internet. These advances have made it possible to work on more than one gene at a time, ask complex questions and increase the pace of biological discovery. However, the progress made in scientific research until now has been recorded in the form of free-text articles readable only by humans and accessible by machine mostly through shallow keyword-based search engines. For improved methods of information access and knowledge discovery it is necessary to automatically map from the unstructured text representation into partially structured forms that provide discovered facts to databases. The large-scale data generated from the experiments in molecular biology needs to be assessed and integrated into the scientific communities' knowledge stores. This has created a need for various kinds of specialized databases. While some existing databases contain only molecular level information (e.g. PDB [ 1 ], SCOP [ 2 ]) others (e.g. BIND [ 3 ], SWISS-PROT [ 4 ], MINT [ 5 ]) contain literature associated with molecular entities. These literature databases contain a higher level of relationships (e.g. functional modules, interaction networks, gene products and disease phenotypes), are more informative and can be mined for further knowledge discovery (e.g. G2D [ 6 ]). At the same time hand curation of these databases is limiting their growth and reducing the accuracy of the information provided. This is where information extraction (IE) has an important role to play. Previous research in IE for biology has focused intensively on the recognition of named entities (NE) from scientific texts [ 7 - 9 ], i.e. the identification and classification of technical terms such as proteins, genes, drugs or cell types. Recently, the focus of research has been moving to higher levels of IE such as co-reference resolution and event extraction [ 10 - 18 ] which involves the filling of an event template that makes use of the results from NE recognition. However, significant challenges remain at all levels of biology IE due to the complexity of biological terminology and sentence structure. From the early days of research into computational linguistics it has been known that scientific sublanguages have special properties that make them different from general language [ 19 ]. These differences are notable at the level of vocabulary, semantic relationships and sometimes even syntax [ 20 ] and often require specialized knowledge sources to aid in analysis. In this article we focus on differences at the semantic and syntactic levels and we will provide motivating examples throughout the following discussion. Predicate-argument structure (PAS) analysis seeks to formally describe 'frames' for predicates (usually verbs) and the roles of their arguments (parts of the sentence surrounding it). Such roles usually need to be specified according to several factors including meaning and obligation. Meaning can be determined in several ways such as a domain or predicate-specific fashion such as catalyst and reaction being catalyzed in the case of the first and second arguments to the predicate catalyze . Alternatively, functional roles can be employed such as thematic relations that try to express some linguistically motivated aspect of the argument's behavior such as agent , location or experiencer . Traditional IE systems that use regular expressions based on shallow chunking at the phrase level (e.g. noun phrase, verb phrase, preposition phrase) capture weak notions of 'argument' for event predicates and their linear precedence. Such approaches seem to be inadequate to the goal of achieving high completeness and accuracy in event extraction. In recognition of this several major projects [ 21 - 24 ] have now begun based on newswire and balanced text collections which examine the relations that exist between the constituents in a sentence with the key assumption that those arguments correspond to major objects in events of interest. Although PAS frames seem to be expensive to construct by hand in terms of time and effort, particularly where this requires insights from domain specialists, we believe that this is justified as they provide a systematic reference guide for improving performance compared to ad-hoc pattern-building approaches. For PAS to be practically realized within IE three major knowledge components will be required: (1) a hierarchy of concept categories for objects of interest; (2) a definition of predicate-argument frames and the semantic labels of their arguments; and (3) the mapping rules that define how to transform the relevant parts of a surface sentence to the arguments in the PAS frame. Currently (1) is already quite advanced with several controlled vocabularies such as MeSH [ 25 ] or Gene Ontology [ 26 ] now in wide-scale use. At a more modest level core domain specific ontologies for individual annotation schemes such as the GENIA project [ 27 ] have also been proposed. To the best of our knowledge, however, nobody has yet made a proposal for (2) which will then serve as the basis on which to develop annotated resources for machine learning approaches to (3). This is the approach we intend to follow and this paper focuses on (2). It is of course possible to approach the task of PAS definition from a machine learning approach, and also to follow a path of hand-built heuristic mapping rules but we believe that both of these approaches may prove to be more costly in terms of time than the one we advocate here. In this work we introduce the concept of semantic analysis of argument roles in biological texts and propose the construction of PAS for molecular biology (PASBio). We have analyzed and annotated sentences from MEDLINE abstracts and full-text journal articles for building PASBio. The working scheme is similar to the PropBank project [ 22 , 23 ]. Results of our analysis are available online as a knowledge base of predicates and their respective argument sets at PASBio's web page [ 28 ]. By specifying PASBio we hope to enhance the event extraction system for accuracy (i.e. the ability to extract only relevant facts) by means of corpus-based semantic interpretation. To achieve this the intended IE system consists of 4 steps: (1) creation of a semantic lexicon (PASBio); (2) semantic annotation of texts using PASBio as a reference resource; (3) building an automatic semantic interpretation model using the annotated texts as a machine learning training corpus; (4) embedding this automatic semantic interpretation module into an IE system. This paper focuses on the key PASBio creation step by discussing the influential processes and choice points and a comparison to other schemes. The annotation task has been done on more than 300 sentences as the result of a preliminary analysis to support in defining PAS frames. This amount of annotation is unlikely to be sufficient for machine learning purposes, so further corpus annotation as well as the machine learning task needs to be carried out in order to reach the final step. It should be noted that other event extraction approaches [ 14 , 17 , 18 ] and also other text analysis applications (e.g. machine translation (MT), NE recognition tasks, text summarization [ 29 , 30 ]), requiring the use of semantic relations between a verb and its argument in their processing, would be able to take advantages of PASBio. In this article we first give a short introduction to IE and PAS. Next, we describe the approach taken in the PropBank project. Then, we discuss and exemplify how the specification of predicate-argument frames needs to be extended to meet the requirements for extracting molecular events. The second half of the paper is devoted to explaining the methodology used to define the PAS and discussing results of our analysis and its comparison with those of PropBank. Finally, we describe how the PAS frames can be exploited by showing their place in the IE system for molecular biology and discussing existing IE systems used for event extraction in molecular biology. Results and discussion Information extraction IE systems aim to provide instances of structured knowledge representations from unstructured free-form text. IE, based on the Message Understanding Conference (MUC) tradition of task segmentation [ 31 ] works fundamentally by using predefined frames and slots in agreement with a specific scenario describing user requirements. Such systems typically use regular expressions to match facts for the event to be extracted in each sentence. Each logical form is founded upon the syntactic relationship between components in each sentence. To take an example from the newswire domain: if we wanted to extract facts relating to a scenario ( company outlook ) then patterns such as "np (stock index) + vp (driven up) + integer (number %)" and "np (company) + vp (bid) + np (stock)" could be developed as a template. Sentences in documents which (1) contain a noun phrase (np) describing stock index , together with a verb phrase (vp) driven up , and followed by a number ; or (2) contain a noun phrase representing a company name , followed by a verb phrase with bid , plus a noun phrase of stock index should be extracted. The difficulties are compounded because a single event can nearly always be written in a variety of syntactic forms due to linguistic processes such as passive voice, (pro-) nominalization, raising, etc. The following simple example involves a linguistic phenomenon sometimes called locative alternation or spray alternation by Levin [ 32 ]. The verb spray may express its arguments in at least two different ways, i.e. (a) " Peter sprayed water on his flowers. " and (b) " Peter sprayed his flowers with water. " Thus, two syntax-based regular expressions plus some information about NE as "np (people) + vp (spray) + np (object1) + pp (on) + np (object2)" and "np (people) + vp (spray) + np (object2) + pp (on) + np (object1)" are required. Surface level extraction patterns can be hand built [ 33 ] or based on machine learning (ML) from a sample of annotated text (a corpus) [ 10 ] or from a few patterns which are known to be good indicators of the topic of interest (seed patterns) [ 34 , 35 ] to reduce the cost and time in constructing patterns manually. However, to extract the relations between objects in the complex sentences that frequently occur in technical and scientific texts requires deeper semantic knowledge. Reported systems [ 15 - 18 ] generally use a set of rules relevant to syntactic roles (e.g. subject, object, and modifier) obtained from parsers together with surface level patterns to extract the interactions between genes or gene products from the biological literature. Although extending the systems with syntactic roles or syntactic functions can achieve better performance compared to the pure pattern-matching approach, some errors resulting from a lack of semantic understanding still remain. For example, [ 15 ] mentions that their system will incorrectly extract a protein interaction between " Msp1p " and " Dec1p " from a sentence " These findings suggest that Msp1p is a component of the secretary vesicle docking complex whose function is closely associated with that of Dec1p . ", because it conforms to the pattern "A associate with B" predefined within the system. In this respect we consider that deeper knowledge, describing the semantic relationship between verbs and their arguments, encoded in PAS are needed. Predicate-argument structures An event is described in a sentence by a composition of a verb and its arguments. A verb, which indicates a particular type of event conveyed by a sentence, can exist in its verbal form, its participial modifier format or its nominal form. For example, the normal form of a verb used to describe the event "making something active" would be activate , its participial modifier format would be activating or activated , and its nominal format would be activation . Beyond a verb, sentence constituents holding semantic roles to complete the meaning of an event indicated by the verb are called arguments. The semantic roles played by the set of arguments with respect to the particular verb are represented in the PAS frame of that verb. Recently several major projects have been proposed that provide resources in the form of an English predicate-argument lexicon. These projects include VerbNet [ 24 ], FrameNet [ 21 ], and PropBank [ 22 , 23 ]. There are significant differences in approach among these 3 projects. For example, PAS of verbs sell and rent are proposed as two distinct structures in the case of PropBank and only a single structure for both verbs in the case of VerbNet and FrameNet (Figure 1 ). VerbNet defines general PAS for a group of verbs that share similar syntactic behavior, underlying Levin's alternations theory [ 32 ]. VerbNet's PAS for give contains sell and rent as members. Argument roles for all of the give verb members are assigned for agent , theme , and recipient illustrated by example sentences 1 and 2. In the case of FrameNet, PAS is defined based on the underlying principal of what users or applications expect to see for a specific event definition. FrameNet's PAS for event Commerce_sell shown in Figure 1 expects only argument seller and goods from the event driven by any verb in a set of verb members. Considering the annotation on sentence 1 in these 3 projects, "All Brownstein" is annotated as seller , agent , and seller in PropBank, VerbNet, and FrameNet respectively. Similarly, there is also an argument to support the annotation of "it" in all projects. But, only the PropBank scheme has an argument labeled price paid to support element "$60 a bottle" of sentence 1 which is likely to be an important participant of the event describing a selling activity. Moreover, a constituent "a week" in sentence 2 is considered to be an argument labeled as term only by the PropBank scheme. We consider that arguments like price paid for the events involving the verb sell , and an argument term for events involving the verb rent , are important for down-stream user applications. In contrast to VerbNet and FrameNet, PropBank defines individual verb-specific PAS frames which are likely to contain more detailed specifications of arguments than are possible for verb groupings. Moreover, PAS construction in a more verb-specific manner than either VerbNet or FrameNet would assist explicitly in discovering rules for mapping from surface syntactic structures to underlying semantic propositions. Hence, we utilize PropBank's scheme as a basic starting point and examined sentences containing interesting verbs from a variety of molecular biology journal articles such as MEDLINE abstract [ 36 ] and full-text journal articles as EMBO [ 37 ], PNAS [ 38 ], NAR [ 39 ] and JV [ 40 ]. The verbs were analyzed and compared to frames proposed by PropBank, which were created based on an analysis of the Wall Street Journal corpus. At least one PAS frame per verb was defined. The verbs were chosen based on both their frequency in the articles and also based on their importance in a number of major event types such as gene expression, molecular interactions and signal transduction. In PropBank a verb may get more than one PAS frame if the verb sense and its argument set differ, reflecting the fundamental assumption that syntactic frames are directly related to the underlying semantics. For example, PropBank defines three distinctive PAS frames (Figure 2 ) for the verb run on account of sense variation. Each structure contains its own set of arguments labeled with semantic roles. A semantic role of an argument represents a semantic relationship between the argument and its related verb. It is possible that in any particular sentence a complete set of semantic roles or a set of arguments for each sense will not all occur together. The example sentence in Figure 2(a) illustrates this point i.e. only Arg0 and Arg1 occur in this sentence without the occurrence of Arg2 , Arg3 , and Arg4 though all arguments are defined as core arguments of the PAS. In each PAS, arguments are labeled ranging from Arg0 up to Arg5 with a mnemonic label indicating its predicate-dependent role. Besides these core arguments defined in PAS are adjuncts which are traditionally not defined in PAS because they can potentially take multiple values and not required to minimally define the event. PropBank does consider adjuncts when annotating sentences, and provides labels such as ArgM plus tags such as TMP for temporal information, LOC for locative information, PRP for a reason or motivation, etc. Covering the full working details of PropBank is out of the scope of this paper and we refer interested readers to [ 22 , 23 ] for more information. After manually defining PAS, PropBank has annotated the Penn TreeBank II Wall Street Journal corpus, which contains constituency and dependency information from the TreeBank project [ 41 ]. Events in molecular biology According to the Gene Ontology (GO) [ 42 ], the term biological process refers to a broad category of biological tasks accomplished via one or more ordered assemblies of molecular entities (gene products). It often involves transformation, in the sense that something goes into a process and something different comes out of it. Examples of biological processes are cell growth and maintenance, signal transduction, metabolism and biosynthesis etc. A biological process can be subdivided into temporal and spatial molecular events. Each molecular event is carried out by a gene product or well-defined assemblies of them. For example, phosphorylation of a protein molecule by a protein kinase is a molecular event, which is a part of the cellular signalling process or transcription of a gene by a polymerase is a part of the gene expression process. Hence, by definition a molecular event or a disruption of it will have a local effect in terms of the process that it is a part of and an observable or phenotypic effect in terms of overall effect of disruption of the entire process. For example, a mutation in the coding region of a gene that introduces a stop codon into the open reading frame would lead to a pre-mature termination of transcription considered as the local effect and may be responsible for a disease state of an organism due to deficiency of that protein as the phenotypic effect. Different events are described by different verbs (Figure 3 ) using its associated sets of arguments. Need for semantic relationships in molecular event extraction As we exemplified previously for the newswire domain, similar issues of syntactic variants will inevitably be encountered in scientific domains. The following examples from our analysis (Figure 4 ) illustrate these points. The sentences (1)–(3) in Figure 4 show some different instances of the event eliminate taken from our corpus of MEDLINE [ 36 ] and EMBO [ 37 ] Journal articles. Here, we consider 3 different pieces of information to be extracted, i.e. A – causal agent of the event, B – the entity being removed, C – location at molecular (sequence) or cellular level where the entity is being removed. In Figure 4 , sentence (1) shows simple indicative form of which the syntactic-based extraction pattern would be "A eliminates B in C" (where A = One mutation , B = the BamHI site and C = exon7 ); sentence (2) shows the passive form, without mention of A and C, for which a syntactic-based extraction pattern would be "B are eliminated" (where B = all three sites ); sentence (3) shows a form, using a different preposition compared to sentence (1) in order to mention C, for which the syntactic-based extraction pattern would be "A would eliminate B within C" (where A = a 3-bp in-frame deletion , B = an asparagines residue and C = a kinase domain of the product ). Examples of sentences describing the event express are shown as sentences (4)–(6). Information slots consist of A – entity expressed, B – physical property of the expressed entity, and C – location referring to organelle, cell or tissue. In sentence (4), (where A = the enzyme , B = two mRNA isoforms of 2.4 and 4.0 kb , C = brain ) the information needed to describe the event with respect to slot B is marked by using a prepositional phrase, but in sentence (5), (where A = two equally abundant mRNAs for il8ra , B = 2.0 and 2.4 kilobases in length , C = neutrophils ) using an appositive form, seemingly not playing an important role in the description of the event in which it participates. Sentence (6), (where A = RNA and protein for all four transgenic TCR proteins and C = T cells , without mentioning B) shows a different kind of problem that arises because biologists generally would not think of "T cells" as an agent in this context, perceiving it as information about location. On the other hand, without deep domain knowledge one may understand "T cells" as an agent of the express event instead of its intended role as a cell or tissue. These examples show that using regular expressions around syntactic information of the surface texts would not be adequate for IE to make sense of the complex surface structure. PAS represents information describing verb arguments and the semantic roles these arguments play in conveying a certain event. Different surface forms describing the same event can be mapped into the same PAS. To illustrate this point we return to the example mentioned earlier, (a) " Peter sprayed water on his flowers ." and (b) " Peter sprayed his flowers with water ." Both sentences can be mapped into the PAS of a verb spray , which indicates the particular event "apply thin liquid to surface" with 3 required arguments (i.e. agent, liquid, surface). The sentence's constituent " Peter " in both sentences is perceived from its verb-specific semantic role to be an agent that does the action. " water ", when it is either a direct object as in sentence (a) or an object of a preposition as in (b), is perceived as the liquid used in the event, and " his flowers " is perceived as the surface getting wet. Similarly, a surface text from molecular biological corpus such as " One exon is spliced out of the MLC3 nm transcript in smooth muscle to give an alternative product. " could be conceptualized into PAS relationship as shown at the topmost level in Figure 5 . Figure 5 illustrates understanding a sentence from the surface text level up to the PAS level. The sentence's constituents " One exon ", " is spliced out ", " of the MLC3 nm transcript ", " in smooth muscle ", and " to give alternative product " have their syntactic categories as noun phrase , verb , prepositional phrase , prepositional phrase , and verb phrase respectively. At the syntactic relations level, " One exon " shows its role as the surface subject of the passive form verb " is spliced out " and other constituents play the role of complements . Beyond the syntactic level of description, there are semantic levels including argument categories level and predicate-argument relations level. At the argument categories level " One exon ", " the MLC3 nm transcript ", " smooth muscle " and " alternative product " constituents pertain to the domain concept classes of a gene product (RNA) , tissue and alternative mRNA respectively. At the highest level of our scheme the representation contains the most abstract information. Semantic roles played by other constituents to the verb indicating the event are represented at this level. Thus, the process of removal of an exon from mRNA (alternative splicing) is indicated by the verb splice out . Here, the verb arguments play the semantic roles of lost component ("One exon"), entity getting spliced ("the MLC3 nm transcript"), location referring to tissue ("smooth muscle"), and secondary predication – showing purpose or reason in this example ("to give an alternative product"). The semantic role secondary predication is assigned to the argument " to give an alternative product " because this by itself is capable of instantiating a PAS frame and is considered in our scheme to possibly be a core argument. The semantics of a sentence relate in complex ways to the syntax of the sentence, as we can see from the illustration of semantic and syntactic levels in Figure 5 . Using this layered approach different surface forms describing the same event can be mapped into the same PAS. Thus, PAS could be helpful for IE to overcome the syntactic variation problem. After we describe the PAS frames constructed for molecular biology (PASBio), we provide an explanation about how to apply this knowledge in PASBio for event extraction. Defining predicate-argument structures for molecular biology In molecular biology, a gene and its products are at the center of the study, as a set of these molecular entities dictate, and their products carry out, different functions at the cellular level and the combined effects can be seen at the organism level. Hence, in the literature a gene or a gene product is possibly described as an agent participating in some events, with the help of various appropriate verbs indicating the specific events. Different molecular-level or phenotypic effects are described as the other arguments of such events. As described above, PAS is a representation of semantic relationships between arguments with specified roles and a verb relating to a particular event narrated in a sentence. Thus, PAS would be a natural choice for IE, especially event extraction in molecular biology. Guidelines to define PAS We use PropBank's scheme (with necessary adaptations) to define PAS for the molecular biology domain. To define PAS for any verb, a survey about the usages of the verb from a set of sample sentences in a representative corpus is made. Examining the usage of an individual verb will indicate if it needs to be divided into several senses. In PASBio, these senses are divided with the aim of obtaining fine-grained semantic senses using the WordNet [ 43 ] lexical database. Each of PASBio's PAS contains a set of core arguments. A core argument is an argument shown by its usage to be important to complete the meaning of the event. Nevertheless, if an argument is considered important but there is no evidence to show that the argument exists together with the predicate in at least 20% of our selected sentences, this predicate may not be assigned as a core argument. There are two different types of core argument: the first type plays a role during the main event denoted by the predicate while the second type plays a role after the main event and aims to express results or consequences of the main event. Further details are given in the next section (Figure 6 -Frame 1) illustrated with the PAS for mutate . Arg X (with X , a cardinal number, starting from 0 and then incremented for each additional argument) is used for labeling the first type of core argument and ArgR is used for the second type. A mnemonic label is added after Arg X and ArgR in order to give a short description of the semantic role played by the argument. Biological function and usage of the argument are used to describe the semantic role in PAS. No attempt is made to ensure the consistency of mapping between argument labels (argument name) and the roles (the mnemonic labels) played by the arguments across verb frames, except Arg0 . Arg0 is reserved for only the argument playing the semantic role of agent . In some cases, this agent argument is not found in the usage of some verbs. Thus, PAS frames of such verbs will contain the core argument from Arg1 . See PAS frames for mutate (Figure 6 -Frame 1), express (Figure 9 ) and transform.02 (Figure 10 -Frame 9) as examples. In addition to annotating a sentence's constituents corresponding to core-arguments with the tag Arg X or ArgR , the sentence's constituents which do not play the role of core arguments but fall into three types, i.e. adverbial, negation and modality, are annotated with the tag ADV or MAN in the case of an adverbial, NEG in the case of negation, and MOD in the case of modality. At the current stage of this project, only adverbials in terms of adverbs are considered to be annotated as MAN (for a manner adverb) or ADV (for other types of adverbs). If any adverbials in terms of phrases or clauses are mandatory for expressing events indicated by particular predicates, these adverbials will be defined as core arguments within PAS frames. For example, an adverbial phrase playing the role of locative modifier is included in the set of core arguments in the frame for predicate initiate . (Refer to example sentence "Apparently HeLa cells either initiate transcription at multiple sites within RPS14 exon 1 ."). Moreover, we are interested in distinguishing only the adverb playing the roles of manner modifiers (e.g. normally , genetically , etc.) from other adverbs. A manner adverb deserves special distinction from other adverb types because it shows how a certain action is performed which is very important to understand facts in a biological sentence. For example, " normally " in the sentence "Mice have previously been shown to develop normally " is necessary for IE in order to understand that there is no problem in the development of the mice. Other types of adverbs for example play the roles of aspectual modifiers that give information about whether some event or state of affairs is completed or is still going on, and so forth (e.g. " still " in the sentence "Wanda still would like to talk about the music festival."), adverbs playing roles as frequency modifiers that indicate the frequency of a certain type of event (e.g. "always" in the sentence "One always hears rumors."), adverbs playing roles as focusing modifiers that consist of the four words even , only , also , and too (e.g. "The transcription is initiated only in female blastoderm embryos."), and so on will be all tagged as ADV . In case of negation and modality, NEG and MOD are given directly to a negator word (i.e. not or n't) and a modal verb (i.e. will, may, can, shall, must, might, should, could and would) respectively. Though negations (operating at the sentence level) and modality (operating at various levels) are not defined as core arguments (mandatory arguments) within any PASBio's PAS frames because linguistically both of them cannot even be considered as any types of predicate's arguments, they are all worth annotating from an IE perspective if they exist in the same clause where a focused predicate exists. Similarly, adverbials which are not mandatory enough to be core arguments are also considered worthy of being annotated when found in the text. We consider that they should not be ignored because they can significantly alter or even reverse the meaning of the sentence. Examples of defined PAS In this subsection, we show some examples of PASBio's PAS frames and discuss how each frame is defined by examples of sentences relevant to it. There are three important cases that we examine in comparison to PropBank: (1) verbs that are rarely used in general language (e.g. splice ) or have a unique biological interpretation (e.g. express , translate , etc.), (2) verbs that have a similar meaning used in the newswire domain and biology domain but show different patterns of usage (e.g. alter , initiate , etc.), and (3) verbs that are used with the same meaning and usage style in both domains (e.g. abolish , delete , etc.). The usage of different verbs in biology influence PAS for biological domain falls into four groups: A – same sense, more arguments; B – same sense, fewer arguments; C – same sense, same structure; D – different sense or does not occur. Table 1 shows some verbs for each group. We give PAS of two verbs as examples of each group. Group A Verbs in this group have been used in biology documents with the same semantic sense as in PropBank, but they required more core arguments in their structures. Consider the event of mutation, one of the most important biological events and a general cause behind genetic diseases. The verb mutate is used to describe the changes in an entity (gene or gene product) and mutations can be natural or engineered. PropBank defines two arguments for this verb which are Arg0: agent and Arg1: entity undergoing mutation , but from analysis we propose four arguments for the PAS frame of the verb mutate . As mentioned in the Guidelines section, Arg0 is reserved only for the argument playing the semantic role of agent. From all the examples we observed, passive forms are used to describe mutate events which mean that the agent does exist in the event but it is unnecessary to be explicitly stated because it is commonly known by the domain experts. This results in PASBio's core arguments for mutate starting from Arg1 and we leave a position for agent which possibly could be mentioned in other biological sub-domains. The PASBio's Arg2 describing event participating entities (referred to as 'Name Entities') is analogous to PropBank's Arg1 . Thus PASBio's Arg1 , Arg3 , and ArgR are extra arguments compared to PropBank. The arguments Arg1 and Arg3 are captured conforming to linguistic criterion [ 44 ] which considers that a sentence element which plays a particular role to a predicate will be considered to be a core argument in a PAS frame even though it does not exist in every sentence in which the predicate appears. In sentences where such an element is omitted we infer that it is implied by the existence of the predicate. For example, in the sentence "John is eating" we infer the existence of a core argument of eat which denotes a type of food. Similarly, Figure 6 -Frame 1 shows that Arg1 and Arg3 do not exist in all sentences 1.1 to 1.3, but are assigned as core arguments by their intuitive presence in the domain models of biologists. Noticeably, consequences of the event driven by verb mutate are often seen in examples. Apart from "changes at molecular level" assigned as Arg3 , the consequence, "changes at phenotype level" is suggested as ArgR (explained below). Sentence 1.1, 1.2, and 1.3 support this explanation. The argument ArgR:results/consequences is an argument giving information about consequences after the event denoted by the predicate occurs. For mutate , most of the example sentences describing this event contain an ArgR argument, revealing the necessity of it. The requirement of this argument from an observation perspective coincides with biologist's viewpoint, thus we consider this as a core argument (more precisely an IE core argument) and named as ArgR instead of Arg X (a core argument from a purely linguistic perspective). We make this distinction under the rule that Arg X has to play a role during the event but not after the event. This condition is depicted by a formula like "mutation event = ( Arg X + mutation + Arg X ) + ArgR ". Empirically, we find that this result argument ( ArgR ) is used with verbs relating to an abnormal biological phenomenon. Examples of other verbs that need this argument are skip , delete , etc. Verb initiate also takes additional arguments as core arguments. As shown in Figure 6 -Frame 2, Arg2 (sentences 2.1 and 2.2) describes the point of transcription initiation and Arg3 provides information about the tissue/cell where the gene (or product) is expressed. In PropBank, the sentence's segments defined by the parser with functional tag as LOC (location) will be considered as non-required elements. However, the extraction of spatial information is very important from the perspective of biological description. Furthermore, another interesting point that can be seen from the examples in Figure 6 -Frame 2 is that authors in biology not only put the agent but also various other kinds of semantic roles in the subject position. In Sentence 2.1 " HeLa cells " is syntactically the subject which seems to be the agent of an initiate event, but domain knowledge suggests that the agent can be only a protein (usually polymerases bound to the gene being transcribed) in this case. " HeLa cells " is annotated as Arg3:location as tissue or cell instead of Arg0:agent . In sentence 2.2, " I kappa B-epsilon translation " is also a subject as in the previous example, but it is "entity created" assigned as Arg1 . Only in Sentence 2.3 (describing initiation of signaling event), the subject of the sentence fills the semantic role "agent", so a subject " RTKs " can be annotated as Arg0 . Additionally, the point to note is "the entity created" in sentence 2.3 is different from sentence 2.1 and 2.2 as it is a signaling event that is initiated, but not a transcription or translation event. Group B Verbs in this group have been used in biological texts with the same semantic sense as in PropBank, but they required fewer arguments in their structures in our PAS Verb block both in biomedical texts and in business news texts has very similar semantics. However, an event described by verb block in the biomedical domain may not mention information about secondary predication and instrument most of the time. The semantic role secondary predication is assigned to the argument that is in itself capable of instantiating another PAS frame. The sentence " [ John Arg0 ] blocked [ Mary Arg1 ] from [ completing her dissertation Arg2 ] with [ his constant pestering Arg3 ]." is annotated by PropBank's PAS frame. An argument Arg2-secondary predication is annotated for "completing her dissertation" because this contains in itself the PAS of the verb complete . From this PropBank example, the meaning of the event denoted by block cannot be completely understood if the sentence just states as " [ John Arg0 ] blocked [ Mary Arg1 ]." as it is necessary to mention the action being stopped. In contrast in the biology domain, by mentioning only the entity being stopped (Sentence 3.1–3.3), the expert reader can understand that the event which applies to that entity is being stopped without providing an explanation of the action being stopped at the position of secondary predication. Similarly, an instrument used to block is encoded in the nature of an agent or causer. The structure of block and its examples are given in Figure 7 -Frame 3. Only core arguments as defined in the structure exist in Sentence 3.1 and 3.2 (the agent is not mentioned). In sentence 3.3, MAN is used to label "specifically" as this adverb plays the role of a manner modifier. In Figure 7 -Frame 4 the PAS frame of generate is similar to that of block . Only Arg0-agent and Arg1-entity created are expressed in all observed sentences from our biology corpus. Group C Verbs in this group have been used in biological documents with the same semantic sense as in PropBank. Moreover, their usage in both the biology corpus and PropBank indicates that their PAS frames are identical. Specialization of domain does not seem to affect verbs in this group. In Figure 8 , Frame 5 and Frame 6 show PAS for confer and lead . In both biology and newswire corpora, confer is used with semantic "to give (as a property or characteristic) to someone or something" and lead to is used in the sense of "to tend toward or have a result". Group D Verbs in this group have been used in biology documents with a different semantic sense compared to PropBank, or PAS frames for them are not found in PropBank. More than one semantic sense is found in our corpus for some verbs. PAS frames for express and transform are presented in Figures 9 , 10 , respectively to illustrate predicate-argument structures for this group. The verb express is used in the biology domain with the meaning "to manifest the existence of a gene or gene product" (or detection of the same by the experimenter) unlike its normal usage with the meaning of "give an opinion or send quickly". The PAS of express is given as Figure 9 . In the case of transform , two senses are used in biology papers: "to cause (a cell) to undergo genetic (or neoplasmic) transformation" as shown in Figure 10 -Frame 8 and "to transfer a gene from source organism into target organism" as shown in Figure 10 -Frame 9. Even though the first meaning of transform found in our corpus is similar to the sense of "change" found by PropBank, there is still a huge gap between them. In the biological literature, illustrated by examples in sentences 8.1–8.3, this genetic transformation mentions only the agent or causer, what entity is getting transformed, and what will be the effect after transformation. It will not mention the start state of the entity undergoing transformation because it is known from the expert reader's domain 'common sense' knowledge that the start state refers to a normal condition of that entity. Transform in the second sense always occurs in a sentence connected by preposition into , and in the passive voice form in which no mention is made with regard to the agent. Complexities in biology texts In the discussion so far we have assumed that the predicate is the center of semantic information. Here we intend to show that the argument contents can change the event description specified by the predicate, by examining sentences that describe an 'alternative splicing' event. Alternative splicing is used to generate multiple transcripts from a single gene and hence is a helpful event for increasing the functional complexity of eukaryotic systems. Consider the following example of a set of sentences that talk about the 'expression' of a single type of mature mRNA generated from 'splicing' of pre-mRNA and generation (and expression) of multiple mature mRNA transcripts with different properties from the single type of pre-mRNA. Sentences annotated follow PASBio's frame for express : (a) " Northern blot analysis with mRNA from eight different human tissues demonstrated that [ the enzyme Arg1 ] was expressed exclusively [ in brain Arg3 ], [ with two mRNA isoforms of 2.4 and 4.0 kb Arg2 ]." and (b) "[ A complementary DNA clone Arg1 ] encoding the large subunit of the essential mammalian pre-messenger RNA splicing component 2 snRNP auxiliary factor(U2AF65) has been isolated and expressed [ in vitro Arg3 ]." Sentence (a) is considered as a sentence denoting the alternative splicing event but sentence (b) is considered as a negative (not describing alternative splicing) sentence, which talks about expression of an mRNA of a splicing factor. It would be difficult, based on word contents or regular expression methods, to put these two examples into different 'bins' for alternative splicing events. But the discussion about the length of the two different transcripts in Arg2 (with two mRNA isoforms of 2.4 and 4.0 kb) in the first sentence can be helpful to understand it as a sentence discussing about alternative splicing. On the other hand, the later sentence contains all the interesting words (e.g., mRNA, express and splicing) but misses Arg2, hence describes just an expression event. Utilization of PASBio Construction of PAS frames by expert introspection may be considered as a time-consuming process, however domain-specific PAS frame definitions have valuable uses in several applications as discussed below. Each PAS frame in PASBio provides a set of semantic relationships between arguments participating in an event and a verb conveying the event. Although we focus on applying PASBio for event extraction in the molecular biology domain, information processing applications that require semantic understanding of a sentence will be able to take advantage of this knowledge. For example, machine translation (MT) requires encoding a surface sentence of a source language into a language independent logical form of clause meaning, and then generating from this logical representation a surface sentence in a target language. PAS would be one practical choice to be used as such a logical representation in MT [ 29 , 30 ]. In the case of a text summarization application, PAS frames could naturally be employed as the basic unit of a discourse representation, before being summarized [ 45 ]. PASBio is available online for the wider research community in the molecular biology domain for exploitation in such applications. With respect to our molecular event extraction system, as we discussed in the introduction, PASBio takes on the role of a reference source in the stage of corpus annotation for creating training examples for machine learning. The planned IE system is composed of 4 activities: (1) construction of PASBio semantic lexicon; (2) annotation of full-text journal in terms of semantic represented in PASBio's frames; (3) construction of the module for automatically transforming an unseen sentence into a logical form of semantic relationships drawn within each particular PASBio frame; (4) integration of the resultant automatic semantic interpretation module within the event extraction system. So far, manual annotation and machine learning have not been completed yet and will be reported elsewhere. For a description of an IE system that can make use of such an annotated corpus we refer readers to the work of for example Surdeanu et al. [ 46 ] who uses PAS defined for the newswire domain to extract market change events. Apart from our corpus-based semantic interpretation approach, several other research groups have proposed systems for event extraction from the biological literature, especially about the interaction information between genes and genes product. Related work so far can be summarized into two sets. The first set of methods use regular expressions and rely on syntactic patterns. These methods may use statistical models of the surface words [ 12 , 13 ], rules of the sentence elements' precedence order [ 11 ], shallow knowledge like part of speech tags, syntactic roles of constituents [ 15 , 16 ], gene/protein name dictionaries and domain knowledge (e.g. a template slots for the particular event) about the events they intend to extract [ 17 , 18 ]. A template used in this research group consists of only a simple set of slots for a simple predicate (i.e. the predicate relating only two arguments: subject and object) and only a shallow notion of the predicate-argument structure has been considered (i.e. consider one argument as subject and another as object, but not consider as arguments' semantic roles). The only work in the second set, that has taken into account a large number of linguistic and deeper semantic aspects is, that of Novichkova et al. [ 14 ]. The approach described in Novichkova et al., is to construct a biology IE system (MedScan) containing two components: an NLP engine deducing the semantic structure of a sentence, and a configurable information extraction component to validate and interpret results produced by the NLP engine, in order to achieve a flexible and efficient IE system. In one of its steps, the authors propose to transform the syntactic tree of a whole sentence into a normalized semantic tree, which represents the logical relationships between the words in a sentence. To carry out the transformation, a set of semantic frames describing predicate-argument structures, are required. However, the MedScan system's semantic interpretation process is still under development and not precisely specified. As mentioned above, most of the approaches, whether a deep notion of predicate-argument relations is taken [ 14 ] or a shallow notion [ 17 , 18 ], do require a reference resource of PAS frame for each predicate. In this respect, we believe that PASBio's description of PAS frame for each predicate would make a useful complement to other approaches. Recently, another research group [ 47 ] reported the aim of annotating a biological corpus with semantic knowledge in the form of PAS. While this work appears to be at an early stage it again shows the importance of the definition of predicate-argument frames and the semantics of their arguments as a key knowledge for IE in the molecular biology domain. Conclusions With the explosion of molecular data, tools developed by computer scientists are gradually being applied and integrated in the domain of biology to aid in information access and knowledge discovery. Text data appearing as reports about biological discoveries demands automated IE methods for faster knowledge discovery. Traditional IE systems that use regular expressions based on shallow chunking at the phrase level (e.g. noun phrase, verb phrase, preposition phrase etc.) capture weak notions of 'argument' for event predicates and their linear precedence. Such approaches seem to be inadequate to the goal of achieving high accuracy in event extraction in molecular biology. PAS which is used as a representation of the semantic relationship between a verb and its arguments participating in the event has the potential to support deep knowledge acquisition from a sentence within the extended system framework that is now being proposed within the IE community. Due to the importance of PAS and the lack of a specific PAS frame resource for the domain of molecular biology, we have proposed the analysis of PAS for molecular biology in this article. We have analyzed sentences for 30 verbs (and different frames per senses of the verb) from MEDLINE abstracts and full-text journal articles where the sentences contain each verb in its verbal form and its participial modified form for building PASBio. Our analysis suggests in some cases a significant difference in the predicate frames compared to those obtained from analyzing news articles by the PropBank project. In addition to the significance of verb senses used in the molecular biology domain, syntactic constructions also differ markedly; such as the use of passives allowing the semantic subject to be omitted where they are part of the common-sense understanding in the domain. Human readers are required to have domain knowledge in order to understand the texts. Our result frames and examples are available to the wider research community as a knowledge base at PASBio's webpage. In the future, we intend to utilize knowledge from the PASBio frames for annotating a corpus to be used as training examples to achieve automatic annotation of PAS semantics into sentences. Furthermore, we aim to complete analyzing PAS for more verbs related to molecular events and afterwards to extend our analysis to sentences containing the nominal forms of verbs. Methods Selection of verbs The English language used in research articles of biological and biomedical sciences is a sublanguage of written natural language. While most of its vocabulary is similar to that of general English, some verbs are domain-specific in nature. Our main focus here is the verbs that are used for describing molecular events in biology. Various researchers have different areas of interest and new concepts are added in the literature continuously. However, the areas of cellular signaling, gene expression, regulation and disruption of expression events are very important for the larger community of investigators involved in basic biomedical research and those involved in high throughput analysis. They are discussed throughout different parts of papers as possible cause of normal and disease states of different organisms. Hence, ignoring the normal distribution (frequency) of different verbs in the literature we choose the verbs from those involved in the above-mentioned processes (events). Most of the verbs are shown in Figure 3 . Selection of example sentences Information extraction work is still largely carried out using PubMed abstracts. Using abstracts is advantageous because they contain the highest density of keywords compared to other section of research articles but our intuition is that bio-text mining should scale-up to analyze full journal articles where the most detailed results are contained along with supporting evidence, comparisons to others work and background information, etc. [ 48 ] Recent investigations have shown that Introduction and Discussion sections apart from paper abstracts may be viewed as interesting sources of important biological information [ 49 ]. We thus define our PAS by analysis on sentences from MEDLINE [ 36 ] and from all other sections except the Method section on EMBO [ 37 ]. Furthermore, we inspect the usage of some verbs in other journals such as PNAS [ 38 ], NAR [ 39 ] and JV [ 40 ] in order to achieve usage agreement and good PAS. Sentences from the Method section are not used in this analysis as they are limited in terms of biomedical information, have generic written styles and verb sense usage tend to overlap with general language. Sentences were carefully chosen to cover a broad usage of each verb under study from the MEDLINE and full text journal corpora as described before. We tried to choose equal numbers of sentences containing a particular verb in its verbal format and its participial modifier format. Before starting an analysis on each sentence, a sentence was parsed using Connexor Parser [ 50 ] that uses Functional dependency Grammar (FDG), to give parse tree, word, lemma, syntactic function and dependency links between words in order to help in determining the boundary of each argument exists in a sentence. This parse tree served as a useful guide in hand analysis, but was not considered by any means as a gold standard. At least 10 sentences were selected to determine PAS of the verb under study. The use of the parser considerably reduces the manual labors involved in defining arguments. Authors' contributions This work was directed by NC. TW carried out the analysis of the predicate-argument structures with technical support from NC and biological knowledge from PKS. PKS chose the predicates and the sentences analyzed from the MedLine corpus. Sentences from other corpuses were complemented by TW. TW prepared the figures (except fig 3 by PKS). All authors contributed during the whole length of the project and writing of the paper. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535924.xml

539059

The Structure of a Rigorously Conserved RNA Element within the SARS Virus Genome

We have solved the three-dimensional crystal structure of the stem-loop II motif (s2m) RNA element of the SARS virus genome to 2.7-Å resolution. SARS and related coronaviruses and astroviruses all possess a motif at the 3′ end of their RNA genomes, called the s2m, whose pathogenic importance is inferred from its rigorous sequence conservation in an otherwise rapidly mutable RNA genome. We find that this extreme conservation is clearly explained by the requirement to form a highly structured RNA whose unique tertiary structure includes a sharp 90° kink of the helix axis and several novel longer-range tertiary interactions. The tertiary base interactions create a tunnel that runs perpendicular to the main helical axis whose interior is negatively charged and binds two magnesium ions. These unusual features likely form interaction surfaces with conserved host cell components or other reactive sites required for virus function. Based on its conservation in viral pathogen genomes and its absence in the human genome, we suggest that these unusual structural features in the s2m RNA element are attractive targets for the design of anti-viral therapeutic agents. Structural genomics has sought to deduce protein function based on three-dimensional homology. Here we have extended this approach to RNA by proposing potential functions for a rigorously conserved set of RNA tertiary structural interactions that occur within the SARS RNA genome itself. Based on tertiary structural comparisons, we propose the s2m RNA binds one or more proteins possessing an oligomer-binding-like fold, and we suggest a possible mechanism for SARS viral RNA hijacking of host protein synthesis, both based upon observed s2m RNA macromolecular mimicry of a relevant ribosomal RNA fold.

Introduction The virus that causes SARS, like other pathogenic coronaviruses and astroviruses, possesses a linear plus-sense strand RNA genome that has a 5′ methylated cap and 3′ poly-A tail. The viral replicase is translated directly from the genomic sense-strand RNA, and it then creates a full-length complementary (minus-sense strand) copy of the genomic RNA, as well as a nested set of shorter, subgenomic mRNAs having common 3′ UTRs. These 3′ UTRs all share with the genomic SARS RNA a 32-nucleotide element, immediately upstream of the 3′ poly-A tail (residues 29,590–29,621) [ 1 ], originally termed the stem-loop II motif (s2m) in human astroviruses [ 2 ]. The s2m element is the most highly conserved RNA element within the coronaviruses and astroviruses that contain it ( Figure 1 ). Figure 1 The Primary, Secondary, and Tertiary Structures of the SARS s2m RNA (A) Phylogenetic comparisons of s2m RNA sequences from various coronavirus and astrovirus species. The SARS RNA sequence is color-coded to match the color scheme used throughout. Conserved sequences are highlighted as bold letters, and co-varying sequences involved in conventional RNA helical base-pairing are indicated in italics. Sequence complements are indicated using color-coded brackets. (B) The 2.7-Å experimental SIRAS platinum-phased and solvent-flattened electron density map contoured at 1.25 root mean square deviation. The map allowed unambiguous tracing of the RNA molecule because the density was unambiguous for all backbone atoms and all nucleotide bases except U(25), U(30), and U(48). (C) A corresponding ribbon diagram highlighting the unusual fold. (D) Schematic representation of the s2m RNA secondary structure, with tertiary structural interactions indicated as long-range contacts. The schematic diagram is designed to approximate the representation of the fold. The GNRA-like pentaloop structure is shown in yellow, A-form RNA helices are shown in blue and purple, the three-purine asymmetric bulge is in red, and the seven-nucleotide bubble is in green. Long-range tertiary contacts are indicated by thin red and yellow lines. Standard structural genomics analyses focus upon obtaining the three-dimensional structures of proteins encoded within a genome, and on identifying unknown protein function based on three-dimensional homology to protein structures of known function [ 3 ]. However, it is also imperative to identify and to elucidate the three-dimensional structures of non-protein gene products, including the various RNAs required for mRNA processing, protein synthesis, and other cellular functions [ 4 ]. In the case of viruses that possess an RNA genome, including such pathogens as HIV and SARS, it becomes critical to expand the scope of structural genomics analyses even further to include biologically relevant RNA tertiary interactions that occur within the RNA genome itself. Those genomic RNA elements having the greatest degree of conservation are the most likely to be crucial to the evolution, growth, and replication of these viruses, and therefore demand the most attention from those seeking to understand RNA viral pathogenesis and to design appropriate anti-viral drugs. Using X-ray crystallography, we have solved the three-dimensional structure of the SARS virus s2m RNA to 2.7-Å resolution. The structure reveals a dramatic 90° bend and several additional novel tertiary interactions. Although the sequence and three-dimensional structure of the s2m RNA are both unique, comparison of the global fold of the SARS s2m RNA to known RNA tertiary structures reveals that the backbone fold of the s2m RNA mimics that of the 530 loop of 16S rRNA, permitting us to hypothesize that the biological function of s2m in SARS and related viruses is based upon macromolecular mimicry of this region of ribosomal RNA. The ribosomal RNA 530 loop and the proteins that bind to it are involved in translational initiation, suggesting that the role of the s2m in SARS may also involve translation initiation. Specifically, we propose, based on structural homology arguments, that the SARS s2m RNA might bind to the host's eukaryotic translation initiation factor 1A (eIF-1A) to hijack the host's translational machinery for use by the virus, or to bind other translational regulation proteins having similar folds for similar purposes. Results Sequence Analysis of the Conserved s2m Element We aligned the most recent available genomic sequences of coronaviruses and astroviruses and analyzed conservation patterns within the s2m element ( Figure 1 ). Remarkably, about 75% of this sequence is absolutely invariant between viral species (nucleotides shown in boldface in Figure 1 A) and much of the variation that does occur preserves secondary structural elements (nucleotides shown in italics in Figure 1 A). In addition, we analyzed 38 sequenced SARS variants and found that the motif is absolutely conserved within all of them. No insertions or deletions appear to be tolerated, indicating that this region forms a highly conserved RNA tertiary structure that is universally required for viral function [ 1 , 2 , 5 ]. The Crystal Structure of the s2m RNA Element of SARS Using in vitro transcription, we prepared and crystallized a 48-nucleotide construct containing the 45-nucleotide s2m element. We solved the crystal structure to 2.7-Å resolution using a single platinum isomorphous/anomalous derivative and obtained a readily interpretable solvent-flattened electron density map ( Figures 1 B– 1 D and 2 A). The quality of the electron density enabled us to fit the s2m RNA sequence unambiguously to the map and to build a model of the unusual tertiary structure. The initial map was virtually indistinguishable from the final 3Fo–2Fc map calculated using phases from the refined RNA structure, indicating that the single isomorphous replacement with anomalous scattering (SIRAS) experimental phases initially obtained were quite accurate ( Tables 1 and 2 ). Two well-ordered hydrated Mg 2+ complexes bound to the phosphate backbone of the RNA are also readily observable in the initial electron density map ( Figure 2 B). Figure 2 Stereo Representations of the SARS s2m RNA Structure (A) The overall SARS s2m RNA three-dimensional structure and (B) a detailed view of tertiary contacts the and [Mg(H 2 O) 5 ] 2+ binding sites in the context of the experimentally phased electron density map (dark blue). The [Mg(H 2 O) 5 ] 2+ complex ions, depicted as white octahedra, bind to the pro -R and pro -S phosphate oxygen atoms of A(12). An extensive network of potential hydrogen bonds between the metal-coordinated water molecules and the RNA is shown as yellow dotted lines. Table 1 Crystallographic Data Collection All X ray intensity data to 2.7 Å were processed without imposing a cutoff, and all amplitude data for which F ≥ 0.0 were used for model refinement and electron density map calculations Table 2 Phasing and Refinement Data for which F ≤ 2σ, for which isomorphous differences were greater than five times the root mean square isomorphous difference, or for which anomalous differences were greater than 3.5 times the root mean square anomalous differences were excluded from the initial phase estimation only. The crystallographic spacegroup is P6 5 22 and the cell dimensions are a = b = 93.2184 Å and c = 128.109 Å. There is one RNA molecule per asymmetric unit, consistent with a 73% solvent content. The number of non-hydrogen atoms in the refined structure is 1,037, the number of Mg[(H 2 O) 5 ] 2+ complex ions is two, and one well-ordered water molecule that interacts with the metal complexes was explicitly modeled rms, root mean square The crystal structure of the s2m domain of the SARS RNA reveals several novel tertiary structural elements ( Figure 3 ). Three regions of canonical A-form RNA are indicated in various shades of blue, and three regions of unusual structure, including tertiary interactions, are represented in green, red, and yellow. The actual three-dimensional fold of the RNA is illustrated in Figure 1 C, with Figure 1 D designed to represent this fold schematically as well as the secondary and tertiary structural contacts that stabilize it. Figure 2 A shows a corresponding stereo diagram in which all non-hydrogen atoms are present. Figure 3 Tertiary Structural Interactions in the SARS s2m RNA (A) Close-up of the pentaloop structure together with the augmenting helix, shown in yellow, and the perpendicular junction formed with the A-form stem, shown in cyan. The pink hydrogen bonds indicate base-quartet hydrogen bonding, as shown in (B). The 90° kink thus formed is facilitated by a very sharp bend in the backbone involving unpaired residues 29 and 30. (B) Formation of the junction of two perpendicular helices is facilitated by a base quartet composed of two G–C pairs. (C) The unusual pairing between A(17) and G(34) facilitates formation of a long-range tertiary contact between A(33) of the three-purine asymmetric bulge and G(11) and A(12) of the seven-nucleotide asymmetric bubble. A(38) forms a base triple with C(39) and G(13), forcing G(11) and A(12) out of the main helix. (D) Space-filling representation of the region shown in (C), but rotated approximately 180°. A tunnel is created by the tertiary contacts between A(33) of the purine asymmetric bulge (red), G(11) and A(12) of the seven-nucleotide bubble (green), and the helical region between them (purple). The non-bridging phosphate oxygens of G(11) and A(12) line the surface of the cavity, creating a negatively charged region into which Mg 2+ ions are observed to bind. The Fold of the s2m RNA, the Pentaloop, and a Nucleotide Quartet The overall structure of the s2m SARS RNA consists of two regions that are defined by two perpendicular RNA helix axes (see Figures 1 and 2 ). The larger region contains several non-helical motifs involved in long-range tertiary contacts (see Figure 3 ). The smaller region (residues 20–30, shown in yellow in Figure 3 ) forms a stem-loop structure in which a pentaloop (residues 22–26) is structured similar to a conventional GNRA tetraloop motif but has an extra residue (U[ 25 ]) bulged out of the stack formed by A(23), G(24), A(26), and the augmenting helical stem (residues 20–21 and 27–28). This is similar to what is observed in a spliceosomal stem-loop structure [ 6 ]. The base of U(25) is disordered in the structure, and little side-chain density is apparent in an otherwise well-defined electron density map. Residues 29 and 30 are unpaired and are involved in forming a rather severe backbone reversal that accompanies the 90° kink in the helix axis. The phylogenetic comparisons shown in Figure 1 A reveal that the pentaloop sequence is highly conserved. Although the structure of the pentaloop is very similar to the standard GNRA tetraloop structure [ 7 , 8 ], the “extra” U(25) insertion between R and A is always present. The unusual perpendicular helical junction is stabilized by the formation of an RNA base quartet involving two adjacent G–C pairs wherein the G(19)/C(31) pair shares four hydrogen bonds with the C(20)/G(28) pair (shown as pink dotted lines in Figure 3 A and 3 B). The RNA sequences required to preserve these G–C pair interactions are present in all but one of the viral sequences analyzed (avian nephritis virus), implying that the base quartet serves a significant structural role in SARS and most related viruses. All previously characterized RNA base quartets are purine tetrads [ 9 , 10 , 11 , 12 , 13 , 14 , 15 ] and do not occur within double-helical structures; the G–C quartet thus appears to be another novel structural feature present within the s2m element of SARS and related viruses. A Three-Purine Asymmetric Bulge An asymmetric bulge in the s2m SARS RNA secondary structure containing A(17), A(33), and G(34) (highlighted in red in Figure 3 C) is absolutely conserved in SARS and all other related viruses analyzed (as shown in Figure 1 ). A(17) pairs with G(34), involving the Watson–Crick base-pairing faces of both purines. This mode of interaction is rather distinct from the more usual “sheared” G–A pairings involving the Hoogsteen faces of these purines, and has the effect of significantly widening the RNA helix from the standard A-form geometry. As a consequence, A(33) is able to adopt a very unusual conformation in which it becomes completely excluded from the helical stack, and instead forms long-range tertiary interactions with G(11) and A(12). G(34), in addition to forming a Watson–Crick-like base pair with A(17), hydrogen bonds to C(18) as well as to G(21), thereby stabilizing the unusual pentaloop-stem conformation and 90° helical kink. A Seven-Nucleotide Asymmetric Bubble Interacts with the Purine Bulge The remaining non-canonically base-paired region of secondary structure (residues 10–13 and 38–40), highlighted in green in Figure 3 C, contains mostly conserved nucleotides including an absolutely conserved pair between C(10) and A(40), and a Watson–Crick pair within an otherwise highly distorted helical region between conserved residues G(13) and C(39). A base triple forms between A(38) and this G–C pair, a variant of the adenosine platform motif [ 16 ], and consequently G(11) and A(12) are rotated out of the helical structure completely. A(33) forms long-range tertiary interactions with G(11) and A(12) by hydrogen bonding to the N3 of G(11) and the ribose of A(12). Substitutions at position 12 are thus tolerated, as is a single instance of purine substitution at position 11 (which will preserve the N3 hydrogen-bonding interaction with A[ 12 ]). Together, these interactions superficially resemble those observed in domain IV of 4.5S RNA of the signal recognition particle [ 17 , 18 ], but the structural details are completely different. G(11), A(12), and A(33), despite their extrusion from the helical base-pair stack, form a well-defined structure that is highly ordered, judging by electron density in the initial map as well as the comparatively low temperature factors these residues have in the refined structure. They conspire with the remaining residues in the asymmetric bubble and the helical region above it to form a rather wide tunnel whose channel runs approximately perpendicular to the main helical axis. The phosphates of G(11) and A(12) are turned inward, creating a negatively charged environment within the tunnel cavity. Consequently, the tunnel forms a binding site for two [Mg(H 2 O) 6 ] 2+ ions in the native structure (see Figure 2 B), and the tunnel is also the binding site for cis -[(NH 3 ) 2 Cl 2 Pt(IV)] 2+ and [Ru(NH 3 ) 6 ] 3+ metal complexes that were introduced for heavy atom isomorphous replacement phasing. These highly structured and rigorously conserved features allow us to suggest that SARS pathogenesis might be inhibited by a drug designed to bind to s2m and disrupt one of these structures. Chemical Probing of the Solution Structure To compare the crystal structure with the solution structure of s2m, we performed chemical modification experiments. The results are consistent with the crystal structure, and in some cases enable us to verify that long-range tertiary interactions observed in the crystal structure also occur in solution. Dimethyl sulfate (DMS) modification patterns ( Figure 4 A) of the N1 atomic position of A and the N3 of C residues are consistent with the observed fold in the crystal structure ( Figure 4 B). A and C residues that are solvent-exposed in the tertiary structure, such as A(12), A(23), and C(27), are among the most heavily modified by DMS (along with A[ 44 ] and A[ 45 ] near the helical terminus). These modification sites are shown as red spheres in Figure 4 B. Although A(33) is quite exposed in the tertiary structure, the N1 is protected from modification by DMS (shown as a green sphere in Figure 4 B), consistent with the involvement of the N1 of A(33) in a 2.8-Å hydrogen bond with the exocyclic N2 of G(11) (white atom and dotted line in Figure 4 B) in the crystal structure. We therefore conclude that this tertiary structural interaction observed in crystals of s2m RNA is likely to be quite similar to what occurs in solution. G(11) is the only G residue of the s2m RNA detectably modified by kethoxal (data not shown), which reacts with nitrogens at the N1 and N2 positions. The N1 modification site is highlighted as an orange sphere and is consistent with the observed tertiary structure formed by G(11), A(12), and A(33) that exposes G(11) to the solvent. U(30) is solvent-exposed in the crystal structure and is reactive to 1-cyclohexyl-3-(2-morpholinoethyl) carbodiimide metho-p-toluene sulfonate (CMCT; magenta spheres in Figure 4 B; data not shown), as are the non-conserved 3′-terminal uridines (probably due to helix fraying in solution). U(25), which is not well ordered in the crystal but which we expect is also solvent-exposed, appears not to be reactive. Figure 4 Chemical Probing of the SARS s2m RNA in Solution (A) An autoradiogram of DMS modification of the s2m RNA in solution. (B) Mapping the results of DMS, kethoxal, and CMCT modifications onto a stereo representation of the RNA structure. Red spheres represent strongly reactive N1 positions of adenosines and N3 positions of cytidine residues in the presence of DMS, and yellow spheres represent weaker reaction. Green spheres represent positions that appear to be protected from DMS. The orange sphere represents reaction with kethoxal at the N1 position of G(11), and magenta spheres represent CMCT reactions with uridines. (C) The most extensive crystal packing interaction involves stacking of G(11) upon its symmetry mate, G(11)′. (D) Temperature factors mapped onto all non-hydrogen atoms (left) and the phosphate backbone (right) of the s2m RNA crystal structure. U(25) is the most disordered residue in the structure and has the highest temperature factor. Density of the base of U(25) is not apparent even after refinement. Most of the rest of the structure is rather well ordered. Discussion The intricate three-dimensional structure of the SARS s2m RNA, along with its rigorous sequence conservation, is compelling prima facie evidence for its biological importance in coronaviruses and astroviruses. The structure by itself, however, does not indicate what the function of this motif must be. Hence, comparison of this unique fold with those of known RNA structures is of particular value for formulating testable hypotheses regarding potential biological functions of the s2m RNA. In addition, identification of novel and rigorously conserved tertiary structures that are unique to the viral RNA is of critical importance for future rational design of anti-viral therapeutic agents that specifically target SARS and other coronaviruses and astroviruses. Biological Relevance of the s2m Sequence and Crystal Structure The s2m RNA sequence we crystallized was originally identified from the genomic sense strand within a rigorously conserved region of the 3′ UTR of the RNA. However, because RNA replication and transcription take place via a full-length negative-strand RNA intermediate, it is formally possible that the conserved sequence instead corresponds to a conserved structure at the 5′ end of the anti-sense RNA. We believe this to be improbable because of the energetically unfavorable tertiary structures that would be required to form from the sequence complement. For example, the variant of the energetically stable and rather common GNRA loop structure (GAGUA) would have to be replaced with an energetically unstable and rare CUCAU loop. Similar arguments apply to the other non-Watson–Crick regions of the structure. Crystal packing interactions may potentially distort RNA structures. This effect is sometimes observed for small stem-loop sequences, which often crystallize as duplex dimers rather than as monomeric hairpins. The s2m RNA structure is sufficiently large, and apparently contains enough stabilizing secondary and tertiary interactions, to offset any energetic advantage that might come from crystallizing as a duplex. In addition, the 73% solvent content of the s2m RNA crystals ensures that most of the crystallized RNA is solvent-exposed, rather than involved in extensive packing interactions. At least three inter-molecular contacts are required to form a crystal. The most extensive contact is the base of residue G(11); it stacks upon that of its 2-fold symmetry mate ( Figure 4 C). It is likely that these nucleotide bases become oriented in such a way as to optimize this stacking interaction. The nonessential nucleotide G(1) forms a weak (3.4-Å) hydrogen-bonding interaction with A(29) of an adjacent molecule, but most of this packing interaction appears to be due to shape complementarity and is thus expected to have little distorting effect. The remaining interaction is a nonspecific, presumably cation-mediated backbone parallel helical interaction, again unlikely to result in significant distortions. Crystallographic temperature factors provide direct physical evidence for the relative flexibility or mobility of various regions of a macromolecule. Figure 4 D shows relative temperature factors color-coded on all non-hydrogen atoms (left) and on the RNA phosphate backbone atoms (right). Blue atoms have the lowest relative temperature factors and red atoms have the highest. Consistent with the observed electron density map, by far the most flexible region of the RNA is U(25). U(30) and the 5′-terminal triphosphate are also moderately disordered. Much of the rest of the structure appears to be rather rigid and well defined, including the three-purine asymmetric bulge and the seven-nucleotide asymmetric bubble, along with the hydrated magnesium complex ions that bind to the non-bridging phosphate oxygens of A(12). The phosphate backbone atoms of these non-Watson–Crick regions are among the most ordered in the structure. Therefore, based on our chemical probing data, analysis of crystal packing interactions, and consideration of the crystallographic temperature factors, along with the ability to rationalize the sequence conservation pattern and intolerance for nucleotide insertions or deletions based on the structure, we conclude that the crystal structure of s2m is likely to be a close representation of the structure that forms in solution and in the context of the SARS virus RNA genome. Functional Implications of the s2m Three-Dimensional Structure The several unique features and unanticipated tertiary contacts we identified in the SARS s2m RNA crystal structure allowed us to reexamine genomic sequences and previously determined RNA tertiary structures for similar motifs with additional constraints imposed by knowledge of the tertiary structure. Our analysis of the human genome, other animal and viral genomes, and the currently available database of RNA three-dimensional structures revealed that the s2m element is found only in astroviruses and coronaviruses; no cellular homologs are immediately apparent. The G(11) to A(33) tertiary contact in the s2m RNA is homologous to the G(1,452) to A(1,486) contact in Domain III of the 23S ribosomal RNA, but the context of the interaction in the ribosome is completely different, and the sequence is not conserved between Escherichia coli and Thermus thermophilus . However, if we relax the sequence constraints and focus attention upon the conformation of the RNA backbone, we find that the phosphodiester backbone fold accompanying the 90° kink in s2m RNA mimics that found in the 530 stem-loop of 16S ribosomal RNA [ 19 ] ( Figure 5 A). The latter binds to the S12 protein found at the interface between the small and large ribosomal subunits. The 530 stem-loop, and the S12 protein that binds to it, have been implicated in EF–G-independent ribosomal translocation [ 20 ]. Remarkably, superposition of the s2m RNA upon the 530 stem-loop within the 30S ribosome in which prokaryotic initiation factor 1 (IF-1) has been added [ 21 ] reveals plausible modes of s2m RNA binding to both the S12 protein and to IF-1 ( Figure 5 B). Both S12 and IF-1 have eukaryotic homologs; the structure of IF-1 and its eukaryotic analog, eIF-1A, possess almost identical RNA oligomer binding (OB) folds [ 22 , 23 ]. Based upon these structural homology arguments, we propose that the SARS s2m RNA is a functional macromolecular mimic of the 530 loop of the small subunit ribosomal RNA (which is conserved in eukaryotes). Mechanisms of translation and protein synthesis regulation via macromolecular mimicry are in fact well established [ 24 , 25 ]. We propose, on the basis of the similarity between the 530-loop fold and the s2m fold, that the s2m RNA of SARS may be capable of binding one or more eukaryotic proteins whose structures resemble S12 or the OB folds typical of these ribosomal proteins, and that each would do so in a manner similar to that shown in Figure 5 B. This proposal leads us to formulate two separate, testable hypotheses regarding the function of the s2m RNA in SARS. Figure 5 SARS Virus RNA Macromolecular Mimicry (A) The SARS s2m RNA structure (red) is superimposed upon the 530 loop of 16S rRNA (cyan), revealing the similar stem-loop folds. (B) The IF-1 (magenta) and S12 protein (blue) that bind to the 16S rRNA 530 loop (now hidden) are shown relative to the same s2m RNA superposition, suggesting that their eukaryotic homologs might plausibly bind to the s2m RNA. Does s2m Macromolecular Mimicry Facilitate Viral Hijacking of Protein Synthesis? eIF-1A, like IF-1, possesses an OB fold. Our first hypothesis is that eIF-1A may bind to the 90° bend of the SARS s2m RNA. In addition, we suggest that the function of the s2m RNA of SARS and related viruses might involve viral hijacking [ 26 ] of the cell's protein synthesis machinery, either facilitating mRNA circularization and ribosome re-initiation, in gross analogy to viral internal ribosomal entry site–mediated mechanisms [ 27 , 28 ], or perhaps even more simply by titrating eIF-1A away from the host initiation complexes and thus inhibiting host cell protein synthesis in favor of viral protein synthesis by sequestering a factor required by the host. Does s2m Bind to the nsp9 SARS Protein to Facilitate Virus Transcription? Recently, two protein structural genomics investigations of SARS revealed the structure of a so-called nonstructural protein, nsp9, that is believed to be involved in viral RNA synthesis and to interact with the viral polymerase in an unspecified manner [ 29 , 30 , 31 ]. The crystal structure of nsp9 reveals it to be a variant of the OB fold, a protein structural motif not previously recognized to be involved in viral replication. The authors demonstrate nonspecific single-strand RNA binding affinity for nsp9. We propose that nsp9, by virtue of its OB fold, may bind specifically to s2m in a manner similar to that illustrated in Figure 5 B, and may thus facilitate viral polymerase RNA transcription, translation, or replication. From Structure to Functional Predictions Our structural genomics analysis of the SARS RNA has thus enabled us to formulate specific, experimentally testable hypotheses regarding the function of a highly conserved RNA motif whose importance has been evident [ 2 ] but whose biological activity hitherto was completely unknown. The possibility that the 90° bend of the s2m RNA binds to an OB-like protein permits us to propose two potential mechanisms of interaction relevant to the two main functions of the SARS virus (protein synthesis and viral replication). The possibility of additional interactions with proteins at the S12-like site and in the highly structured and rigorously conserved tunnel region formed by the three-purine bulge and the seven-nucleotide bubble should also not be overlooked, as these both are likely sites for RNA–protein or RNA–RNA interactions that are crucial to the function of the SARS virus, and therefore also merit further attention. The s2m RNA Tunnel Is an Attractive Target for the Design of Anti-SARS Drugs Figure 3 C and 3 D dramatically illustrates the most striking and unique structural feature within the SARS s2m RNA. A tunnel is created by the tertiary contacts between A(33) of the purine asymmetric bulge (red), G(11) and A(12) of the seven-nucleotide bubble (green), and the helical region between them (purple). The non-bridging phosphate oxygens of G(11) and A(12) line the surface of the cavity, creating a negatively charged region into which Mg 2+ ions are observed to bind. It is likely that in the context of the virus, this invariant feature of the s2m structure is involved in binding interactions with highly conserved proteins or other components of the host cell that interact specifically with the negatively charged cavity. Because this tunnel structure is unique to coronaviruses and astroviruses and because the sequence comprising this structure is invariant, it is reasonable to propose that by designing a drug that specifically targets this structural feature and binds tightly to it, an anti-SARS therapeutic might be obtained that avoids the pitfall of being toxic to uninfected host cells while escaping the usual problem of drug resistance that develops in rapidly mutating RNA viruses. Materials and Methods Crystals of a 48-nucleotide T7 RNA transcript containing the conserved s2m RNA element were obtained via hanging-drop vapor diffusion by equilibrating a solution containing equal volumes of the RNA sample and the reservoir solution against 1-ml of the reservoir solution. The RNA sample solution contained 4.5 mg/ml s2m RNA dissolved in 30 mM Tris (pH 7.6), 100 mM NaCl, and 60 mM MgCl 2 . The reservoir solution contained 50 mM MES (pH 5.6), 100 mM Mg(OAc) 2 , and 20% MPD. Data from a native crystal diffracting to 2.7-Å resolution, and 3.0-Å cis -(NH 3 ) 2 (Cl) 2 Pt(IV)–derivative single-wavelength anomalous dispersion data, were collected at Beamline 9.1 at Stanford Synchrotron Radiation Laboratory on a 3 × 3 CCD detector using 0.98-Å wavelength X rays and crystals that were cryoprotected in the reservoir solution spiked with 12% glycerol and maintained at 100 K. The native and platinum derivative data were processed using CCP4's MOSFLM and reduced and scaled within CCP4 version 5.0 [ 32 , 33 ]. A single platinum heavy atom site was found in both isomorphous- and anomalous-differences Patterson-map Harker sections calculated using data from 10- to 5-Å resolution. Phase calculation, solvent flattening, phase extension, and simulated annealing refinement were carried out within CNS version 1.1 [ 34 ]. The initial SIRAS map was uninterruptible in spacegroup P6 1 22 but was unambiguous in P6 5 22, permitting the hand of the space group to be determined. A 47-nucleotide poly-C model was built into the SIRAS map using O, the actual nucleotide-sequence register was then confirmed by inspecting the electron density, and residues 1–47 were built in using O [ 35 ]. The phosphate for residue 48 is clearly present in the electron density map, but the density for the remainder of U(48), as well as that for the bases of U(25) and U(30), was rather disordered. The final refinement was performed using CCP4's refmac [ 36 ], and the figures were produced using MacPymol [ 37 ]. All crystallographic computations were performed on the Mac OS X platform. Details of data processing, phasing, and refinement are provided in Tables 1 and 2 . The crystal structure of the SARS s2m RNA was compared to others in the RCSB Protein Data Bank using the program MC-Annotate [ 38 , 39 ] and by visual inspection. Sequence comparisons prior to obtaining the s2m tertiary structure were performed using the UCSC Genome Browser [ 40 ], and were subsequently supplemented with tertiary constraints imposed by the crystal structure using the programs PatScan [ 41 ] and RNABOB [ 42 , 43 ]. Transcripts containing s2m for solution structure analysis were prepared using plasmid templates cleaved downstream, so that the s2m element was present at the 5′ end of the transcript and contained an RNA tail consisting of plasmid sequences. Chemical probing experiments were carried out according to established methods [ 44 ]. Primer extension was performed as described previously [ 45 ] using a primer complementary to sequences 3′ of the s2m element. Supporting Information Coordinates, native and derivative amplitudes, and experimental phases have been deposited in the RCSB Protein Data Bank ( http://www.rcsb.org/pdb/ ) under accession number 1XJR and are also available with other supplementary materials at http://www.chemistry.ucsc.edu/%7Ewgscott/sars . Accession Numbers The RCSB Protein Data Bank accession number for the SARS s2m RNA structure reported here is 1XJR. The RCSB Protein Data Bank accession numbers for the other protein and RNA structures discussed in this paper are as follows: the 30S ribosome (1J5E), the 30S ribosome in which prokaryotic IF-1 has been added (1HR0), the eukaryotic analog of prokaryotic IF-1 (1D7Q), and the crystal structure of nsp9 (1QZ8 and 1UW7).

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539059.xml

555752

Estimating a preference-based index for a menopause specific health quality of life questionnaire

Background The aim of the study was to develop a menopause-specific, preference-based health-related quality-of-life (HRQoL) index reflecting both menopausal symptoms and potential side-effects of Hormone Replacement Therapy (HRT). Methods The study had three phases: the development of a health state classification, a prospective valuation survey and the estimation of a model to interpolate HRQoL indices for all remaining health states as defined by the classification. A menopausal health state classification was developed with seven dimensions: hot flushes, aching joints/muscles, anxious/frightened feelings, breast tenderness, bleeding, vaginal dryness and undesirable androgenic signs. Each dimension contains between three and five levels and defines a total of 6,075 health states. A sample of 96 health states was selected for the valuation survey. These states were valued by a sample of 229 women aged 45 to 60, randomly selected from 6 general practice lists in Sheffield, UK. Respondents were asked to complete a time trade-off (TTO) task for nine health states, resulting in an average of 16.5 values for each health state. Results Mean health states valued range from 0.48 to 0.98 (where 1.0 is full health and zero is for states regarded as equivalent to death). Symptoms, as described by the classification system, can be rank-ordered in terms of their impact (from high to low) on menopausal HRQoL as follows: aching joints and muscles, bleeding, breast tenderness, anxious or frightened feelings, vaginal dryness, androgenic signs. Hot flushes did not significantly contribute to model fit. The preferred model produced a mean absolute error of 0.053, but suffered from bias at both ends of the scale. Conclusion This article presents an attempt to directly value a condition specific health state classification. The overall fit was disappointing, but the results demonstrate that menopausal symptoms are perceived by patients to have a significant impact on utility. The overall effect is modest compared to the more generic health state descriptions such as the EQ-5D. The resultant algorithm generates a preference-based index that can be used economic evaluation and that reflects the impact of this condition.

Background The increasing demand for economic evaluation of health care interventions has lead to a corresponding rise in the derived demand for evidence on the key parameter inputs into cost effectiveness models. One of those inputs is the health state utility value used to estimate the quality adjusted life years (QALYs) associated with an intervention. This article is concerned with estimating a preference-based measure for generating utility values for menopausal health states. Most preference-based measures of health such as the EQ-5D, SF-6D and the Health Utilities Index 3, have a generic health state descriptive system [ 1 - 3 ]. However, general measures of health have been found to be inappropriate or insensitive for some medical conditions [ 4 ], and it has been found that these instruments are not sufficiently sensitive to the impact of menopausal symptoms [ 5 , 6 ]. There has been increasing interest in estimating preference-based indices from condition specific measures. These have often involved mapping from condition specific measures onto preference-based measures [ 4 , 7 - 9 ]. While this approach is useful, it is a second best solution for studies that did not use a generic measure and the aim is to estimate generic preference scores. However, for some conditions generic measures may not be appropriate and in this case a better solution would be to elicit preference weights for the condition specific measure [ 10 ]. There have been a number of studies published recently that have estimated conditions specific preference scores, including for Rhinitis [ 11 ], Erectile Dysfunction [ 12 ], asthma [ 13 ] and Prostate symptoms [ 14 ]. This is the first attempt to estimate a preference-based measure for menopausal symptoms. The aim of this study is to quantify the impact of menopause-related health problems on health-related quality of life as indicated by a "strength-of-preference" index. This project had three components. The first was to construct a health state classification system for menopausal symptoms based on work by Zoellner and others [ 15 ]; secondly a sample of menopausal health states defined by the latter were then valued by means of the time trade-off; and then modelled the health state values using regression techniques to produce an algorithm for valuing all states described by the menopausal health state system. Methods The menopause-specific health state classification A menopause-specific quality-of-life questionnaire has been developed by Zoellner and colleagues [ 6 , 15 ]. Initially, a pool of 39 menopause-related items – identified as being important on the grounds of two focus group sessions of peri- and postmenopausal women, literature review, and expert opinion – underwent intensive analysis to determine the degree of fulfilment of standard psychometric criteria of re-test reliability, face validity, construct validity and convergent validity. The application of these criteria resulted in a questionnaire with 22 items grouped into 6 domains, namely (1) psychosocial, (2) physical, (3) vasomotor, (4) sexual, (5) menstrual, and (6) androgenic complaints. In order to derive a health state classification from the former, the most robust item(s) were chosen from each domain. As it was felt important to cover potential side-effects of Hormone Replacement Therapy (HRT), the menstrual domain is represented with two items – 'breast tenderness' and 'vaginal bleeding' – in the classification systems; the latter hence consists of the following seven domains of menopausal health (see table 1 ): hot flushes, aching joints/muscles, anxious/frightened feelings, breast tenderness, bleeding, vaginal dryness and undesirable androgenic signs. The assignment of the number as well as the descriptor of levels was performed according to the frequency distributions observed in the screening section of the postal survey (n = 785, Table 1 ). Each dimension contains between three and five levels and defines a total of 6,075 health states. Table 1 The Menopause health state classification 1. hot flushes 1) You have no hot flushes 2) You get 1–3 hot flushes per day 3) You get 4 or more hot flushes per day 2. aching joints or muscles 1) You have no aching joints or muscles at all. 2) You have 1–3 episodes of aching joints or muscles per week . 3) You have 4 or more episodes of aching joints or muscles per week . 4) You have mild to moderate constant pain in your joints or muscles. 5) You have severe constant pain in your joints or muscles. 3. anxious or frightened feelings 1) You do not have anxious or frightened feelings. 2) You have anxious or frightened feelings 1–3 times per week . 3) You have anxious or frightened feelings 4 or more times per week . 4. breast tenderness 1) You have no breast tenderness. 2) You have mild to moderate breast tenderness. 3) You have severe breast tenderness 5. bleeding 1) You have no bleeding 2) You have mild regular (monthly) bleeding 3) You have mild irregular bleeding 4) You have intense regular (monthly) bleeding 5) You have intense irregular bleeding 6. undesirable cosmetic signs (facial or body hair growth, greasy skin or acne) 1) You have no undesirable cosmetic signs. 2) You have mild to moderate undesirable cosmetic signs 3) You have severe undesirable cosmetic signs. 7. vaginal dryness 1) You have no vaginal dryness. 2) You have mild to moderate vaginal dryness. 3) You have severe vaginal dryness. The valuation survey The design of the survey was to elicit values for a sample of states defined by the menopausal health state classification using a variant of the Time Trade-off (TTO) on a sample of women aged 45 to 60. The key design issues were the sample of health states to be valued, the sample of respondents, the valuation technique and the interview. Selection of health states It is not possible to value all the states defined by the menopausal specific health state classification. However, there is currently little guidance of the selection of states for valuation [ 16 ]. Based on past practice, the states used in this survey were selected using the orthoplan programme in SPSS. This programme generates an orthogonal array of states that need to be valued in order to estimate an additive model. This programme indicated that 49 states were necessary to estimate an additive model. It was decided to enhance these states in order to ensure some degrees of freedom and to permit some examination of interactions. Therefore, the programme selected another 47 as 'hold out' states drawn at random. This resulted in a total of 96 health states valued out of a potential of 6,075 defined by the menopausal health state classification system. Each respondent was asked to value a sample of eight states. These states were selected from the larger sample of 96 using a stratified sampling technique to ensure that each respondent has a mix of mild, moderate and more severe states. The severity of the states has been assessed by summing the dimension levels. The states were then ranked using this sum score and divided into quartiles to identify four severity groups. Two health states have been selected at random without replacement from each severity group to form a set of eight health states. This was done another 11 times to create 12 sets of states. These 12 sets were used an equal number of times in order to ensure that each of the 96 states would be valued an equal number of times. The original aim was to interview 150 respondents, where each respondent valued eight health states. This would mean undertaking 24 sessions with between 6–8 respondents at each session and would have resulted in 1200 observations and an average of 12.5 valuations per state. Selection of respondents A previous survey was undertaken in Sheffield (UK) with the main aim of validating the new questionnaire designed to assess the health of women in mid -life. One thousand and eighty women aged 45 to 60 were randomly selected from the lists of 6 GPs in Sheffield (180 women per GP List) and sent the postal questionnaire concerning their menopausal symptoms. Of these 790 (73%) were returned. Five were dropped from further analysis due to incomplete response, so the total number was 785 responders. All responders were sent a summary of the study "Women's Health in Mid life" in December 2001 and asked if they would be interested in participating in a second phase of the study. Out of these 417 women replied saying they would like more information of the phase 2 study. The 6 GP practices signed a consent form agreeing to these women being invited to participate in the valuation survey. Invitation letters were sent by each practice with a Patient information sheet and a Patient consent form. Out of the 417, 229 (55%) attended the interviews and completed a questionnaire. Valuation technique Health states were valued using a variant of the time trade-off technique (TTO). This technique asks the respondent to choose between a fixed period of time (t) in the health state to be valued compared to a shorter period in full health (x). The amount of time spent in full health is varied until the respondent is indifferent between the two alternatives. The value of the health state is then x/t for states better than dead. This study used a self-completed variant of TTO developed by Gudex that uses a titration procedure shown in Table 2 , where the respondent is presented with two lists of values [ 17 ]. Each row has a value of 25 for t and a declining value for x, where the value of x declines by one year between each row. Twenty-five years was chosen to represent a reasonable life expectancy for this sample of respondents. The respondent is asked to indicate all the cases where they are confident they would choose A (i.e. the health state to be valued), all the cases where they would choose B (i.e. full health) and the put an equals against states where they cannot choose. There was no allowance for states worse than death, but this was felt to be an unlikely scenario for states defined by the menopausal health state classification. Table 2 The time trade-off question Choice A ---- Choice B 25 years 25 years 25 years 24 years 25 years 23 years 25 years 22 years 25 years 21 years 25 years 20 years 25 years 19 years 25 years 18 years 25 years 17 years 25 years 16 years 25 years 15 years 25 years 14 years 25 years 13 years 25 years 12 years 25 years 11 years 25 years 10 years 25 years 9 years 25 years 8 years 25 years 7 years 25 years 6 years 25 years 5 years 25 years 4 years 25 years 3 years 25 years 2 years 25 years 1 year 25 years 0 years Please put an " A " against all cases where you are CONFIDENT that you would choose Choice A . Please put a " B " against all cases where you are CONFIDENT that you would choose Choice B . Please put an " = " against the case where you cannot choose between Choice A and Choice B . Interviews Respondents were invited to attend interview sessions held in a room at the Institute of General Practice in Sheffield (UK). Two researchers experienced in interviewing patients coordinated these sessions. The interview began with the researchers explaining the purpose of the survey and to explain the TTO task. Patients were asked to complete an example and encouraged to ask questions in order to aid in their understanding. Once the respondents were ready, they were then asked to complete all remaining questions on their own. There were 33 such interview sessions in the survey. Respondents were reimbursed for their time and travel with a voucher for £10. The questionnaire had questions on age, occupation, education level, general health, whether or not they had stopped menstruating and if so when, and finally whether or not they continued to take HRT. They were then asked to complete the menopausal health state classification. They then undertook a practice TTO question followed by eight TTO questions. The health states were presented in a random order to avoid the risk of an ordering effect. Finally they were asked to value their own health using a TTO question. Modelling The overall aim is to construct a model for predicting health state valuations based on the menopausal health state classification. The data generated by the valuation survey described above has a complex structure, as they are skewed and health state valuations are clustered by respondent. Disentangling the respondent effect is a complex task and can only be tackled at the individual level, where each valuation is regarded as a separate observation, rather than using the mean value for each health state. The former has the advantage of greatly increasing the number of degrees of freedom available for the analysis (from 96 to over 1200) and enabling the analysis of respondent background characteristics on health state valuations. A number of alternative models have been proposed for estimating preference functions from health data [ 12 , 14 , 15 ]. The general model has been defined elsewhere as [ 12 ]: y ij = g ( β ' x ij + θ ' r ij + δ ' z j ) + ε ij (1) where i = 1, 2, ..., n represents individual health state values and j = 1,2, ..., m represents respondents. The dependent variable, y ij , is the TTO score for health state i valued by respondent j . x is a vector of binary dummy variables (x δλ ) for each level λ of dimension δ of the classification. Level λ = 1 acts as a baseline for each dimension, so in a simple linear model, the intercept represents state 1111111, and summing the coefficients of the 'on' dummies derives the value of all other states. The r term is a vector of terms to account for interactions between the levels of different attributes. z is a vector of personal characteristics that may also affect the value an individual gives to a health state, for example, age, sex and education. The role of personal characteristics is not discussed in this paper. g is a function specifying the appropriate functional form. ε ij is an error term whose autocorrelation structure and distributional properties depend on the assumptions underlying the particular model used. This is an additive model, which imposes no further restrictions on the relationship between dimension levels of the classification. For example, it does not enforce an interval scale between the levels of each dimension and does not impose ordinality on the levels. OLS assumes a standard zero mean, constant variance error structure, with independent error terms, that is cov( ε ij ε i ' j ) = 0, i ≠ i '. This specification ignores the clustering in the data and assumes that each individual health state value is an independent observation, regardless of whether or not it was valued by the same respondent. An improved specification, which takes account of variation both within and between respondents, is the one-way error components random effects model. This model explicitly recognises that n observations on m individuals is not the same as n × m observations on different individuals. Estimation is via generalised least squares (GLS) or maximum likelihood (MLE). Analysis of first order interactions alone is problematic, since the large number of possible interactions means there is a risk of finding some are significant purely by chance. We have therefore adopted the approach used in other studies of using summary terms for describing interactions [ 16 , 18 ]. Extreme level dummies were created to represent the number of times a health state contains dimensions at the extreme ends of the scale [ 18 ]. Least severe is defined as level 1 on each dimension. Most severe is defined as the bottom level of each dimension. These are used to create dummy variables LEAST and MOST which take a value of 1 if any dimension in the health state is at the least (most) severe level, and 0 otherwise. Finally we consider alternative functional forms – g in (1) – to account for the skewed distribution of health state valuations. Four functional forms are used. Firstly, a Logit transformation and two complementary log-log transformations suggested by Abdalla and Russell. [ 19 ] These are chosen to map the data from the range (-1,1) to the range (-∞,∞) via the unit range (0,1). Secondly, a Tobit transformation which, although designed to deal with truncated data, can approximate for the left skew in this data, where 25% of the values lie between 0.9 and 1. Specifying a Tobit model with upper censoring at 1 does this. All modelling will be done using STATA 7.0 and SPSSWin. Results Respondents The characteristics of the 229 interviewed women are presented on Table 3 . Their mean age was 54 with a range between 46 and 61. Seventy four percent had stopped menstruating and the average time to since they last menstruated was 64 months. A third had taken HRT in the last month. The respondents reported their general health to be in the mid-range of the excellent to poor scale. The seven menopausal symptoms were highly prevalent, with two thirds experiencing aching joints and muscles, nearly half reporting hot flushes and vaginal dryness and around one third experiencing anxiety or fright, breast tenderness and cosmetic signs. The mean valuation of their current health state by the TTO was 22.8 (SD = 4.4) which translated into a health state utility value of 0.91. Table 3 Characteristics of respondents Full sample n = 229 Age: mean (s.d) 53 (SD) Highest qualification % Degree 26 A levels 11 Other 63 Self-rated general health: % Excellent 7 Very good 44 Good 31 Fair 14 Poor 4 Reporting the following: % Hot flushes 45 Aching joints or muscles 74 Anxious or frightened 37 Breast tenderness 31 Bleeding 23 Cosmetic signs 35 Vaginal dryness 45 TTO own valuation 22.8 (4.4) Stopped menstration 74 Average time since stopped menstruation (months) 64 (72) Taken HRT in last month 35 Thirty respondents were excluded from the modelling data set, leaving 199. Respondents were excluded due to ambiguity in the responses to the (self-administered) questionnaire. The main sources of ambiguity were the mixing up of responses (e.g. ticks and crosses appearing the wrong way around) and large gaps between the responses with no indication of the appropriate point of indifference. There were also a number of individual responses elicited from 199 respondents that had to be excluded due to similar ambiguities. These exclusions left 1580 health state values across the 96 health states for modelling, a final completion rate of 86% of all questions asked at interview. Health state values Descriptive statistics for 50 of the 96 states are presented on Table 4 . Each health state is valued on average 16.5 times, which exceeds the original target. Mean health state values range from 0.48 to 0.98 with large standard deviations. The median values usually exceed the mean values, reflecting the highly skewed nature of the data. This skewness is even more apparent at the individual level, as shown in the histogram presented in Figure 1 . Very few health states values were 1.0 (32/1580) indicating that that most respondents were willing to trade time for quality of life, however 31% were at the next possible value of 24.5 years. At the other end of the scale, only three had a value of zero where it might have been possible that respondents regarded these states as worse than death. Table 4 Descriptive Statistics for 50 health state valuations State Mean n s.d. Median maximum Minimum 1112311 0.93 16 0.08 0.98 0.98 0.78 1112422 0.87 16 0.16 0.90 1.00 0.38 1112433 0.79 15 0.24 0.94 0.98 0.26 1113122 0.88 18 0.16 0.94 0.98 0.42 1113232 0.79 17 0.23 0.82 1.00 0.02 1113512 0.82 15 0.21 0.94 0.98 0.38 1113531 0.80 17 0.22 0.90 0.98 0.38 1121223 0.83 18 0.16 0.88 0.98 0.42 1121331 0.86 16 0.17 0.90 1.00 0.42 1211522 0.87 15 0.19 0.98 1.00 0.38 1322231 0.71 18 0.31 0.84 0.98 0.00 1323123 0.81 14 0.22 0.90 0.98 0.22 1323231 0.72 18 0.27 0.84 0.98 0.22 1331412 0.80 15 0.23 0.90 0.98 0.22 1331432 0.56 13 0.22 0.54 0.94 0.06 1332213 0.77 18 0.23 0.88 0.98 0.26 1333112 0.84 18 0.19 0.92 0.98 0.42 1413211 0.68 13 0.22 0.74 0.90 0.10 1413513 0.78 15 0.22 0.82 0.98 0.22 1423133 0.76 18 0.21 0.80 0.98 0.26 2221511 0.67 13 0.25 0.74 0.98 0.06 2223131 0.83 17 0.14 0.82 1.00 0.54 2231312 0.89 15 0.13 0.98 1.00 0.54 2233333 0.48 13 0.28 0.42 0.98 0.06 2311511 0.87 18 0.16 0.92 1.00 0.42 2313123 0.96 16 0.04 0.98 1.00 0.86 2321122 0.70 13 0.24 0.74 0.98 0.10 2332413 0.54 13 0.26 0.54 0.98 0.06 2421212 0.92 17 0.09 0.98 1.00 0.70 2421322 0.82 18 0.19 0.92 0.98 0.42 2521323 0.65 18 0.30 0.78 0.94 0.00 2523423 0.71 16 0.23 0.74 0.98 0.34 2532323 0.83 18 0.24 0.88 0.98 0.00 2532531 0.71 16 0.26 0.72 0.98 0.14 2533311 0.77 15 0.31 0.94 1.00 0.04 3121111 0.92 17 0.15 0.98 1.00 0.42 3121533 0.81 18 0.20 0.86 0.98 0.34 3132211 0.83 16 0.19 0.90 1.00 0.34 3133412 0.79 17 0.21 0.90 0.98 0.34 3133521 0.67 15 0.26 0.78 0.98 0.14 3232433 0.77 15 0.27 0.90 0.98 0.18 3311433 0.75 17 0.24 0.82 0.98 0.42 3312321 0.89 16 0.14 0.94 0.98 0.46 3412222 0.80 18 0.19 0.86 0.98 0.42 3413111 0.83 18 0.29 0.94 0.98 0.06 3422113 0.82 16 0.14 0.82 0.98 0.56 3422412 0.78 15 0.29 0.90 1.00 0.06 3431133 0.80 16 0.19 0.86 0.98 0.38 3433532 0.71 18 0.32 0.80 0.98 0.02 3511211 0.80 17 0.23 0.86 0.98 0.38 Please put an " A " against all cases where you are CONFIDENT that you would choose Choice A . Please put a " B " against all cases where you are CONFIDENT that you would choose Choice B . Please put an " = " against the case where you cannot choose between Choice A and Choice B . Figure 1 Histogram for TTO values. Modelling Basic models: main effects The Breusch-Pagan test for individual effects suggests these are important ( χ 2 = 25585.15, P = 0.000) and Hausman's test suggests random rather than fixed effects is the appropriate specification ( χ 2 = 27.11, P = .035), therefore only Random Effects (RE) and mean models are presented in Table 5 . The main effects dummies in each model represent levels of each dimension of the menopausal health state classification. These are expected to have a negative sign and to increase in absolute value within each dimension. It would be inconsistent with the scale for the absolute value to decrease when moving to a worst level within a dimension. Table 5 Models Main effects only Interaction effects (1) (2) (3) (4) RE Mean RE Mean c 0.912 0.917 0.925 0.879 HF2 0.007 -0.008 0.006 -0.005 HF3 -0.006 0.008 -0.002 0.013 AJ2 -0.016 -0.013 -0.016 -0.0106 AJ3 -0.026 -0.062 -0.024 -0.062 AJ4 -0.023 -0.022 -0.022 -0.021 AJ5 -0.070 -0.085 -0.066 -0.08 EM2 -0.012 -0.018 -0.012 -0.018 EM3 -0.034 -0.057 -0.029 -0.051 BT2 -0.018 -0.002 -0.018 0.000 BT3 -0.033 -0.039 -0.028 -0.032 BL2 -0.041 -0.026 -0.039 -0.024 BL3 -0.057 -0.025 -0.057 -0.022 BL4 -0.066 -0.058 -0.068 -0.054 BL5 -0.062 -0.043 -0.059 -0.037 COS2 -0.004 0.010 -0.003 0.014 COS3 -0.015 -0.028 -0.011 -0.024 VAG2 0.006 -0.008 0.006 -0.006 VAG3 -0.024 -0.035 -0.02 -0.029 MOST -0.026 -0.013 LEAST 0.002 0.035 N 1580 96 1580 96 adj R 2 0.040 0.178 0.039 0.164 inconsistencies 3 2 2 3 MAE 0.056 0.053 0.065 0.0552 No > |0.05| 37 36 47 36 No > |0.10| 14 15 17 15 t(mean = 0) -0.334 † -0.344 † JBPRED 34.789 20.587 36.089 17.028 LB 214.99 124.92 218.04 150.44 Estimates shown in bold are significant at t 0.10; † Mean error is zero by definition. In the RE model (1), the coefficients have the expected negative sign for all main effect dummies except HF2 and VAG2, but neither of these is significant. There are 13 significant coefficients, including the constant term. There are three inconsistencies involving significant coefficients, AJ3 to AJ4, BL2 to BL3 and BL4 to BL5. The mean model has a better explanatory power than the OLS model (not shown), but has only seven significant coefficients that produce just two inconsistencies. The ability of the mean and RE main effects models to predict health state valuations within the data set is presented at the bottom of Table 5 . The main effects models have similar mean absolute errors, though it is slightly lower for the mean model. The proportion of errors greater than 0.05 and 0.1 is also very similar at 39% and 15% respectively. The JB test found evidence for non-normality of errors for both the models. An important problem has been identified by the Ljung-Box statistics that reveal significant autocorrelation in the prediction of all the models. Plots of actual against predicted errors reflect a tendency to over predict at the lower end and under predict at the upper end. The model was re-estimated using a Tobit procedure, but this did not improve the predictive performance of the model. Applications of the logit and complementary log functions also did not improve model performance. Interactions The RE and mean models in Table 5 include dummy variables for MOST and LEAST, which take the value of 1 if any dimension is at the most or least severe level respectively. The coefficients associated with these dummies suggest a further negative impact when any dimension is at its worst level and a slight positive impact from having any dimension at the least severe level. These coefficients were significant for some models. However, the coefficients on the main effects have been slightly reduced by these additional dummies, particularly the worst levels of AJ5 and BL5. Furthermore, the addition of these variables has not significantly improved either model Discussion The paper presents the results of a study aiming to estimate preference functions for the menopausal health state classification. The preference models look credible in terms of the coefficients, though there are a number of problems with the models predictive performance. This paper supports the findings of other studies, that it is the feasible to estimate condition specific preference-based indices [ 11 - 14 ]. It was perhaps more ambitious than other published studies in that it attempted to estimate values for a large health state classification, where it was not possible to directly value all states. It was the first study using statistical inference to model health state values. The explanatory power of the models is not high. This is due to the high variability around the health state means, which may have been a result of the self-completed format of the TTO task. It may also have been due to comparatively low number of observations per state. The classification describes health states that are mild compared to the full range of states described by descriptive systems such as the EQ-5D or HUI3 that reflects the nature of the conditions. The specific values found for our instrument have tended to be more skewed at the upper end than the generic measures, such as the EQ-5D. This was also found for a number of other conditions, with the lowest value for Erectile Dysfunction being 0.74 [ 12 ] and 0.87 for prostate symptoms [ 14 ]. However this seems to be a consequence of the comparatively mild impact of this condition, because a preference scale for Asthma had a lowest value of 0.04 [ 13 ] and 0.15 for Rhinitis [ 11 ]. For milder conditions a valuation technique such as TTO that relies on trading quality with survival may be rather insensitive for some respondents, which is reflected in the higher proportion of people indicting the first response choice down the scale. This might suggest that more effort needs to be made to develop variants of the TTO and SG that allow milder states to be valued with sufficient sensitivity. One approach would be the use of chaining, where each mild state is valued against full health and a lower anchor that is better than being dead, which in turn has been valued against full health [ 20 ]. However, this has been shown to produce biased estimates [ 21 , 22 ]. A further problem may have arisen from the descriptive system. The 2 or 3 inconsistencies between coefficients may be due to possible ambiguities in the health state classification. The ranking of AJ3 and AJ4 is ambiguous, since it is possible for 4 or more episodes per week of pain to be worse than mild to moderate constant pain. Also for BL, some people may regard irregular bleeding as better than regular bleeding. Such differences of opinion in the population in the ranking ordering of some levels would reduce the fit of the models. Of more concern is the evidence for systematic patterns in the residuals resulting in over prediction at the lower end and under predicting at the upper end of the range. The MVH group was able to solve this problem in their valuation of the EQ-5D by the inclusion of an interaction term. The inclusion of interaction terms in this study had little impact on the problem. The application of various transformations to the dependent variable also did not solve this problem. The models nonetheless provide a basis for valuing menopausal health states using this health state classification. The coefficients are consistent with the cordiality of the health state classification and the size of the mean absolute error of 0.055 to 0.065 is comparable to that achieved in other models [ 16 ]. The addition of interaction terms did not improve the model and tended to offset the main effects, therefore it is not proposed to recommend the models with interactions. The choice of models is between the random effects and mean main effects models (i.e. (1) and (2)). Given the mean model is slightly better in terms of fit and numbers of inconsistencies this is the one recommended for use. The estimation of preference weights for condition specific quality of life has been questioned by some health economists as to its value [ 23 ]. The argument for using condition specific descriptive systems is that they are likely to be more sensitive to changes in the condition than generic measures and more relevant to the concerns of patients. On the other hand, condition specific measures often focus on symptoms and it could be argued this concentrates the mind of the respondent on the negative aspects of the conditions. This may have a framing effect that produces lower values because the respondents are not thinking about other aspects of their lives unaffected by the condition. However, the risk of this was reduced by selecting women in the age range of 45–60, most of whom had experienced menopausal symptoms and would have a realistic view of the likely impact of the condition. The argument for using condition specific descriptive systems is that they are going to better reflect the impact of the condition on a patient's life. However, provided the descriptive system is valued on the same full health – death scale using the same variant of the same valuation technique using a comparable population sample, then the valuations should be comparable. Any remaining differences in values should be a legitimate consequence of the descriptive system. However, this assumes that the value of a dimension is independent of those dimensions outside of the descriptive system and this requires empirical testing. Despite these arguments, there has been increasing interest in estimating condition specific preference measures of health because the analyst often only has condition specific data and wishes to use them to undertake an economic evaluation, or the analyst feels a generic measure is not appropriate for the condition. Conclusion The advantages of using a condition specific descriptive system over a generic are that it should be more sensitive to improvements in health. However, the overall fit was disappointing. The results demonstrate that menopausal symptoms are perceived by patients to have a significant impact on utility, but the overall effect is modest compared to the more generic health state descriptions such as the EQ-5D. This research has also demonstrated the problems that can be encountered when trying to value a comparatively mild condition. The resultant algorithm generates a preference-based index that can be used economic evaluation and that reflects the impact of this condition. Authors' contributions JB led the project, including the design of the valuation study and undertaking much of the analysis. MP contributed to the design of the valuation survey and undertook the interviews. JR provided a key input into the econometric analyses. YZ designed the menopausal health state classification and contributed to the overall design of the study. All authors contributed to the writing of the paper.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555752.xml

544887

Sicily statement on evidence-based practice

Background A variety of definitions of evidence-based practice (EBP) exist. However, definitions are in themselves insufficient to explain the underlying processes of EBP and to differentiate between an evidence-based process and evidence-based outcome . There is a need for a clear statement of what Evidence-Based Practice (EBP) means, a description of the skills required to practise in an evidence-based manner and a curriculum that outlines the minimum requirements for training health professionals in EBP. This consensus statement is based on current literature and incorporating the experience of delegates attending the 2003 Conference of Evidence-Based Health Care Teachers and Developers ("Signposting the future of EBHC"). Discussion Evidence-Based Practice has evolved in both scope and definition. Evidence-Based Practice (EBP) requires that decisions about health care are based on the best available, current, valid and relevant evidence. These decisions should be made by those receiving care, informed by the tacit and explicit knowledge of those providing care, within the context of available resources. Health care professionals must be able to gain, assess, apply and integrate new knowledge and have the ability to adapt to changing circumstances throughout their professional life. Curricula to deliver these aptitudes need to be grounded in the five-step model of EBP, and informed by ongoing research. Core assessment tools for each of the steps should continue to be developed, validated, and made freely available. Summary All health care professionals need to understand the principles of EBP, recognise EBP in action, implement evidence-based policies, and have a critical attitude to their own practice and to evidence. Without these skills, professionals and organisations will find it difficult to provide 'best practice'.

Background The Sicily statement on evidence-based practice "Knowing is not enough; we must apply. Willing is not enough, we must do" [ 1 ] Health care delivered in ignorance of available research evidence, misses important opportunities to benefit patients and may cause significant harm [ 2 - 4 ]. Providing evidence-based care is recognised as a key skill for health care workers from diverse professions and cultures [ 5 - 10 ]. The ability to deliver evidence-based practice promotes individualisation of care and assures the quality of health care for patients today as well as those of tomorrow [ 11 ]. A variety of definitions of evidence-based practice (EBP) have been proposed. However, definitions are in themselves insufficient to explain the underlying processes of EBP and to differentiate between an evidence-based process and evidence-based outcome . Towards this goal, we propose three points to clarify and promote the realisation of EBP: 1) A clear statement of what EBP means. 2) A description of the minimum skill set required to practise in an evidence-based way. 3) A curriculum that outlines the minimum standard educational requirements for training health professionals in EBP. This statement was conceived by the delegates of the second international conference of Evidence-Based Health Care Teachers and Developers held in Sicily in September 2003 ("Signposting the future of EBHC", [ 12 ]). In response to a request from the delegates at this conference's final plenary session the steering committee prepared the first draft. The proposed statement and a topic questionnaire were then circulated to all 86 attendees of the Sicily conference for suggestions and clarifications. Eighteen professions allied to health from 18 countries were represented. Suggestions were incorporated and a final paper approved by consensus. Discussion Increase in medical information During the last century there has been an exponential growth of research and knowledge [ 13 , 14 ]. The growth of health care information has been particularly rapid in diagnostic and therapeutic technologies. The volume of medical papers published doubles every 10 to 15 years [ 15 ]. Electronic searching of this expanding evidence base was initiated by the National Library of Medicine in 1966 [ 16 ]. Electronic access to full text articles and journals started to become available in 1998 [ 17 ]. Increasingly, specialist databases of utility for health professionals are being developed, such as the Physiotherapy Evidence Database [ 18 ] and the C2-SPECTR [ 19 ]. Regular use of these resources is identified as one marker for lifelong learning among physicians [ 20 ], but the process is not easy [ 21 ]. Identification of the best methods to understand and integrate patient values, such as decision aids or patient-centred consultations, is still at the early stages of development [ 22 ]. With this expansion of information, our knowledge should be greater and our practice should be more effective. Unfortunately this is too often not the case [ 23 ]. This recognised gap between best evidence and practice is one of the driving forces behind the development of EBP. Clinical decision making Good practice including effective clinical decision making – step 4 of the EBP process – requires the explicit research evidence and non-research knowledge (tacit knowledge or accumulated wisdom). Clinical decision making is the end point of a process that includes clinical reasoning, problem solving, and awareness of patient and health care context [ 24 ]. This process is uncertain and frequently no "correct" decision exists. EBP can help with some of the uncertainties in this decision process by using the explicit knowledge obtainable from research information. But to do so the research information must be transformed into clinicians' knowledge. Information can be defined as data that has been sorted, analysed, & displayed and communicated through language, graphic displays, or numeric tables. Explicit knowledge is then the meaning people create using this information and its application through action in specific settings [ 25 ]. For example clinician's knowledge should include the need to evaluate quickly the patient with chest pain to take advantage of the research proven window of opportunity for treatment of acute coronary syndrome. Step 4 also requires the tacit knowledge which comes from the wisdom of experience, informed by evidence and outcomes, and which is consequently harder to share. An example is the recognition of a sick child. Research may develop a list of clinical features that, when present, denote severe illness in a child. While this list will help the inexperienced junior doctor, nurse, or midwife, the experienced health practitioner has a tacit knowledge of "sickness" in a child that comes from both knowledge of the features list and assimilation with experience, thereby speeding up the recognition of "sickness" in a child. Principles & development of evidence-based practice The term "Evidence-based medicine" was introduced in the medical literature in 1991 [ 26 ]. An original definition suggested the process was "an ability to assess the validity and importance of evidence before applying it to day-to-day clinical problems" [ 27 , 28 ]. The initial definition of evidence-based practice was within the context of medicine, where it is well recognised that many treatments do not work as hoped [ 29 ]. Since then, many professions allied to health and social care have embraced the advantages of an evidence-based approach to practice and learning [ 5 - 8 , 30 ]. Therefore we propose that the concept of evidence-based medicine be broadened to evidence-based practice to reflect the benefits of entire health care teams and organisations adopting a shared evidence-based approach. This emphasises the fact that evidence-based practitioners may share more attitudes in common with other evidence-based practitioners than with non evidence-based colleagues from their own profession who do not embrace an evidence-based paradigm. EBP evolved from the application of clinical epidemiology and critical appraisal to explicit decision making within the clinician's daily practice, but this was only one part of the larger process of integration of evidence into practice. Initially there was a paucity of tools and programmes to help health professionals learn evidence-based practice. In response to this need, workshops based on those founded at McMaster by Sackett, Haynes, Guyatt and colleagues were set up around the world. During this period several textbooks on EBP were published accompanied by the development of on-line supportive materials. The initial focus on critical appraisal led to debate on the practicality of the use of evidence within patient care. In particular, the unrealistic expectation that evidence should be tracked down and critically appraised for all knowledge gaps led to early recognition of practical limitations and disenfranchisement amongst some practitioners [ 31 ]. The growing awareness of the need for good evidence also led to awareness of the possible traps of rapid critical appraisal. For example problems, such as inadequate randomisation or publication bias, may cause a dramatic overestimation of therapeutic effectiveness [ 32 ]. In response, pre-searched, pre-appraised resources, such as the systematic reviews of the Cochrane Collaboration [ 33 ], the evidence synopses of Clinical Evidence [ 34 ] and secondary publications such as Evidence Based Medicine [ 35 ] have been developed [ 36 ], though these currently only cover a small proportion of clinical questions. Process of Evidence Based Practice The five steps of EBP were first described in 1992 [ 37 ] and most steps have now been subjected to trials of teaching effectiveness (indicated by references) 1. Translation of uncertainty to an answerable question [ 38 ] 2. Systematic retrieval of best evidence available [ 39 ] 3. Critical appraisal of evidence for validity, clinical relevance, and applicability [ 40 ] 4. Application of results in practice [ 41 ] 5. Evaluation of performance [ 42 ] This five-step model forms the basis for both clinical practice and teaching EBP, for as Rosenberg and Donald observed, "an immediate attraction of evidence-based medicine is that it integrates medical education with clinical practice" [ 43 ]. Curricula outline of minimum standard educational requirements Different practitioners at different levels of responsibility within evidence-based organisations will require different skills for EBP and different types of evidence. It is a minimum requirement that all practitioners understand the principles of EBP, implement evidence-based policies, and have a critical attitude to their own practice and to evidence. Without these skills and attitidues, health care professionals will find it difficult to provide 'best practice'. Teachers, commissioners, and those in positions of leadership will require appraisal skills that come with higher training and continued use [ 44 ]. The wider knowledge and use of these skills will help health professionals meet some of Hurd's list of desired educational outcomes [ 45 ] in being able to: • distinguish evidence from propaganda (advertisement) • probability from certainty • data from assertions • rational belief from superstitions • science from folklore Curricula that outline the minimum standard educational requirements for practitioners Evidence-based practitioners need additional skills to supplement traditional knowledge. Health care graduates should "be able to gain, assess, apply and integrate new knowledge and have the ability to adapt to changing circumstances throughout their professional life" [ 46 ]. Observational studies suggest that one way to 'future-proof' health care graduates, is to train them in the necessary skills to support life-long learning through the five-step model of EBM [ 47 ]. Learning has three components: knowledge, skills and attitudes. It is said that "attitudes are caught, not taught" [ 48 ]. Attitudes, such as comfort with managing uncertainty and reflective learning, provide the psychological framework in which evidence is appraised and applied, described by Sackett as "the conscientious , explicit and judicious use of current best evidence in making decisions about the care of individual patients" [ 49 ]. This presents a challenge, as EBP is rarely taught well [ 50 ] and is applied (and observed) irregularly at the point of patient contact [ 51 ] where professional attitudes are formed, and students learn to incorporate theory into practical skills for patient care. Patient involvement in decision making is part of the process of being an effective practitioner. The degree of involvement and the methods by which this is achieved will depend on the setting, the patients and the practitioner. The curriculum framework for EBP should consider the importance of all steps shown in Table 1 . Often courses focus on one of these elements, most commonly critical appraisal, but a balance of skills in each of the steps is needed to take a student from question through to application. Indeed, the most difficult step (sometimes dubbed "step 0") is to get students and colleagues to recognise and admit uncertainties. As Table 1 suggests, learning should be focused on educational outcome, which in turn needs to reflect the clinical setting. This practical orientation means that EBP teaching and assessment needs to consider the real-time setting of practice, and hence searching and appraisal need to be done in minutes rather than hours or days. Table 1 provides examples of established methods of teaching and assessment for each step, but further compilation, innovation, development, and testing are needed. Future research should be informed by the movement in best evidence medical education (BEME) [ 52 ]. Table 1 Description of evidence for aspects of Evidence-Based Practice teaching and assessment Educational outcome Examples of methods of teaching Examples of methods of assessment Translation of uncertainty into an answerable question. The student identifies knowledge gaps during the course of practice and asks foreground questions to fill these gaps, The student should ask focused questions that lead to effective search and appraisal strategies. Presenting clinical scenarios or asking for students to share a problem encountered in clinical practice. Framing a focussed, answerable question in a structured format [38]. Several formats are taught: 3 part (patient-intervention-outcome), 4 part (patient-intervention/exposure-comparator-outcome), or 5 part (patient-intervention/exposure-comparator-outcome-time) questions. The skills can be assessed by presenting a clinical scenario and asking the student to form a focussed, answerable question (included in the Fresno test) [53]. Search for and retrieval of evidence. The student can design and conduct a search strategy to answer questions. The strategy should be effective and comprehensive: likely to retrieve all relevant evidence. The student understands the strengths and weaknesses of the different sources of evidence. Theoretical instruction backed by a supervised practical session with online connection [39]. A variety of databases should be shown such as Cochrane, MEDLINE, CINAHL, Evidence-Based Medicine, SumSearch, tripdatabase.com with the relative benefits discussed. Computer based OSCE has been used to test the abilities of framing questions, searching, and retrieving appropriate evidence [54]. Critical appraisal of evidence for validity and clinical importance. The student can appraise the validity of a study. The appraisal will include: the suitability of the type of study to the type of question asked, the design of the study and sources of bias, the reliability of outcome measures chosen, and the suitability and robustness of the analysis employed. The student can appraise the importance of the outcomes and translate them into clinically meaningful summary statistics, such as number needed to treat (NNT). This is probably the most widely taught skill [55]Examples include the Critical Appraisals Skills Program [56]. Tests for critical appraisal of validity include the Berlin Questionnaire [57] and the Fresno test. Application of appraised evidence to practice The student can assess the relevance of the appraised evidence to the need that prompted the question. The student can explore the patient's values and the acceptability of the answer. Examples include applying the identified evidence to the specific context that led to the quest for evidence. This requires exploration of the generalisability of the evidence to the specific scenario, and 'particularising' outcomes by adjusting for patient-specific risks[58]. Objective structured clinical examination involving clinical application and interaction with patient after reading supplied evidence [59]. Evaluation of performance . The student asks focussed questions, searches sources of evidence, appraises or uses pre-appraised evidence and applies these in practice. The student reflects on how well these activities are performed. Role modelling by EBP teachers. The encouragement of adult learning styles. Journal clubs [60]. Use of a questionnaire to assess knowledge, attitude and behaviour [61]. Recommendations The term 'EBM' has evolved into a larger phenomenon, as increasing numbers of practitioners in various disciplines recognise the importance of evidence to inform all types of health care decisions. Furthermore, greater patient choice and complexity of care mean that many professionals practise as a team. In recognition of the importance of a united commitment to the principles of 'best practice', we propose that the term 'evidence-based practice' (EBP) be used to describe all aspects of this discipline. To ensure that future health care users can be assured of receiving 'best practice' regardless of the type or location of the care received, we make the following recommendations for education: 1. The professions and their colleges should incorporate the necessary knowledge, skills and attitudes of EBP into their training and registration requirements. 2. Curricula to deliver these competencies should be grounded in the "five-step model" (Table 1 ). 3. Further research into the most effective and efficient methods for teaching each step should be fostered, and linked with ongoing systematic reviews on each step. 4. Core assessment tools for each of the steps should be developed, validated, and made freely available internationally. 5. Courses that claim to teach EBP should have effective methods for teaching and evaluating all components. Evidence-Based Practice (EBP) requires that decisions about health care are based on the best available, current, valid and relevant evidence. These decisions should be made by those receiving care, informed by the tacit and explicit knowledge of those providing care, within the context of available resources. Finally, EBP requires a health care infrastructure committed to best practice, and able to provide full and rapid access to electronic databases at the point of care delivery. We believe that without the skills and resources for all the relevant components of this framework, the practice of a health care professional, or a health care organisation, cannot be said to provide their users with evidence-based care. Summary 1. This consensus statement is from an international working group representing both organisations and individual teachers and developers of evidence-based practice. 2. Evidence-Based Practice (EBP) requires that decisions about health care are based on the best available, current, valid and relevant evidence. These decisions should be made by those receiving care, informed by the tacit and explicit knowledge of those providing care, within the context of available resources. 3. All health care professionals need to understand the principles of EBP, recognise it in action, implement evidence-based policies, and have a critical attitude to their own practice and to evidence. Without these skills professionals will find it difficult to provide 'best practice'. 4. The teaching of EBP should, as far as possible, be integrated into the clinical setting and routine care so that students not only learn the principles and skills, but learn how to incorporate these skills with their own life-long learning and patient care. Competing interests All of the authors have received honoraria for teaching EBP FP is a consultant of Lilly Deutschland GmbH. The International Conferences of EBHC Teachers and Developers do not accept sponsorship from health technologies (including pharmaceutical) manufacturers. Authors contributions MD, WS & PG wrote the original draft. AC, JM, KH, FP, AB & JO contributed to the concept and all revised drafts of the statement. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544887.xml

423149

Beyond Therapy …

In response to the Blackburn and Rowley essay on the President's Council on Bioethics, several thought-provoking opinions on ethical challenges in biomedical research are expressed by prominent stakeholders

It is indeed regrettable that a distinguished and thoughtful scientist such as Elizabeth Blackburn should have been dismissed from the President's Council on Bioethics. Scientific perspectives such as hers are surely needed on this committee. Her dismissal was apparently the consequence of her disagreement with some of the text of the Council's report, “Beyond Therapy: Biotechnology and the Pursuit of Happiness” (2003). The thrust of this report is that some of the directions of current biological research will, if carried to fulfillment, result in major changes in the nature of human life—changes that the report regards with foreboding. In their essay, Drs. Blackburn and Rowley (2004) try to bypass these concerns with the argument that we really are not able to accomplish any of these changes yet and, indeed, some may never be possible. I would suggest that as scientists we should face these issues forthrightly. We should not seek refuge in presentday uncertainties. The authors of the report are not naïve nor ignorant. Yes, if these lines of research are successful, their outcome will change the nature of human life. As an example, consider current research into the causes of aging. Clearly, we do not at present know how to achieve major increases in the human life span (although we are able to do so in lower life forms). But it is plausible that we will learn how to do so. And surely a, say, doubling of the human life span would change the nature of human life. Likewise, if we learn to modify the human gene pool so as to produce exceptional individuals or to alter human capabilities, or if powerful drugs are developed that may commandeer the human psyche, the nature of human life will be altered. But so be it. The nature of human life has changed repeatedly and profoundly in the past—with the invention of agriculture, with the invention of writing, with the development of machines and mechanical power, with the advent of modern science and medicine. The nature of human life is different in 2004 a.d. from what it was in 1000 a.d. or 46 b.c. or 5000 b.c. or 10,000 b.c., and it will change again in the future. The concerns expressed in the report are earnest, and they should be confronted in earnest.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC423149.xml

539065

Cobalamin-Independent Methionine Synthase (MetE): A Face-to-Face Double Barrel That Evolved by Gene Duplication

Cobalamin-independent methionine synthase (MetE) catalyzes the transfer of a methyl group from methyltetrahydrofolate to L- homocysteine (Hcy) without using an intermediate methyl carrier. Although MetE displays no detectable sequence homology with cobalamin-dependent methionine synthase (MetH), both enzymes require zinc for activation and binding of Hcy. Crystallographic analyses of MetE from T. maritima reveal an unusual dual-barrel structure in which the active site lies between the tops of the two (βα) 8 barrels. The fold of the N-terminal barrel confirms that it has evolved from the C-terminal polypeptide by gene duplication; comparisons of the barrels provide an intriguing example of homologous domain evolution in which binding sites are obliterated. The C-terminal barrel incorporates the zinc ion that binds and activates Hcy. The zinc-binding site in MetE is distinguished from the (Cys) 3 Zn site in the related enzymes, MetH and betaine–homocysteine methyltransferase, by its position in the barrel and by the metal ligands, which are histidine, cysteine, glutamate, and cysteine in the resting form of MetE. Hcy associates at the face of the metal opposite glutamate, which moves away from the zinc in the binary E·Hcy complex. The folate substrate is not intimately associated with the N-terminal barrel; instead, elements from both barrels contribute binding determinants in a binary complex in which the folate substrate is incorrectly oriented for methyl transfer. Atypical locations of the Hcy and folate sites in the C-terminal barrel presumably permit direct interaction of the substrates in a ternary complex. Structures of the binary substrate complexes imply that rearrangement of folate, perhaps accompanied by domain rearrangement, must occur before formation of a ternary complex that is competent for methyl transfer.

Introduction Methionine synthases catalyze the transfer of a methyl group from N5-methyl-5,6,7,8-tetrahydrofolate (CH 3 -H 4 folate) to L -homocysteine (Hcy), the terminal step in the biosynthesis of methionine. Two apparently unrelated families of proteins catalyze this reaction: cobalamin-dependent methionine synthase (MetH; EC 2.1.1.13) and cobalamin-independent methionine synthase (MetE; 5-methyltetrahydropteroyltriglutamate–homocysteine methyltransferase; EC 2.1.1.14) Organisms that synthesize or transport B 12 encode the cobalamin-dependent enzyme whereas organisms that cannot obtain B 12 encode only the cobalamin-independent enzyme. Escherichia coli and many other species of bacteria express both enzymes, but mammals utilize only cobalamin-dependent methionine synthase while plants and yeasts utilize only the cobalamin-independent enzyme. MetH and MetE both face the same mechanistic challenge. They must catalyze the transfer of a very poor leaving group from the tertiary amine, CH 3 -H 4 folate, to a relatively poor nucleophile, the sulfur of Hcy. MetH facilitates this transfer by using cobalamin as an intermediate methyl carrier [ 1 ]. Cobalamin accepts a methyl group from CH 3 -H 4 folate at one active site and donates it to Hcy at a second site [ 2 ]. In contrast, MetE appears to catalyze the direct transfer of the methyl group from CH 3 -H 4 folate to Hcy [ 3 ]. This latter strategy seems to offer a less satisfactory answer to the mechanistic problems: measured k cat values for MetE are smaller than those for MetH by a factor of approximately 50–100. MetE and MetH both activate Hcy by binding the thiolate form of the substrate to Zn +2 [ 4 ]. A similar strategy for alkylation of thiol groups is employed in protein farnesyltransferase [ 5 ], geranylgeranyltransferase [ 6 ], methanol:CoM methyltransferase (MtaA) [ 7 ], the E. coli DNA repair Ada protein [ 8 ], and betaine–Hcy methyltransferase (BHMT) [ 9 ]. However, the sets of zinc ligands and the structures that house the zinc-binding sites are not conserved within this functional family. In particular, the metal ligands and their positions in the sequence are not the same in MetH and MetE. Three cysteines bind the essential zinc in MetH; the first cysteine ligand resides at the end of strand 6 of a (βα) 8 barrel, and the remaining vicinal cysteine ligands follow strand 8. A histidine and two cysteines have been identified as metal ligands in E. coli MetE by a combination of mutagenesis experiments [ 10 , 11 ] and extended X-ray absorption fine structure (EXAFS) measurements [ 12 ]. The relative positions of these residues in the sequence led to the prediction that in a (βα) 8 MetE barrel the histidine and cysteine ligands would reside at the ends of strands 5 and 8 [ 4 ]. In contrast, the sequences of MetE enzymes give few if any clues to the strategy for binding and activation of folate by MetE. Thus, the mode of folate binding is a key question to be addressed by structure analysis. In both MetE and MetH, activation of the leaving group is thought to involve the protonation of CH 3 -H 4 folate in a ternary complex, E·Hcy·CH 3 -H 4 folate in MetE, or E·cob(I)alamin·CH 3 -H 4 folate in MetH [ 4 ]. However, the residues that may facilitate protonation have not been identified for either enzyme. MetE appears to have evolved through gene duplication of a sequence encoding a domain of approximately 340 residues that binds and activates Hcy. Within the family of MetE enzymes ( Figure 1 ), the N- and C-terminal halves exhibit significant sequence homology. The C-terminal half is more highly conserved than the N-terminal half and has homologs in archae and elsewhere. Among these thiol methyltransferases are several enzymes that are approximately half the size of MetE and utilize corrinoid proteins, rather than folates, as methyl donors. Taken together, these observations suggested that the MetE gene arose as the result of a primordial gene duplication event followed by loss of zinc- and Hcy-binding determinants from the duplicated sequence [ 11 ]. If this hypothesis is correct, the two halves of the MetE sequence should display structural homology, and the N-terminal domain should be more closely related to the C-terminal domain than to any other protein in the database. Figure 1 Multiple Alignment of MetE from T. maritima (METE_THEMA), E. coli (METE_ECOLI), Saccharomyces cerevisiae (METE_YEAST), and A. thaliana (METE_ARATH) Conservation in the N-terminal domain is indicated in aqua while conservation in the C-terminal domain is shown in yellow. Zinc ligands His618, Cys620, Glu642, and Cys704 are highlighted in green. The conserved repeat at β4 is marked by asterisks. Main barrel elements are designated β(1–8)F and α(1–8)F and β(1–8)H and α(1–8)H for the N- and C-terminal barrels, respectively. Extension elements are labeled alphabetically and numbered based on the β strand that they follow. For example, α1AF follows β1F and precedes α1F. To determine how MetE has assembled an active site for catalysis of direct methyl transfer from CH 3 -H 4 folate to Hcy, we have solved the crystal structure of Thermotoga maritima MetE at 2.0 Å resolution, along with structures of the binary substrate complexes with Hcy and folate. Difficulties in crystallization of the E. coli enzyme were circumvented by analyzing the MetE from T. maritima . This thermophilic bacterium encodes orthologs of E. coli MetH and E. coli MetE. T. maritima MetE (TM1286) is 41% identical to the E. coli enzyme and is only 19 residues shorter than E. coli MetE ( Figure 1 ), making it an excellent prototype for the MetE family. MetE comprises two (βα) 8 barrels. To our knowledge, it is the first example of a dual-(βα) 8 barrel enzyme in which the active site is located between barrels arranged in a head-to-head orientation. MetE also provides a rare example of a catalytic zinc site in which four residues serve as metal ligands. Repetition of features within the structure supports the idea that MetE evolved through gene duplication of a primordial zinc/Hcy (βα) 8 barrel. Results Description of the Fold and Its Evolution We have determined the structures of several forms of MetE from T. maritima ( Table 1 ), including the zinc-replete binary substrate complexes with folate (the substrate in these experiments was N5-methyl-5,6,7,8-tetrahydropteroyl-(tri)-γ- L -glutamate [CH 3 -H 4 PteGlu 3 ]) at 2.59 Å resolution and with Hcy at 2.20 Å resolution. The two (βα) 8 barrels are formed by residues 1–351 (folate barrel) and 387–734 (Hcy barrel) and joined by an extended inter-domain linker ( Figure 2 A). An unusual feature is that many of the binding determinants for folate actually reside in the C-terminal barrel that binds Hcy. Figure 2 The Fold of MetE and Similarities between the Two Barrels (A) MetE folds into two (βα) 8 barrels. The N-terminal barrel (aqua) is joined to the C-terminal barrel (yellow) by a 35-residue inter-domain linker (gray) that spans 65 Å. Except for the α1 helix (at the right), the linker residues are in extended conformations. This view is along the approximate 2-fold axis that relates the two barrels. The drawing is based on coordinates for zinc-replete MetE in complex with CH 3 -H 4 folate ( Table 1 ). The zinc ligands, zinc, and CH 3 -H 4 folate are shown in ball-and-stick representation. This figure and all subsequent figures were prepared using RIBBONS [ 43 ]. See Figure 1 for the nomenclature used to describe secondary structures. (B) A side-by-side view of the barrels of MetE, arranged to show the similarities of the β–α loop extensions. Major extensions that follow the first four β strands of the barrels are shown in cyan and gold for the N-terminal and C-terminal barrels, respectively. The drawing is based on coordinates for the zinc-replete binary complex with folate (not shown). The active site is located in the C-terminal barrel (on the right) between the extra-barrel β hairpin of the β2–α2 loop and the C-termini of the barrel strands. Zinc is gray and the zinc ligands, His618, Cys620, Glu642, and Cys704, are shown in ball-and-stick mode. Table 1 Data Collection and Refinement Statistics for TM1286-HIS a Statistics calculated using the programs XDS and DENZO/SCALEPACK b Number of unique data assigned as test c Determined experimentally at APS n/a, not applicable In (βα) 8 (triose phosphate isomerase [TIM]) barrel enzymes, the active site is usually located near the C-termini of the inner barrel strands, with catalytic residues contributed by the β–α segments (loops) that join these strands to the outer helices. These barrels are topologically polar with their “tops” decorated by insertions that extend the β–α loop segments. MetE is the first example of a dual-(βα) 8 barrel in which the decorated tops of the two barrels face each other to form a single active site that lies between the domains. A deep cleft between the barrels permits entry of the substrates ( Figure 2 A). As a result of the arrangement of the barrel domains, residues 352–386 of the inter-barrel linker must span approximately 65 Å to connect the bottoms of the two barrels ( Figure 2 A). The N- and C-terminal barrels share a number of strikingly similar features that provide structural evidence for gene duplication. Both barrels incorporate long extensions in the first four β–α loops but, with the exception of α8AF ( Figure 2 B), lack insertions in the last four loops. A pseudo 2-fold axis superimposes these similar extensions ( Figure 2 ). The helical extension at the start of β1–α1, α1AH, augments the side of the C-terminal barrel and is repeated in the N-terminal barrel. The β2–α2 loops that appear in both barrels are the longest extensions in the structure, and we refer to them as the “long hairpin loops.” In the C-terminal barrel that binds Hcy and Zn +2 , the long β2–α2 loop begins with helix α2AH and then forms an antiparallel excursion that harbors a number of conserved residues, some of which are involved in binding folate. The β3–α3 loops both include a short helix, α3A, that carries folate-binding determinants in the Hcy (C-terminal) barrel. The β4–α4A segments of both barrels incorporate a conserved sequence, identified by asterisks in Figure 1 . In E. coli MetE, this sequence, Gln-Ile-Asp-Glu-Pro-Ala, is identical in both barrels. Despite the pseudosymmetry that relates the β–α loops of the two barrels, there are significant differences in the sequences and conformations of these connecting loops that distinguish the functional roles of the two barrels. The major binding determinants for the folate substrate lie primarily in the second, third, and fourth β–α loops of the C-terminal barrel. The equivalent binding site in the duplicated N-terminal barrel has been obliterated. Although the long β2–α2 hairpin of the N-terminal barrel resembles the corresponding hairpin of the C-terminal (Hcy) barrel, the potential binding groove for the folate tail is closed by extensive hydrophobic contacts between α1AF, α1F, and α2CF and by interaction with the long hairpin of the C-terminal barrel. The positioning the long β2 hairpin significantly closer to the barrel top than in the C-terminal barrel occludes the pterin-binding site ( Figure 2 B). Several other adaptations enhance the barrel–barrel interface and appear to influence the relative orientations of the barrels. The β8–α8 loop containing helix α8AF is 16 residues longer than the corresponding loop in the C-terminal domain and makes numerous inter-barrel contacts. Likewise, loop β4–α4 in the C-terminal domain is extended by six residues, increasing contacts with the N-terminal barrel. The single-domain archaeal Hcy methyltransferases that are homologous to the C-terminal domain are missing these six residues, supporting the notion that this insert evolved to enlarge the inter-domain interface. The Zinc Site The zinc-binding site of MetE is distinguished from most catalytic zinc sites by the presence of four protein ligands to zinc. Mutagenesis of E. coli MetE had previously shown that zinc is bound by a histidine and two cysteines [ 11 ] that are equivalent to residues His618, Cys620, and Cys704 in T. maritima MetE. EXAFS studies of the E. coli enzyme indicated a fourth oxygen or nitrogen coordinated to zinc that seemed likely to be a water oxygen. However, the structure of zinc-replete enzyme with folate bound ( Table 1 ) clearly reveals the presence of a fourth protein ligand to zinc. A carboxylate oxygen of the invariant Glu642, which had not previously been identified as a metal ligand, is coordinated to zinc. In the zinc-replete complexes of MetE with folate that provide resting-state structures of the zinc site, the metal–ligand cluster adopts tetrahedral geometry. EXAFS measurements on substrate-free E. coli MetE are also consistent with tetrahedral coordination [ 3 , 11 ] with bond lengths of 2.31 Å for two Zn–S bonds and 2.04 Å for two nitrogen or oxygen ligands. The observed Zn–S bond lengths in our structure are 2.30 Å, Zn–N is 2.07 Å, and Zn–O is 2.14 Å, in good agreement with the EXAFS measurements. The residues that bind zinc are located near the ends of barrel strands 5, 6, and 8 ( Figure 3 ). His618 is the C-terminal residue of β5 and Cys620 is located on the following β–α loop. Glu642 is at the C-terminus of β6, and Cys704 resides on the loop following strand β8. The β–α loops that contain the two cysteine residues are drawn together to form the metal-binding site, distorting the barrel ( see Discussion ). Figure 3 The Resting-State Zinc Is Coordinated in a Tetrahedral Fashion by Four Protein Residues The Binary Complex with Hcy In the complex of Hcy with zinc-replete MetE ( Table 1 ), Hcy is positioned by numerous interactions with conserved protein residues ( Figure 4 ). The amino group is coordinated by hydrogen bonds to Asp577, Glu462, and the carbonyl of Ile409. The carboxyl group of Hcy is bound by the backbone amide and the side chain hydroxyl of Ser411, and hydrophobic contact to the Hcy sulfur is provided by Met468. These interactions with the Hcy substrate are reminiscent of those observed in MetH [ 2 ] ( see Discussion ). Figure 4 The Geometry at the Zinc Ion in Complexes with Hcy (A) Interactions of Hcy in the MetE·Hcy binary complex. The amino group of Hcy is bound by hydrogen bonds to Asp577, Glu462, and the backbone carbonyl of Ile409; the Hcy carboxyl group interacts with the backbone amide and side chain hydroxyl of Ser411. The Hcy sulfur is coordinated to zinc via a long (3.15 Å) bond, which is eclipsed in this view. (B) Superposition of the MetE resting state (gray) and the Hcy binary complex (yellow). Upon Hcy binding, zinc and His618 move away from Glu642 and closer to Hcy, and the zinc site adopts trigonal bipyramidal geometry with three strong equatorial ligands (Zn–N His618 , 2.07 Å; Zn–S Cys620 , 2.23 Å; Zn–S Cys704 , 2.24 Å) and two distant axial ligands (Zn–O Glu642 , 2.90 Å; Zn–S Hcy, 3.13 Å). Zinc moves 0.75 Å toward the substrate in the Hcy complex. Full inversion at zinc, upon tight binding of Hcy to MetE, would displace the metal ion approximately 1.5 Å. (C) The MetH·Hcy complex. The zinc configuration in substrate-free MetH is opposite to that found in MetE; binding of Hcy occurs without inversion in MetH and in BHMT. (D) Difference electron density for the MetE·Hcy complex, showing the geometry at the metal-binding site. The map was computed after simulating annealing and refinement of a model omitting the zinc and its five neighbors. The long Zn–Hcy and Zn–Glu642 interactions are indicated with dashes. Contour levels are 2σ (green) and 6σ (orange). Hcy binding induces significant changes at the metal site ( Figure 4 ). Binding of the substrate sulfur to zinc does not proceed by a simple dissociative reaction in which sulfur is substituted for the oxygen of Glu642. Instead, Hcy approaches the metal ion from the side opposite the Glu642 ligand. This mode of association is unusual; in most catalytic zinc sites the incoming substrate replaces a dissociable fourth ligand without inversion [ 13 ]. In the structure of the binary complex determined at pH 5.2, displacement of Glu642 and inversion at Zn +2 are incomplete. Zinc coordination changes from tetrahedral to distorted trigonal bipyramidal geometry ( Figure 4 ), and the substrate sulfur and glutamate oxygen both exhibit unusually long ligand–metal distances: 3.15 Å and 2.9 Å for Zn–S and Zn–O, respectively. Comparison with the zinc-replete complex with folate ( Figure 4 B) shows that zinc and His618 have moved 0.76 Å and 0.98 Å toward the substrate while Cys620, Glu642, and Cys704 remain essentially fixed. The Zn–S(Cys) and Zn–N distances are not significantly altered, but the zinc and these three ligands become more nearly coplanar ( Figure 4 B). Complete inversion would result in geometry resembling the Hcy complex of MetH ( Figure 4 C). The geometry at the metal site in the Hcy complex has been confirmed by omit refinement and by tests with restrained models. The density around the zinc atom in omit maps is well resolved from Hcy but continuous with that of the cysteine ligands ( Figure 4 D), consistent with a Zn–Hcy distance that is significantly longer than the Zn–Cys bond distances. Imposing tetrahedral restraints in refinements with Hcy as the fourth ligand results in large difference Fourier peaks that also indicate a long Zn–Hcy distance. Cysteinyl tRNA-synthetase incorporates an active-site zinc with the same set of ligands as MetE and provides a precedent for an inversion at the metal center induced by substrate binding. Cysteinyl tRNA-synthetase uses this zinc ion not for catalysis, but to discriminate against serine, exploiting the strong zinc-thiolate interaction with its substrate [ 14 , 15 ]. In the absence of substrate the zinc site displays geometry that is intermediate between tetrahedral and trigonal bipyramidal. As in MetE, the cysteine substrate binds opposite a glutamate residue and does not displace the glutamate ligand directly. Upon cysteine binding, zinc moves away from glutamate and forms a 2.5 Å bond to cysteine. Folate Binary Complex Structures of CH 3 -H 4 folate bound to MetE from T. maritima have been determined in both the reduced (zinc-replete) enzyme and the oxidized (disulfide-bonded) form ( Table 1 ). CH 3 -H 4 folate is bound in identical fashion in both structures. The novel feature of these structures is that the MetE·folate complex fails to comply with the classic picture of substrate binding in a TIM barrel. The folate substrate binds in a deep cleft between the two barrels, with its glutamate tail accommodated by a groove in the enzyme surface ( Figure 5 ). The pterin is displaced from the top of the N-terminal barrel, and simultaneously shifted away from the axis of the C-terminal barrel that binds zinc and Hcy (see Figure 2 A). It is thus inappropriate to call the N-terminal barrel the “folate-binding domain.” An animation, in which Figure 5 is rotated about its vertical axis, provides a more complete view of the structure and its bound ligands ( Video S1 ). Figure 5 Interactions of CH 3 -H 4 folate with MetE (A) A stereoview of T. maritima MetE showing the substrate and metal-binding sites. This is a composite picture in which Hcy from the MetE·Hcy complex has been positioned by superposition on the structure of the MetE·CH 3 -H 4 folate binary complex. The substrates and metal ligands are displayed in ball-and-stick mode; Hcy is in green. (B) A stereoview of the zinc site and bound CH 3 -H 4 folate. Folate is bound by conserved residues in the N-terminal barrel (aqua) and the C-terminal barrel (yellow) with the N5-CH 3 facing away from the zinc, at a distance of almost 14 Å. The pterin interacts with the long hairpins of both barrels and with the extra-barrel helices α3AH and α4AH. The groove that binds the glutamate tail of the substrate is bordered by α1AF, by the β hairpin (β2BH and β2CH) and α1AH of the C-terminal domain, and by the conserved DMV sequence that begins α2AH. The pterin ring of CH 3 -H 4 folate is positioned by stacking and by hydrogen bonding with conserved residues ( Figure 5 ). Glu583 makes a bidentate interaction with the 2-NH 2 and N3 groups of CH 3 -H 4 folate, an arrangement found in several other folate-binding sites. Interaction of N3 with an acidic group has been shown to be important for catalytic activity in dihydrofolate reductase [ 16 ], thymidylate synthase [ 17 ], and MetH [ 18 ]. In the binary complexes with MetE that we report here, the pterin ring of the folate is stacked against Trp539, Lys104 hydrogen bonds to O4, and the folate N5 is hydrogen bonded to a water molecule. However the N8 and N1 positions of the pterin are exposed to solvent. The subsite that binds the glutamate tail is a groove lined by conserved basic residues ( Figure 5 ). Arg15, Lys18, Arg493, and Arg496 interact with the first glutamyl residue, which is the only one of the three γ-linked glutamate residues that is ordered in the binary complex. Weak binding of the other tail residues is surprising for an enzyme that exhibits an absolute requirement for polyglutamylated folate, but glutamate tails have displayed disorder in other structures where they also contribute to strength of binding [ 19 ]. The orientation of bound folate in the binary complexes does not allow transfer of the methyl group to Hcy. As can be seen in Figure 5 , the N5-methyl carbon faces away from zinc and Hcy; it is 11 Å from the sulfur of Hcy when the binary complexes are superimposed. A rotation about the folate N10–C4′ dihedral angle, with the interactions of the para -amino benzoyl moiety and glutamate acting to anchor the substrate, would position the methyl group correctly with respect to Hcy, decreasing the distance between groups that react to 6.0 Å. This distance is still long, and additional protein or substrate rearrangements would be necessary to close the gap between the sulfur of Hcy and the methyl carbon. An alternative route to a ternary complex that supports methyl transfer would be complete dissociation and reassociation of the CH 3 -H 4 folate. However, because so many interactions with conserved residues are observed in this binary complex, it seems likely to represent an initial intermediate rather than a dead-end complex ( see Discussion ). The 467 Asp-Met-Val Sequence Mediates Interaction between the Substrate-Binding Sites One of the fascinating features of the binary substrate complexes is the evidence for communication between the substrate-binding sites, mediated by the invariant 467 Asp-Met-Val (DMV) sequence, which forms the N-terminal turn of helix α2AH. When Hcy binds, the side chain of Met468 alters its position in concert with backbone displacements of the aspartic acid and methionine residues that start the helix. Comparison of the substrate-free structure with the Hcy binary complex shows how the DMV region moves toward the zinc center when Hcy binds ( Figure 6 ). These changes in turn affect the interactions and orientation of Trp539, favoring the conformation in which Trp539 can stack against the pterin ring. In the absence of substrates, Trp539 can adopt another conformation that would overlap the binding site for the pterin ring. Figure 6 Superposition of the Hcy Binary Complex (Yellow) and Substrate-Free (Gray) Enzymes Showing Local Changes in the DMV Region Met468 moves toward the Hcy substrate, rearranging the start of helix α2AH, and a new hydrogen bond is formed between the carbonyls of Met468 and Thr531. This position of Met468 stabilizes a rotamer of Trp539 that favors folate binding. In the binary complex with CH 3 -H 4 folate, the DMV loop is also recruited to the position it occupies when Hcy is bound. The changes that are induced by binding of folate are reproduced in the complex of folate with the oxidized enzyme. Thus, binding of either substrate favors a conformation that would be expected to increase the affinity of MetE for the other substrate. These small but significant conformation changes observed in the binary complexes precede larger rearrangements that must be induced by binding of both substrates to form a competent ternary complex. Cooperativity in substrate binding could increase the concentrations of the ternary complex and thereby increase turnover in a system that is already plagued by slow chemistry [ 4 ]. Discussion Comparisons with MetE from Arabidopsis thaliana A very recent paper, which appeared after the submission of our manuscript, has described the structure of MetE from A. thaliana [ 20 ]. Although the folds of the enzymes from A. thaliana and T. maritima are obviously similar, there are some significant differences in the reported features of the zinc-binding sites. In the complexes of A. thaliana MetE with Hcy or methionine, the distances between zinc and substrate or product sulfur are long, as is the case in the Hcy complex of MetE from T. maritima, but water rather than glutamate has been assigned as the ligand opposite to Hcy or Met. In contrast, the electron density in omit maps of the Hcy complex of T. maritima MetE shows no evidence for a water intervening between glutamate and zinc (see Figure 4 D). In substrate-free A. thaliana MetE, the metal–ligand bonds are all very long and the geometry is highly distorted, suggesting some disordering or partial oxidation of the metal site under the conditions used for crystallization. To study folate binding, PteGlu 5 and CH 3 -H 4 PteGlu 5 were added to crystals of the Hcy or methionine complexes of A. thaliana MetE [ 20 ]. In the resulting structures, the pterin ring is flipped relative to its position in the binary folate complex of T. maritima MetE, and adopts an orientation that is similar to what we anticipated from model building. Although the occupancy of the reduced folate appears to be low, it is estimated that the methyl group of the CH 3 -H 4 PteGlu 5 is about 7 Å from the sulfur of Hcy, too distant for transfer to Hcy. Thus, both structure analyses suggest that additional conformation changes must occur to form a reactive ternary complex. Comparisons of MetE with MetH MetE and MetH display no detectable sequence homology and have different sets of zinc ligands. Comparison of the barrels from MetE with the corresponding domains of MetH that bind folate or Hcy reveal that the N-terminal barrel of MetE, which carries some folate-binding determinants, differs in significant ways from the folate barrel of MetH, whereas the Hcy barrels share many similar features. Two other (βα) 8 barrels that bind CH 3 -H 4 folate have been described: methyltetrahydrofolate corrinoid/iron-sulfur protein methyltransferase [ 21 ] and the folate-binding module of MetH [ 2 ]. These homologous barrels both bind the CH 3 -H 4 folate substrate at the top of the folate barrel and use similar interactions with residues contributed by the C-termini of the inner barrel strands. In MetE, CH 3 -H 4 folate is displaced from the N-terminal barrel and bound primarily by residues in the long extra-barrel β hairpin of the C-terminal Hcy domain (see Figures 2 A and 5 ). Dissimilarities of the decorating loops in the N-terminal barrel of MetE and the folate barrel of MetH are documented by poor statistics for sequence matches and for alignments with the structures of corrinoid/iron-sulfur protein methyltransferase or MetH. The Hcy barrels of MetE and MetH are compared in Figure 7 , which shows how each structure accommodates metal binding. Strand distortions in the Hcy barrels, which have been associated with construction of a metal-binding site [ 2 , 9 ], are related but not identical in MetE and MetH (or BHMT [ 9 ]). In MetH, strand β7 is extruded from the barrel and strands β6 and β8 are pinched together to bind the zinc [ 2 ]. In MetE, strands β6 and β7 are displaced relative to their positions in MetH, allowing strands β5 and β8 to approach one another. In both enzymes, distortion of β8 is accompanied by a splay in strand β1; only one classic hydrogen bond is made between strands β1 and β2. Conserved residues in MetH, MetE, and BHMT stabilize inter-strand interactions by forming side-chain-to-main-chain hydrogen bonds. Figure 7 Superposition of MetE and MetH Zn +2 /Hcy Barrels In MetH (gray) the sulfur of Hcy is positioned close to the center of the barrel for interaction with methylcobalamin. In MetE (yellow) the α1–β1 and α8–β8 connectors make large incursions across the top of the barrel, displacing Hcy to the other side of the barrel. Major differences in the connecting loops β1–α1 and β8–α8 displace the zinc and Hcy sites in MetE relative to MetH by approximately 6 Å so that the sulfur of Hcy is no longer on the barrel axis but is shifted toward one wall of the barrel ( Figure 7 ). The altered positions of the ligands and the binding of zinc by Glu642 lead to inversion of the zinc center relative to its configuration in MetH. Displacement of the Zn +2 /Hcy site and the unusual mode of folate binding seem to have evolved to allow methyl transfer in a ternary E·Hcy·CH 3 -H 4 folate complex. Despite the translation and inversion of the metal and its ligands, the orientation and local interactions of Hcy are almost identical in MetE and MetH ( Figure 7 ). Both enzymes use conserved carboxylate residues to interact with the amino group of Hcy, forming salt bridges (see Figure 4 ). In both MetH and BHMT, the carboxyl group of Hcy is bound by a pair of backbone amides located at the beginning of the β1–α1 extension. In MetE the corresponding interactions of COO − are made by the backbone amide and the side chain hydroxyl of Ser411, again located on the extension following β1. Curiously, it appears that distortions of the barrel strands and connecting loops need not be undone when metal binding is lost through evolution. Distortions of the barrel strands and their downstream loops are retained in the N-terminal barrel of MetE despite loss of metal and Hcy binding, and similar distortions were first observed in uroporphyrinogen decarboxylase [ 22 ], which is also not a metalloenzyme. Despite the lack of detectable sequence homology, uroporphyrinogen decarboxylase is the closest structural relative of the Hcy domain of MetE in the current protein database: alignment using DALI [ 23 ] matches the two folds with a similarity score that is higher than that for the Hcy barrels from MetE and MetH. Uroporphyrinogen decarboxylase may have evolved from a Zn +2 /Hcy barrel, despite the fact that it is no longer a zinc-dependent thiol alkyltransferase. Substrate Binding and Activation: Inferences from the Structures In the structure of the MetE·Hcy complex, the zinc site adopts distorted trigonal bipyramidal geometry with long bonds from the metal ion to glutamate and Hcy (see Figure 4 ). In contrast, EXAFS measurements on the Se–Hcy complex of E. coli MetE at pH 7.2 are best fit to a tetrahedral ligand environment with two sulfurs (2.33 Å), one nitrogen or oxygen (2.02 Å), and one selenium (2.433 Å), in which the longer Zn–Se distance reflects the increased radius of selenium relative to sulfur [ 3 ]. EXAFS studies of the related methyltransferase MT2-A, which has the same set of zinc ligands as MetE, also indicate tetrahedral geometry in the binary complex with coenzyme M (2-mercaptoethanesulfonic acid) [ 7 ]. Both of these studies concluded that the geometry at zinc does not change appreciably upon binding of the thiolate substrate. Because the crystals of the Hcy binary complex of MetE were equilibrated at pH 5.2, it is possible that Hcy may be protonated in the X-ray structure. Thus, the long S–Zn bond and partial inversion at zinc might be explained by the relatively weak interaction between the thiol and the metal ion, as documented for thiol and thioether ligands in model compounds [ 24 , 25 , 26 ]. X-ray studies of Hcy binding at neutral pH and EXAFS measurements at lower pH will be necessary to determine whether pH is a key parameter that affects the metal–ligand geometry. It is possible that complete inversion of zinc geometry will be observed in the structure of the MetE·Hcy complex at neutral pH. A long Zn–S − bond may be functionally important. It has been suggested that a long bond and/or distorted geometry would optimize the reactivity of zinc-dependent thiol alkyltranferases [ 27 ] by increasing the charge on the thiolate sulfur, and would avoid trapping of a lower-energy tetrahedral species. Since zinc is known to have flexible coordination geometry [ 28 ], a five-coordinate state seems plausible, and the observed structure of MetE·Hcy may indeed afford a glimpse of an intermediate or transition-like state that is poised to attack the N5 methyl group of the folate substrate. The motion of Zn +2 that accompanies Hcy binding to MetE is unique among the Hcy methyltransferases with known structures. There is no evidence for inversion of configuration at zinc in MetH, and inversion is precluded in BHMT, where a leucine occupies the position analogous Glu642 of MetE. Zinc motion provides a novel way to alter the distribution of charge among the zinc and its ligands, thereby modulating thiolate reactivity. The effects on the electronic structure could be larger than those resulting from the changes in bond lengths observed in other zinc-dependent alkyltransferases [ 29 , 30 ]. By analogy with a proposal for the reaction cycle of protein farnesyltransferase [ 30 ], reassociation of Glu642 could promote dissociation of the methionine product following methyl transfer. An unexpected feature of the MetE structures is the binding mode of CH 3 -H 4 folate, with the pterin ring incorrectly oriented for methyl transfer. It is possible that folate binds initially in this manner to avoid blocking access to the Hcy binding site, which lies between zinc and CH 3 -H 4 folate. Space-filling models show that the Hcy-binding site remains accessible in the MetE·CH 3 -H 4 folate binary complex. The observed binding mode thus permits random addition of substrates but does not rule out a kinetically preferred order of binding. In both MetH and MetE, protonation of methyltetrahydrofolate is thought to occur in the ternary complexes but not in the binary folate complexes [ 4 ]. The immediate proton donor has not yet been identified by enzymatic or biochemical studies of E. coli MetE. Formation of the binary Hcy complex results in release of a proton to solvent (Z. S. Zhou and R. G. Matthews, unpublished data). Thus Hcy is not a likely proton donor. It is more likely that an active-site residue may serve as an acid catalyst. Structures of the binary complexes do not implicate a particular residue as a general acid catalyst, but suggest possible candidates. His111 could serve as a proton donor through water if protonation were to occur before folate rearrangement. His672, part of a conserved 670 Asp-Ile-His-Ser-Pro sequence, or Asp467, located in the DMV loop, may be positioned to serve as donors if protonation occurs after folate rearrangement. All three of these candidate residues are invariant in multiple sequence alignments. Gene Duplication of a Sequence Encoding the Hcy Barrel Earlier comparisons of sequences had suggested that MetE evolved by gene duplication. The two domains of E. coli MetE share a conserved Gln-Ile-Asp-Glu-Pro-Ala repeat [ 31 ], and the sequences display 39% identity (50% similarity) within the regions now seen to span β3 to α4A of the two barrels. The pseudosymmetry of the structural features decorating the barrels provides compelling evidence for a close relationship between the two halves of MetE, verifying the inferences based on sequence alignments. The cores of (βα) 8 barrels, though they display characteristic distortions and can be classified into subgroups [ 32 ], do not provide as strong evidence for relatedness as do the similarities of regions inserted in intervening loops. It is difficult to ascertain whether the regions responsible for folate binding were inserted before or after gene duplication. The long β2–α2 loop in the N-terminal barrel and corresponding sequences in archaeal relatives of MetE favor the notion that precursors of folate-binding segments were present before duplication. In any case, folate-binding determinants have developed or been retained primarily in the C-terminal Hcy barrel but not in its N-terminal replicate. The idea that the zinc/Hcy barrel is the ancestral fold is based on several lines of evidence. This barrel shows significant structural homology to a broad family of zinc-dependent thiol methyltransferases, including not only MetH and BHMT but also the single-domain archaeal transferase enzymes that react with methylcobalamin. Indeed it seems likely that the three enzymes that convert Hcy to methionine, MetE, MetH, and BHMT, are all descended from a primordial zinc/Hcy barrel. In contrast, DALI searches that assess structural similarities [ 23 ] reveal that the N-terminal barrel is more similar to the C-terminal barrel of MetE than to any other known protein, suggesting that its immediate precursor is the Hcy barrel of MetE. Gene duplication of a Zn +2 /Hcy barrel would have replicated the sites that bind Zn +2 and Hcy, but these sites have been disabled in the N-terminal barrel of contemporary MetE. Disruption of zinc and Hcy binding is effected by both residue mutations and backbone conformational changes ( Figure 8 ). Mutation of the equivalent of Cys620 in the Hcy barrel to Tyr232 leads to stacking with the adjacent Tyr233 (phenylalanine in most MetE sequences) and results in a major backbone conformational change. The hydroxyl group of the Tyr232 forms a hydrogen bond to Asn199 of the conserved Leu-Val-Asn-Glu-Pro-Ala sequence at the β4–α4 loop and thus removes a crucial Hcy-binding determinant. Although Cys309, the equivalent of Cys704 of the C-terminal barrel, is retained in most MetE sequences, mutation of the other zinc ligands and many of the Hcy-binding residues leads to a complete overhaul of the binding site for Hcy. Figure 8 Disruption of Hcy and Zinc Binding in the N-Terminal Barrel Overhaul of the Zn +2 /Hcy site in the N-terminal domain (aqua) following gene duplication is accomplished through mutation and small backbone displacements. The competent Zn/Hcy site from the C-terminal barrel is at the left; the remodeled site from the N-terminal barrel is at the right. Important substitutions that disable Hcy and zinc binding are Trp146, Phe230, Tyr232, and Asp253 (see text). Why Is a Second Barrel Recruited for a Ternary Complex Mechanism? Although an entire barrel domain is recruited to elicit direct methyl transfer from folate to Hcy, the structure reveals that the functional groups of this domain are mostly disabled. The binding determinants and potential catalytic residues are located primarily in extensions and excursions of the C-terminal Zn +2 /Hcy barrel rather than in the “new” N-terminal domain. Duplication and divergence of an entire barrel is an elaborate strategy to accommodate a second relatively large substrate at a single active site, and MetE provides the first instance of the face-to-face barrel construction that is required to build such an active site. A more common strategy to accommodate a ternary complex is exemplified by the related Hcy methyltransferase, BHMT. In this enzyme the site for the rather small second substrate, betaine, is constructed in part from β–α barrel extensions and in part from a dimerization arm that is appended to the barrel and contributes to the active site of the partner chain of the functional dimer. The N-terminal domain of MetE may nevertheless be essential for ternary complex formation and catalysis. In the family of thiol alkyltransferases, quenching of opposite charges on Hcy and the alkyl donor is believed to drive the reactions, and a hydrophobic environment [ 33 ] and desolvation [ 34 ] of reactants may be critical for reactivity. Rearrangement of folate to form a viable ternary complex may be accompanied by rearrangement or closure of domains around the substrate-binding cleft that would reposition key residues and isolate the active site from solvent. Hcy binding is accompanied by a contraction of the top of the C-terminal barrel that alters the relative positions of the N- and C-terminal barrels. This small domain shift clearly gives hints of inter-domain flexibility. On the other hand, the commissioning of the N-terminal domain may reflect an evolutionary strategy in which gene duplication is utilized as the most facile way to recruit additional sequences. Structures of ternary complexes, obtained using mutant enzymes and/or substrate analogs, should provide further insights into the methyl transfer reaction and the functional roles of the N-terminal domain. Materials and Methods Cloning and purification The TM1286 gene was PCR amplified from T. maritima genomic DNA ( ATCC, Manassas, Virginia, United Stats) and cloned into pET-151D/TOPO (Invitrogen, Carlsbad, California, United States). The expression construct contains an N-terminal leader sequence consisting of a 6X histidine tag followed by a V5 epitope, rTEV cleavage site, and residues 2–734 of the coding sequence of TM1286. This vector was overexpressed in BL21(DE3)Star (Invitrogen) by induction with 0.8 mM IPTG for 8 h in LB media supplemented with zinc sulfate. The histidine-tagged protein was purified by a 10-min 70 °C heat step, which precipitates heat-labile protein, followed by affinity chromatography on Zn(II)-NTA and elution with a 50 mM to 1.5 M glycine gradient. The protein was dialyzed against 50 mM Tris (pH 7.4) and 500 μM Tris(2-carboxylethyl)phosphine (TCEP) and concentrated to 20 mg/ml. Selenomethionine-labeled protein was expressed in M9 medium supplemented with amino acids and zinc sulfate. Crystallization TM1286 was crystallized by the vapor batch (microbatch) method under oil utilizing 96-well Douglas vapor batch plates and a 1:1 mixture of silicon:paraffin oil (Hampton Research, Alliso Viejo, California, United States). Orthorhombic crystals of space group P2 1 2 1 2 ( a = 163.57 Å, b = 158.76 Å, c = 64.16 Å, α = β = γ = 90°) were grown by mixing 20 mg/ml protein 1:1 with 25% poly(ethylene glycol) 4000, 0.2 M ammonium sulfate, and 0.1 M sodium acetate (pH 4.6). Selenomethionine-labeled protein was crystallized by mixing 1:1 with 12% poly(ethylene glycol) 4000, 0.2 M ammonium sulfate, and 0.1 M sodium acetate (pH 4.6). Anapoe detergents (Anatrace, Maumee, Ohio, United States) were used as additives and seen to have a favorable effect on crystal morphology. Crystals in a cryoprotectant solution consisting of 15% poly(ethylene glycol) 4000, 0.12 M ammonium sulfate, 0.1 M sodium acetate (pH 5.2), and 12.6%–14.1% meso-erythritol were flash-cooled in liquid nitrogen. These crystals were found to be depleted of zinc with a disulfide bond connecting the zinc ligands, Cys620 and Cys704. Zinc-replete crystals were obtained by soaking in cryoprotectant solution containing 500 μM zinc sulfate and 500 μM TCEP for 4–16 h prior to flash cooling. The enzyme:folate binary complex was formed by soaking crystals in cryoprotective solution with added 3.5 mM CH 3 -H 4 PteGlu 3 (a gift from Rebecca E. Taurog) for several hours. The enzyme:Hcy binary complex was formed by soaking crystals pre-equilibrated with zinc sulfate and TCEP in cryoprotective solution containing 10 mM L -Hcy (a gift from Rebecca E. Taurog) for several hours. Phasing and refinement All datasets were collected at the Advanced Photon Source (APS) at Argonne National Laboratory. Data collected at the DND-CAT beamline on a Mar225 detector were processed with XDS [ 35 ] whereas those collected at COM-CAT on a Mar165 detector were processed with DENZO/SCALEPACK [ 36 ]. Statistics for the datasets appear in Table 1 . Experimental phases to 2.8 Å were derived from selenium SAD measurements at the selenium peak using heavy atom sites located by a three-wavelength selenium MAD experiment at lower resolution. Thirty of the 36 expected selenium sites were found and phases were determined using SOLVE version 2.05 [ 37 ], and statistical density modification was performed in RESOLVE [ 38 ]. The initial model from RESOLVE was rebuilt in MI-fit [ 39 ] and partly refined, and was used with the molecular replacement program EPMR [ 40 ] to determine the higher resolution structure of oxidized MetE at 2.00 Å. A search model derived from the refined oxidized structure was subsequently used to solve the substrate complexes with EPMR. All models were developed by refitting and rebuilding in MI-fit alternated with refinement in CNS version 1.1 [ 41 ]. Refinement protocols included simulated annealing with torsional dynamics, coordinate minimization, and adjustment of individual B-factors. In late rounds of refinement of the binary complexes, weak restraints were applied to maintain the geometry at the zinc site. For the MetE·CH 3 -H 4 folate complexes (the resting state of the zinc center), restraints were based on ideal tetrahedral geometry; for the Hcy complex, restraints were chosen using bond valence sums analysis [ 42 ] to make the bond lengths compatible with the known +2 oxidation state of zinc. The resting-state zinc site (Zn–N His618 , 2.07 Å; Zn–S Cys620 , 2.31 Å; Zn–S Cys704 , 2.29 Å; Zn–O Glu642 , 2.14 Å) gives a bond valence sum of 1.88, consistent with the known +2 oxidation state of zinc, and a net contraction of Zn–S Cys bonds from 2.30 to 2.24 Å upon Hcy binding is enough to maintain the bond valence sum for the five-coordinate state with long axial bonds. Supporting Information Coordinates of the structures have been deposited in the Research Collaboratory for Structural Bioinformatics' Protein Data Bank ( http://www.rcsb.org/pdb/ ) with accession codes 1T7L (substrate-free oxidized), 1XDJ (zinc and Hcy complex), 1XPG (zinc and methyltetrahydrofolate complex), and 1XR2 (oxidized methyltetrahydrofolate complex). Video S1 Video of MetE Showing the Substrate and Metal-Binding Sites (7.9 MB WMV). Click here for additional data file. Accession Numbers The SwissProt ( http://www.ebi.ac.uk/swissprot/ ) accession numbers for the gene products discussed in this paper are A. thaliana MetE (SwissProt O50008), E. coli MetE (SwissProt P25665), S. cerevisiae MetE (SwissProt P05694), T. maritima MetE/TM1286 (SwissProt Q9X112), and MetH (SwissProt P13009).

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539065.xml

555746

Functional characterization of two newly identified Human Endogenous Retrovirus coding envelope genes

A recent in silico search for coding sequences of retroviral origin present in the human genome has unraveled two new envelope genes that add to the 16 genes previously identified. A systematic search among the latter for a fusogenic activity had led to the identification of two bona fide genes, named syncytin-1 and syncytin-2, most probably co-opted by primate genomes for a placental function related to the formation of the syncytiotrophoblast by cell-cell fusion. Here, we show that one of the newly identified envelope gene, named env P(b), is fusogenic in an ex vivo assay, but that its expression – as quantified by real-time RT-PCR on a large panel of human tissues – is ubiquitous, albeit with a rather low value in most tissues. Conversely, the second envelope gene, named env V, discloses a placenta-specific expression, but is not fusogenic in any of the cells tested. Altogether, these results suggest that at least one of these env genes may play a role in placentation, but most probably through a process different from that of the two previously identified syncytins.

Findings Endogenous retroviral sequences represent approximately 8% of the human genome. These sequences (called HERVs for Human Endogenous Retroviruses) share strong similarities with present-day retroviruses, and are the proviral remnants of ancestral germ-line infections by active retroviruses, which have thereafter been transmitted in a Mendelian manner (reviewed in [ 1 - 3 ]). The 30,000 HERV elements have been grouped according to sequence homologies into more than 80 distinct families (each originating from the same founder element), based on a systematic listing of human repeats in the Repbase database [ 4 ]. Most of these elements are non-coding due to the accumulation of mutations, deletions, and/or truncations. A screening of the human genome for retroviral envelope genes with coding capacity, based on a specific envelope protein motif and on the HERV families described in Repbase, has revealed 16 fully coding envelope genes, transcribed in several healthy tissues [ 5 , 6 ], among which two (syncytin-1 and syncytin-2) possess a fusogenic activity [ 7 , 8 ]. Using another approach, based on BLAST searches with various retroviral sequences as queries, a recent elegant study has analyzed the coding potential of human retroviral sequences and two additional fully coding envelope genes have emerged from this screen [ 9 ]. These two envelope genes do not belong to the HERV families listed in Repbase. The first one was designated "HERV-W/FRD-like" env , due to partial homology with syncytin-1 and syncytin-2, encoded by proviruses of the HERV-W and HERV-FRD families, respectively [ 7 , 8 ]. The second one was designated "ZFERV-like" env , due to its homology with the envelope protein encoded by a provirus recently discovered in the zebrafish genome [ 10 ]. The sequences and predicted hydrophobic profiles of the two proteins (renamed here EnvV and EnvP(b) respectively, see below), disclose the characteristic signature of retroviral envelope proteins, with a putative proteolytic cleavage site between the SUrface (SU) and TransMembrane (TM) moieties, and a hydrophobic transmembrane domain within the TM subunit which permits its anchorage to the membrane (Figure 1A ). Figure 1 A) Hydrophobicity profile and predicted features of the EnvV (formerly W/FRD-like env) and EnvP(b) (formerly ZFERV-like env) proteins. The SU (Surface Unit) and TM (TransMembrane) moieties of the envelopes are delineated, with the position of the putative proteolytic cleavage site (consensus, R/K-X-R/K-R) between the two subunits and the « CWLC » motif (consensus, C-X-X-C) indicated. The hydrophobic regions associated with the fusion peptide and the transmembrane region are shaded in light gray, and the putative immunosuppressive domain (ISU) in dark gray. B) Phylogenetic tree of retroviral envelopes and position of the newly identified genes. The tree is based on an alignment of approximately 180 amino acids corresponding to the extracellular and transmembrane domains of the TM subunit of envelope proteins. The protein alignment, phylogenetic tree and bootstrap analysis were performed with the ClustalW program (neighbour joining option). The tree was viewed by using the TreeView program. The scale bar indicates 10% aa sequence difference. The phylogenetic tree determined by the parsimony method was congruent with the neighbour joining tree (data not shown). The two "new" V and P(b) env genes are represented in red, ERV env genes from other species and exogenous retroviruses in blue. The sequences used for the alignments were those of the consensus element of each family, or the coding env gene when present. The consensus sequences of the HERVK(HML-9), HERVFXA21B1 and HERVFXA21B2 families, which are not listed in Repbase, were each inferred from the comparison of 3–6 sequences. Abbreviations: MoMLV, Moloney Murine Leukemia Virus; FeLVA, Feline Leukemia Virus strain A; PERVC, Pig Endogenous Retrovirus strain C; GALV, Gibbon Ape Leukemia Virus; MPMV, Mazon-Pfizer Monkey Virus; MMTV, Mouse Mammary Tumor Virus; JSRV, Jaaksiekte Sheep Retrovirus; HTLV, Human T cell leukemia Virus; BLV, Bovine Leukemia Virus; HIV, Human Immunodeficiency Virus. C) Genomic organization of the env V and env P(b) loci. The envelope ORF (open box) with gag- and pol- related sequences (hatched boxes) and long terminal repeats (black boxes), Alu (dark gray boxes) and MER51B (light gray boxes) retroelements are indicated. Consensus PBS sequences (obtained from two sequences for the HERV-V family and from four sequences for the HERV-P(b) family) are indicated above the corresponding provirus, together with the PBS for the Val and Pro-tRNA, respectively. D) env V and env P(b) mRNA expression in a panel of 19 healthy human tissues, as determined by real-time quantitative RT-PCR. RNAs from human tissues were prepared as described in [6]. The reaction was performed using Sybr Green Master Mix (Applied Biosystems). PCR was developed using an ABI PRISM 7000 sequence detection system. Primer sequences (5'-3') were as follows: (CATGACTTTGGAAAAGGAGG) and (GCCAAAGAGGAAAAGTAAGAGT) for env V; (CAAGATTGGGTCCCCTCAC) and (CCTATGGGGTCTTTCCCTC) for env P(b). The transcript levels were normalized relative to the amount of 18S mRNA (as determined with the primers and TaqMan probe from Applied Biosystems). Samples were assayed in duplicate. PBL, peripheral blood lymphocytes. E) Assay for fusogenicity of env V and env P(b). XhoI containing primer sequences (5'-3') were as follows: (ATCACCTCGAGACACTCCATCGAACCACTTCAT) and (ATCACCTCGAGGGCTGTTCTAGGATGGGTTATT) for env V; (ATCACCTCGAGAGAAGAGAAACTTGAACCGTCC) and (ATCACCTCGAGGGGCTGATAGATGAATGGGTAT) for env P(b). The PCR products were cloned into the phCMV-G vector, opened with XhoI, and the constructs were verified by partial sequencing. Cell lines and fusion assays are as described in [12], except for the SH-SY5Y neuroblastoma cell line (ATCC number CRL-2266). Since these genes belong to previously uncharacterized HERV families, we first analyzed their phylogenetic relationship with known HERV families and animal retroviruses. We generated a phylogenetic tree of endogenous and exogenous retroviruses based on the env gene, namely on the alignment of a conserved domain of the transmembrane (TM) subunit [ 3 , 5 ]. In this tree (Figure 1B ), the "HERV-W/FRD-like" env gene is closely related to that of MER66, MER84 and Z69907 families. This gene seems to be part of a very degenerate proviral structure, with only the LTR being identifiable (see below and Figure 1C ). As mentioned in [ 9 ], a highly homologous gene (95.7% identity at the nucleotide level) encoding an envelope protein truncated due to a frameshift can be found 40 kb downstream. This cognate env gene is unambiguously part of a proviral structure, displaying just upstream of it the 1.6 kb open reading frame of a gag gene, followed by a pol -like non coding region (data not shown. The flanking sequences of both proviruses are distinct. No other provirus or env gene belonging to this "family" can be found in the human genome by a BLAST search on the Ensembl database. Approximately 4 kb upstream of each of these two env genes, as expected, the RepeatMasker program that screens DNA sequences for interspersed repeats present in mammalian genomes identifies 5' LTR sequences (or fragments of LTR sequences). 3' LTRs are also found just downstream of the envelope genes (see Figure 1B for the map of the fully coding env gene locus). The analysis of the PBS (Primer Binding Site) region located downstream of the two 5' LTRs of this family reveals a high degree of homology to the PBS for Val-tRNA (Figure 1C ), so we propose to name this new family HERV-V. The "ZFERV-like" env gene clusters, in the TM-based tree, with the "HERV-I superfamily", which indeed also includes the ZFERV env from zebrafish (see Figure 1B ). As indicated in the retrosearch database , this envelope gene is part of an identifiable provirus (see Figure 1C ). A BLAST query on the Ensembl database using the provirus sequence showed that this new HERV family contains three additional members. All four HERV elements, harbouring a proviral LTR-gag-pol-env-LTR structure (although the only coding gene is the env gene described in [ 9 ]), are close to – but yet unambiguously distinct from – the HERV-IP family. The analysis of the PBS region of these four proviruses reveals a high degree of homology to the PBS for Pro-tRNA (see Figure 1C ), so we propose to name this new family HERV-P(b) (since the HERV-P family already exists, [ 11 ]). To determine whether these two genes could play a role in human placentation, we then characterized their expression pattern and fusogenic properties, as previously performed for the 16 coding envelope genes already identified [ 6 , 8 ]. To get insight into their expression profile, we used a Real-Time RT-PCR strategy as described in [ 6 ]. In this study, specific primers had been designed for Sybr Green amplification in such a way that only env genes with an open reading frame would be amplified among all the envelope genes of a given family, by positioning them within domains of maximal divergence between the coding and the non-coding copies. For the HERV-V coding envelope, the primer pair was designed in the 3' part of the gene, where the two env V genes are the most divergent (79% identity in the last 200 nt). An additional primer pair was also designed to monitore the expression of the truncated HERV-V env gene. To assess the specificity of each primer pair for the corresponding env gene, the PCR products obtained upon amplification of genomic DNA were cloned into a pGEM-T vector and 6 clones per amplicon were sequenced. In each case, the 6 sequences corresponded to the expected env gene. Analysis of the expression level of the coding env P(b) and env V genes was achieved on a series of 19 healthy human tissues, and the results are represented in Figure 1D . The expression pattern of env V was found to be placenta-specific. Interestingly, the truncated envelope of the HERV-V family is highly expressed in the placenta as well, but poorly in other tissues (data not shown). Env P(b) expression, on the other hand, was observed at a rather low level in almost all the tissues tested, without any specificity for the placenta. Among the 16 coding env genes of the human genome tested in [ 8 ], only two, namely env W (syncytin-1) and env FRD (syncytin-2), had been found to be fusogenic in an ex vivo assay. As these two env genes were highly and specifically expressed in the placenta, it was suggested that they are involved in a major physiological process within this organ, namely fusion of the cytotrophoblast cells to form the syncytiotrophoblast layer. The two newly identified env genes were therefore similarly tested. To do so, they were first cloned and introduced into a eukaryotic expression vector. The env P(b) gene was PCR-amplified from the DNA of BAC RP11-828K24 by using a proofreading DNA polymerase and running a 15-cycle PCR reaction, whereas the env V gene -not available as BAC DNA- was PCR amplified from the genomic DNA of a Caucasian individual using the Expand long template enzyme mix (Roche Applied Science). Both env genes were then assayed for cell-cell fusion on a large panel of mammalian cells (known to express on the whole the receptors for all retroviral envelopes identified to date) using a transient transfection assay and two clones from each construct. As shown in Figure 1E , cell-cell fusion was observed in five out of nine cell lines tested for env P(b), and in none of them for env V. The truncated envelope protein member of the HERV-V family was also tested and, as expected, was not fusogenic (data not shown). In some respect, these results are surprising. Indeed, the putative protein encoded by env P(b) is fusogenic despite the absence of a canonical fusion peptide, i.e. of a hydrophobic region located at the N-terminus of the putative TM subunit, just downstream of the SU-TM cleavage site (see Figure 1A ). Conversely, the env V gene product, notwithstanding its canonical sequence, is not fusogenic (at least in the panel of cells tested). To check that the lack of fusogenicity of the latter gene is not due to a fortuitous gene polymorphism of the env V gene from the selected individual, we PCR-amplified, cloned and assayed the env V gene from two other individuals (for both the complete and the truncated env V genes): no cell-cell fusion was observed either (data not shown). Finally, we identified and cloned the chimpanzee orthologous env V gene (which is fully coding as well): neither did it display any fusogenic activity in our assay (data not shown). In conclusion, the present analysis shows, rather paradoxically, that the envelope protein with fusogenic properties is not placenta-specific, whereas the one which is exclusively expressed in the placenta -a characteristic pattern of the two previously described fusogenic syncytin-1 and syncytin-2 gene products- is not fusogenic. In this respect, these results suggest that the two newly identified env V and env P(b) genes are most probably not "syncytin-like" genes, sensu stricto . Additional experiments should now be devised (e.g. search for conservation among primates, search for Single Nucleotide Polymorphisms) to assess their role -if any- in human physiology. List of abbreviations HERV, human endogenous retrovirus; TM, transmembrane; LTR, Long Terminal Repeat; PBS, Primer Binding Site. Competing interests The author(s) declare that they have no competing interests. Authors' contributions SB carried out the cloning of the env genes and the cell-cell fusion assays. NdP analyzed the sequences, constructed the phylogenetic tree, designed and carried out the Real-Time RT-PCR experiments, and drafted the manuscript. TH conceived the study.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555746.xml

544878

Blood pressure demographics: nature or nurture ... ... genes or environment?

Hypertension is a growing worldwide problem associated with an increased risk of cardiovascular morbidity and mortality. However, the rates of prevalence of hypertension are higher in some populations than others. Although ethnic and genetic factors have been implied in the past to explain this, the environmental influence and psychosocial factors may play a more important role than is widely accepted. Examining the non-genetic influences in future hypertension research may be necessary in order to clearly define the local blood pressure demographics and the global hypertensive disease burden.

Hypertension is a common problem, with a consistent and continuous risk of cardiovascular disease and stroke associated with rising blood pressure levels [ 1 ]. Furthermore, effective treatment of blood pressures has been shown to cause reductions in morbidity and mortality from cardiovascular disease and stroke. The modern management of hypertension is even more complex, with the emergence of newer therapies, ageing populations and new clinical trial evidence, as well as the need for multiple agents to achieve target blood pressures, which are much lower than they used to be in the past [ 1 ]. The consequences of poor blood pressure control are huge. As high blood pressure is the most important risk factor for cardiovascular disease, it has been calculated that by achieving the target of 140 mmHg, there would be a reduction of 28–44% in stroke and 20–35% in ischaemic heart disease depending on the age. This would prevent approximately 21400 stroke deaths and 41400 ischaemic heart disease deaths each year – and these translate to approximately 42800 strokes and 82800 ischaemic heart diseases saved, making a total of 125600 events saved per year in the United Kingdom alone [ 2 ]. Even white coat hypertension is by no means a benign condition [ 3 ]. By 2020, the world population would be an estimated 7.8 billion people and hypertension currently is 'estimated' to affect about 1 billion worldwide – this figure will be rising. The growing numbers and the lack of concerted effort to tackle the burden of hypertension makes depressing reading. Nonetheless, what is more intriguing and perhaps still not fully explained, is why some populations seem to have a much higher population prevalence of hypertension as compared to others. For instance, the prevalence and incidence of hypertension differs between the non-westernised and westernised populations. Even within the western world, the Afro-Caribbean or African-American black population has a higher prevalence of hypertension and target organ damage related to it, as compared to white Europeans or Americans [ 4 ]. Differences also exist within the same region, for example, with people of Eastern European origins having a higher prevalence of hypertension compared to elsewhere in Europe [ 5 ]. Understanding the reason(s) behind these geographical and ethnic differences would help devise effective measures in primary prevention. Cooper et al [ 6 ], writing in BMC Medicine , address the issue of whether there is a truly genetic predisposition or perhaps an environmental influence is to blame for higher rates of prevalence of hypertension seen in some of these ethnic populations. In a well-designed pooled analysis, incorporating eight studies involving 8 white and 3 black populations from the North American, European and African populations – a dataset of nearly 85,000 patients – Cooper et al [ 6 ] examined patterns of blood pressure distribution in the different ethnic groups across three continents. They found a wide variation in hypertension prevalence among white and black racial groups, and the rates among blacks were not unusually high when viewed internationally. They therefore suggest that the impact of environmental factors among black and white populations may have been under-appreciated. Specifically, and perhaps contrary to expectations, the prevalence of hypertension was lower amongst the white peoples in Northern America and Canada, as compared to Europe. Does this take us back to the drawing board? Perhaps environmental factors do play a more major role in developing hypertension than is widely accepted. Indeed, does urbanisation per se together with the unhealthy lifestyle and diet in the western world increase the risk of hypertension, compared to the rural, 'low stress', healthier lifestyle and dietary habits in Africa? Perhaps the genotype of black subjects was not idealised for the 'pro-hypertension' environment of the western world, leading to the greater risk of developing hypertension amongst blacks in the western world. This 'genetic predisposition' of certain ethnic groups, coupled with the 'wrong' environment, leads to an unhealthy combination that predisposes to cardiovascular disease [ 7 ]. However, the sociological definition of an ethnic group would be "people of the same race or nationality who share a common and distinctive culture", as it is impossible to consistently classify people by race. Genetic analyses have found more genetic variation within one ethnic group than between one group and another [ 8 ]. Therefore, race or ethnicity may appear to be more defined by customs, traditions, language and history than purely by genotype alone. Indeed, classification of race or ethnicity or skin colour, for example, is pretty subjective, imprecise and unreliable. Evidence for this exists in the differences in coronary risk factors in Indians, Pakistani and Bangladeshi populations in a British city, eventhough together these people might have been classed as 'Indo-Asian' but are clearly different [ 9 ]. Similarly, a Scottish Highland crofter is quite different from a Swede businessman, who would again be quite different from a Greek fisherman, although all would be ethnically classified as 'white caucasian'. Can this 'wrong genotype in the wrong environment' hypothesis be applied to hypertension in black African-Americans? Efforts have already been made to understand the reasons behind the higher prevalence of hypertension in African Americans, with the underlying assumption that there may a genetically determined predisposition as compared to their white counterparts; however, no convincing data are available [ 10 ]. This may be because multiple genes determine human hypertension, at least in the vast majority of cases [ 11 ], and genetic factors have not been able to fully account for the differences in blood pressure prevalence between ethnic groups. Furthermore, elsewhere, black populations migrating to countries like UK have been shown to have similar blood pressures compared to the white UK population [ 12 ]. Numerous potential explanations for the higher prevalence of hypertension in blacks have been proposed. Genetic mechanisms have been used to explain familial aggregation of hypertension in Jamaican blacks and the intra-class correlation of systolic blood pressures amongst black twins [ 13 , 14 ]. Low renin levels noted in the USA black population have been hypothesised to be the result of a genetic 'maladaptation' which though benefited their earlier black ancestors to survive the torment of a transatlantic voyage under slavery, later turned out to be detrimental to survival due to the resultant avid salt retention [ 15 ]. Indeed, increased sodium sensitivity, abnormalities in sodium transport, increased vascular responsiveness to pressor stimuli, association between stresses of low socio-economic status and hypertension, and insulin resistance have also been suggested. Furthermore, ethnic differences in response to anti-hypertensive therapy are also well documented [ 4 ], with a potent effects of diuretics and calcium antagonists in black patients, compared to a relatively poor response to beta blockers and drugs that act on the renin-angiotensin system (such as the ACE inhibitors and angiotensin receptor blockers). Notwithstanding the shortcomings of a retrospective, pooled study, with varying criteria for inclusion, and the intra-, inter- and cross-study observational errors, Cooper et al [ 6 ] have shown that the burden of hypertension seems to be more amongst the white population in Europe and that the global rates of hypertension amongst the black population are less in comparison. This study therefore sets the stage for a closer examination of data across geographic areas and calls for more stringent, standardised and possibly nationalised surveillance of blood pressure trends. If nothing else, the data suggests that inferences from cross-sectional studies done in certain geographic areas of a differing socio-economic stature cannot be extrapolated as logical benchmarks for other areas. Unintentionally perhaps, it beckons 'the scientist' to take a second look into the effect of other non-genetic mechanisms to explain the paradoxical findings. Surely more studies of a larger size are needed to confirm these implications. Evidence that environmental and particularly psychosocial factors are important in the development of hypertension also comes from a series of epidemiological studies conducted in the early part of the last century, which have shown that urbanisation and adoption of a westernised life style leads to blood pressure rises. Many of these studies have been conducted in Africa, where the rural populace has a relatively low prevalence of hypertension. For example, primitive black populations living in more frugal circumstances in rural Africa have been shown to have low blood pressures. Evidence suggests that there are some populations that exist who are naïve to hypertension and related morbidity. Interestingly, their blood pressures hardly rise with increasing age, a common response to age in many other (urbanised?) communities [ 16 - 18 ]. The most important characterising feature of these populations were that, their lifestyle was traditional and they had not adopted or been under the influence of beliefs, customs or practices of another culture alien to theirs; the so-called 'unacculturated societies', with a close resemblance to the 'hunter-gatherer' lifestyle of primitive man. The other interesting feature observed was the constancy of electrolyte intake in the diet in these populations, which was in sharp contrast to the more 'developed' Western populations [ 19 ]. Migration, not withstanding the complexity of the studies in these populations, has been shown to significantly affect blood pressures. A study examining the migrant islanders into New Zealand showed raised population blood pressures, as well as an increased slope of the age-blood pressure relationship [ 20 ]. The Kenyan Luo migration study examined migration of rural tribes to the capital city Nairobi, also found that urbanised populations had higher mean blood pressures [ 21 ]. Chronic and excessive alcohol ingestion may also adversely affect blood pressure [ 22 ]. The relationship of hypertension with obesity has been demonstrated but weight loss seemed to have more pronounced blood pressure reductions in whites rather than blacks. The issue of the influence of colour of the skin to blood pressures is even more complex. On the one hand, studies in America have shown relationship between the dark skin and blood pressures, leading to some suggesting that the link is genetic. In contrast, some argue that it is a manifestation of the stress and social pressure of having a dark skin that causes this. In Cuba, for example, where communist principles are considered to have broken the racial barriers, the ethnic differences in blood pressures were shown to be small, supporting the latter argument [ 23 ]. In addition, there may even be an effect of neighbourhoods or the social environment on blood pressure and cardiovascular disease [ 24 , 25 ]. Thus, the process by which a society becomes more economically advanced or "developed" seems to be associated closely with rates of hypertension prevalence. Indeed, lifestyle and dietary changes related to the so-called "development" seem linked to the prevalence of hypertension. In the INTERHEART population case-control study [ 26 ], which was an investigation into the association of psychosocial risk factors in patients with acute myocardial infarction, examining 11119 cases and 13648 controls from 52 countries, demonstrated higher prevalence of all four 'stress factors' (stress at home, at work, financial stress and major life events) in these patients, with consistency across regions, ethnicity and gender. Though data implicating stress as being contributory to the development of high blood pressures are limited and perhaps even hard to establish, a causal relationship between stress and developing high blood pressure does not appear to be an illogical assumption. There remain many uncertainties to the relative importance and contribution of environmental versus genetic influences on the development of blood pressure – there is more than likely an influence from both. However, there is now evidence to necessitate increased attention in examining the non-genetic influences on blood pressure, a neglected area of hypertensive research but perhaps a goldmine for establishing causal influences. As stated earlier, the future of hypertension research should focus on more standardised and comparable protocols, with comparable designs in data collection. Multi-centric data collection with a view to establishing local or national blood pressure demographics is crucial for the formal assessment of the global hypertension burden and the implementation of cost-effective primary preventive measures. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544878.xml

551534

Accelerated evolution associated with genome reduction in a free-living prokaryote

Prochlorococcus sp. are marine bacteria with very small genomes. The mechanisms by which these reduced genomes have evolved appears, however, to be distinct from those that have led to small genome size in intracellular bacteria.

Background The size of bacterial genomes is primarily the result of two counteracting processes: the acquisition of new genes by gene duplication or by horizontal gene transfer; and the deletion of non-essential genes. Genomic flux created by these gains and losses of genetic information can substantially alter gene content. This process drives divergence of bacterial species and eventually adaptation to new ecological niches [ 1 ]. In some cases, gene deletion may prevail over gene acquisition, leading to genome reduction. This process has occurred several times during evolution and has been well documented for cellular organelles [ 2 , 3 ], obligate pathogens such as Mycoplasma genitalium [ 4 ] or phytoplasmas [ 5 ] and symbionts such as the insect endosymbiont Buchnera [ 6 - 8 ] or the hyperthermophile Nanoarchaeum equitans [ 9 ]. In the case of organelles, the degree of genome reduction can be extensive as a result of massive gene transfer into the host nucleus, allowing maintenance of the corresponding functions in the resulting composite organism. Mitochondrial or chloroplast genomes, for instance, can be as small as 6 kilobases (kb) [ 10 ] and 35 kb [ 11 ], respectively. In the case of obligate host-dependent bacteria, the reduction is more limited because the relationships with their hosts are less intimate than for organelles in eukaryotic cells. Thus, obligatory pathogens need to retain a minimum of functions that allow them to infect new hosts and to avoid host defenses, and obligate endosymbionts carry genes which are absolutely necessary for host survival. For instance, a substantial part (approximately 10 %) of the Buchnera genome is devoted to biosynthesis of amino acids which are essential to its host [ 6 ]. So far, all characterized examples of genome reduction have been associated with a change from a free-living to a host-dependent lifestyle [ 12 ]. It is therefore intriguing that a similar phenomenon of genome reduction has occurred within the free-living marine cyanobacterial genus Prochlorococcus [ 13 - 15 ]. The latter is present at high abundance (often over 10 5 cells/ml) in all nutrient-poor areas of the world's oceans between 40°N and 40°S and is probably the most abundant photosynthetic organism on Earth [ 16 , 17 ]. It has been shown that two major ecotypes exist within this genus [ 18 ]. The first is adapted to grow at the base of the illuminated layer and displays a high divinyl-chlorophyll b to a ratio; the second inhabits the upper layer of the ocean and has a low divinyl-chlorophyll b to a ratio [ 19 ]. The genome of one high-light-adapted (HL) strain, Prochlorococcus marinus MED4 [ 14 ], and of two low-light-adapted (LL) strains, P. marinus SS120 [ 13 ] and Prochlorococcus species MIT9313 [ 14 ], have recently been sequenced and annotated. Phylogenetic trees based on 16S rRNA sequences [ 18 ] or 16S-23S ribosomal internal transcribed spacer sequences [ 20 ] show that Prochlorococcus sp. MIT9313 branches at the base of the Prochlorococcus radiation, close to the Synechococcus group [ 21 ]. In contrast, the Prochlorococcus HL clade, encompassing the MED4 strain, appears to be the most recently evolved Prochlorococcus group, consistent with the fact that this clade is much less diversified than are the LL clades. Despite the close relatedness of these strains, their genomes vary widely in terms of size, G+C content and the number of protein-coding genes (Table 1 ). While the general characteristics of the MIT9313 genome are very similar to those of the Synechococcus sp. WH8102 genome [ 22 ], MED4 has the smallest genome for a photosynthetic organism known to date and the SS120 genome is only 90 kb larger. Furthermore, this genome reduction is clearly accompanied by a drift in G+C content, a phenomenon that commonly occurs during the evolution of host-dependent genomes [ 23 ]. However, the evolutionary mechanisms involved in the genome reductive process are most probably different from those that have occurred in host-dependent organisms. Using comparative sequence analyses of the four genomes of marine picocyanobacteria published to date, we have attempted to better understand the causes and consequences of this phenomenon and to address the relationships between genome reduction and niche adaptation in marine picocyanobacteria. Results Synteny and genome stability Alignments of whole genomes show a strong conservation of the gene order between MED4 and SS120 (Figure 1a ). There are only five inversions larger than 20 kb between these two genomes. In contrast, the large number of inversions and translocations and the shorter size of the colinear segments between SS120 and MIT9313 on the one hand and MIT9313 and WH8102 on the other hand (Figure 1b,c ) indicate that extensive genome rearrangements have occurred not only between Synechococcus and Prochlorococcus but also between MIT9313 and the two other Prochlorococcus strains (see also Figure 2 in [ 14 ]). The degree of synteny observed between the four marine picocyanobacteria genomes strengthens the hypothesis of a more recent divergence of the clades containing MED4 and SS120 than of the clade containing MIT9313. Overall genome composition The downsizing of MED4 and SS120 genomes during evolution is associated with a genome-wide adenine (A) and thymine (T) enrichment (Table 1 ). The bias is most pronounced at neutral sites such as intergenic regions (MED4, 76.6% A+T; SS120, 69.3% A+T) and third-codon positions of protein-coding genes (MED4, 79.7% A+T; SS120, 73.85% A+T). This bias has little effect on ribosomal RNA genes (5S, 16S and 23S) which have a G+C content greater than 50% in all four picocyanobacterial genomes. In both MED4 and SS120, the single rRNA gene cluster can easily be spotted as a G+C-rich anomaly compared to the rest of the genome (see for example, Figure 1 in [ 15 ]). In direct contrast, protein-coding genes are strongly affected by the extreme base composition of these genomes. First, the bias influences codon usage since, for a given amino acid, AT-rich codons are preferentially used (Figure 3a ). Second, the amino-acid composition of the proteins themselves is affected (Figure 3b ). Indeed, when compared to Prochlorococcus sp. MIT9313 and Synechococcus sp. WH8102, the genes of P. marinus MED4 and SS120 contain fewer amino acids encoded by G+C-rich codons (for example, alanine or arginine) and more amino acids encoded by A+T-rich codons (for example, isoleucine or lysine). Orthologous gene pool size A total of 1,306 orthologs belonging to all major functional categories are common to the four genomes (see Additional data file 1) and probably constitute an estimate of the core of genes conserved in all marine picocyanobacteria. This is sensibly more than the pool of around 1,000 orthologs identified by W.R. Hess [ 15 ]. The difference certainly results from the use by the latter author of a low E-value threshold (10e -12 ) for BLAST comparisons. In contrast, our analysis is based on identification of reciprocal best hits without the use of any particular threshold (apart from the default BLAST threshold) and consequently allows the detection of orthologous relationships whatever the gene lengths or the level of similarity. Still, our ortholog identification process is rather strict and the set of orthologs identified in this study probably corresponds to a lower estimate of the actual number of orthologs shared by the four genomes. This set of genes represents a substantial percentage of the total pool of all protein-coding genes in P. marinus MED4 (73.2%) and SS120 (69.2%) and about half of the gene set in Prochlorococcus sp. MIT9313 (56.2%) and Synechococcus sp. WH8102 (51.1%). These percentages are consistent with the differences in the respective number of genes within these genomes (Table 1 ) and are compatible with the assumption that a massive gene loss has occurred in MED4 and SS120 during their evolution from a Prochlorococcus ancestor with a larger genome [ 13 - 15 ]. Accelerated rate of evolution of protein-coding genes in Prochlorococcus Because biased base composition seems to constrain amino-acid usage in the Prochlorococcus genomes, we have investigated whether it also affects the rate of protein sequence evolution in these genomes. We used the 1,306 orthologs common to the four genomes to estimate the amino-acid substitution rate in each genome. Branch lengths calculated for a given tree topology (the same topology as for the 16S rRNA gene tree; see Figure 2 ) are 0.46, 0.22, 0.16 and 0.14 amino acid substitutions per site for branches dA, dB, dC and dX, respectively. Using Synechococcus sp. WH8102 as the outgroup, we tested the rate-constancy hypothesis and computed the ratios of branch lengths. Relative rate tests (two-cluster and branch length tests) indicate that protein sequences evolved at significantly different rates ( P < 0.001) between MED4, SS120 and MIT9313. Therefore the hypothesis of a constant evolutionary rate between these strains can be rejected for protein-coding genes. The calculation of branch-length ratios reveals that the amino-acid substitution rate is 2.04-fold higher inMED4 than in SS120 (dA/dB) and 3.81-fold higher in MED4 than in MIT9313 ((dA+dX)/dC). This rate is also 2.31-fold higher for SS120 than for MIT9313 ((dB+dX)/dC). Computation of branch lengths for each functional category shows that the increased rate of amino-acid replacement in protein sequences concerns every category (Figure 4 and Table 2 ). These results imply that the rate of amino-acid substitution increased during evolution of the Prochlorococcus genus concomitantly with genome reduction and increase in A+T content. Synonymous and nonsynonymous substitutions The ratio of the rate of nonsynonymous substitutions ( d N ) to the rate of synonymous substitutions ( d S ) is commonly used to measure the relative rate of purifying selection acting at the protein level. We determined d S and d N for each gene pair of every group of orthologs and their values were averaged for each genome. Surprisingly, we observed saturation at synonymous sites for all genome pairs ( d S > 2) and the calculation of the d N / d S ratio was thus impossible. Still, the average d N was higher between MED4 and SS120 (0.36) than between SS120 and MIT9313 (0.32). The lowest d N was observed between MIT9313 and WH8102 (0.24), a finding which is consistent with the relative acceleration of amino-acid substitutions in MED4 and in SS120. DNA-repair systems A shift in base composition may reflect the loss of DNA-repair genes and we therefore determined the presence or absence of genes involved in these mechanisms. As the mutational pressure is toward a high A+T content in both MED4 and SS120, we looked more closely at those genes whose absence could increase the frequency of G:C to A:T mutations. Among the genes putatively encoding DNA-repair enzymes identified in MIT9313 and WH8102, a few are missing in SS120 and/or MED4 (Table 3 ). Both MED4 and SS120 lack the ada gene, which encodes 6-O-methylguanine-DNA methyltransferase, which repairs alkylated forms of guanine and thymine in DNA. Such alkylations generate lesions that can lead to G:C to A:T transversions [ 24 ]. Interestingly, the MED4 genome is the only one among the four picocyanobacteria not to encode the A/G-specific DNA glycosylase MutY, as previously noted by Rocap and co-workers [ 14 ]. This enzyme acts with MutT (NTP pyrophosphohydrolase) and MutM (formamido-pyrimidine-DNA glycosylase) in the GO system to avoid misincorporation of oxidized guanine (8-oxoG) in DNA and to repair the base mismatches A:8-oxoG [ 25 ]. In Escherichia coli , knocking out both mutM and mutY translates into a 1,000-fold increase of G:C to A:T transversions in comparison to the wild-type strain [ 26 ]. In addition to MutT and MutY, MIT9313 and WH8102 encode a third enzyme of the NUDIX hydrolase family that is missing in MED4 and SS120. This hydrolase could act to prevent mutations. However because of the broad substrate specificity of this family, one cannot know with certainty the function of this protein. Likewise, two genes coding for enzymes of the RecF pathway have been lost either by both MED4 and SS120 (DNA helicase RecQ) or only by MED4 (exonuclease RecJ). Discussion The process of genome reduction which has occurred within the Prochlorococcus radiation has to our knowledge never been observed so far in any other free-living prokaryote. Since Prochlorococcus sp. MIT9313 has a genome size very similar to that of Synechococcus sp. WH8102 (2.4 megabase-pair (Mbp)), as well as several other marine Synechococcus spp. (M. Ostrowski and D. Scanlan, personal communication), it is reasonable to assume that the common ancestor of all Prochlorococcus species also had a genome size around 2.4 Mbp. Under this hypothesis, the genome reduction which has occurred in MED4 would correspond to around 31%. By comparison, the extent of genome reduction in the insect endosymbiont Buchnera , as compared to a reconstructed ancestral genome, is around 77% [ 27 ]. The genome of P. marinus SS120 - and a fortiori the MED4 genome - is considered to be near minimal for a free-living oxyphototrophic organism [ 13 ]. It would seem that genome reduction in these organisms probably cannot proceed below a certain limit, corresponding to a gene pool containing all the essential genes of biosynthetic pathways and housekeeping functions (probably including most of the 1,306 four-way orthologous genes identified in this study) plus a number of other genes, including genus-specific as well as niche-specific genes. For instance, MED4 encodes a number of photolyase-related proteins, a few specific ABC transporters (for cyanate, for example; [ 14 ] and data not shown). These specific compounds might be critical for survival in the upper water layer, which receives high photon fluxes, UV light and is nutrient-depleted, but less so for life deeper in the water column. If both Prochlorococcus lineages and host-dependent organisms have undergone genome reduction associated with accelerated substitution rates, these phenomena must have arisen from very different causes as the resulting gene repertoires of the two types of organisms differ tremendously. Indeed, the genome evolution of endosymbionts and obligatory pathogens is driven by two main processes which have mutually reinforcing effects on genome size and evolutionary rates. Being confined inside their host, these bacteria have tiny population sizes and are regularly bottlenecked at each host generation or at each new host infection. Consequently, they experience a strong genetic drift [ 28 ] involving an increase in substitution rate. This acceleration results in the accumulation at random of slightly deleterious mutations in protein-coding genes [ 8 , 29 ] as well as in rRNA genes [ 29 , 30 ]. This genetic drift enhances the downsizing of the genome through inactivation and then elimination of potentially beneficial but dispensable genes. Among these, there have been a number of DNA-repair genes, the disappearance of which could have further increased the mutation rate [ 6 , 31 - 33 ]. Furthermore, a number of genes may be subject to a relaxation of purifying selection which is therefore rendered less effective in maintaining gene function. This relaxation particularly affects genes which have become useless because they are redundant in their host genome, such as genes involved in the biosynthesis of amino acids, nucleotides, fatty acids and even ATP [ 4 - 6 , 8 , 9 , 32 ]. Selection pressure is also reduced for genes involved in environmental sensing and regulatory systems, such as two-component systems, because of the much buffered environment offered by the host [ 6 ]. In the free-living genus Prochlorococcus , the very large size of field populations [ 34 ] means that these populations are subject to much lower genetic drift and their genomes are subject to much stronger purifying selection than are those of endosymbionts and pathogens [ 35 ]. Consequently, the observed accelerated rate of evolution probably results merely from the increase in the mutation rate, which in turn is probably due to the loss of DNA-repair genes, even if one should note that, in P. marinus SS120 only two such genes are missing (Table 3 ). We observed a similar acceleration of amino-acid substitutions for all functional categories (Figure 4 ). This finding is more consistent with a global increase in the mutation rate than with relaxed selection, the latter being unlikely to occur to the same extent at all loci. We also assume that most amino-acid substitutions that have occurred in Prochlorococcus proteins are neutral; that is, they have not altered protein function. Indeed, populations of the HL clade which, like MED4, have the most derived protein sequences of all Prochlorococcus species, appear to be the most abundant photosynthetic organisms in the upper layer of the temperate and inter-tropical oceans [ 16 ]. Such an ecological success would hardly be possible for organisms handicapped by a large number of slightly deleterious mutations, especially given the fact that most genes are single copy, and so compensation of gene function is generally not possible. The effect of the maintenance of a high level of purifying selection on counteracting deleterious substitutions is particularly obvious in the rRNA genes. Contrary to the protein-coding genes, relative rate tests did not show any significant differences in the rates of evolution of the 16S rRNA genes in the four marine picocyanobacterial genomes, and thus there is no evidence that either SS120 or MED4 could have accumulated mutations destabilizing the secondary structure of their 16S rRNA molecule. One noteworthy consequence of the acceleration in the rates of evolution of protein-coding genes in Prochlorococcus is that phylogenetic reconstructions based on protein sequences are biased. Indeed, this leads to much longer branches for these two strains than for MIT9313. The resulting tree topology most often does not support that obtained with the 16S rRNA gene, for which the molecular clock hypothesis holds true according to our analyses. Thus, rRNA genes are likely to be among the few genes that will give reliable estimates of the phylogenetic distances between Prochlorococcus strains. If it is neither the relaxation of purifying selection nor an increase in genetic drift that has been the main factor causing Prochlorococcus genome reduction, an alternative possibility is that the latter could be the result of a selective process favoring the adaptation of Prochlorococcus to its environment. The apparently better ecological success in oligotrophic areas of Prochlorococcus species compared to their close relative Synechococcus [ 16 , 34 ], strongly suggests that the reduction of Prochlorococcus genome size could provide a competitive advantage to the former. Indeed, extensive comparisons of the gene complements of these two organisms show very few examples - at least among genes for which function is known - of the occurrence of specific genes in MED4 which could explain its better adaptation (data not shown). One noteworthy exception is the presence in Prochlorococcus , but not Synechococcus , of flavodoxin and ferritin, two proteins that possibly give Prochlorococcus a better resistance to iron stress. Apart from that, Synechococcus appears more like a generalist, in particular with regard to nitrogen or phosphorus uptake and assimilation [ 22 ], and should a priori be more suited to sustain competition. Hence, we assume that the key to the success of Prochlorococcus resides less in the development of a specific complex or pathway to cope better with unfavorable conditions than in the simplification of its genome and cell organization, which can allow this organism to make substantial economies in energy and material for cell maintenance. The mere reduction in genome size per se is a potential source of substantial economies for the cell, as it reduces the amount of nitrogen and phosphorus, two particularly limiting elements in the upper part of the ocean, which are necessary, for instance, in DNA synthesis. Another advantage is that it allows a concomitant reduction in cell volume. It has been previously suggested (see, for example [ 36 ]) that, for a phytoplanktonic organism, a small cell volume confers two selective advantages by reducing self-shading (the package effect) and by increasing the cell surface-to-volume ratio, which can improve nutrient uptake. The first advantage would improve the fitness of the LL strains, whereas the second would offer an advantage to the HL strains living in nutrient-depleted surface waters. Finally, cell division is less costly for a small than for a large cell. On the basis of these observations, we assume that the major driving force for genome reduction within the Prochlorococcus radiation has been the selection for a more economical lifestyle. The bias toward an A+T-rich genome in MED4 and SS120 is also consistent with this hypothesis, as it can be seen as a way to economize on nitrogen. Indeed, an AT base-pair contains seven atoms of nitrogen, one less than a GC base-pair. With this hypothesis in mind, we propose a possible scenario for the evolution of Prochlorococcus genomes. Using a rate of 16S rRNA divergence of 1% per 50 million years [ 37 ], one can estimate that the differentiation of these two genera is as recent as 150 million years, as the molecular clock hypothesis holds for this gene in Prochlorococcus and Synechococcus . The ancestral Prochlorococcus cells must have developed in the LL niche, a niche probably left free by other picocyanobacteria. Given the considerable difference in genome size between the LL strains MIT9313 and SS120, it appears that genome reduction itself must have started in one (or possibly several) lineage(s) within the LL niche some time after Prochlorococcus differentiation from its common ancestor with marine Synechococcus species. Why the selection has affected only one (or some?) and not all Prochlorococcus lineages remains unclear. Examination of the gene repertoire of P. marinus SS120 [ 13 ] suggests that this genome reduction must have concerned the random loss of dispensable genes from many different pathways. At some point during evolution, some genes involved in DNA repair have been affected; these would include the ada gene, which may be responsible for the shift in base composition, but also possibly several others, not necessarily involved in GC to AT mutation repair (see Table 3 ). Loss of these genes may have led to an increase in the mutation rate and therefore in the rate of evolution of protein-coding genes, accompanied by a more rapid genome shrinkage and a shift of base composition toward AT. It is worth noting that one likely consequence of this genome-wide compositional shift is the absence of the adaptive codon bias in the genomes of Prochlorococcus species MED4 and SS120. AT-rich codons are preferentially used whatever the amino acid (Figure 3a ). Thus, codon usage in these genomes appears to reflect more the local base-composition bias than the selection for a more efficient translation through the use of optimal codons. The same conclusion has been drawn for other small genomes with high A+T content [ 28 , 38 ]. Later during evolution (around 80 million years ago, according to the degree of 16S rRNA sequence divergence between MED4 and SS120) one LL population which probably already had a significantly reduced cell and genome size must have progressively adapted to the HL niche and eventually recolonized the upper layer. How this change in ecological niche was possible is still hard to define. Comparison of the gene set that differs between the LL-adapted SS120 and the HL-adapted MED4 shows that very few genes might be sufficient to shift from one to the other niche, including a multiplication of hli genes [ 39 ] and the differential retention of genes which were present in the common ancestor of Prochlorococcus and Synechococcus , (such as the photolyases and cyanate transporters mentioned above) and were secondarily lost in the LL-adapted lineages. Conclusions Genome evolution in the free-living genus Prochlorococcus has similar features to that in host-dependent prokaryotes: genome reduction, bias toward a low G+C content, acceleration in the evolution rate of protein-coding genes, and loss of DNA-repair genes. In contrast to the latter organisms, however, in Prochlorococcus this evolution does not appear to be the result of genetic drift or relaxed selection being exerted on some gene categories. Indeed, purifying selection is very efficient in Prochlorococcus , as rRNA genes have evolved at a similar rate in all genomes. Despite the decrease in G+C content and an accelerated rate of evolution of protein-coding genes, purifying selection must also act on these genes and avoid potentially deleterious mutations. We hypothesize that a reduction in genome size (which allows a concomitant reduction in cell size and substantial economies in energy and nutrients) can constitute a selective advantage for life in the open ocean, both at depths where photon energy is low and in surface waters where nutrients are scarce. Genome shrinkage in Prochlorococcus has led to populations highly specialized to narrow ecological niches, at the expense of versatility and competitiveness in changing conditions. Indeed, not only is the distribution of the Prochlorococcus genus limited to low latitudes (40°N and 40°S, see [ 34 ]) but the different ecotypes are themselves more or less confined to a restricted part of the euphotic layer [ 40 ]; for example, they experience only limited changes in temperature and salinity. Paradoxically, because warm oligotrophic areas constitute a very large part of the world's oceans, the ecological niches (both LL and HL) occupied by Prochlorococcus species are huge, and thus this organism appears globally, despite its specialization, as one of the most successful oxyphototrophs on Earth. Materials and methods Genome sequence data The complete genome sequences and annotations of Prochlorococcus marinus MED4, P. marinus SS120, Prochlorococcus sp. MIT9313 and Synechococcus sp. WH8102 (accession numbers: NC_005071, NC_005072, NC_005042 and NC_005070 respectively) were downloaded from the Genome division of the NCBI Entrez system. A few additional genes which were modeled in at least one genome and were present in the other genomes but not modeled (because of their small size, for example) were included in our dataset (see Additional data file 2). Alignment of whole genomes Genome sequences translated in their six reading frames were aligned with the Promer program of the MUMmer 3.0 system [ 41 ]. Codon and amino-acid usage Codon usage was computed for every open reading frame (ORF) of each genome with the EMBOSS program cusp. Amino-acid usage was derived from the results produced by cusp. Identification of orthologous proteins We used a sequence-similarity based approach which is similar to the procedure used for the cluster of orthologous groups (COGs [ 42 ]). For each genome pair, all-against-all BLAST [ 43 ] comparisons were performed using protein sequences and reciprocal genome-specific best hits were identified. We considered genes as being probable orthologs when they were included in groups of size four in which each gene was the best hit of the three others. From similarity searches against the COG database, orthologs were assigned to functional categories according to those defined for the COG system. Because of the lack of a particular category for photosynthesis genes, the latter were assigned to the 'energy production and conversion' COG category. Other genes which fell into more than one of the 19 COG categories have been assigned to a supplementary category called 'miscellaneous'. Phylogenetic branch length estimations Protein sequences from each of the groups of four orthologous genes were aligned using ClustalW [ 44 ] with default parameters. After exclusion of all gap sites, individual alignments were concatenated in one super-alignment of 388,120 sites. Gamma distances [ 45 ] with an alpha parameter of 1 were estimated between each pair of sequences of the super-alignment. Phylogenetic branch lengths were calculated from distances with the ordinary least-squares method [ 45 ]. Relative rate tests (two-cluster test and Branch length test) were applied in order to test the constancy of amino-acid substitution rates between the three Prochlorococcus genomes (hypothesis of the molecular clock). The same analysis was applied to orthologs of each functional category. Estimate of synonymous and nonsynonymous substitution rates Nucleotide sequences of each group of orthologs were aligned with Protal2dna according to alignments of their corresponding amino-acid sequences [ 46 ]. Pairwise estimates of the synonymous ( d S ) and non-synonymous ( d N ) substitution rates were obtained from the Yn00 program of the PAML 3.13 package [ 47 ]. Additional data files The following additional data are available with the online version of this article. Additional data file 1 lists the orthologous genes classified by functional category. Orthologous genes were assigned to the functional categories of COG system. Photosynthesis genes were assigned to the 'energy production and conversion' COG category. Genes falling in more than one of the 19 COG categories have been assigned to a supplementary category called 'miscellaneous'. Additional data file 2 is a fasta file of orthologous genes which were modeled in at least one genome and present but not modeled in the other genomes. Supplementary Material Additional data file 1 The orthologous genes classified by functional category Click here for additional data file Additional data file 2 A a fasta file of orthologous genes which were modeled in at least one genome and present but not modeled in the other genomes Click here for additional data file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551534.xml

551520

Nuclear localization is required for Dishevelled function in Wnt/β-catenin signaling

Background Dishevelled (Dsh) is a key component of multiple signaling pathways that are initiated by Wnt secreted ligands and Frizzled receptors during embryonic development. Although Dsh has been detected in a number of cellular compartments, the importance of its subcellular distribution for signaling remains to be determined. Results We report that Dsh protein accumulates in cell nuclei when Xenopus embryonic explants or mammalian cells are incubated with inhibitors of nuclear export or when a specific nuclear-export signal (NES) in Dsh is disrupted by mutagenesis. Dsh protein with a mutated NES, while predominantly nuclear, remains fully active in its ability to stimulate canonical Wnt signaling. Conversely, point mutations in conserved amino-acid residues that are essential for the nuclear localization of Dsh impair the ability of Dsh to activate downstream targets of Wnt signaling. When these conserved residues of Dsh are replaced with an unrelated SV40 nuclear localization signal, full Dsh activity is restored. Consistent with a signaling function for Dsh in the nucleus, treatment of cultured mammalian cells with medium containing Wnt3a results in nuclear accumulation of endogenous Dsh protein. Conclusions These findings suggest that nuclear localization of Dsh is required for its function in the canonical Wnt/β-catenin signaling pathway. We discuss the relevance of these findings to existing models of Wnt signal transduction to the nucleus.

Background The specification of cell fates during embryonic development frequently depends on inductive interactions, which involve transmission of extracellular signals from the cell surface to the nucleus. In the transforming growth factor β (TGFβ) signal transduction pathway, Smad proteins that are initially associated with TGFβ receptors move to the nucleus to regulate target genes [ 1 ]. Another example of a direct link between the cell surface and the nucleus during embryonic development is the proteolytic cleavage and nuclear translocation of the cytoplasmic fragment of the Notch receptor [ 2 ]. In contrast, multiple steps appear to be required for a Wnt signal to reach the nucleus. In this molecular pathway, signals from Frizzled receptors are transduced to Dishevelled (Dsh), followed by inactivation of the β-catenin degradation complex that includes the adenomatous polyposis coli protein (APC), Axin and glycogen synthase kinase 3 (GSK3) [ 3 , 4 ]. Stabilization of β-catenin is thought to promote its association with members of the T-cell factor (Tcf) transcription factor family in the nucleus, resulting in the activation of target genes [ 5 , 6 ]. As well as the canonical β-catenin-dependent pathway, Frizzled receptors also activate small GTPases of the Rho family, protein kinase C and Jun-N-terminal kinases (JNKs) to regulate planar cell polarity in Drosophila and convergent extension cell movements and tissue separation in Xenopus [ 7 - 12 ]. Thus, the Wnt/Frizzled pathway serves as a model for molecular target selection during signal transduction. Dsh is a common intracellular mediator of several pathways activated by Frizzled receptors and is composed of three conserved regions that are known as the DIX, PDZ and DEP domains [ 13 ]. Different domains of Dsh are engaged in specific interactions with different proteins, thereby leading to distinct signaling outcomes [ 13 ]. Daam, a formin-related protein, promotes RhoA activation by Dsh [ 9 ], whereas Frodo, another Dsh-binding protein, is required for Wnt/β-catenin signaling in the nucleus [ 14 ]. These interactions may take place in various cellular compartments, linking specific activities of Dsh to its distribution inside the cell. Dsh is found in a complex with microtubules and with the actin cytoskeleton [ 15 - 17 ]. Dsh is also associated with cytoplasmic lipid vesicles, and this localization was shown to require the DIX domain [ 7 , 16 , 18 ]. Overexpressed Frizzled receptors can recruit Dsh to the cell membrane in Xenopus ectoderm, and this redistribution requires the DEP domain [ 7 , 18 , 19 ]. The DIX domain is essential for the Wnt/β-catenin pathway, whereas the DEP domain plays a role in the planar cell polarity pathway [ 7 , 8 , 16 , 18 , 20 , 21 ]. Thus, the specific subcellular localization of Dsh may be crucial for local signaling events. The current study was based on our initial observation that a Dsh construct lacking the carboxy-terminal DEP domain was found in cell nuclei. We have now identified a nuclear export signal in the deleted region and also discovered that Dsh proteins accumulate in the nuclei of Xenopus ectodermal cells and mammalian cells upon inhibition of nuclear export. Dsh also accumulated in the nuclei after stimulation of mammalian cells with Wnt3a-containing culture medium. By analyzing various mutant Dsh constructs in Xenopus ectoderm, we show that the signals responsible for Dsh nuclear localization reside in a novel domain and that the nuclear translocation of Dsh is essential for its ability to activate Wnt/β-catenin signaling. Results and discussion A nuclear export signal in Dsh is responsible for the cytoplasmic localization of Dsh We studied the subcellular distribution of fusions of Dsh with green fluorescent protein (GFP) in Xenopus ectodermal cells. In contrast to Dsh-GFP, which is localized in punctate structures within the cytoplasm [ 7 , 18 ], the Ds2 construct, lacking the carboxy-terminal region, accumulates in the nucleus (Figure 1a–c ), indicating that the carboxyl terminus contains sequences for nuclear export. Indeed, we found a potential leucine-rich nuclear export signal (NES) in Dsh at positions 510–515, corresponding to the conserved consensus M/LxxLxL (single letter amino-acid code, where x is any amino acid) [ 22 , 23 ]. When leucines 513 and 515 in this putative NES were substituted with alanines, the mutated Dsh fusion construct, DsNESm, was localized predominantly in the nucleus (Figure 1a,d ), demonstrating that the sequence is a functional nuclear export signal. To examine whether inhibition of nuclear export abrogates Dsh activity, we compared the abilities of DsNESm and wild-type Dsh-GFP to induce secondary axes in frog embryos. Although the molecular mechanism operating during axis induction remains to be elucidated, this assay faithfully reflects the biological activity of Dsh in the canonical Wnt/β-catenin pathway [ 14 , 16 , 18 , 24 ]. DsNESm, which was expressed at similar levels to the wild-type Dsh-GFP (data not shown), induced secondary axes at least as efficiently as Dsh-GFP (Table 1 ). Induced axes contained pronounced head structures with eyes and cement glands (Figure 1e–g ). These results suggest that Dsh may function in the nucleus to trigger dorsal axial development. Nuclear localization of Dsh in cells treated with nuclear export inhibitors Accumulation of DsNESm in the nucleus implies that the wild-type Dsh shuttles between the nucleus and the cytoplasm. We therefore studied the subcellular distribution of Dsh in Xenopus embryonic cells under conditions in which nuclear export is blocked. When ectodermal cells expressing Dsh-GFP were incubated with N-ethylmaleimide (NEM), an inhibitor of the nuclear export receptor CRM1/exportin [ 25 , 26 ], Dsh-GFP was detected predominantly in the nucleus, compared to the punctate cytoplasmic pattern of Dsh-GFP in untreated cells (Figure 2a,b ). This effect was specific to full-length Dsh-GFP, as Ds3, a Dsh construct that lacks 48 amino acids adjacent to the PDZ domain (Figure 1a ), did not accumulate in the nucleus after NEM treatment (Figure 2e,f ). The nuclear retention of Dsh-GFP was also observed using leptomycin B (LMB), another inhibitor of CRM1-dependent nuclear export [ 22 , 23 ] (Figure 2c,d ). These results indicate that Dsh shuttles between the cytoplasm and the nucleus, and that its abundance in the cytoplasm is due to highly efficient nuclear export. To ensure that the Dsh-GFP fusion behaves similarly to the endogenous Dsh protein, we examined the localization of endogenous Dvl2, a mammalian homolog of Dsh, in human and rat tissue culture cells. Human embryonic kidney (HEK) 293 cells treated with LMB accumulated Dvl2 in the nucleus, contrasting with the cytoplasmic localization of Dvl2 in untreated cells (Figure 3a–c ). We also evaluated the subcellular localization of endogenous Dvl2 in Rat-1 fibroblasts, which are known to respond to Wnt signaling. Fractionation of cells into nuclear and cytoplasmic protein pools revealed only a small amount of endogenous Dvl2 in intact nuclei, whereas after NEM treatment, Dvl2 was localized predominantly in the nuclear fraction (Figure 3d ). The efficiency of subcellular fractionation was controlled for by staining with antibodies to glyceraldehyde phosphate dehydrogenase (GAPDH) and nuclear lamins. These proteins remained exclusively cytoplasmic or nuclear, respectively, in both untreated and NEM-treated cells (Figure 3d ). Thus, our data show that Dsh translocates into the nucleus and is actively exported into the cytoplasm of both Xenopus ectodermal cells and mammalian fibroblasts. Identification of sequences responsible for Dsh nuclear localization To identify specific amino-acid sequences that direct the transport of Dsh to the nucleus, we studied the subcellular distribution of mutated Dsh-GFP fusion constructs (Figure 4a ). The removal of the DIX and PDZ domains (Ds1) did not eliminate nuclear translocation in response to NEM or LMB (Figure 4a–d ), indicating that these two domains are not required for the nuclear import. Similarly, the DEP domain is not required for Dsh nuclear localization (Ds2; Figure 1a,c ). Comparison of Ds1 and Ds2 (see Figure 4a ), both capable of nuclear localization, reveals a short stretch of shared amino acids located between the PDZ and DEP domains. Strikingly, the removal of just this 48 amino-acid region abrogated nuclear import of Dsh in the presence of NEM or LMB (Ds3; Figures 2e,f and 4a ). Together these experiments identify amino acids 333–381 as the region required for nuclear localization of Dsh. Although this short sequence is highly conserved in all Dsh homologs from Hydra to humans (Figure 4j ), it does not bear detectable similarity to nuclear localization signals characterized in other proteins [ 27 ]. This sequence may interact directly with components of the nuclear import machinery or bind to a protein that itself binds a karyopherin/importin and mediates the nuclear import of Dsh by a piggyback mechanism. Interestingly, this region overlaps a novel proline-rich domain identified by mutational analysis of Dsh in Drosophila [ 28 ]. To define further the specific amino acids necessary for nuclear localization, a panel of Dsh constructs with point mutations spanning the conserved region was examined (data not shown). Nuclear import was eliminated with the substitution of three amino acids, converting IVLT into AVGA (DsNLSm; Figure 4a,e–g,j ), indicating that these three amino acids are critical. Dsh nuclear translocation is crucial for its function in the β-catenin pathway To determine whether nuclear localization of Dsh is required for its activity, we compared the abilities of DsNLSm and wild-type Dsh to induce secondary axes in frog embryos. We also assessed activation of a luciferase reporter construct for Siamois [ 29 ], an immediate target of Wnt/β-catenin signaling. DsNLSm had impaired ability to induce secondary axes and to activate the Siamois reporter when compared with wild-type Dsh (Figure 5a,b ; Table 1 ). Furthermore, DsNLSm failed to stabilize β-catenin (Figure 5c ). This difference was not due to differences in protein expression, as both constructs were present in embryo lysates at similar levels (Figure 5c ). Thus, these findings indicate that the nuclear localization of Dsh is critical for its functional activity in the β-catenin pathway. Not only was the function of DsNLSm in the β-catenin pathway impaired, but we found that this construct behaved as a dominant inhibitor of Wnt signaling and prevented the activation of the Siamois reporter by Xwnt3a and Xwnt8 RNAs (Figure 6a,b ). Consistent with these observations, another construct lacking the region responsible for the nuclear localization (Ds3; see Figure 4a ) also suppressed Wnt signaling (Figure 6b ). Despite these inhibitory properties, dorsally injected DsNLSm RNA, like Xdd1, a dominant negative deletion mutant [ 24 ], did not suppress primary axis formation (data not shown). Impaired activity of the DsNLSm construct may be due to its inability to translocate to the nucleus, or due to a coincidental elimination of a binding site for an essential cofactor that functions together with Dsh in the cytoplasm. To exclude the latter possibility, the IVLT sequence of Dsh NLS was replaced with KKKRK, an unrelated NLS from SV40 T antigen [ 27 ]. This construct, DsSNLS, relocated to the nucleus even in the absence of nuclear export inhibitors (Figure 4a,i ). Notably, all activities of wild-type Dsh, including induction of complete secondary axes, activation of the Siamois promoter and β-catenin stabilization were significantly restored in DsSNLS (Figure 5a–c ; Table 1 ). In contrast to DsNLSm, DsSNLS did not inhibit the ability of Wnt ligands to activate pSia-Luc (Figure 6b ), consistent with its being a positive regulator of the Wnt pathway. We note that the signaling activity of DsSNLS was not enhanced compared to wild-type Dsh, suggesting that the rate of the nuclear translocation of Dsh rather than its steady state levels in the nucleus is critical for target gene activation. It is also possible that other nuclear components, rather than Dsh, become rate-limiting for signaling. Overall, the simplest interpretation of our data is that the nuclear import of Dsh is essential for its activity. We next examined the ability of DsNLSm to bind critical Wnt signaling components, such as casein kinase 1ε (CK1ε), a positive regulator of the β-catenin pathway [ 30 , 31 ], or Axin, a negative regulator [ 20 , 32 - 36 ], both of which are known to bind Dsh. Both DsSNLS, enriched in the nucleus, and DsNLSm and Ds3, which do not enter the nucleus, bound CK1ε and XARP, a Xenopus Axin-related protein [ 20 ] (Figure 7 ). Thus, these mutated Dsh constructs retain the ability to associate with critical components of the Wnt/β-catenin pathway, arguing that defective nuclear translocation of DsNLSm is likely to be responsible for its inability to activate β-catenin signaling. Suppression of Dsh nuclear import does not affect noncanonical signaling Besides the β-catenin pathway, Dsh also functions in a planar cell polarity (PCP) pathway, which involves Rho GTPase and JNK and controls morphogenetic movements in early embryos [ 8 , 9 , 37 - 39 ]. We asked whether mutations in DsNLSm influence the β-catenin pathway exclusively or affect the PCP pathway as well. First, we observed that both Dsh-GFP and DsNLSm-GFP were efficiently recruited to the cell membrane by overexpressed Xfz8, a Frizzled family member [ 40 ] (Figure 8a ). As Dsh relocalization to the cell membrane in response to Frizzled is associated with its ability to signal in the PCP pathway [ 7 , 8 ], this observation suggests that DsNLSm can respond to Frizzled signaling independent of β-catenin. In Xenopus , the PCP pathway involving Dsh is implicated in the control of convergent extension movements [ 24 , 41 , 42 ]. Overexpression of the Xdd1 deletion mutant leads to the development of short embryos when expressed in dorsal marginal cells ([ 24 ]; Figure 8b ). Severe convergent extension defects (Figure 8b ) were observed in 22%, and mild defects were observed in 28% of the embryos injected with Xdd1 RNA (N = 35). In contrast, only mild morphogenetic defects were observed in embryos coinjected with Dsh (15%; N = 40) or DsNLSm RNA (18%; N = 39), indicating that both Dsh and DsNLSm partially rescued the effect of Xdd1. This indicates that DsNLSm is active in noncanonical PCP-like signaling. We also examined whether DsNLSm activates c-Jun N-terminal kinase (JNK), which is thought to function downstream of Dsh in the PCP pathway [ 8 , 37 - 39 ]. Both DsNLSm and Dsh activated JNK at equivalent levels (Figure 8c ), suggesting that nuclear localization of Dsh is not required for its function in noncanonical signaling. Nuclear accumulation of Dsh following Wnt3a stimulation Our findings are consistent with a scenario in which Wnt signaling may cause nuclear translocation of Dsh followed by formation of a stable β-catenin/Tcf3 complex and transcriptional activation of target genes. In support of this hypothesis, Dsh was reported to move to the nucleus in response to Wnt signaling in primary embryonic kidney cells [ 17 ]. In Rat-1 cells, we did not detect a significant change in Dsh distribution in response to Wnt signals (data not shown), possibly due to highly efficient nuclear export of Dsh in these cells. But immunofluorescence staining for Dvl2 revealed the nuclear accumulation of the protein in HEK293 and MCF7 cells after 3–6 h stimulation with Wnt3a-containing medium (Figure 9a , and data not shown). The effect was quantified by measuring nuclear to cytoplasmic (N/C) ratios of fluorescence intensity. The N/C ratio averaged 28% after 6 h treatment with the control medium, but increased to 91% after stimulation with Wnt3a-conditioned medium (Figure 9b ). These observations are consistent with the view that Dsh regulates Wnt-dependent gene targets in the nucleus. A role for Dsh in the nucleus In the current view, Wnt signaling causes inactivation of the β-catenin degradation complex, leading to stabilization and nuclear translocation of β-catenin [ 3 ]. Given that Dsh is genetically upstream of the β-catenin degradation complex [ 3 , 4 ] and that β-catenin degradation is thought to occur in the cytoplasm [ 43 ], Dsh nuclear import is unexpected. Nevertheless, our data demonstrate that Dsh shuttles between the cytoplasm and the nucleus and that its presence in the nucleus is critical for signaling. One explanation of these results is that β-catenin degradation may occur in the nucleus. Consistent with this possibility, APC, Axin and GSK3, components of the β-catenin degradation complex, have also recently been found to shuttle between the cytoplasm and the nucleus [ 22 , 23 , 44 - 47 ]. Moreover, Frat/GBP, a positive regulator of β-catenin, has been reported to expel GSK3 from the nucleus [ 47 ]. We show that the ability of Dsh constructs to enter the nucleus correlates with their ability to stabilize β-catenin (Figure 5c ). These observations indicate that Wnt/β-catenin signaling may depend on the nuclear localization of pathway components. Alternatively, nuclear localization of Dsh may affect β-catenin stability indirectly, by regulating protein interactions that sequester β-catenin in the nucleus, thereby preventing its cytoplasmic degradation [ 48 ]. Although we did not detect a significant change in nuclear import of β-catenin-GFP in Xenopus ectoderm cells overexpressing Dsh (data not shown), this process may be cell-context-dependent. On the other hand, we recently showed that Frodo, a nuclear Dsh-interacting protein, associates with Tcf3 and influences Tcf3-dependent transcription [ 49 ]. It is thus possible that Frodo links Tcf3 and Dsh to regulate Wnt target genes. Future studies should examine molecular components critical for the nuclear function of Dsh. Materials and methods DNA constructs GFP-tagged Dsh constructs were all derived from the DshGFP-RN3 plasmid that encodes the Xdsh protein fused at amino acid 724 to the first amino acid of GFP (Figures 1a , 4a ). Ds1 lacks the first 332 amino-terminal amino acids. Ds2 is the carboxy-terminal deletion of Xdsh, starting with amino acid 383. Ds3 lacks amino acids 334–381. In DsNLSm, the IVLT residues at positions 334–337 were replaced with AVGA, whereas in DsSNLS the same region is replaced with KKKRK, the SV40 T antigen NLS [ 27 ]. In DsNESm, L513 and L515 were substituted for alanines. To generate these constructs, DshGFP-pRN3 was used as a template. The in-frame deletion in Ds3 was made by PCR. Other GFP fusion constructs were synthesized with specific primers and PfuI DNA polymerase followed by DpnI digestion of the template [ 50 ]. The following primers were used: 5'-GTCCATAAACCGGGGCCCGCAGTCGGCGCCGTGGCCAAATGCTGG-3' for DsNLSm; 5'-ACACTAGGCCGCAGAATGCCCATTGTCCTGACCGTG-3' for Ds1; 5'-TCCATAAACCGGGGCCAAAGAAGAAGCGAAAGGTGGCCAAATGCTGGGA-3' for DsSNLS; 5'-TTCCCAGTGTACCCCGGGGCCATGGTGAGCAAGGGC-3' for Ds2, and 5'-GAGAACTATGACCAACGCTAGCGCGAATGACAACGATGGAT-3' for DsNESm. All constructs were verified by sequencing. Myc-tagged Dsh mutant constructs were made by replacing mutated regions with corresponding regions of Myc-Dsh [ 24 ]. Cloning details are available as an Additional data file with the online version of this article. Embryo culture, axis-induction assay and axis-extension assay In vitro fertilization, culture and microinjections of Xenopus eggs were essentially as described previously [ 24 ]. Stages were determined according to Nieuwkoop and Faber [ 51 ]. Axis induction was carried out by injecting mRNAs encoding different Dsh constructs (1 ng) into a single vegetal ventral blastomere at the 4–8-cell stage and assessed when the injected embryos reached stage 36–40. To monitor axis extension defects, 0.6 ng of Xdd1 RNA was injected alone or with 2 ng of Dsh or DsNLSm RNA into two dorsovegetal blastomeres of 4-cell embryos and the injected embryos were allowed to develop until sibling embryos reached stage 32. GFP fluorescence and luciferase assay For subcellular localization of Dsh-GFP constructs, mRNAs were injected into the animal pole region of 2–4-cell embryos. Animal cap explants were dissected at stages 9–10.5, incubated for 60 min in 10 mM N-ethylmaleimide (NEM; Sigma, St Louis USA) in 0.8 × MMR (Marc's Modified Ringer's solution, 1 × MMR: 100 mM NaCl, 2 mM KCl, 1 mM MgCl2, 2 mM CaCl2, 5 mM HEPES, pH 7.4), or in control (0.8 × MMR), then fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) for 30–45 min, washed three times in PBS, and mounted in 70% glycerol, 30% PBS containing 25 mg/ml of diazabicyclo(2,2,2)-octane (Sigma). Leptomycin B was used at 50 ng/ml in low-calcium medium (76 mM NaCl, 1.4 mM KCl, 0.2 mM CaCl 2 , 0.1 mM MgCl 2 , 0.5 mM Hepes, 1.2 mM sodium phosphate, (pH 7.5), 0.6 mM NaHCO 3 and 0.06 mM EDTA) for one hour prior to fixation. In some experiments, nuclei were stained by addition of 1 μg/ml 4,6-diamidino-2-phenylindole (DAPI) to the final PBS wash. For membrane localization studies, Xfz8 RNA was coinjected with RNAs encoding the Dsh constructs in the animal-pole region; animal-cap explants were dissected at stage 9–9.5 and mounted for observation. Fluorescence was visualized using a Zeiss Axiophot microscope. For luciferase assays, pSiaLuc reporter plasmid (20–40 pg) was coinjected with mRNAs encoding Xwnt3a [ 52 ] or Xwnt8 [ 53 ] and different Dsh constructs into one or two animal-ventral blastomeres or into one ventral-vegetal blastomere at the 4–8-cell stage. Luciferase activity was measured as described [ 29 ]. Tissue culture, immunocytochemistry and subcellular fractionation Rat-1 fibroblasts, human embryonic kidney (HEK) 293 cells and MCF7 human breast carcinoma cells were cultured in 1 × Dulbecco's Modified Eagle Medium (DMEM; Gibco/Invitro-gen, Carlsbad, USA) supplemented with 10% fetal calf serum and 5 μg/ml gentamicin. Conditioned medium was prepared from L cells stably transfected with Wnt3a as described [ 54 ], with the medium from untransfected L cells serving as a control. For immunocytochemistry, HEK293 cells were treated with 50 ng/ml LMB for 14 h while MCF7 cells were treated with Wnt3a or control conditioned medium for 1, 3, 6 or 8 h. Cells were fixed with 4% paraformaldehyde, immersed in methanol, and incubated with anti-Dvl2 antibodies and then Cy3-conjugated anti-rabbit IgG. Nuclei were stained by addition of 1 μg/ml DAPI as described for animal-cap cells. Fluorescence was observed under the Zeiss Axiophot microscope; 10–15 cells from each group were randomly picked up for measurement of the nuclear and cytoplasmic staining intensity using Image-Gauge software (Fuji Film, Tokyo, Japan). For subcellular fractionation, confluent cultures of Rat-1 cells were harvested by scraping plates and resuspended in hypotonic lysis buffer containing 1 mM EGTA, 1 mM EDTA, 2 mM MgCl 2 , 10 mM KCl, 1 mM DTT, 10 mM β-glycerophosphate, 1 mM sodium orthovanadate, 1 μg/ml leupeptin, 1 μg/ml aprotinin, and 1 μg/ml pepstatin. Cells were swollen for 30 min, and broken open with 25 strokes in a tight fitting Dounce homogenizer. Lysates were layered into tubes containing 1 M sucrose in hypotonic lysis buffer, and spun at 1600 × g for 10 min. Supernatant remaining above the sucrose cushion was used as the cytoplasmic fraction. The pellet, containing nuclei, was resuspended in an equivalent volume of hypotonic lysis buffer. Immunoprecipitation and western blotting Immunoprecipitation and western analysis were carried out with cell and embryo lysates as described [ 14 ]. To prepare embryo lysates at stage 10+, four animal blastomeres of 4–8-cell embryos were injected with RNAs encoding different forms of Dsh, ΔRGS-Axin [ 32 ], Flag-β-catenin [ 55 ], CK1ε [ 30 ] and HA-XARP [ 20 ]. To generate anti-Xdsh polyclonal antibodies, rabbits were immunized with a carboxy-terminal half of Xdsh (amino acids 301–736) fused to GST. First, GST beads were used for purification of anti-GST antibodies. Subsequently anti-Xdsh antibodies were affinity-purified on GST-Xdsh (301–736) beads. Polyclonal anti-Dvl2 antibody was generated in rabbits and affinity-purified on PVDF membrane blotted with human Dvl2 (79–249) [ 56 ]. A small aliquot of anti-human Dvl2 was obtained from M. Snyder (Yale University, New Haven, USA). Anti-GAPDH antibody was a gift from A. Stuart-Tilley and S. Alper (Beth Israel Deaconess Medical Center, Boston, USA), anti-lamin antibody was from F. McKeon (Harvard Medical School, Boston, USA). Anti-β-tubulin antibodies were from BioGenex (San Ramon, USA), anti-Flag M2 antibody was from Sigma and anti-CK1ε antibodies were from BD Biosciences (Palo Alto, USA). Anti-Myc and anti-HA monoclonal antibodies are hybridoma supernatants of 9E10 and 12CA5 cells (Roche Applied Science, Indianapolis, USA). JNK assay Four-cell embryos were injected with 4 ng Dsh or DsNLSm RNA into four animal blastomeres. Embryo lysates were prepared at stage 10.5 and in vitro kinase assays were carried out essentially as described [ 57 ], except that phosphorylated c-Jun-GST was detected with anti-phospho-c-Jun-specific antibodies (Cell Signaling Technology, Beverly, USA) by western blotting rather than with autoradiography. Additional data files The following is provided as an additional data file with the online version of this article. Additional data file 1 , containing cloning details of Dsh mutant constructs. Supplementary Material Additional data file 1 Cloning details of Dsh mutant constructs Click here for additional data file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551520.xml

529470

A comparison of vas occlusion techniques: cautery more effective than ligation and excision with fascial interposition

Background Vasectomy techniques have been the subject of relatively few rigorous studies. The objective of this analysis was to compare the effectiveness of two techniques for vas occlusion: intraluminal cautery versus ligation and excision with fascial interposition. More specifically, we aimed to compare early failure rates, sperm concentrations, and time to success between the two techniques. Methods We compared semen analysis data from men following vasectomy using two occlusion techniques. Data on intraluminal cautery came from a prospective observational study conducted at four sites. Data on ligation and excision with fascial interposition came from a multicenter randomized controlled trial that evaluated the efficacy of ligation and excision with versus without fascial interposition. The surgical techniques used in the fascial interposition study were standardized. The surgeons in the cautery study used their customary techniques, which varied among sites in terms of type of cautery, use of fascial interposition, excision of a short segment of the vas, and use of an open-ended technique. Men in both studies had semen analyses two weeks after vasectomy and then approximately every four weeks. The two outcome measures for the analyses presented here are (a) time to success, defined as severe oligozoospermia, or <100,000 sperm/mL in two consecutive semen analyses; and (b) early vasectomy failure, defined as >10 million sperm/mL at week 12 or later. Results Vasectomy with cautery was associated with a significantly more rapid progression to severe oligozoospermia and with significantly fewer early failures (1% versus 5%). Conclusion The use of cautery improves vasectomy outcomes. Limitations of this comparison include (a) the variety of surgical techniques in the cautery study and differences in methods of fascial interposition between the two studies, (b) the uncertain correlation between sperm concentrations after vasectomy and the risk of pregnancy, and (c) the use of historical controls and different study sites.

Background Vasectomy techniques have been the subject of relatively few rigorous studies. The Royal College of Obstetricians and Gynecologists [ 1 ] noted that more research is needed to compare different methods of vasectomy. Experts have recommended the use of fascial interposition or cautery [ 1 ], cautery with clips [ 2 ], and cautery with fascial interposition [ 2 , 3 ]. We have shown in a randomized controlled trial that fascial interposition can reduce failure rates by about half when the occlusion method is suture ligation with vas excision [ 4 ]. However, the failure rates defined by semen analyses in that study were relatively high, with a failure rate of 5.7% even in the fascial interposition group. At an experts' meeting organized by Family Health International (FHI) and EngenderHealth in April 2001, experts reviewed data on vas occlusion techniques, including the preliminary pooled results from the randomized controlled trial of fascial interposition mentioned above. Given the apparently high rate of vasectomy failures in that study, the experts recommended that an observational study be done to assess sequential sperm concentrations after vasectomies with cautery to see if there was a qualitative difference between sperm concentrations after vas occlusion by cautery and sperm concentrations after occlusion by ligation and excision. Some participants suggested that if such a study showed a clear difference in the rate of success or the frequency of apparent recanalizations between cautery and ligation and excision with fascial interposition, then this might provide sufficient evidence for service providers to consider switching to the use of a cautery technique [ 5 ]. Based on that recommendation, we conducted an observational study of vasectomy by cautery [ 6 ] and then conducted this comparative analysis. Even if some researchers decide that additional research is needed, the data from this comparative analysis should provide a strong basis for planning further research. The objectives of the analysis presented here were to compare early failure rates, sperm concentrations, and time to success for vas occlusion by ligation and excision with fascial interposition versus those for vas occlusion by cautery. Methods The methods of the fascial interposition and cautery studies have been previously described [ 4 , 6 ]. In brief, the fascial interposition study involved eight sites in seven countries. It was a randomized controlled trial comparing two occlusion techniques: ligation and excision with versus without fascial interposition. All surgeons used the no-scalpel approach to the vas and a standardized occlusion technique. The vas was occluded using two silk sutures, and an approximately 1-cm segment of vas between the ligatures was excised. For the fascial interposition technique, a suture was used to contain the testicular end of the vas inside the fascial sheath; the prostatic end remained outside [ 4 ]. The study was halted following a planned interim analysis that demonstrated a clear benefit from the use of fascial interposition [ 7 ]. Of 419 men who had fascial interposition in that study, 410 were included in this comparative analysis. Nine men were excluded because of lack of follow-up data. The cautery study involved four sites in four countries. It was designed to estimate the effectiveness of cautery as currently performed and to describe the trends in sperm counts after vas occlusion by cautery. The surgeons used their customary cautery occlusion techniques, which differed among the sites. At two sites, surgeons used electro-cautery without fascial interposition: one with and one without excision of a short segment of the vas. At the other two sites, they used thermal cautery with fascial interposition: one with and one without an open-ended technique and excision of a short segment of the vas. Three sites used the no-scalpel approach to the vas. Graphic depictions of the four methods used have been published [ 6 ]. Of 400 men enrolled, 389 are included in this comparative analysis. Eleven men were excluded because of lack of follow-up data. Follow-up and semen analysis methods Both studies included frequent semen analyses, beginning at two weeks after vasectomy. In the fascial interposition study, subsequent semen analyses were conducted every four weeks until a man had provided two consecutive azoospermic specimens, was declared a vasectomy failure, or reached the end of study follow-up at 34 weeks. In the cautery study, after the first sample at two weeks, subsequent semen analyses were conducted at weeks 5, 8, 12, 16, 20, and 24, regardless of semen analysis findings. In both studies, participants were asked to record all ejaculations between semen analyses on a wallet-sized card, which they gave to study personnel at each follow-up visit. Semen analyses methods for both studies were based on World Health Organization recommendations, but the methods differed somewhat between the two studies. Freshly collected semen was examined in the fascial interposition study, and data were obtained on sperm concentration, motility, and viability. For the cautery study, two of the four sites did not routinely collect fresh specimens, so semen analysis data from those two sites were limited to sperm concentrations. Therefore, for this comparative analysis, we did not consider sperm motility as an outcome measure. In addition, specimens showing azoospermia or very low sperm concentrations were centrifuged in the fascial interposition study but not in the cautery study. During both studies, the laboratories conducted periodic quality-control tests. Outcome measures In both studies, we used frequent semen analyses rather than pregnancy as the vasectomy effectiveness outcome measure, to minimize the risk of pregnancy, sample size, and study duration. Vasectomy success is commonly defined as two azoospermic specimens [ 2 ]. However, small numbers of nonmotile sperm may persist for many months in some men. Surgeons' experience [ 8 ] and guidelines recently published by the British Andrology Society [ 9 ] suggest that low concentrations of nonmotile sperm (<100,000 sperm/mL) are of less concern than higher concentrations. We found in the fascial interposition study [ 4 ] that severe oligozoospermia (<100,000 sperm/mL) was a more robust measure of success than was azoospermia, at least for research purposes. In both studies, men of different ages tended to reach severe oligozoospermia at about the same time, but older men took longer to reach azoospermia than did younger men. Consequently, we used two definitions for vasectomy success for this comparative analysis. Our primary definition of success was severe oligozoospermia, defined as <100,000 sperm/mL in two consecutive specimens taken at least two weeks apart. Our alternate definition of success was the occurrence of two consecutive azoospermic specimens taken at least two weeks apart, with no subsequent samples showing sperm concentrations of 100,000 sperm/mL or higher. Motility was not considered, for reasons mentioned earlier. The date of success was the date of the first of the two oligozoospermic or azoospermic semen samples. For early failure, we used a criterion of >10 million sperm/mL at week 12 or later, regardless of motility. This is an adaptation of Alderman's criteria specifying 5 million motile sperm/mL or more as evidence of "overt failure" [ 10 ]. This definition is different from the definitions of failure used by each of the two studies, but it was necessary for a comparative analysis because some sites in the cautery study did not measure sperm motility. Thus, the failure rates reported here may differ slightly from the failure rates reported by each study. In addition, to avoid bias from the two studies' different lengths of follow-up, we included semen analysis data from the fascial interposition study through only 26 weeks of follow-up. The data collection forms, study monitoring, and laboratory quality-control procedures were similar for both studies, though only one research site was common to both. Both studies were organized and managed by researchers and staff at FHI and EngenderHealth, and both received approval from FHI's institutional review board and from institutional review boards at the study sites. Statistical methods Kaplan-Meier product-limit estimates of the probabilities of severe oligozoospermia, at each scheduled week of follow-up through week 24, and their 95% confidence intervals (CIs) were produced overall, by study group (i.e., fascial interposition and cautery), and by study group and age group (i.e., <35 years and 35 years and older). Peto's standard error [ 11 ] was used to compute the 95% CIs. The Kaplan-Meier probabilities were compared between the study groups using a two-sided log-rank test with an alpha of 0.05. Given that the participants in the fascial interposition study had a longer follow-up period than the participants in the cautery study (34 versus 24 weeks), the information on the fascial interposition participants was censored at the 26-week visit for the purpose of this comparative study. The comparison of failure rates between the two study groups was based on a Fisher exact test with a two-sided alternative hypothesis and an alpha of 0.05. In addition, a logistic model was fit to estimate an age-adjusted odds ratio of the failure rates and its 95% CI. Unlike in the cautery study, participants in the fascial interposition study were discontinued after azoospermia was confirmed or after vasectomy failure was declared. Therefore, for the purpose of comparing the distribution of the participants in the different sperm concentration categories by week of follow-up, we kept azoospermic cases in the azoospermic category for all follow-up weeks after their discontinuation due to confirmed azoospermia. Similarly, we kept participants with a declared vasectomy failure in the sperm concentration category that they were in at the moment of discontinuation, for all subsequent follow-up weeks. Results Detailed results for the two studies have been reported [ 4 , 6 ]. We report here the results of the comparison of the semen analysis data from the two studies. Baseline population data Among the baseline population characteristics (Table 1 ), age distribution was somewhat different between the two studies, with a younger population in the fascial interposition study. Marital status, number of children, use of condoms, and years of education were similar between the two studies. Analysis of early failures We found significantly fewer early failures in the cautery study than in the fascial interposition group from the randomized controlled trial: 1.0% (4/389) versus 4.9% (20/410) (p = 0.0014 by the Fisher exact test). The adjusted odds ratio was 4.8 (95% CI, 1.6–14.3), indicating nearly a five-fold higher risk of early failure in the fascial interposition study than in the cautery study. No significant age effect was detected (data not shown). Sperm concentrations The distribution of sperm concentrations by week is shown for the two studies (Figure 1 ). The difference in early failures can be appreciated by examining the percentages of men with high sperm concentrations. In the fascial interposition study, the percentage of men with sperm counts of 10 million or more stayed about the same from 6 to 26 weeks. However, in the cautery study, the percentage decreased dramatically from 5 to 8 to 12 weeks. This difference was probably due to recanalizations, which become apparent in the first 6 to 10 weeks after the procedure. Time to success Life-table analyses of time to success showed that the participants in the cautery study reached severe oligozoospermia significantly more rapidly than did those in the fascial interposition study (p = 0.0049) (Figure 2 ). Ninety-seven percent of the men in the cautery study had reached success by 12 weeks, while only 91% in the fascial interposition study had reached success by 14 weeks. The analyses of the data stratified by age group showed similar results (data not shown). Using the time to azoospermia outcome, the difference between the two groups was also significant (p < 0.0001) (data not shown). Discussion The difference in the observed failure rates suggests that vas occlusion techniques that include cautery are significantly more effective than ligation and excision plus fascial interposition, at least based on semen analysis. We believe that most of the failures in the fascial interposition group were due to early recanalizations within the first two to three months after vasectomy (data not shown). Possible explanations for the surprisingly high failure rate among the fascial interposition group have been previously presented and discussed [ 4 ]. Recent reports suggest that pregnancy rates may be higher in low-resource settings, where semen analysis is usually not available and where most vasectomies are done by ligation and excision [ 12 , 13 ]. The use of cautery devices may have the potential to reduce failure rates in low-resource settings. Which cautery technique is best? When a ligation and excision technique is used, we have shown that fascial interposition provides an important improvement in effectiveness [ 4 ]. The addition of fascial interposition may be less important when cautery is used as the occlusion method, but this study was not designed to answer that question. Schmidt was a pioneer in the use of cautery for vasectomy. His preferred technique was thermal cautery of about 5 mm of each end of the vas, combined with fascial interposition and no excision [ 3 ]. Two of the sites in the observational cautery study used techniques very similar to Schmidt's technique. The other two sites used electro-cautery without fascial interposition, an approach that has been used for many years at the Elliot-Smith Clinic in Oxford [ 14 ] and similar to one used for many years by Marie Stopes clinics with excellent results [ 15 ]. Little evidence is available to support one type of cautery over another. Schmidt [ 16 ] preferred thermal cautery to electro-cautery based on histological examination of specimens at vaso-vasostomy, and Li [ 17 ] found a lower failure rate with thermal cautery than with electro-cautery, but the difference in Li's study was not statistically significant. The length of the vas segment that is cauterized can vary by the type of cautery. In the United States and Canada, marketed thermal cautery "vasectomy tips" have a functional length of about 0.8 cm, so it would be difficult to cauterize more than 1 cm of each end. The cautery tip of the Sturgeon cautery device used by Schmidt was 0.5 cm long. However, the tips available for electro-cautery devices are much longer, which could permit cauterization of longer segments of the vas and potentially cause more difficulty in the event of a request for reversal. Since several reports suggest that the combination of thermal cautery plus fascial interposition is one of the most effective methods available [ 18 ], this procedure can be recommended with few reservations. However, there is at least one other reason to consider the inclusion of fascial interposition in the vasectomy procedure, especially in low-resource settings. If providers in low-resource settings adopted a cautery technique, cautery instruments could occasionally become unavailable for various reasons. In those cases, a provider might want to be able to perform fascial interposition as part of a ligation and excision procedure. Schmidt suggested that cautery should not be combined with suture or clip ligation of the vas [ 3 ]. He noted that while blood vessels will thrombose after ligation, the vas will remain open. Thus, using a ligature in addition to cautery could reduce the value of cautery by causing necrosis of some or the entire cauterized end, potentially reducing the benefit from cautery. None of the cautery techniques in this study included the use of ligatures or clips on top of a cauterized vas. Study limitations Limitations of this comparison include (a) the variety of vas occlusion techniques used in the cautery study and differences in methods of fascial interposition between the two studies, (b) the uncertain correlation between post-vasectomy sperm concentrations and the risk of pregnancy, (c) the lack of sperm motility data and centrifugation for the cautery study, and (d) the use of different study sites and surgeons in the two studies. The difference in time to vasectomy success could in part be related to the lack of centrifugation in the cautery study. However, the difference in centrifugation between the two studies would not have affected the detection of early failures, especially since the cautery study gathered semen samples throughout the 24-week follow-up period (i.e., men continued providing semen samples even after they reached azoospermia). Another potential limitation of this comparison was the difference in follow-up schedules between the two studies. In addition, other factors could have an unknown impact on the comparability of the data. Even though the results are encouraging for the use of cautery in vasectomy, we have to be cautious about making definitive statements based on this nonrandomized comparison. Future vasectomy research Results from this comparative analysis suggest that cautery may be a more robust and less technique-dependent method than is fascial interposition. However, additional research would be useful to directly compare the effectiveness of the following standardized techniques in a randomized controlled trial: cautery without fascial interposition, cautery with fascial interposition, and ligation and excision with fascial interposition. Additional research would also be of interest to compare an open-ended procedure with a closed-end procedure. Several investigators have suggested that leaving the testicular end open reduces post-vasectomy pain, but no randomized controlled trials have examined this issue [ 1 , 18 ]. Conclusions We compared data from two prospective multicenter studies conducted using similar methodologies. We found that the use of cautery as part of the vasectomy procedure significantly reduced vasectomy failure rates compared with ligation and excision plus fascial interposition as part of the procedure. It is unclear from our results and those of others whether fascial interposition used with cautery improves vasectomy success rates when compared with cautery alone. Competing interests The authors declare that they have no competing interests. Authors' contributions DS participated in the conception, design and analysis of the study, and drafted the manuscript. BI participated in the conception, design, and analysis of the study, and was primarily responsible for managing the study implementation. MC participated in the design of the study and was primarily responsible for the statistical analysis. ML participated in the conception, design and analysis of the study. MB participated in the conception, design, management and analysis. All authors reviewed and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529470.xml

212695

V(D)J Recombination and the Evolution of the Adaptive Immune System

In order for the immune system to generate its vast numbers of receptors, B- and T-cell receptor genes are created by recombining preexisting gene segments. This well- coordinated set of reactions is explained

The immune system needs to be able to identify and ultimately destroy foreign invaders. To do so, it utilizes two major types of immune cells, T cells and B cells (or, collectively, lymphocytes). Lymphocytes display a large variety of cell surface receptors that can recognize and respond to an unlimited number of pathogens, a feature that is the hallmark of the “adaptive” immune system. To react to such a variety of invaders, the immune system needs to generate vast numbers of receptors. If the number of different types of receptors present on lymphocytes were encoded by individual genes, the entire human genome would have to be devoted to lymphocyte receptors. To establish the necessary level of diversity, B- and T-cell receptor (BCR and TCR, respectively) genes are created by recombining preexisting gene segments. Thus, different combinations of a finite set of gene segments give rise to receptors that can recognize unlimited numbers of foreign invaders. This is accomplished by a supremely well-coordinated set of reactions, starting with cleaving DNA within specific, well-conserved recombination signal sequences (RSSs). This highly regulated step is carried out by the lymphocyte-specific recombinationactivating genes ( RAG1 and RAG2 ). The segments are then reassembled using a common cellular repair mechanism. For foreign invaders and their proteins (antigens) that are not part of the host to elicit an immune response, the immune system must be able to recognize countless numbers of antigens. For obvious reasons, an unlimited number of unique antigen receptors cannot be genetically encoded. Rather, the necessary diversity in receptors is achieved by creating variations in the antigen-recognition regions of the receptors of both B cells and T cells. These regions are created by the pairing of two different protein segments, called polypeptide chains (heavy [H] and light [L] chains in the case of the BCR and α and β chains in the case of the TCR), which form a cleft that provides a binding site for the antigen. The mechanism that generates variation in the antigen-binding pockets of these receptors involves mixing and matching variable (V), diversity (D), and joining (J) gene segments in a process called V(D)J recombination. To assemble a single functional receptor, preexisting V, D, and J gene segments are rearranged to yield a contiguous V(D)J region, just upstream of another element of the receptor, the constant (C) region ( Figure 1A ). Figure 1 V(D)J Recombination Takes Place within the BCR and TCR Loci (A) Schematic of a receptor locus.V, D, and J segments are found just upstream of the constant region. (B) A cartoon view of a VJ recombination reaction. V segments (red) are flanked by RSSs with 12 bp-long spacers (green), while the J segments are flanked by RSS with 23 bp-long spacers (orange). Breaks are introduced directly between the heptamer and the coding sequence, and a CJ is formed between a V and a J segment, while the RSS ends are put together to form an SJ within a circular DNA that is later lost. Symbols: P, promoter; E, enhancer. The BCR H chain and the TCR β chain consist of V, D, and J segments, while BCR L chains and the TCR α chain are comprised of only V and J segments. The number of each type of segments within the chains allows for a large but finite combinatorial possibility in rearrangement, a phenomenon termed combinatorial diversity ( Table 1 ). However, variations generated by V(D)J recombination are uncountable because they do not simply rely on the number of gene segments. Further diversity is introduced because the junctions between rearranged gene segments contain small insertions and deletions (junctional diversity). Finally, both BCR and TCR are heterodimers (consisting of two unmatched polypeptides), so the possibilities of different pairing between the chains can also increase variation. Successful V(D)J rearrangement is clearly useful in terms of antigen recognition, and it is absolutely required for the development and survival of B and T cells. Table 1 Diversification of BCRs and TCRs Number of V, D, and J segments contributes to combinatorial diversity. Further changes are introduced by junctional diversity, to give the total number of BCR and TCR repertoires V(D)J Recombination: A Cut-and-Paste Reaction In the first part of the “cut-and-paste” reaction, breaks within both strands of the DNA helix (double-stranded breaks) are made within the RSS sites; in the second part, the newly created breaks are repaired by the cell's general DNA repair pathway. In the initial phase, two lymphocyte-specific proteins that are encoded by the recombinationactivating genes ( RAG1 and RAG2 ) work together to recognize and bind RSSs. The complex consisting of these two proteins, RAG1–RAG2 (henceforth RAG), cuts the DNA between the rearranging DNA segments and the adjacent RSS motifs ( Figure 1B ). The second step of the reaction glues together the ends of the chromosome containing the rearranging segments, which will ultimately code for the receptor and are called coding joints (CJs). The portion of DNA between the rearranged segments is shed from the genome, but it too gets glued together in a minicircle (a signal joint [SJ]). Typically, SJs are rapidly and precisely fused, but CJs are ligated more slowly, in part because their fusion is not precise—small insertions are present quite often, and even deletions can be detected. (For detailed reviews, see Fugmann et al. 2000 ; Gellert 2002 ). Both pasting reactions are necessary for creating the receptors as well as for preventing havoc within the genome. RAGs: Indispensable for V(D)J Recombination RAG proteins carry out the first enzymatic step of the reaction—site-specific cleavage of DNA ( van Gent et al. 1995 ). Artificial expression of RAGs in mammalian cells other than B- or T-lymphocytes suggests that RAG is the only lymphocyte-specific factor required for this recombination event to occur ( Schatz and Baltimore 1988 ). Indeed, in mice whose RAG genes have been deleted ( RAG −/− ), V(D)J recombination is completely abolished, and these mice have neither mature B nor T cells ( Mombaerts et al. 1992 ; Shinkai et al. 1992 ). A similar type of immunodeficiency, called Omenn syndrome, is seen in people with mutations in their RAG genes ( Villa et al. 1998 ). In test-tube experiments, purified RAG proteins are sufficient for cleavage of a synthetic DNA containing the appropriate RSS ( McBlane et al. 1995 ). This reaction can be subdivided into two stages. First, a nick is made in the DNA, at a specific site within the RSS, leaving specific chemical modifications at the ends ( Figure 1B ). Then, one free end with a specific (3′-hydroxyl) group forms a new chemical (diester) bond with a different chemical (phosphoryl) group on the complementary nucleotide of the opposite strand (this is called a transesterification reaction). This results in the formation of a hairpin at the coding end, while leaving the signal end blunt ( McBlane et al. 1995 ). RSSs: The Targets of the Reaction RSSs are found next to every variable (V), diversity (D), and joining (J) segment. They consist of three distinct elements: a heptamer and a nonamer sequence, separated by a spacer element—either 12 or 23 bp long ( Figure 2 ) ( Tonegawa 1983 ; Akira et al. 1987 ). Although the two RAG proteins work together in a protein complex, they do have unique functions. RAG1 binds both the 12- and 23-RSSs with equal affinities, while RAG2 does not bind either RSS sequence. This suggests that RAG1 forms the initial complex with DNA, which then recruits and is stabilized by RAG2 ( Fugmann et al. 2000 ). Figure 2 RSSs Consist of a Fairly Conserved Heptamer and Nonamer Sequence, Separated by a Spacer Element Heptamer is shown in red and nonamer in green. Conserved nucleotides are shown in bold. The spacer is either 12 or 23 bp long. Both the heptamer and the nonamer contain nucleotides that are absolutely required for efficient V(D)J recombination. The first three nucleotides within the heptamer are conserved, whereas mutations in the next four positions affect but do not abolish the reaction. Within the nonamer, positions 5 and 6 are conserved, while variations are tolerated elsewhere in the sequence ( Figure 2 ). The length of the spacer (but not its sequence) was thought to play an important role in regulating which elements could be recombined ( Gellert 2002 ): a 12mer differs from a 23mer by a single turn of the DNA helix, providing for the proteins that bind RSSs to remain in the same rotational phase ( Gellert 2002 ). The paper in this issue of PLoS Biology by Lee et al. (2003) now challenges this paradigm. They report that not only the length of the spacer but also its sequence are important, so that changes in the spacer sequence directly affect rates of recombination. Based on these data, they have developed an algorithm that accurately predicts the relative efficiencies of RSS binding, cleavage, and rearrangement based on nucleotide sequence. Breaks: Necessary Precursors of Recombination After RAG cuts the DNA at the RSS, the two sides of the break are different. The coding ends are closed to resemble hairpins, while the RSS ends are open and blunt. These blunt RSS ends are rejoined rapidly, forming SJs, but before the coding ends can be fused, the hairpins must be opened ( Gellert 2002 ). Normally, hairpins are opened either at the tip or on the side. In either case, an enzyme called terminal deoxynucleotidyl transferase can add a small amount of random (nontemplated) nucleotides to freed ends, and this phenomenon of N-nucleotide addition contributes to junctional diversity of the CJs ( Bassing et al. 2002 ). When the hairpin is opened on the side, the reaction leaves a short overhang on one strand. When the overhangs are filled, palindromic sequences are generated, and these modifications are referred to as “P-nucleotides” ( Bassing et al. 2002 ). These, too, contribute to the diversity of the CJs and ultimately, when the rearranged gene produces protein, to more variability in the antigen-binding pocket of the receptor. Resolution: The Final Step As mentioned, RAG −/− mice do not have mature B and T cells. This causes a severe combined immunodeficiency syndrome (SCID) characterized by a complete block in B- and T-cell development, but no other defects ( Mombaerts et al. 1992 ; Shinkai et al. 1992 ). However, there are other molecular deficiencies that also have a SCID phenotype. One of these is mapped to the enzyme DNA protein kinase (DNA-PKcs), which is required for the proper joining of DNA ends ( Bosma et al. 1988 ). Mice deficient in DNA-PKcs can initiate V(D)J recombination, but cannot form the CJs ( Gao et al. 1998 ). These mice are also sensitive to processes that induce DNA double-stranded breaks, such as ionizing radiation ( Gao et al. 1998 ). Hence, the repair pathway responsible for fixing DNA breaks caused by radiation also creates CJs. Indeed, along with DNA-PKcs, other proteins of this nonhomologous end-joining repair pathway are important for the completion of V(D)J recombination (such as Ku70 and Ku80, Artemis, XRCC4, and DNA ligase IV) ( Bassing et al. 2002 ). An Evolutionary Model of V(D)J Recombination From the discovery of the RAG genes on, investigators have suspected that V(D)J recombination may be the result of the landing of a transposable genetic element (a “jumping gene” or transposon) into the vertebrate genome. The clues were many. Firstly, the compact organization of the RAG locus resembles a transposable element ( Schatz et al. 1989 ). Secondly, RAGs cut the DNA after binding RSSs throughout the BCR and TCR loci ( Gellert 2002 ). RSSs resemble the ends of other transposable elements. Biochemically, the reaction shares characteristics with enzymes found in other transposable elements ( Gellert 2002 ). Finally, these genes appear abruptly in evolution: they are present in the jawed vertebrates (like the shark), but not in more ancient organisms ( Schluter et al. 1999 ). A current model is that an ancient transposon containing the RAG genes flanked by RSS ends “jumped” into an area of the vertebrate lineage containing a primordial antigen receptor gene, separating it into pieces. When the transposon lifted off, it left the RSSs behind. Multiple rounds of transposition and duplication eventually gave rise to our TCR and BCR loci. Mechanistically, transposition could happen even now, if the DNA segment containing the SJs is not ligated into a minicircle. These unsealed SJs are able to insert into heterologous sequences in vitro, and if they are allowed to float free in the cell, they have the potential to invade the genome. This phenomenon could give rise to certain types of chromosomal translocations prevalent in some types of B- and T-cell cancers ( Shih et al. 2002 ). Generally, repair processes in the cell actively suppress this phenomenon by rapidly ligating SJs. Conclusion V(D)J recombination is absolutely crucial for the adaptive immune response. In its absence, our immune system is compromised. When it is not properly controlled, it gives rise to chromosomal translocations and B-and T-cell cancers. The elucidation of all steps of the reaction and attempts to understand exactly how these steps are regulated to avoid disastrous side effects are areas of study that have occupied researchers in the past and will continue to do so in the future.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC212695.xml

544893

Hypoxia-mediated apoptosis in oral carcinoma cells occurs via two independent pathways

Background We are attempting to elucidate the mechanism of apoptotic cell death induced by hypoxia in oral cancer cells. Since hypoxia can render solid tumors more resistant to radiation and chemotherapy, understanding the pathways involved in hypoxia-induced apoptosis of oral cancer cells would be of significant therapeutic value. Results Here we showed that oral cancer cells from primary tumor and lymph node metastasis undergo apoptosis after 24 to 48 h of hypoxia. During hypoxic growth, an increase in caspase-3 proteolytic activity was observed, accompanied by the cleavage of PARP (poly (ADP-ribose) polymerase) indicative of caspase activity. In addition, hypoxic stress also lead to activation of caspase-8, -9, and -10 but not -1, elicited the release of cytochrome C into the cytosol, and resulted in internucleosomal DNA fragmentation. Conclusion These results show that hypoxia-induced apoptosis in oral carcinoma cell lines relies on both intrinsic (mitochondrial) and extrinsic (cell death receptor mediated) pathways. This novel evidence will assist in designing more efficient combination chemotherapy approaches as promising strategy for the treatment of oral cancers.

Background Oral cancer is one of the 10 most frequently occurring cancers worldwide, and its incidence in Europe and the United States ranges from 2% to 6% among all cancer patients [ 1 , 2 ]. The 5-year survival rate of less than 50% has not substantially improved over the past several decades, since many oral carcinomas respond poorly to chemotherapy approaches and their responses to radiation therapy have been highly variable. Hypoxia, a reduction in the level of tissue oxygen tension, occurs during acute and chronic vascular disease, pulmonary disease and cancer, and can lead to apoptotic or necrotic cell death [ 3 , 4 ]. Fast growing tumors become hypoxic because newly developed blood vessels are inefficient and have poor blood flow. Although hypoxia is toxic to both cancer cells and normal cells, tumor cells can undergo genetic and adaptive changes in response to hypoxia that allow them to survive and proliferate [ 5 ]. Thus, hypoxic growth can result in a tumor with more aggressive growth characteristics and more malignant phenotype [ 3 ]. Although micro-environmental irregularities in solid tumors have been well documented, little is known about how different types of tumor cell phenotypes tolerate and respond to these conditions. Apoptotic cell death is controlled by pro-apoptotic caspases, proteases that are synthesized as inactive precursors and activated by proteolytic processing [ 6 ]. The apoptotic cascade can be initiated via two major pathways, involving either the release of cytochrome C from the mitochondria (mitochondrial pathway) [ 7 ] or activation of death receptors in response to ligand binding (death receptor pathway) [ 8 ]. Upon triggering of either pathway, caspases, the final executioners of apoptosis, are activated, causing degradation of cellular proteins and leading to typical morphological changes such as chromatin condensation, nuclear shrinkage, and the formation of apoptotic bodies [ 9 ]. Both pathways are differentially involved in the cellular response to diverse apoptotic stimuli [ 10 , 11 ]. The majority of chemotherapeutic agents trigger the mitochondrial pathway, but the death receptors have also been reported to be involved in chemotherapy-induced apoptosis [ 12 ]. Death ligands such as TNF-α or CD95L recruit, via the adapter molecule FADD, cytoplasmic monomeric initiator caspase-8 to their surface receptors, resulting in dimerization and activation of caspase-8 [ 13 , 14 ]. Active caspase-8 cleaves and activates downstream effector caspases including caspase-3, -6 or -7, which degrade a broad range of cellular proteins and trigger the appearance of the apoptotic morphology [ 6 , 15 ]. On the other hand, mitochondria are important regulatory sites of the apoptotic process [ 16 ]. Defects in mitochondrial function result in release of cytochrome C, which can associate with Apaf-1 (apoptosis protease activating factor) and pro-caspase-9. The observation that chemical inhibition of caspase-9 blocks hypoxia-induced apoptosis points to a role of the complex in hypoxia-induced apoptosis [ 17 , 18 ]. This activation complex results in auto-processing of caspase-9 and further activation of downstream caspases, such as caspase-3 [ 19 , 20 ]. Activation of caspase-3 has been linked to the proteolytic cleavage of cellular substrates including poly-ADP-ribose-polymerase (PARP) [ 21 ], and is also necessary for the nuclear changes and chromatin condensation associated with apoptosis [ 22 ]. The low oxygen tension in hypoxic tumors is known to interfere with the efficacy of chemotherapy or radiotherapy. Also, hypoxia-induced apoptosis may impose a selection pressure favoring growth of more resistant tumor cells. However, the factors leading to hypoxia-induced apoptosis and their relative contribution to intrinsic and extrinsic apoptotic pathways are not well characterized. In the present study, we determined which factors in the mitochondria-dependent and -independent apoptosis pathways are activated in oral cancer cells. We observed that hypoxia-induced apoptotic cell death occurs through activation of caspase 8, but also cytochrome C release, caspase-9 activation, and results in caspase 3 processing, PARP cleavage, and DNA fragmentation. These results suggest that hypoxia-induced apoptosis in oral carcinomas cells relies on both intrinsic (mitochondrial) and also extrinsic (cell death receptor mediated) pathways. Results Hypoxia condition Direct measurements of oxygen tensions in human tumors show a range of median oxygen tensions from 1.3 to 3.9 % (10–30 mm Hg), with readings recorded as low as 0.01 % (0.08 mm Hg) under severe hypoxia, whereas in normal tissues O 2 levels can range from 3.1 to 8.7 % (24–66 mm Hg) [ 25 ]. Severe hypoxia stimulates cells to undergo apoptosis, whereas oxygen levels above 0.5 % prevent cell death, indicating tight regulation of cellular responses to the microenvironment. Critical O 2 levels (hypoxic thresholds) characterize the upper limit of the hypoxic range below which activities and functions progressively become restricted. These levels can encompass O 2 partial pressures from 35 mm Hg (start of reduced cell death in conventional photodynamic therapy or restricted efficacy of some immunotherapy) to 0.02 mm Hg; below this level, cytochromes aa3 and c are no longer fully oxidized [ 25 ]. During severe hypoxia or anoxia, a cascade of events is initiated that leads to global apoptotic cell death, thereby preventing the accumulation of cells with hypoxia-induced regulatory responses or mutations [ 26 ]. Thus, a condition of less severe hypoxia (1 %) was chosen to be able to monitor the response of viable cells under this condition. Morphological changes and DNA fragmentation during hypoxia Trypan Blue dye exclusion was used to quantitate the number of viable cells after 24 and 48 h of hypoxia. TUNEL assays were used to visualize apoptotic cells, and DNA fragmentation into oligonucleosome-sized fragments, indicative of apoptotic cellular death, was monitored by gel electrophoresis [ 27 ]. The percentage of viable cells was steadily reduced in hypoxia-treated cells compared to normoxic control cells (Figure 1 ). Whereas approximately 5 – 15 % reduction of viability was seen at 24 h, all hypoxic cell lines declined further at 48 h to 85 – 75 % viability compared to normoxic growth. Hypoxic growth induced morphological alterations typical for apoptotic cell death, as determined by TUNEL assays (Figure 2 ). The hypoxia-treated cells showed extensive nuclear dye staining indicative for DNA breakage and cell death. This effect was detectable at 24 h (data not shown) and very apparent after 48 h of hypoxia, when most of the treated cells showed the typical morphology of apoptotic nuclear condensation. In contrast, nuclear staining was much weaker and less pronounced for normoxic cells up to 48 h incubation. Hypoxic conditions induced the chromatin alterations typical for apoptotic cell death, as demonstrated by fragmentation of chromosomal DNA into nucleosomal DNA ladders (Figure 3 ). Exposure to hypoxia for 24 or 48 h resulted in a time-dependent increase in DNA fragmentation in all four oral cancer cells, whereas no internucleosomal DNA fragmentation was observed in the normoxic control cultures (Figure 3 ). It is also apparent that the two metastatic lines 686Ln and 1386Ln had less extensive DNA fragmentation compared to their respective primary tumor lines 686Tu and 1386Tu. Figure 1 Viability assay for oral carcinoma cells under hypoxia. The numbers of viable (Trypan blue-excluding) cells were determined after 24 or 48 hours, and the percent of viability of hypoxic cells plotted relative to normoxic control cells; viabilities under normoxia were 100% for all cells. The results show the mean (± SD) of three independent experiments. 686Tu = open square; 686Ln = closed square; 1386Tu = open circle; 1386Ln = closed circle. Figure 2 Hypoxia-induced nuclear TUNEL staining in oral carcinoma cells. The cells were incubated for 48 h in hypoxic or normoxic conditions, and photographs were taken after TUNEL staining of cells with DAB. Few apoptotic nuclei were observed in normoxic cells, but exposure to hypoxia for 48 h induced nuclear DNA condensation and fragmentation. Cells with nuclei showing strong chromatin condensation and nuclear fragmentation were considered apoptotic. Figure 3 Effect of hypoxia on apoptosis induction as determined by DNA fragmentation. Nucleosomal DNA fragments in 686 (A) and 1386 (B) cells were analyzed by gel electrophoresis after 24 or 48 h hypoxia. Apoptosis was confirmed by the appearance of a ladder of oligonucleosomal DNA. M, molecular standard; P, positive DNA ladder control. Processing of caspases, PARP cleavage and cytochrome C release The effects of hypoxia treatment on activation of key caspases and PARP in the four cell lines 686Tu/Ln and 1386Tu/Ln was determined by Western blotting using antibodies that recognize both full-length and cleaved proteins (Figures 4 and 5 ). Growth of cells under hypoxia caused a time-dependent processing of caspase-3, -8 and -9. For all four cell lines, hypoxia resulted in enhancement of procaspase-3 (32 kD) cleavage into the two immunoreactive fragments of ~20 and ~11 kD at the 24 and 48 h time points (Figures 4A and 5A ). This treatment also resulted in cleavage of the 47 kD procaspase-9 to yield fragments of ~37 and ~20 kD, in parallel to caspase-3 cleavage (Figures 4C and 5C ). Furthermore, caspase 8 was present primarily as ~55 kD pro-form in normoxic cells, whereas exposure to hypoxia resulted in its time-dependent processing to the ~32 kD active form (Figures 4B and 5B ). These observations point towards involvement of both caspase-9 and caspase-8 in hypoxia-mediated cleavage of caspase-3 in all four cell lines. They suggest that a cascade of caspase activation occurs through the mitochondrial and also through the cell death receptor pathway in these cells in response to hypoxia. Figure 4 Response of apoptosis-related proteins to hypoxia. Western blot analysis of 686Tu (Tu) or 686Ln (Ln) cell extracts (30 μg each lane) after 24 or 48 hours of hypoxia (H) or normoxia (N) treatment. A: Cleavage of procaspase-3 (32 kD) into a 20 and 11 kD species; B: Cleavage of procaspase-8 (55 kD) into the 32 kD product (23 kD product not shown); C: Cleavage of procaspase-9 (47 kD) into lower mol. weight products; D: Processing of PARP (113 kD) into the typical 89 kD protein and a lower molecular weight product (not shown); E: cytochrome C (15 kD) release into cytosolic fraction; F: Re-probing for β-actin as an internal loading control. Data are representative of at least two independent experiments with similar results. Figure 5 Response of apoptosis-related proteins to hypoxia. Western blot analysis of 1386Tu (Tu) or 1386Ln (Ln) cell extracts (30 μg each lane) after 24 or 48 hours of hypoxia (H) or normoxia (N) treatment. Other legend details are as for Figure 4. To gain further insight into the role of mitochondria in this process, the extent of cytochrome C release under hypoxia was analyzed. Translocation of cytochrome C from the mitochondria to the cytosol was detected in all four cell lines after 24 h, and more pronounced after 48 h, of hypoxia (Figures 4E and 5E ). This release of cytochrome C was a controlled event and not due to physical disruption of mitochondria, since no signal for intra-mitochondrial cytochrome oxidase could be detected in the same cytosolic fractions under these conditions (data not shown). Thus, these results demonstrate that hypoxia induced the release of cytochrome C from intact mitochondria. Since we observed that hypoxia activated caspase-3 in the oral carcinoma cells, we investigated the cleavage of the caspase-3 substrate PARP under hypoxic versus normal growth. Clearly, cleavage of PARP, as indicated by a decrease in the full-length 113 kD protein and appearance of the 85 kD cleaved PARP product, was prominent in hypoxic cells, whereas it was almost completely absent in normoxia cells (Figures 4D and 5D ). A small amount of cleaved PARP was already found after 24 h hypoxic conditions, and this effect was much more pronounced at 48 h. Only very small amounts of PARP cleavage product could be detected in the normoxic 1386 cell line pair, whereas for the 686 pair PARP cleavage appeared undetectable. Caspase activities during hypoxia-mediated apoptosis As caspases are early effectors for triggering apoptosis, assays to determine caspase enzymatic activities further substantiated our findings that hypoxia-induced apoptosis occurs through both intrinsic (mitochondrial) and also extrinsic (cell death receptor mediated) pathways. We examined caspase proteolytic activities in cell extracts using fluorogenic peptide substrates specific for individual caspases. These substrates are conjugated with AFC or AMC and have aspartic acid residues at P1 positions, a requirement for caspase proteolysis. Cleavage of these substrates after the aspartic acid residue results in release of unbound AFC or AMC which can be monitored fluorometrically. Detergent extracts prepared from cells after exposure to hypoxia or normoxia for either 24 or 48 h were tested for caspase cleavage activities, and specific inhibitors for control measurements were used as described in Materials and Methods. N-Ac-DEVD-AFC is cleaved by caspase-3 and -7, but may also be cleaved by other caspases, N-Ac-YVAD-AFC is cleaved by caspase-1, N-Ac-IETD-AMC by caspase-8, N-Ac-LEHD-AFC by caspase-9, and N-Ac-AEVD-AFC by caspase-10. Activities of caspase-3, -8, -9 and -10 were clearly and consistently elevated during hypoxia treatment for up to 48 h compared to normoxic growth for the 686 (Figure 6 ) and 1386 (Figure 7 ) cell line pairs. In contrast, only low caspase-1 activities were found for all four cell lines in normoxic condition, and these were not substantially altered after hypoxia challenge at any of the times examined (Figures 6A and 7A ). Figure 6 Hypoxia-stimulated caspase activities in 686 oral cancer cells. The 686Tu (open bars) and 686Ln (closed bars) cells were exposed to hypoxia or normoxic control growth for 24 or 48 hours, and induction of caspase activities were assayed as described in Materials & Methods. A: caspase-1; B: caspase-3; C: caspase-8; D: caspase-9; E: caspase-10. The means ± S. D. of three independent experiments are shown. Figure 7 Hypoxia-stimulated caspase activation in 1386 oral cancer cells. The 1386Tu (open bars) and 1386Ln (closed bars) were analyzed; other legend details are as for Figure 6. Overall, significant induction of hypoxia-mediated cleavage activities for the N-Ac-DEVD-AFC, N-Ac-IETD-AMC and N-Ac-LEHD-AFC substrates was detected in all four cell extracts, and induction of these activities correlated well with the levels observed for caspase-3, caspase-8 and caspase-9 protein expression and processing. Thus, induction of apoptosis was preceded by the activation of activator caspase-8, initiating receptor-mediated apoptosis, and caspase-9, initiating mitochondrial apoptosis, as well as the effector caspase-3. Effects of caspase inhibitors on caspase activity profile We investigated the effects of individual cell-permeable caspase inhibitors on caspase-3, -8 and -9 activities during hypoxic growth for 48 hours (Figure 8 ). These inhibitors can enter viable cells and are covalently and irreversibly bound to their target caspases. Z-VAD-fmk was a pan-caspase inhibitor for all caspases analyzed; z-DEVD-fmk was inhibitor for caspase-3, z-LEHD-fmk for caspase-9, and z-IETD-fmk for caspase-8 activity. The presence of Z-DEVD-fmk clearly inhibited activity of its target protease caspase-3, but not caspases 8 or 9. The pan-caspase inhibitor z-VAD-fmk, as expected, diminished activities of all three tested caspases. Z-LEHD-fmk as inhibitor for caspase 9 did not affect activity of caspase-8, but did partially decrease caspase-3 activity. Finally, the z-IETD-fmk caspase-8 inhibitor also clearly decreased caspase-3 activity in addition to the target caspase. These data showed that both caspase-8 and caspase-9 contribute to the overall caspase-3 activity during hypoxic cell growth, and that those are the main caspases involved in hypoxia-mediated apoptosis activation pathways of these oral cancer cells. They suggest that both caspase-8 and caspase-9 activation pathways contributed to the activation of the major executioner caspases, such as caspases-3 and possibly caspase-7. Figure 8 Prevention of hypoxia-stimulated caspase activities by intracellular caspase inhibitors. The 686Tu (open bars) and 686Ln (closed bars) cells were grown under hypoxia for 48 hours in the presence of different cell-permeable caspase inhibitors, and caspase activities were assayed as described in Materials & Methods. A: caspase-3 activity; B: caspase-8 activity; C: caspase-9 activity. The vertical dashed columns represent cell growth in the presence of the following caspase inhibitors: z-DEVD-fmk, caspase-3; z-VAD-fmk, pan-caspase; z-LEHD-fmk, caspase-9; z-IETD-fmk, caspase-8. The means ± S. D. of three independent experiments are shown. Discussion The aim of this study was to identify factors which contribute to hypoxia-induced cell death in human oral cancer cells. The involvement of caspase pathways in induction of apoptosis of oral cancer cells during hypoxia was not previously determined. In the present study, we provide novel evidence for the participation of both initiator and effector caspases in this process (Figure 9 ). We showed that exposure to hypoxia elicits apoptotic cell death, and that this process relies on both intrinsic (mitochondrial) and also extrinsic (cell death receptor mediated) pathways. Our data showed that caspase-3, caspase-8, caspase-9, and caspase-10, but not the pro-inflammatory caspase-1, are activated during hypoxic growth. Activation of the executioner caspase-3 can be blocked in hypoxic cells by inhibitors of upstream caspases 8 or 9 during cell growth. We also observed that hypoxia-mediated apoptosis of oral cancer cells is associated with controlled cytochrome C release from mitochondria, proteolytic cleavage of PARP, and DNA fragmentation. Our results are in agreement with data on hypoxia-induced apoptosis in other cells [ 28 - 30 ]. In line with our data, others have observed activation of both caspase-9 and caspase-8 following hypoxic stress in animal models of brain ischemia [ 31 , 32 ]. Studies with caspase-9 knock-out mice demonstrated that caspase-9 is a critical upstream activator of the caspase cascade in vivo and may be essential for the processing of caspase-3 [ 33 , 34 ]. Also, earlier reports showed that chemical inhibition of caspase-9 protects against hypoxia-mediated effects [ 17 , 18 ]. Figure 9 Potential pathways leading to apoptosis induction during hypoxia treatment. Hypoxia-induced apoptosis in oral carcinoma cell lines relies on both intrinsic (mitochondrial pathways) and also extrinsic (cell death receptor mediated) pathways. Key steps are activation of procaspase-8 or procaspase-9, then procaspase-3, and the subsequent cleavage of PARP by activated caspase-3, resulting in the induction of apoptosis. On the other hand, it was suggested that key elements of the death receptor pathway are essential for hypoxia-induced apoptosis. The extrinsic pathway of apoptosis is initiated by death ligands, such as the Fas ligand or TRAIL (TNF-α related apoptosis inducing ligand), leading to the activation of caspase-8 and caspase-3 [ 35 , 36 ]. Recent studies indicate that DISC (Death Inducing Signaling Complex) formation precedes formation of Fas surface clusters, and that such clustering is dependent on DISC-generated active caspase-8 [ 37 ]. Also, TRAIL can induce receptor-mediated cell death selectively in tumor cells and is not active in non-malignant cells [ 38 , 39 ]. It was shown previously that in some tumor cells, only the receptor-independent mitochondrial pathway is activated during hypoxia without caspase-8 involvement [ 4 ]. On the other hand, there is recent evidence that TRAIL receptor-mediated apoptosis induction can be maintained and functional during hypoxic growth of tumor cells [ 39 ]. In view of the equal contribution of the caspase-8 and caspase-9 pathways established here, future work needs to examine the detailed mechanisms of receptor-mediated caspase activation with respect to TRAIL and death receptor-DISC-caspase-8 cascade, as well as the mitochondria-cytochrome C-caspase-9 cascade, and the possible involvement of HIF-1α as activator of caspases in OSCC cells. It was reported that hypoxia can induce upregulation of cell death receptors or death receptor ligands [ 40 ], and that inhibition of caspase-8 or FADD may interfere with hypoxia-induced apoptosis [ 18 , 41 ]. Our data also suggest that the activator caspase-8 is an integral component of the cell death-inducing mechanism in oral cancer cells, in agreement with other studies [ 32 ]. In receptor-mediated apoptosis, activation of caspase-8 represents a point of commitment to cell death. Thus, our data clearly show that in oral carcinoma cells two types of pathways are activated (Figure 9 ). Hypoxia-induced caspase-3 activation and DNA fragmentation have been described by others recently [ 42 , 43 ], as well as caspase activation accompanying cytochrome C release from mitochondria [ 28 - 30 ]. Such findings correlate well with our studies showing that caspases-3, -8, and -9 activity and expression was significantly higher in hypoxic than in normoxic cells, and similar caspase activation was observed in the hypoxic cerebral cortex of newborn piglets [ 34 ]. In our cell system, PARP cleavage was observed within 24 h of hypoxia treatment and was accompanied by the appearance of a ~11 kD procaspase-3 cleavage product, suggesting activation of caspase-3. Caspase-3 is an executioner caspase that can be activated by a mitochondrial pathway involving release of cytochrome C [ 44 ]; alternatively, caspase-3 can also be activated by caspase-8 [ 45 , 46 ]. The results of the present study indicate that hypoxia-induced cleavage of procaspase-3 appears to be mediated by both caspase-9 and caspase-8 pathways. Although cleavage of procaspase-9 was evident as early as 20 h into hypoxia treatment, it is possible that its activation is mediated by other caspases at earlier time points. Currently, possible involvement of other Bcl-2 family of apoptosis regulating proteins (e.g. Bad, Bag, Bak, Bik, etc.) in hypoxia-induced activation of the mitochondrial caspase cascade cannot be ruled out. The key regulator of hypoxia-induced cellular response is believed to be hypoxia inducible factor 1 (HIF-1). For all cell lines used here, we observed recently that there were several-fold increases in HIF-1α expression during hypoxia compared to normoxia (Wickramasinghe N, Banerjee K, Nagaraj N, Vigneswaran N and Zacharias W, manuscript submitted). HIF-1 can initiate apoptosis by inducing pro-apoptotic proteins such as BNIP3 or NIX, which will inhibit Bcl anti-apoptotic activity. It can also cause stabilization of wild-type p53 tumor suppressor, an effect that is lost in cells with pre-existing p53 mutations [ 26 , 47 ]. On the other hand, anti-apoptotic proteins, such as IAP-2, can be induced during hypoxia, whereas the pro-apoptotic protein Bax can be downregulated, leading to decreased accumulation of Bax in the mitochondria and thus decreased mitochondrial leakage and cytochrome C release [ 26 , 48 ]. It is apparent that during hypoxia, an intricate balance exists between factors that induce or counteract apoptosis, or even stimulate proliferation. More detailed studies are needed to define the precise mechanism for hypoxia-induced cleavage of procaspase-9 and procaspase-8; however, our results clearly demonstrate involvement of both caspase-8 and caspase-9 in hypoxia-mediated cleavage of caspase-3 and PARP. Because caspase-3 is a critical mediator of apoptosis [ 49 ] and correlates with the onset of apoptosis in oral cancer cells, it may be a potential marker for predicting response or resistance to chemotherapeutic agents in oral cancer. The detailed temporal and spatial relationship of these events to other components of the apoptotic pathway including downstream caspases remain to be determined. Recently, the targeted elimination of oral squamous cell carcinoma cells by inducing apoptosis has emerged as a valued strategy to combat oral cancer [ 50 ]. Increased mitochondrial permeability is a crucial event in many types of chemotherapy-induced apoptosis and leads to release of cytochrome C from the mitochondrial intermembrane space. Our study confirmed that the release of cytochrome C was actually augmented during hypoxic growth, indicating a possible role of cytochrome C in hypoxia-mediated apoptosis. However, it also has been reported that certain anticancer drugs induce apoptosis in oral cancer cells but do not trigger cytochrome C release, thereby suggesting that cytochrome C can be an inducer-dependent phenomenon [ 51 ]. In some of the caspase cleavage assays, slightly lower activities were found for the metastatic Ln cells compared to the corresponding primary Tu cells. Also, the final appearance of nucleosomal DNA ladders is much more pronounced in both Tu cells than in their Ln counterparts. Although some of those differences were only minor, they are in line with previous studies which demonstrated much higher resistance of metastatic OSCC lines to TRAIL-induced cell death [ 38 ] and also to TNF-α-induced apoptosis [ 52 ] than their corresponding primary tumor lines. Such differential apoptosis sensitivity has also been observed recently in a different matched cell line pair form head & neck primary and metastatic carcinoma (UMSCC101A versus UMSCC101B; unpublished observations from this lab). On the other hand, a very apparent difference among the four cells is that caspase-8 activity is in general several-fold higher in the 686Tu/Ln pair than in the 1386Tu/Ln pair, which presumably is a reflection of the different pathologic histories of the two patients from which the respective tumor tissues were derived. Conclusions In summary, we have reported that hypoxia directs apoptosis through mitochondria and cell death receptor mediated signaling pathways in oral cancer cells. We believe that this is the first report on caspase-dependent mechanisms during hypoxia in human oral cancer cells. Exposure to hypoxia lead to the activation of procaspase-9, -8, -3, and -10, cytochrome C release from mitochondria, with subsequent cleavage of PARP by activated caspase 3, finally resulting in the induction of apoptosis. The detailed molecular and sequential mechanisms of such hypoxia-induced caspase activation leading to apoptosis need further investigation. However, the knowledge of the relevant signaling cascades participating in this process can provide important insights in the mechanisms of acquired apoptotic deficiencies during malignant progression in poorly oxygenated oral carcinomas. It is well established that poor oxygenation of solid tumors is associated with poor prognosis. This may not only be due to direct effects of hypoxia on the efficacy on certain tumor treatment modalities, but also due to the evolvement of resistant tumor cells during the ontogenesis of a tumor under hypoxic conditions [ 39 ]. Our novel evidence, showing that hypoxia can induce apoptosis through both pathways, will assist in designing more efficient combination chemotherapy approaches as promising strategy for the treatment of oral cancers. Methods Cell lines MDA-686Tu (686Tu) and MDA-686Ln (686Ln) cell lines were derived concurrently from the primary tumor and lymph node metastasis, respectively, of OSCC involving the left tonsillar fossa and posterior portion of the tongue in a 49 year old man (tumor stage T3N3B). MDA-1386Tu (1386Tu) and MDA-1386Ln (1386Ln) cell lines were obtained from the primary tumor and lymph node metastasis, respectively, of a 71 year old male patient with primary hypopharynx tumor (tumor stage T4N3B). All cell lines were generous gifts from Dr Peter Sacks, New York University, New York [ 23 ]. The cell lines were routinely maintained in DMEM/F12 50/50 mix (Cambrex BioScience, Walkersville, MD) containing 10% fetal bovine serum and 0.4 μg/ml hydrocortisone at 37°C with 5% CO 2 . All protocols for the use of human cell lines in this work were approved by the Institutional Review Boards of The University of Louisville and the University of Texas at Houston. Hypoxia exposure Hypoxic conditions were produced by placing logarithmic phase subconfluent monolayer cultures, grown on 100 mm dishes, in a modular incubator chamber and equilibrating for 30 minutes with humidified gas containing 1 % oxygen, 5 % CO 2 and 94 % nitrogen. The cell lines were maintained under hypoxic conditions for periods of 24 or 48 hours. Control cells were grown in normal oxygen conditions for the same duration. After incubation, media collection and cell harvesting were done immediately within 2–3 minutes to avoid adaptation of the cells to re-oxygenation. Cell viability assays Determination of cell viability was done by Trypan Blue dye exclusion assay. Cells were grown in six-well plates (2 × 10 4 cells/well) in 3 ml medium to 70 % confluence, then washed and treated for hypoxia in DMEM/F12 medium with 10 % fetal bovine serum. For viability counting, cultures containing both dead and live cells from each well were collected, centrifuged, and resuspended in 0.5 ml FBS-free DMEM/F12. An aliquot of 0.1 ml was taken and incubated with 0.1 ml of Trypan Blue dye (0.4 %) for 5 min. Both live (unstained) and dead (blue) cells were counted in triplicate measurements from randomly selected fields in a hemocytometer. Protein extractions and Western blotting Cultured cells were rinsed with PBS, gently scraped into 1 ml of PBS, and centrifuged at 4,000 rpm for 3 min. The pellets were resuspended into RIPA buffer (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 % Triton X-100, 0.1 % SDS, and 1 mM EDTA) containing fresh protease inhibitors (0.5 mM phenylmethylsulfonyl fluoride, 10 μg/ml aprotinin, and 2 μg/ml of both leupeptin and pepstatin) (all from Sigma, St. Louis, MO). Then, cell extracts were sonicated (Model 550 Sonic Dismembrator, Fisher Scientific, Pittsburgh, PA) for 1 min (1.0 sec on/0.5 sec off pulses) and cell debris was removed by centrifugation. Proteins were quantified using the Bradford protein assay kit (Bio-Rad, Hercules, CA) and compared with a γ-globulin standard curve. Equal amounts of total proteins were separated on a SDS-polyacrylamide gel and transferred onto a nitrocellulose membrane by electroblotting overnight at 20 V. Membranes were blocked in TBS-T (10 mM Tris-HCL, 150 mM NaCl, 0.25 % Tween 20, pH 7.5) with 5 % fat-free powdered milk at room temperature for 1 h. After rinsing membranes in TBS-T, the following primary antibodies were used: rabbit polyclonal IgGs for caspase-3 (H-277), poly(ADP-ribose) polymerase (PARP) (H-250), caspase-8 (H-134), caspase-9 (H-170) (all from Santa Cruz Biotechnology, St. Cruz, CA), or mouse monoclonal β-actin antibody (Sigma, St. Louis, MO). After incubation overnight at 4°C or 1 h at room temperature, the membranes were washed four times, 10 min each, in TBS-T. Secondary antibodies used were either horseradish peroxidase-conjugated goat anti-rabbit IgG or goat anti-mouse IgG (ICN, Costa Mesa, CA), followed by five washes with TBS-T. Bands were detected using the enhanced chemiluminescence ECL substrate (Amersham Biosciences, Piscataway, NJ). For β-actin detection, previously probed membranes were soaked in stripping buffer (70 mM Tris-HCl, pH 6.8, 2 % SDS, 0.1 % β-mercaptoethanol) at 60°C for 30 min and incubation as above. Caspase assays After hypoxia treatment for 24 or 48-hours, treated and control cell cultures were rinsed once in cold PBS and collected in cold PBS by scraping. After centrifugation and removal of PBS, cell pellets were kept at -80°C until caspase assays were performed. The frozen pellets were resuspended in caspase lysis buffer (10 mM HEPES, pH 7.4, 2 mM EDTA, 0.1 % CHAPS) supplemented with protease inhibitors (5 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, 10 μg/ml pepstatin A, 10 μg/ml aprotinin, and 20 μg/ml leupeptin). Freeze-thaw cell lysis cycles were performed by alternatively transferring the samples from an ethanol/dry ice bath to a 37°C water bath five times. The supernatant was collected after 20 min of centrifugation at 12,000 rpm in a cold microcentrifuge. Assays were performed in caspase buffer (10 mM PIPES, pH 7.4, 2 mM EDTA, 0.1 % CHAPS, 5 mM dithiothreitol), to which 50 μM of substrate and 5 μl of protein extract were added to yield a final volume of 100 μl. Peptide substrates for caspase-3, N-acetyl-Asp-Glu-Val-Asp-AFC (DEVD-AFC), caspase-1, N-acetyl-Tyr-Val-Ala-Asp-AFC (YVAD-AFC), caspase-8, N-acetyl-Ile-Glu-Thr-Asp-AMC (IETD-AMC), caspase-9, N-acetyl-Leu-Glu-His-Asp-AFC (LEHD-AFC) (Biomol, Plymouth Meeting, PA) and caspase-10, N-acetyl-Ala-Glu-Val-Asp-AFC (AEVD-AFC) (Alexis, San Diego, CA) were dissolved in dimethyl sulfoxide. The respective specific inhibitors N-acetyl-Asp-Glu-Val-Asp-CHO (DEVD-CHO), N-acetyl-Tyr-Val-Ala-Asp-CHO (YVAD-CHO), N-acetyl-Ile-Glu-Thr-Asp-CHO (IETD-CHO), N-acetyl-Leu-Glu-His-Asp-CHO (LEHD-CHO) (Biomol), and N-acetyl-Ala-Glu-Val-Asp-CHO (AEVD-CHO) (Alexis, San Diego, CA) were used in control assay reactions. Assays were performed in black-wall, clear bottom plates using a Spectramax Gemini XS Microplate Spectrofluorometer (Molecular Devices); reading was at 500 nm after excitation at 405 nm for 7-amino-4-trifluoromethylcoumarin (AFC) and at 380 nm after excitation at 460 nm for 7-amino-4-methylcoumarin (AMC). The results were compared against AFC and AMC standard curves generated in parallel. Specific activity was expressed as units, with 1 unit defined as AFC or AMC release of 1 nMol/hour/μg protein. Cytochrome C release assays Cells were collected at the indicated times and washed once in ice cold PBS. Cell pellets were resuspended in cytosol extraction buffer, and cytosolic extracts were prepared by the method described previously [ 24 ]. Western blotting for cytochrome C was done with mouse monoclonal anti-cytochrome C IgG (BD Biosciences-Pharmingen, San Diego, CA) as described above; the absence of intra-mitochondrial proteins was verified by blotting for mitochondrial cytochrome oxidase with mouse monoclonal anti-cytochrome oxidase IgG (BD Biosciences-Pharmingen, San Diego, CA). Hypoxic growth in presence of caspase inhibitors Oral cancer cells 1386 and 686 were exposed to hypoxia for 48 hours in the presence or absence of individual cell-permeable inhibitors for caspase-3 (z-DEVD-fmk; 10 μM), caspase-8 (z-IETD-fmk; 20 μM), caspase-9 (z-LEHD-fmk; 20 μM), or pan-caspase (z-VAD-fmk; 10 μM) (all Santa Cruz Biotechnology, Santa Cruz, CA), and processed for caspase activity assays as above. Analysis of DNA fragmentation Apoptotic cells were detected by in situ TdT-mediated dUTP nick end labeling (TUNEL) assays using the In Situ Cell Death Detection Kit POD, and nucleosomal DNA fragments detected with the Apoptotic DNA Ladder Kit (both from Roche, Indianapolis, IN). DNA fragments were resolved on 2 % agarose gels for visualizations of apoptosis-indicative DNA ladders. List of Abbreviations AFC, 7-amino-4-trifluoromethylcoumarin; AMC, 7-amino 4-methyl coumarin; DAB, diaminobenzidine; DISC, death-inducing signaling complex; ECL, enhanced chemiluminescence; FADD, Fas-associated death domain protein; fmk, fluoromethylketone; PARP, poly (ADP-ribose) polymerase; TRAIL, TNF-α related apoptosis inducing ligand; TUNEL, TdT-mediated dUTP nick end labeling; z, benzyloxycarbonyl. Authors' contributions NSN carried out the molecular and enzymatic studies and drafted the manuscript. NV participated in the design of the study, interpretation of collected data, and contributed to the manuscript preparation. WZ conceived and directed the study, contributed its design and coordination, and participated in the interpretation and final manuscript preparation. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544893.xml

529316

DNA Methylation Profiling of the Human Major Histocompatibility Complex: A Pilot Study for the Human Epigenome Project

The Human Epigenome Project aims to identify, catalogue, and interpret genome-wide DNA methylation phenomena. Occurring naturally on cytosine bases at cytosine–guanine dinucleotides, DNA methylation is intimately involved in diverse biological processes and the aetiology of many diseases. Differentially methylated cytosines give rise to distinct profiles, thought to be specific for gene activity, tissue type, and disease state. The identification of such methylation variable positions will significantly improve our understanding of genome biology and our ability to diagnose disease. Here, we report the results of the pilot study for the Human Epigenome Project entailing the methylation analysis of the human major histocompatibility complex. This study involved the development of an integrated pipeline for high-throughput methylation analysis using bisulphite DNA sequencing, discovery of methylation variable positions, epigenotyping by matrix-assisted laser desorption/ionisation mass spectrometry, and development of an integrated public database available at http://www.epigenome.org . Our analysis of DNA methylation levels within the major histocompatibility complex, including regulatory exonic and intronic regions associated with 90 genes in multiple tissues and individuals, reveals a bimodal distribution of methylation profiles (i.e., the vast majority of the analysed regions were either hypo- or hypermethylated), tissue specificity, inter-individual variation, and correlation with independent gene expression data.

Introduction DNA methylation is indispensable for vertebrate genome function. It is involved in diverse genomic processes such as gene regulation, chromosomal stability, and parental imprinting ( Bird 2002 ), and interest in the function of DNA methylation is further heightened by the various human diseases associated with epigenetic dysfunction, a notable example being cancer ( Laird 2003 ). However, the DNA methylation profile of the human genome is still largely a mystery. The sequencing of the human genome (IHGSC 2001) and creation of a whole-genome map of single nucleotide polymorphisms (SNPs) ( Sachidanandam et al. 2001 ) laid the foundation for the Human Epigenome Project (HEP). For the HEP, we aim to analyse DNA methylation in the regulatory regions of all known genes in most major cell types and their diseased variants, along with producing high-density snapshots of non-genic regions spread evenly across the human genome. Although genome-wide DNA methylation analyses have been performed previously ( Costello et al. 2000 ; Strichman-Almashanu et al. 2002 ), the HEP is the first systematic whole-genome study of DNA methylation at the sequence level. As a prelude to the HEP, here we report the results of the HEP pilot study: DNA methylation profiling of the human major histocompatibility complex (MHC). The MHC, located on Chromosome 6 (6p21.3), is one of the most gene-dense regions in the human genome, containing genes with a high diversity of function, many of which are involved in the innate and adaptive immune systems. We chose to analyse the MHC for the pilot HEP study for three main reasons. (i) The MHC is associated with more diseases than any other region of the human genome, and therefore the generated data will be of interest to researchers with diverse biomedical interests. (ii) It is also the most polymorphic region in the genome, and therefore the data will allow study of the potential effects of the loss or gain of cytosine–guanine dinucleotide (CpG) methylation sites (due to SNPs) on gene expression and possibly other phenotypes. (iii) At the time when the HEP pilot study was initiated in 1999 ( Beck et al. 1999 ), the MHC was one of the few regions within the human genome for which finished sequence and annotation were readily available ( MHC Sequencing Consortium 1999 ). Using an integrated pipeline involving high-throughput bisulphite DNA sequencing, we have determined the DNA methylation levels within the vicinity of the promoter and other relevant regions, such as CpG islands and first exons and introns of 90 genes within the 3.8-Mb MHC region in multiple tissues and individuals. Our analysis reveals a bimodal distribution of methylation levels, tissue specificity, and inter-individual variation. We have also developed matrix-assisted laser desorption/ionisation mass spectrometry (MALDI-MS) assays for high-throughput epigenotyping of the analysed regions. Finally, we have established a publicly available database for the HEP data ( http://www.epigenome.org ), which integrates, for the first time, epigenetic information with the existing genome annotation. Results/Discussion For the DNA methylation profiling of the human MHC, the following regions of interest (ROIs) were chosen: (i) a potential regulatory region for each gene and (ii) the most CpG-dense region of each gene. It is well established that epigenetic modifications at regulatory regions, in particular promoters, correlate with the transcriptional state of the cognate gene (reviewed in Bird 2002 ). Because the precise locations of promoters within the human MHC were unknown at the time this study was initiated, we surmised that analysing a region from 2 kb upstream to 500 bp downstream of the annotated start codon would, in many cases, include the promoter region. Such regions were designated as “upstream” ROIs. ROIs representing the most CpG-dense region within each gene were defined for the region from 500 bp downstream of the annotated start codon to the end of the gene and did not exceed a total length of 2.5 kb. These ROIs were named “intragenic”. For longer genes, more than one intragenic ROI was chosen. Within each ROI, we used the amplicon with the highest CpG density that could be successfully amplified. Other amplicons, if used, were chosen based on the ranking of their CpG density. Wherever possible, CpG islands associated with genes were included. (CpG islands have been defined by Bird [1986] as a contiguous window of DNA of at least 200 bp in which the G + C content is at least 50% and the ratio of observed over expected CpG frequency is greater than 0.6. We used a slightly stricter definition: regions of at least 400 bp in which the G + C content is at least 50% and the ratio of observed over expected CpG frequency is greater than 0.6.) All known repeat sequences were avoided during amplicon design. Methylation was analysed in seven human tissues—adipose, brain, breast, liver, lung, muscle, and prostate—with multiple samples from different individuals for all tissues (except adipose) (see Table S1 ). Figure 1 shows the locations and the coverage provided by the bisulphite PCR amplicons across the 3.8-Mb human MHC in the context of annotated genes, CpG content, CpG islands, and SNPs extracted from the SNP database ( http://www.ncbi.nlm.nih.gov/SNP ). A total of 253 unique amplicons were successfully analysed ( Table 1 ). On average, the amplicons were 438 bp in length (which is close to the optimum amplicon length for the bisulphite PCR), were relatively GC-rich (average G + C > 50%), and had a high density of CpGs (approximately 1 CpG/31 bp). Ninety genes (i.e., more than 70% of all expressed genes within the MHC) were represented by at least one amplicon. Of the analysed CpG sites, 80% displayed methylation levels that varied (i.e., by more than 20%) either between individuals and/or tissues, suggesting that the potential information content of the selected amplicons was relatively high. Figure 1 Map of the Human MHC Showing Coverage and Locations of the Bisulphite PCR Amplicons for Which Methylation Data Have Been Generated Tracks from top to bottom are as follows. (1) CpG content—the proportion of CpGs in 8-kb windows. The expected proportion of CpG dinucletides is 0.04 based on the background base composition of Chromosome 6 ( Mungall et al. 2003 ). (2) Random SNP density in 1,000-bp windows. (3) Location of predicted CpG islands. (4) Bisulphite PCR amplicons. (5) Location of annotated gene structures. Right and left arrows indicate gene structures on the sense and antisense strand, respectively. Official gene symbols are used where available. Table 1 Amplicon Statistics a Total number of CpG sites for which methylation levels were determined from forward and reverse sequencing b Proportion of unique CpG positions analysed that showed a difference of 50% or greater between the minimum and maximum methylation levels observed at that site across all samples c Proportion of unique CpG positions analysed that showed a difference of 20% or greater between the minimum and maximum methylation levels observed at that site across all samples d Proportion of unique CpG positions analysed that showed a difference of 20% or greater between the minimum and maximum average methylation levels observed at that site across different tissue types. The mean methylation for a unique CpG in a tissue was calculated by averaging the measurements of the individual samples from that tissue type Quantification of DNA Methylation by Direct Sequencing of Bisulphite PCR Products We analysed DNA methylation using bisulphite sequencing ( Olek et al. 1996 ). In the presence of sodium bisulphite, unmethylated cytosines are converted to uracil, whereas methylated cytosines are unreactive under the same conditions. After bisulphite treatment the DNA is subjected to PCR and sequencing. Methylated cytosines are detected as cytosines in the sequencing reaction, whereas all unmethylated cytosines appear as thymidines. Traditionally, bisulphite PCR analysis involves sequencing multiple sub-clones of the bisulphite PCR product. This approach is time-consuming, and there have also been reports of bias ( Grunau et al. 2001 ) and hetero-duplex amplification ( Sandovici et al. 2003 ) during sub-cloning of bisulphite PCR products. We sequenced the bisulphite PCR products directly with the same primers used in the PCR and developed software, called ESME ( Lewin et al. 2004 ), to determine the DNA methylation levels from the sequence trace files. Briefly, ESME performs quality control, normalises signals, corrects for incomplete bisulphite conversion, and maps positions in the trace file to CpGs in the reference sequence. The program calculates methylation levels by comparing the C to T peaks at CpG sites, with the ability to discriminate levels of methylation that differ by as little as 20%. Methylation estimation by ESME at any given CpG site is the average from all the copies generated during PCR and is, therefore, compared to sub-cloning, a more accurate representation of the methylation level. Furthermore, we reanalysed the methylation levels of 77 amplicons by MALDI-MS, which allows for discrimination of methylation levels that differ by as little as 5% ( Tost et al. 2003 ). Figure 2 shows the comparison of the two methods, demonstrating a concordance rate of 88% between ESME and MALDI-MS. Figure 2 Comparison of Methylation Measurements Obtained Using MALDI-MS with Those from ESME Analysis of Directly Sequenced Bisulphite PCR Products (A) Comparison of methylation measurements obtained by MALDI-MS (x-axis) with ESME-processed data from sequencing (y-axis). Methylation rates at CpGs from forward and reverse sequencing were binned into ten intervals from zero to one using corresponding MALDI-MS measurements at the same CpGs and in the same tissue samples. (B) Comparison of methylation measurements obtained from ESME-processed data (x-axis) with measurements from MALDI-MS (y-axis). Methylation rates from MALDI are binned as in (A), using the corresponding methylation values from sequencing. Red lines show the means of the binned rates; bars show the standard deviations. The overall correlation of the data is 0.887. Data points that are not around a methylation rate of zero or one are covered by few measurements because of the bimodal distribution of methylation measurements. The HEP Database To make the data generated in this study a publicly available resource, we have designed a Web-based, ENSEMBL-like genome browser ( http://www.epigenome.org ) that allows easy access to the data from the pilot HEP study ( Figure 3 A and 3 B). Methylation levels calculated by ESME are displayed in a colour-coded matrix. Rows represent the averages of forward and reverse sequences for various tissues while columns represent individual CpG sites. Each matrix square therefore represents the average methylation level at a given CpG site for a given tissue. Multiple data rows are available for all tissues (except adipose). Clicking on a square in the genome browser reveals the level of methylation observed at that particular CpG site (the average of the forward and reverse sequence) and information about the tissue source. Additional annotation includes chromosome coordinates, CpG islands, SNPs, ENSEMBL and high-quality, manually curated Vertebrate Genome Annotation database transcripts, the ROIs, and amplicon and primer sequences. The browser provides a zoom function to view the genomic sequence ( Figure 3 B), and a link to ENSEMBL facilitates access to additional information and the ENSEMBL search engines. The data from the full-scale HEP will be made available via the same browser, providing a novel public resource for the research community. Figure 3 The HEP Database (A) We have created a Web-based, ENSEMBL-like genome browser for displaying HEP data that is publicly available at http://www.epigenome.org . The methylation levels calculated by the ESME software are displayed in the form of a matrix. Each matrix contains the data obtained from all the samples of one amplicon. Each colour-coded square (yellow represents 0% methylation, blue represents 100% methylation, and green represents intermediate levels) within the matrix represents one CpG site. Clicking on a square reveals the tissue source of the sample and the level of methylation observed at that particular CpG site. Grey squares indicate CpG sites for which methylation levels could not be determined. Each row of squares represents all the CpG sites for one sample of a particular amplicon, and the samples are grouped by tissue type. The red bar indicates the genomic region analysed. Also shown are chromosome coordinates, CpG islands, SNPs, and ENSEMBL and high-quality, manually curated VEGA transcript information. The HEP database links to the Ensembl genome browser, providing additional information about the region of interest. The example shows amplicons within the SynGAP 1 gene that correspond to regions that were determined to be hypomethylated (second amplicon from the left), hypermethylated (first and fifth amplicons), and heterogeneously methylated (fourth amplicon). Insufficient data were obtained for the third amplicon. (B) By using the zoom function, the user can view the complete DNA sequence for the analysed amplicon. Methylation Profile Characteristics of the MHC The methylation profile of the human MHC region appears to be strongly bimodal, with over 90% of the amplicons being either relatively hypomethylated (i.e., median methylation of amplicon 30% or less) or relatively hypermethylated (i.e., median methylation of amplicon 70% or greater) ( Figures 3 A and 4 ). Re-analysis of a subset of the data by MALDI-MS confirmed the bimodality of the methylation profile ( Figure 4 ). Extensive bimodality of genomic methylation profiles has been observed by several authors (reviewed in Bird 2002 ). Furthermore, the experiments of Lorincz et al. (2002) suggest that the extremes of methylation profiles may in fact be the most stable states within the genome. Lorincz et al. showed that a high density of methylation at a proviral construct is stably propagated in vivo, whereas a low density of proviral methylation is inherently unstable, with daughter cells harbouring proviral cassettes that are demethylated or de novo methylated. It must be noted that even though the amplicons displayed hypo- or hypermethylated profiles, small variations in the levels of methylation at individual CpG sites within an amplicon were also frequently observed. Although there may be technical reasons for this heterogeneity, numerous studies (using a variety of techniques) have shown that the methylation profile of a given region in vivo is rarely homogenous ( Costello et al. 2000 ; Kondo et al. 2000 ; Grunau et al. 2001 ; Cui et al. 2003 ). The functional outcome of these small variations, particularly when they exist between tissues or individuals, remains to be elucidated. Figure 4 Bimodal Distribution of DNA Methylation within the Human MHC (A) Determined by direct sequencing/ESME analysis (based on 86,374 single CpGs in different tissue samples building the median for measurement repetitions). (B) Determined by MALDI-MS (based on 1,019 MALDI measurements). Comparison of the methylation values for upstream amplicons (median methylation of 10%) versus intragenic amplicons (median methylation of 86%) revealed that upstream amplicons were more likely to be hypomethylated ( p < 0.0001). Interestingly, within the upstream category we found that CpG sites located within the 5′ UTR were less likely to be methylated (median methylation of 7%) than the CpG sites located within 2 kb of the first start codon but not within the 5′ UTR (median methylation of 14%) ( p < 0.0001). Within the intragenic category, we found that CpG sites located within introns (median methylation of 84%) were less likely to be methylated than CpG sites located within exons (median methylation of 89%) ( p < 0.0001). Whether these significant but small differences reflect any bias for the presence of regulatory elements close to the transcriptional start site or within introns, or some other functional consequence, is currently hard to assess. Analysis of Heterogeneously Methylated Regions Fourteen amplicons displayed significant heterogeneous methylation profiles (i.e., median methylation between 30% and 70%) (see Figure 3 A). These might represent differentially methylated regions at which parental alleles display reciprocal methylation profiles that are determined by the parent-of-origin of the allele, or regions that were heterogeneously methylated on both alleles. Our sequencing method could not discriminate between these two possibilities, and none of these regions corresponded to known imprinted sites within the human genome. We therefore sub-cloned the PCR products and sequenced individual sub-clones of ten different heterogeneously methylated amplicons and used polymorphisms to discriminate between the parental alleles. The overall methylation profiles determined by sequencing individual sub-clones were consistent with those obtained by direct sequencing of the bisulphite PCR products. None of the six amplicons for which polymorphisms were found showed allele-specific methylation (data not shown). This is consistent with the fact that so far there have been no reports of imprinted regions within the human MHC. Since these amplicons were heterogeneously methylated to a similar extent in samples from various tissues and individuals, they might represent regions where maintenance of a specific epigenetic state is not essential. It is also possible that these regions are located at the boundaries of hypermethylated regions, and, consequently, the methylation levels are “trailing off”. However, both these possibilities contradict models that suggest that the genome prefers to maintain methylation profiles in bimodal states. Interestingly, a few regions were heterogeneously methylated in some tissues only, suggesting that the tissue sampled was a mosaic of several sub-types among which the methylation profile at certain genes varied, or that the region displays tissue-specific parental imprinting similar to the insulin-like growth factor 2 (IGF2) gene, which is imprinted in all tissues except brain ( Pham et al. 1997 ). Direct sequencing of the heterogeneously methylated amplicons was unsuccessful in a small proportion of cases. Possible reasons include incomplete bisulphite conversion and genetic polymorphisms within the primer binding site. We also noticed a mobility shift of the sequence in a few cases. This occurs because a population of bisulphite PCR products generated from a heterogeneously methylated region contains a mixture of molecules, some with cytosines at certain CpG sites (i.e., initially methylated) and others with thymidines at those CpG sites (i.e., initially unmethylated). When these PCR products are sequenced directly, the cumulative effect of the molecular weight difference between cytosines and thymidines is that some molecules migrate faster than others during capillary electrophoresis. The sequence trace therefore contains two traces that do not perfectly overlap, resulting in erroneous estimation of methylation levels. Such sequences were excluded from further analyses. Analysis of CpG Islands CpG islands are GC-rich regions that contain a high density of CpGs and are positioned at the 5′ ends of many human genes (reviewed in Bird 2002 ). Although most CpG islands remain hypomethylated throughout development in all tissues ( Antequera and Bird 1993 ), regardless of expression state, a small proportion become hypermethylated during development (reviewed in Bird 2002 ), and this correlates with transcriptional silencing of the associated gene. In our study, 27 amplicons overlapped CpG islands, and 22 of these (i.e., 80%) were hypomethylated in all tissues examined. Interestingly, this proportion of hypomethylated CpG islands is similar to that reported by Yamada et al. (2004) , who analysed the methylation status of CpG islands on human Chromosome 21q and found that 103 out of 149 CpG islands (i.e., 70%) were hypomethylated. In our study, CpG island amplicons situated in the upstream ROIs were always hypomethylated, whereas hypermethylated CpG island amplicons were found only in the intragenic regions. Among the intragenic CpG island amplicons, those situated at the 5′ end of the gene (i.e., overlapping exon 1, intron 1, or exon 2) were always hypomethylated. A tissue-specific methylation profile was observed for the CpG island situated within exon 3 of the tenascin-XB ( TNXB ) gene, which was hypomethylated in muscle samples only. This hypomethylation correlates with the temporally regulated and tissue-specific expression of TNXB, which is abundantly expressed in connective tissues. It has been suggested that TNXB has a role in limb, muscle, and heart development ( Burch et al. 1995 ), and, therefore, epigenetic modifications at the TNXB CpG island may have an important regulatory role (tissue specificity of methylation profiles is discussed in more detail below). Interestingly, the CpG island amplicon located within exon 3 of the HLA-G gene spanned a methylation boundary, being hypomethylated at the 5′ end with a sharp transition to a hypermethylated profile at the 3′ end. Overall, the results are consistent with the prevailing model of CpG islands being regions of the genome that are hypomethylated, especially when they occur upstream or within the 5′ end of the gene. Tissue Specificity of the Methylation Profiles DNA methylation profiles are complex and dynamic, and can vary with developmental stage, tissue type, age, the alleles' parent-of-origin, and also phenotype or disease state (reviewed in Bird 2002 ). In particular, the role of DNA methylation in setting up and maintaining tissue-specific expression patterns has received a lot of attention. However, the extent of tissue specificity of DNA methylation profiles is relatively unknown. The HEP pilot study involved the analysis of 32 samples (from different individuals) comprising seven tissues: adipose, brain, breast, liver, lung, muscle, and prostate. Upon comparison of the amplicon profiles, we found that 10% of all amplicons displayed differential methylation between the tissue types (examples are shown in Figure 5 ). Of these amplicons, 31% were located in the upstream regions, a proportion that is in the same range as the total number of upstream amplicons relative to intragenic amplicons analysed in this study (see Table 1 ). We scanned the literature and publicly available gene expression databases to determine whether the cognate genes displayed tissue-specific expression. An example is the complement protein C2 mRNA, which normally has a long 5′ upstream region; in the liver, an additional transcript with a much shorter 5′ upstream region is expressed ( Horiuchi et al. 1990 ). In our study we found that a region that overlaps intron 2 and exon 2 of the C2 gene was hypomethylated in liver samples only (however, this region is downstream of the transcriptional start sites of both forms of C2 mRNA). Another example is DOM3Z, which is ubiquitously expressed but occurs only at very low levels in the lung ( Yang et al.1998 ), and this correlates with a region overlapping exons 4 and 5 of DOM3Z that is hypermethylated in lung (and brain) but hypomethylated in the other tissues examined. It has also been demonstrated that the murine complement factor B utilises differential tissue-specific start sites ( Garnier et al. 1995 ), and in our analysis the human homologue is hypomethylated at a region overlapping exons 3 and 4 only in liver. However, the majority of the genes that were associated with tissue-specific methylation profiles in our study did not show corresponding tissue-specific expression profiles in a previously reported whole human genome expression microarray analysis ( Su et al. 2002 ). Some of these genes are known to be associated with various mRNA isoforms, but detection of such alternative transcripts is quite difficult with conventional microarray analysis and usually requires more detailed analysis. It is also possible that the tissue-specific methylation profiles we observed in adult tissue may hint at tissue-specific expression profiles that existed during early development, or they may be associated with as yet unknown transcripts, e.g., non-coding RNAs. Alternatively, there may be only a modest proportion of genes in which tissue specificity of gene expression is affected by methylation. Figure 5 Example of METHANE Output Showing Regions That Display Tissue-Specific Methylation Profiles The top colour-scale bar refers to the degree of methylation (percent). The bottom colour-scale bar refers to the absolute difference in the methylation level observed between tissues at a given CpG site, and is therefore a measure of the confidence level for a CpG site to be defined as a MVP. (A) The upper matrix represents an amplicon that contains 18 CpG sites within a 386-bp region overlapping exon 3, intron 3, and exon 4 of the complement factor B gene. It is hypomethylated in liver (median methylation is 17%) and hypermethylated in all other tissues examined (median methylation is 100%). The lower matrix shows pairwise comparisons of the methylation values for each CpG site between tissues. (B) The upper matrix represents an amplicon that contains 19 CpG sites within a 550-bp region overlapping exon 3 and intron 3 of the DAXX gene. It is relatively hypomethylated in breast (median methylation is 64%) compared with the other tissues examined (median methylation is 100%). The lower matrix shows pairwise comparisons of the methylation values for each CpG site between tissues. Inter-Individual Variation of Methylation Profiles There is increasing evidence that an individual's epigenetic profile can influence phenotype and susceptibility to various diseases such as cancer, an example of such evidence being a recent report linking the loss of imprinting at the IGF2 locus with an increased risk of developing colorectal cancer ( Cui et al. 2003 ). In our study, nearly all loci displayed some degree of heterogeneity, which probably has no bearing on the differences in genome function among individuals. However, considerable differences in methylation profiles between individual samples within a tissue were observed for a number of amplicons. We calculated a median methylation value for each individual sample and then compared these values within each tissue type for each amplicon. A total of 118 amplicons displayed a difference of greater than 50% between the lowest and highest median methylation values in at least one tissue. Of these amplicons, 76% were intragenic, which is a similar proportion to the overall number of intragenic amplicons (71%; 181 out of 253 amplicons) analysed in the study. This proportion is also similar to the overall proportion of amplicons that showed tissue-specific methylation profiles and were classified as intragenic (69%). Although inter-individual variation for a given amplicon was not observed in every tissue, there was no apparent tissue-specific enrichment for inter-individual variability of methylation profiles. Examples of amplicons that displayed significant inter-individual variation in methylation profiles include a region overlapping the last exon in CYP21A2 that showed considerable inter-individual variation in prostate ( Figure 6 A), and a 5′ upstream region of tumour necrosis factor (LocusID 7124) that varied significantly between individuals in liver ( Figure 6 B). Although the differences could be attributable to the technical variability inherent in our approach or the fact that we did not control for age or sex of the tissue donors, it is also possible that certain genotypes are associated with unique epigenotypes. In a recent study, Van Laere et al. (2003) mapped a porcine quantitative trait locus that affects muscle growth, fat deposition, and heart size to an evolutionarily conserved CpG island within the imprinted Igf2 gene. Pigs inheriting the mutation from their sire had a 3-fold increase in Igf2 expression in postnatal muscle (i.e., the quantitative trait locus is paternally expressed). Furthermore, the mutation abrogated in vitro interaction with a nuclear factor, and this effect was phenocopied following in vitro DNA methylation of the region. Evidence for an interaction between genotype and epigenotype at the IGF2 gene in humans has also recently been reported ( Murrell et al. 2004 ). Of the 3,273 unique CpG sites we analysed, 101 overlapped with known SNPs (relatively evenly distributed over all amplicons), all representing sites at which the CpG was lost (see Figure 1 ; Table 1 ). The SNPs were extracted from dbSNP ( http://www.ncbi.nlm.nih.gov/SNP ) and are annotated in the HEP database. One could postulate that the gain or loss of one or more critical CpG sites may affect the overall methylation profile of a locus and, consequently, promoter activity. Alternatively, non-CpG SNPs located within an epigenetically sensitive regulatory element could also influence the epigenetic makeup of that region. Therefore, mutations in regulatory sequences could influence epigenetic profiles, resulting in altered phenotypes. Figure 6 Example of METHANE Output Showing Regions That Display Inter-Individual Variation of Methylation Profiles (A) Example of a region that displays significant inter-individual variation, especially in prostate. The matrix represents an amplicon that contains 27 CpG sites within a 527-bp region overlapping the last exon of the CYP21A2 gene. (B) Another example of a region that displays significant inter-individual variation. The matrix represents an amplicon that contains 13 CpG sites within a 453-bp region overlapping the 5′ UTR and exon 1 of the tumour necrosis factor gene. Analysis of Methylation Variable Positions by MALDI-MS A major aim of the HEP is to identify genomic regions at which DNA methylation profiles display statistically significant variation due to biological or environmental influences. Therefore, based on the tissue-specific and inter-individual variation in methylation profiles discussed above, we were interested in establishing high-throughput assays for epigenotyping. This involved the identification (manually or using the METHylation ANalysis Engine [METHANE]) of methylation variable positions (MVPs), which we define as CpG sites that have statistical power to discriminate between different biological samples or states. In other words, by assaying the methylation state of just a few select CpG sites within a given region, information can be inferred about the tissue source or disease state. Such a high-throughout MVP epigenotyping method was recently developed based on the GOOD assay ( Tost et al. 2003 ). This recently developed epigenotyping assay allows for accurate discrimination of methylation levels that differ by 5% or more. Furthermore, MALDI-MS is a relatively inexpensive method that offers a high degree of automation and integration and that has no requirement for sample purification. Assays for 231 MVPs in 77 amplicons, including all those that displayed differential methylation profiles between different tissue types or inter-individual variability, were designed and analysed in a triplex format (i.e., methylation levels at three independent CpG sites are analysed in one assay). A subset of 11 MALDI-MS assays is shown in Figure 7 . Figure 7 Comparison of Methylation Values Measured in Five Tissues and Eleven Amplicons Using MALDI-MS and ESME Analysis of Directly Sequenced PCR Products Each column is a tissue sample, each row a CpG site. Data are ordered in blocks by tissue type and amplicons. Positions of measurements for MALDI-MS (A) correspond to those for ESME analysis (B). The methylation values are colour coded from 0% methylation (yellow) to 100% methylation (blue), with intermediate methylation levels represented by shades of green. White indicates missing measurement values. Comparison of Methylation Profiles with Independent Gene Expression Data The primary function of epigenetic modifications is to modulate gene expression: a specific combination of epigenetic modifications at regulatory elements, notably promoters and enhancers, influences the transcriptional state of a gene (reviewed in Bird 2002 ). In many cancers, aberrant epigenetic modifications occur within CpG islands that overlap promoters (some of which are candidate tumour suppressors), which is thought to result in aberrant transcription of the cognate gene, thus contributing to tumour progression. We compared the amplicon methylation profiles with the human genome expression patterns available from the Genomics Institute of the Novartis Research Foundation Gene Expression Atlas database ( http://expression.gnf.org ). This publicly available database contains whole-genome mRNA expression data obtained by Su et al. (2002) using human U95A Affymetrix microarray chips. We calculated a median methylation value for each amplicon (see Materials and Methods ). As mentioned above, the methylation profiles displayed a bimodal distribution, with more than 90% of the amplicons being either hypomethylated (median methylation of 30% or less) or hypermethylated (median methylation of 70% or greater). Therefore, to perform the analyses we divided the amplicons into two categories: hypomethylated (methylation less than 50%) and hypermethylated (methylation greater than 50%) (see Materials and Methods ). We then compared the range of expression values associated with hypomethylated amplicons with those of hypermethylated amplicons. Most genes on the U95 microarray are represented by multiple probes, and, in a few cases, contradictory expression values were obtained for the same gene, in which case the gene was excluded from our analyses. Analyses were performed for liver, lung, and prostate samples only ( Figure 8 ), since appropriate Gene Expression Atlas data were unavailable for the other tissues. For prostate and liver, a significant difference was found between expression levels associated with hypomethylated versus hypermethylated upstream amplicons: hypomethylated upstream amplicons correlated with a wide range of expression levels whereas hypermethylated upstream amplicons correlated with a lack of expression ( p < 0.0001 for prostate and p < 0.01 for liver). The intragenic amplicons did not show any correlation between methylation and expression levels ( p > 0.3 for both prostate and liver). A list of all upstream amplicons included in the analysis is given in Table S2 . Figure 8 Comparison of DNA Methylation with Gene Expression Amplicons generated from prostate (yellow), lung (blue), and liver (green) samples were divided into two categories: “upstream” and “intragenic”. The median methylation values for the amplicons were calculated as described in the text, and these were then classified as hypomethylated (median methylation less than 50%) or hypermethylated (median methylation greater than 50%), and plotted against the cDNA microarray expression data available at http://expression.gnf.org ( Su et al. 2002 ). The expression values are expressed as average difference values (ADVs) for each gene. The average difference value is computed using Affymetrix software and is proportional to mRNA content in the sample, with a value of 200 being a conservative cut-off below which a gene can be classified as being not expressed. The average difference values are the mean of 2 or 3 independent experiments. For prostate and liver, the expression levels associated with the hypermethylated upstream amplicons were significantly lower than the expression levels associated with the hypomethylated upstream amplicons ( p < 0.0001 for prostate and p < 0.01 for liver). For lung, there was no significant difference between the expression levels associated with the hypermethylated upstream amplicons and those of the hypomethylated upstream amplicons ( p > 0.3). There was no correlation between expression and methylation for the intragenic amplicons for any of the three tissues ( p > 0.3). The width of the bars is indicative of the number of amplicons in each category: prostate upstream, hypermethylated ( n = 9); prostate upstream, hypomethylated ( n = 15); prostate intragenic, hypermethylated ( n = 109); prostate intragenic, hypomethylated ( n = 53); liver upstream, hypermethylated ( n = 9); liver upstream, hypomethylated ( n = 14); liver intragenic, hypermethylated ( n = 115); liver intragenic, hypomethylated ( n = 45); lung upstream, hypermethylated ( n = 9); lung upstream, hypomethylated ( n = 13); lung intragenic, hypermethylated ( n = 112); and lung intragenic, hypomethylated ( n = 57). For the lung samples there was no significant correlation between expression and methylation state for amplicons within the upstream or intragenic categories ( p > 0.3 for both categories). Although the lung data show the same trend as the prostate and liver data, the lung hypomethylated data contained a number of outlier data points representing very high expression values (as shown in Figure 8 by the unfilled circles). The overall trend of the data suggests that these data may be artefactual, but there is nothing that indicates these data points are not real. These data points were enough to influence the analysis such that we could not find a significant difference in the expression of between hypo- and hypermethylated lung genes. If the data points are real, the lack of correlation for the lung samples may be due to inconsistencies within the expression or methylation datasets for lung. Alternatively, there may be additional regulatory elements that influence the expression state of the analysed genes in the lung. Overall, the findings are consistent with a model in which the DNA methylation profile of the upstream region of the gene is an informative indicator of the expression of the cognate gene, specifically, in which hypermethylation within the upstream region is associated with transcriptional silencing. Furthermore, the data also suggest that epigenetic modifications within the upstream regions influence the transcriptional state of a significant number of the genes within the MHC. This is supported by the study of Jackson-Grusby et al. (2001) in which they employed homogeneous cultures of primary mouse embryonic fibroblasts and used the Cre-loxP system to conditionally inactivate Dnmt1, an enzyme that methylates DNA. They found that in the absence of Dnmt1, several mouse MHC class I genes showed altered expression profiles. Concluding Remarks One of the principal challenges in the post-genomic era is to provide a holistic view of genome function, a challenge which is currently being addressed by several large-scale studies of the transcriptome, proteome, metabolic networks, and haplotype maps. The HEP is therefore timely, since DNA methylation is an indispensable part of the genome's regulatory mechanisms. Here we have described the pilot study for the HEP—DNA methylation profiling of the MHC region—which is the first systematic large-scale study of methylation profiles at the sequence level within a multi-megabase region of the human genome. For this project, we developed an integrated pipeline for high-throughput methylation analysis using bisulphite DNA sequencing, MVP discovery, and epigenotyping by MALDI-MS, and created an integrated database ( http://www.epigenome.org ) for public access to the data generated by the study. The results from the pilot study demonstrate that a significant proportion of the analysed loci within the MHC show tissue-specific methylation profiles, and inter-individual methylation differences are common. Furthermore, the tissue-specific differences in DNA methylation suggest that epigenetic mechanisms are involved in the use of alternative transcriptional start sites. We have also shown that the generated methylation data allow the identification of MVPs that can be typed with high quantitative resolution and sensitivity using MALDI-MS, providing a tool for large population-based studies and for diagnosing diseases in the future. The study reported here lays the foundation for the HEP, which aims to analyse the methylation state of the regulatory regions of all annotated genes in most major cell types and their diseased variants. In the first phase, which is well underway, we are analysing the DNA methylation profiles of over 5,000 amplicons (representing a 20-fold scale-up relative to the pilot HEP study reported here) associated with nearly all the annotated genes (approximately 3,000) on human Chromosomes 6, 13, 20, and 22. The excellent genomic annotation available for these four chromosomes, e.g., high-quality transcript information and location of SNPs, will enable us to perform comprehensive analyses linking the epigenetic information gained from the HEP with the underlying genetic information. Samples from over 40 different individuals representing 20 tissues will be used in the study. The resulting data will generate a map that complements other large-scale efforts that are linking our knowledge of gene sequence and cellular phenotypes: studies involving DNA sequencing, SNPs, histone modifications, and transcriptome and proteomic analyses. The epigenome map will be invaluable for understanding gene regulation and the interactions between genes in normal and disease states. It will offer new explanations in well-studied areas such as cancer research, and will also provide a basis for novel approaches to research on environmental effects, nutrition, and ageing ( Eckhardt et al. 2004 ). The HEP also promises to provide DNA methylation markers for disease states, and new targets for drug development and diagnostic applications based on DNA methylation research are already emerging ( Cairns et al. 2001 ). Current efforts to target the epigenomic machinery of cells with drugs have global effects ( Besterman and McLeod 2000 ; Lubbert 2000 ; Munster et al. 2001 ), and more refined approaches will become possible with accumulating knowledge in the new field of epigenomics. Materials and Methods Tissue samples. Human tissue samples were obtained from the National Disease Research Interchange (Philadelphia, Pennsylvania, United States) and consisted of tissue material from healthy individuals. Tissue samples included seven different tissue types (adipose, brain, breast, lung, liver, prostate, and muscle) from 32 different individuals ( Table S1 ). DNA was extracted using standard protocols ( Sambrook et al. 1989 ). Bisulphite conversion. Bisulphite treatment of genomic DNA was performed with minor modifications to a method described previously ( Olek et al. 1996 ). Amplicon design and PCR. Primers were designed to be at least 21 bases (G + C ≥ 30%) and to contain at least two bases complementary to bisulphite-converted sequence to increase the specificity. To ensure that the primers were not biased for either hypomethylated or hypermethylated sequences, controls were performed using bisulphite-converted unmethylated or in vitro methylated target sequences. PCRs were performed in 96-well plates on MJ Research (Waltham, Massachusetts, United States) thermocyclers in a final volume of 25 μl containing 250 μM dNTPs, 1X PCR Buffer (Qiagen, Valencia, California, United States), 10 pmol each of forward and reverse primer, 1 U Taq polymerase (Qiagen), and 8 ng of bisulphite-treated genomic DNA. Water-only controls were also included in each 96-well PCR plate. The cycling conditions were 95 °C for 15 min followed by 40 cycles of 95 °C for 60 s, 55 °C for 45 s, and 72 °C for 90 s, and a final extension step of 10 min at 72 °C. The PCR amplicons for some of the genomic regions that displayed heterogeneous levels of CpG methylation were sub-cloned using the pGEM-T Easy Vector System according to the manufacturer's instructions (Promega, Madison, Wisconsin, United States). The clones were sequenced on ABI 3700 capillary sequencers (Applied Biosystems, Foster City, California, United States) using ABI Prism Big Dye terminator V 3.1 sequencing chemistry. Sequencing. PCR amplicons were purified using MultiScreen PCR plates (Millipore, Billerica, Massachusetts, United States) and sequenced directly in forward and reverse directions with the same primers used in the PCR. Sequencing was performed on ABI 3700 capillary sequencers using ABI Prism Big Dye terminator V 3.1 sequencing chemistry. Analysis and database generation. Quantitative methylation rates were estimated from sequence traces using the ESME software ( Lewin et al. 2004 ). This involved appropriate quality control, normalisation of signals, correction for incomplete bisulphite conversion, and mapping of positions in the trace file to CpGs in reference sequences as described in Lewin et al. (2004) . Amplicons were mapped to the human genome assembly (NCBI34) using BLAST ( Altschul et al. 1990 ) and CrossMatch ( http://www.genome.washington.edu/UWGC/analysistools/Swat.cfm ). Using these offsets the positions of CpG dinucleotides were determined in genomic coordinates. Positional data were then loaded into an LDAS database ( http://www.biodas.org ) suitable for serving to applications using the distributed annotation system (DAS) XML format ( Dowell et al. 2001 ). HEP methylation data converted to DAS format could then be incorporated dynamically into any third-party applications (such as ENSEMBL) capable of understanding DAS. A Web-based, ENSEMBL-like genome browser, driven entirely from DAS data, was created for displaying HEP data and is publicly available at http://www.epigenome.org . The data reported in the pilot HEP study are subject to the HEP's data release policy ( http://www.sanger.ac.uk/PostGenomics/epigenome/drp.shtml ). Kolmogorov-Smirnov tests confirmed that the methylation data did not follow Gaussian distribution ( p <0.0001 in all cases). We therefore used the Mann-Whitney U test (a non-parametric test that compares the medians of two unpaired groups of samples that do not follow a Gaussian distribution) to perform three different comparative analyses for methylation profiles: (i) upstream versus intragenic amplicons, (ii) within the upstream amplicon category, CpG sites located within the 5′ UTR versus CpG sites not located within the 5′ UTR but still within 2 kb of the first start codon, and (iii) within the intragenic amplicon category, intronic versus exonic CpG sites. We designed a software package, METHANE, to identify and generate graphical views of MVPs (see Figures 5 and 6 ). This tool uses the same DAS data source as the Web browser to generate graphical views but adds additional analysis facilities to compare and display relative methylation level differences in pairwise comparisons. METHANE provides options to compare CpG methylation level differences calculated from either simple averages or medians per tissue or per site. METHANE can export data as tabular text output or as images in SVG, PDF, or postscript formats and is available on request from the authors. Epigenotyping by mass spectrometry. Assays, based on the GOOD assay for DNA methylation analysis ( Tost et al. 2003 ), were established for amplicons displaying differential methylation between tissue types or inter-individual variability in the sequencing effort. In most cases, three MVPs within an amplicon were simultaneously queried and analysed. Assay volumes started at 3 μl for PCR, with 2-μl additions for shrimp alkaline phosphatase treatment, primer extension, and 5′-phosphodiesterase digest, and the addition of 10 μl of alkylation mix in the respective reaction steps. After dilution with acetonitrile, 0.5-μl samples were transferred to a matrix coated MALDI target plate. All liquid handling was carried out with automated liquid-handling robotics (BasePlate, The Automation Partnership, Royston, United Kingdom). All MALDI-MS analyses were performed with positive ion mode detection on Bruker Autoflex MALDI mass spectrometers (Bruker Daltonics, Billerica, California, United States), equipped with target-plate-changing robots. To ensure accurate quantification and to compensate for the common preferential amplification in bisulphite-treated DNA, triplicate calibration standards from 0% to 100% methylation in 25% increments were included into the analysis. Statistical parameters were defined to compensate for factors complicating quantification by MALDI-MS, such as bad reproducibility of the crystallisation, different rates of ionisation of analytes, and signal-to-signal interactions. Quantitative results were obtained with high accuracy when 200 laser shots were accumulated on a sample spot, and eight preparations accounted for differences in the sample preparations. Thus, in total, a quantitative data point is obtained from the average of 1,600 individual spectra and calibrated with the help of mixtures with a known degree of methylation. Success rates are above 97%, and standard deviation for data points is less than 2%. Comparison of DNA methylation with mRNA expression levels. The methylation data were compared with data in the Gene Expression Atlas database ( Su et al. 2002 ) that contains human transcript data based on the U95 build of Unigene, Affymetrix U95A chip. We calculated methylation values for each amplicon using the median of the methylation values for each CpG site within that amplicon. A quantile plot indicated a strongly non-normal distribution of the data, specifically, platykurtosis (one of two different kinds of kurtosis) demonstrating a bimodal distribution. Therefore, for the purposes of simplified statistical analysis, amplicons were classified as either hypermethylated or hypomethylated dependent on whether their median methylation value was either greater than or less than 50%, respectively. We performed methylation versus expression comparisons after removing the data that corresponded to multiple probes that gave contradictory expression patterns for the same gene on the U95 microarray. Equality of mean expression levels between hypermethylated and hypomethylated datasets was tested using a Welch two-sample (unpaired) t -test. Supporting Information Table S1 Tissues Used in the HEP Pilot Study (63 KB DOC). Click here for additional data file. Table S2 Upstream Amplicons Included in the Comparative Analysis of DNA Methylation with mRNA Expression (32 KB DOC). Click here for additional data file. Accession Numbers The LocusLink ( http://www.ncbi.nlm.nih.gov/projects/LocusLink/ ) accession numbers for the genes and gene products discussed in this paper are C2 (LocusID 717), CYP21A2 (LocusID 1589), DOM3Z (LocusID 1797), HLA-G (LocusID 3135), IGF2 (LocusID 3481), TNXB (LocusID 7148), and tumour necrosis factor (LocusID 7124).

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529316.xml

529458

Review of, "Biostatistics and Epidemiology" by S. Wassertheil-Smoller

If you need to know anything basic about Biostatistics and Epidemiology, this book belongs in your library! Unlike most science and technology books – which, I have concluded after nearly 42 years in this business, tend to be written less to teach and communicate information than to impress the reader with how much the author knows about the subject matter – this one, refreshingly and quite successfully, ensures that the reader comes away with a better understanding of the topics of Biostatistics and Epidemiology, as a result of the authors' genuine desire to convey information in a clear, concise, relevant, understandable, and very readable way. In nine Chapters and nine Appendices, spanning 243 pages (which include ample references and suggested further readings), the indexed material covers a wide range of fundamental principles that form the basis for such subjects as: (1) The Scientific Method; (2) Probability Theory; (3) Statistics; (4) Epidemiology; (5) Screening Methods and Techniques; (6) Clinical Trials; (7) Quality of Life; (8) Genetics; and (9) Biomedical Ethics. A unique feature of the way the book is written, is that individual Chapters can be read out-of-sequence with no loss of continuity. The reader can skip around, omit certain sections, and still glean what he or she needs to know without feeling cheated. That's not easy to do, but this author does it well. I found the material to be written at a very comfortable cognitive level – aimed mainly at medical, upper-level-college, and graduate students – and sequenced in a logical manner, thus making it easy to follow and totally user-friendly. The author has a special talent for reducing complicated concepts to an easily understandable level. For example the Appendix dealing with Genetic Principles gives the best introductory synopsis of this topic that I have seen anywhere; and the "middle Chapters" that address "Mostly About" – Statistics (Chapter 3), Epidemiology (Chapter 4), Screening (Chapter 5), Clinical Trials (Chapter 6), Quality of Life (Chapter 7), and Epidemiology (Chapter 8), are gems!. What are especially nice are the many relevant, easy-to-understand and practical examples, that the author judiciously and effectively intersperses with the theoretical background material; and, the Chapter Summaries that conclude most of them. There is a very well-balanced give-and-take between theory and practice. I also liked the wonderful (sometimes witty) little parables and anecdotes that give the book a charming personality. In all, I would offer my compliments to the author for a "job well done!"

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529458.xml

544844

Evolutionary autonomous agents and the nature of apraxia

Background Evolutionary autonomous agents are robots or robot simulations whose controller is a dynamical neural network and whose evolution occurs autonomously under the guidance of a fitness function without the detailed or explicit direction of an external programmer. They are embodied agents with a simple neural network controller and as such they provide the optimal forum by which sensorimotor interactions in a specified environment can be studied without the computational assumptions inherent in standard neuroscience. Methods Evolutionary autonomous agents were evolved that were able to perform identical movements under two different contexts, one which represented an automatic movement and one which had a symbolic context. In an attempt to model the automatic-voluntary dissociation frequently seen in ideomotor apraxia, lesions were introduced into the neural network controllers resulting in a behavioral dissociation with loss of the ability to perform the movement which had a symbolic context and preservation of the simpler, automatic movement. Results Analysis of the changes in the hierarchical organization of the networks in the apractic EAAs demonstrated consistent changes in the network dynamics across all agents with loss of longer duration time scales in the network dynamics. Conclusion The concepts of determinate motor programs and perceptual representations that are implicit in the present day understanding of ideomotor apraxia are assumptions inherent in the computational understanding of brain function. The strength of the present study using EAAs to model one aspect of ideomotor apraxia is the absence of these assumptions and a grounding of all sensorimotor interactions in an embodied, autonomous agent. The consistency of the hierarchical changes in the network dynamics across all apractic agents demonstrates that this technique is tenable and will be a valuable adjunct to a computational formalism in the understanding of the physical basis of neurological disorders.

Background The conceptual framework by which a neurological syndrome such as apraxia is presently explained is based on a computational understanding of brain function. Briefly, this framework assumes the existence of well-defined, determinate perceptual representations and motor programs that interact in a fashion similar to the way that symbols are manipulated in a computer. Recent theoretical and applied work in natural and artificial systems, however, has conspired to shift the emphasis in neuroscience from the computational paradigm to a dynamical understanding of brain function [ 1 - 5 ]. This approach emphasizes that cognition occurs in an embodied agent interacting dynamically with its environment and avoids the assumptions of determinate motor programs and perceptual representations implicit in the computational framework [ 6 , 7 ]. The identification of the dynamical changes in an apractic nervous system may provide a causally mechanistic explanation for the syndrome that may complement the higher-level descriptive explanation of the standard computational approach. The apraxias constitute a spectrum of movement disorders in which there is impairment in the performance of a skilled, learned movement that cannot be attributed to an elementary motor or sensory deficit. Based of the pioneering work of Liepmann [ 8 ], the apraxias have traditionally been divided into ideational, ideomotor and limb-kinetic apraxia. Limb-kinetic apraxia, which is felt by many to not be a true apraxia, manifests as slowness or clumsiness of distal limb movements with preservation of knowledge of the appropriate action to perform. Ideational apraxia is characterized by loss of knowledge of how an object is used or as impairment in the sequencing of constituent movements in a complex movement. Ideomotor apraxia is usually diagnosed on the basis of spatiotemporal errors that occur on transitive gesture tasks requiring demonstration of the pantomime appropriate to specific object use [ 9 ]. Asking a patient to demonstrate how they would use a comb or a hammer would be typical transitive gesture tasks used in the assessment of the presence or absence of ideomotor apraxia. In many cases of ideomotor apraxia, the spatiotemporal errors improve when the object is actually used rather than when its use is pantomimed. In addition, the movements are often performed normally when they occur spontaneously but are impaired when the patient is instructed to perform the movement. The patient may scratch his nose spontaneously but may be unable to perform this task to command. This voluntary-automatic dissociation, although frequently seen in ideomotor apraxia, is not universal [ 10 ]. The standard conceptualization of apraxia is based on a two-system model of action: a conceptual system, located in the dominant parietal lobe, and a production system localized to the frontal lobe [ 11 ]. Dysfunction of the former would lead to ideational apraxia and dysfunction in the latter would result in ideomotor or limb-kinetic apraxia. With improved knowledge of the multiple frontoparietal circuits that subserve visuospatial transformations for reaching, somatosensory transformations for postural adjustments and the coding of peripersonal space for limb and neck movements, it has been possible to analyze the deficits which occur in apraxia in more detail than afforded by the standard conceptualization. Based primarily on primate studies, specialized circuits responsible for more detailed properties of action have been identified and provide a framework by which the idiosyncratic deficits evident in the apractic patient that defy explanation by the simple two-system model have been explained [ 11 ]. Regardless of whether the standard two-system scheme of action or the more detailed model based on multiple, specialized parietofrontal circuits is used, the standard conceptualization of the origin of apraxia still conforms to the computational paradigm in which a specific motor program is activated based on the conceptual framework in place in the dominant parietal lobe. Similarly, the usual explanation for the automatic-voluntary dissociation frequently seen in ideomotor apraxia relies on this computational framework. This framework postulates that a verbal command establishes a conceptual bias in the parietal lobe that activates the appropriate motor program in the frontal lobe. In ideomotor apraxia there would be a disconnection between the instruction and the effector mechanisms frontally. To explain the preservation of the corresponding automatic movement, alternative pathways not dependent on the parietal lobe would need to remain functional. Evolutionary autonomous agents (EAAs) are robots or robot simulations whose controller is a dynamical neural network and whose evolution is guided by a genetic algorithm. They are embodied agents-either software programs living in a virtual environment or true robots that function in a specified environment. These agents function autonomously in their environments with the agents performing such functions as navigation around obstacles, gathering food, seeking prey or mating partners. [ 12 ] Their development is guided by evolutionary algorithms which utilizes a fitness function to select the most appropriate agents for propagation. Motor or sensory activity, in particular, evolves autonomously in response to the constraints of the fitness function without the organizational restrictions imposed by the notions of determinate motor programs or perceptual representations. Such agents provide a system in which the organization of motor or perceptual activity can be followed and analyzed. Because its nervous system is limited to a small number of neuron-like elements, the analysis of the network dynamics of these agents is also more tractable. Their primary value for the neurosciences is that they provide simple systems unencumbered by the assumptions inherent in present day neurosciences that can serve as a test-bed for thinking about neural processing and techniques for deciphering these processes [ 13 ]. To model the voluntary-automatic dissociation seen in ideomotor apraxia, a lesioned EAA needs to demonstrate a behavioral dissociation between a movement that the agent does automatically and an identical movement that has a symbolic context. An analysis of the change in network dynamics that occurs in the apractic EAA will provide information on the physical basis of the dissociation without the assumptions inherent in a computational formalism. The extrapolation of results in the EAA to human brain function is based on a principle concerning the organization of complex systems that has been emphasized by Herbert Simon. He suggested that the organization of self-organizing complex systems is dependent only on the behavioral characteristics of the system and not the nature of the constituent elements of the system. "My central theme is that complexity takes the form of hierarchy and that hierarchic systems have some common properties independent of their specific content" [ 14 ]. Regardless of whether the system consists of 100 billion interacting living cells or a small number of computer generated input-output units, it is not inconceivable that the organizational structure of the systems will be similar if their demands are identical and if they are allowed to evolve autonomously. In the absence of an evolved agent with language capabilities, a paradigm is needed that captures the essential elements of an inability to move to command with preservation of that same movement if performed spontaneously. It is felt that the ability to move to a target location without ongoing visual feedback from the target represents the simplest activity upon which the dynamics and connectivity of a robotic system could develop cognitive functions such as off-line reasoning [ 15 ]. Since predicative activity such as language may originate in this type of network activity, this particular movement paradigm may be used as a surrogate for a verbal command. The presence of multiple time scales in the dynamics of a neural network is indicative of a temporal hierarchical structure. Simon discussed the presence of higher and lower frequency dynamics in complex systems and associated more executive function with the lower frequency components. He stated that "it is generally believed that the relevant planning horizon of executives is longer, the higher their location in the organizational hierarchy" and that "the average interval between interactions are greater at higher than lower levels [ 14 ]." In the nervous system, it is also expected that such a multiple time scale framework would occur with a hierarchical structure requiring more executive function reflecting lower frequency dynamics evolving as tasks become more complicated. In this paper, the temporal hierarchical structure of the dynamics of the neural network controller of an EAA will be assessed by the analysis of the power spectral distribution and Hurst exponent of all nodes in the network [ 16 ]. The analysis will be applied to an EAA model of the voluntary-automatic dissociation seen in ideomotor apraxia in an attempt to causally explain its physical origin. Methods A simulation platform, WEBOTS (Cyberbotics, Switzerland), was used to simulate the movement of a Khepera robot (K-TEAM Corporation, Switzerland) in a dark 1 m square arena without walls or obstacles. The Khepera is a two-wheeled mobile robot with eight ambient light sensors and a rotation encoder for each wheel. The Khepera also has eight infrared proximity sensors but these were not activated in the present study. The neural network controller was composed of a layer of 5 fully connected radial basis function units or neurons. In addition, each unit projected to the two linear units or motor neurons and received inputs from the eight light sensors and the two wheel encoders. A combined genetic and adaptation algorithm was used similar to that described by Urzelai and Floreano [ 17 ]. Briefly, the first generation at the beginning of the evolution was composed of a hundred individuals with randomly assigned synaptic weights and adaptation rules. A chromosome was made up of all the synaptic weights and their associated adaptation rules. A standard genetic algorithm with cross over and mutation operators were used with the best twenty individuals selected. The roulette wheel selection method, wherein the probability of an individual being selected for a new generation is the individual's fitness as a fraction of the total population fitness, was used. One of five Hebbian adapation rules were associated with each weight, the standard Hebbian, presynaptic, postsynaptic, covariance [ 17 ] and no change. Although the synaptic weights were altered by the adaptation rule during the life of each individual, it was the pre-adaptation weights and rules that were transmitted to the next generation. One individual had a lifespan of 75 seconds. Each sensorimotor cycle was 64 ms allowing adaptation to occur over 1172 iterations. The arena was maintained dark, except at random times when a light of constant intensity would appear at random locations. The robot had to attempt to reach the light. There were two tasks for the robot during each trial. In the first task, the robot simply had to go to the light when it appeared. The light remained on and was extinguished when the robot moved to within 2 body diameters, or 7 cm of the light. After a 1–3 s of random delay (in darkness) another light would appear in the arena at another random location requiring the robot to approach that light. This would continue for a total of 4–6 constant lights. The second task began after this series of constant lights. With the second task the robot had to go to the location of a brief light flash (1 s or 16 sensorimotor cycles in duration) that did not persist while the robot approached the location. Since the robot could not use light intensity to continuously guide its movement to the location, successful completion of this task required the robot to have some notion of objective orientation and distance. Each trial lasted a total of 75 seconds. The fitness function F was defined as: F = T ( s )· n where T ( s ) = 1 for s ≤ 1 and T ( s ) = s for s > 1, n the number of lights reached, d f the average distance of the robot from the flash point once the flash occurred, t f the time between flash onset and end of simulation, and t s the duration of simulation. Mutation rate was in the 0.2 – 0.5% range. Evolution was continued until the fitness plateaued. Three populations were evolved requiring up to 1300 generations before there was a clear plateau. Individuals in each population were screened for their ability to accomplish the tasks. Each screening trial was the same as the trials in which the robots were evolved, that is, a 75 second epoch during which the robot had to approach several sustained light locations followed by a single unsustained light flash. Invariably, individuals appeared first that were able to go to the sustained light but failed to respond to the light flash. Eventually, individuals with high fitness were found that were able to successfully complete both tasks. From this latter group, lesions were introduced into the neural networks and the subsequent individual was screened with the same 75 s trial. Individuals were subsequently identified in whom the lesion resulted in the loss of the ability to move toward the brief light flash with preservation of the ability to move toward the sustained light. The lesions that were introduced into the networks were inactivations of single synapses. Typically, an individual that could successfully accomplish both tasks was screened for apraxia by lesioning each synapse in the network and observing the subsequent behavior of that individual. Although individual neurons could have been inactivated, this approach was avoided because the network only had 5 neurons and the resulting lesioned individual often demonstrated gross motor deficits. Once individuals were identified which could successfully accomplish both tasks and in whom a specific lesion resulted in a dissociation between movement towards the sustained light and movement towards the light flash, their network dynamics were characterized in two ways. First, a fast Fourier transform (FFT) was performed on the activation pattern of all 5 neurons in the network. The time window chosen in the calculation of the FFT was the entire test epoch. The Hurst exponent was then calculated by the average wavelet coefficient method [ 18 ] for all 5 neurons in the network again using the entire test epoch. Results Five agents were identified which were able to successfully accomplish both tasks and in whom a lesion resulted in the dissociation between the two tasks. Analysis was performed on all 5 of the neurons in each individual. Five trials were used in each individual to allow statistical analysis of the Hurst exponent data. Figures 1 and 2 demonstrate prototypical data obtained for one individual which was able to accomplish both tasks prior to lesioning and demonstrated a dissociation after lesion. In figure 1 , before lesioning, only the trajectories to the last constant light stimulus followed by the brief light flash are shown. In figure 2 , after lesioning, the last two trajectories to the sustained light stimuli are shown with the agent failing to approach the brief light flash. Individual neuron activity of all 5 neurons for the entire test epoch before and after lesioning along with the FFT data of each neuron before and after lesioning is also shown. The Hurst exponent for the entire epoch for each neuron is also indicated. Visual inspection of the FFT data showed a consistent trend in all neurons with a relative loss of low frequency components in the apractic robots compared to the normal situation. Figure 1 (a) Trajectory of a robot which was able to accomplish both tasks; only the last 3 movements are shown. The robot begins at the open triangle. It then moves to position 1 (sustained light) and then to position 2 (brief light flash). (b) Activity from each neuron of the network and from the light sensor; data for the entire epoch is shown which constituted movements to 5 different sustained lights followed by 1 brief light flash. (c) FFT data for all five neurons calculated over the entire epoch; the value of the corresponding Hurst exponent for each neuron is on the top right. Figure 2 (a) Trajectory of a lesioned robot which lost the ability to move to the brief light flash; only the last 3 successful movements are shown. The robot begins at the open triangle. It then moves to position 1 (sustained light) and then to position 2 (sustained light). The robot remained at position 2 despite the occurrence of a brief light flash at position 3. (b) Activity from each neuron of the lesioned network and from the light sensor; data for the entire epoch is shown which constituted movements to 6 different sustained lights and inability to move to the brief light flash. (c) FFT data for all five neurons of the lesioned robot over the entire epoch. The value of the corresponding Hurst exponent of each neuron is on the top right. We computed the Hurst exponents of the 25 neurons from the 5 individuals drawn from 3 populations. Five trials were executed for each individual. When comparing the Hurst exponents for each neuron before and after the lesion, highly significant differences were obtained (p < 0.03 to 10 -7 ) for 24 of the 25 neurons, with the remaining neuron showing the same trend but not reaching statistical significance. The grand average of the 125 Hurst exponents (5 trials each for 25 neurons) was1.0957 ± 0.2325 for the normal condition, and 0.8292 ± 0.1833 for the apractic condition. Both the obvious frequency changes in the FFT data and the Hurst exponent calculations show that the loss of the ability to perform a movement that has a symbolic context is associated with a loss in the longer time scales (low frequency components) in the hierarchical organization of the neural network Since the calculation of the Hurst exponent is based on the full test epoch and since most of the robot movement during this period was the movement to the constant light rather than the light flash, this result also demonstrates that the strategy for both movements was influenced by the lesion despite maintained ability to move to the constant light. Discussion Despite significant advances in the understanding of the anatomy and physiology of normal movement and the pathophysiology of movement disorders, their conceptual framework still relies on the computational paradigm used by cognitive science and neuroscience for decades. This approach assumes the existence of determinate structures in the CNS that are manipulated in the same fashion that a computer manipulates symbols or language combines words. In the last 10 years, there has been a shift in focus with appreciation of the limitations of the computational approach and the realization of the importance of embodiment in conceptualization of motor behavior. When the social and the environmental influences on behavior are analyzed without prejudice, it becomes clear that indeterminacy is a hallmark of our functioning [ 19 ] and the idea of determinate perceptual representations and motor programs is an assumption commensurate with the approach of the natural sciences in general. This indeterminacy is captured in the formalism of dynamical systems theory. EAAs provide a system by which development of cognitive structures evolve in an embodied agent constrained only by the nature of the environment and the definition of fitness. They assume no structural characteristics of perception or movement but provide a forum by which sensorimotor structures arise as the organism self organizes in response to its interaction with the environment. Because they possess a simple network controller, the analysis of their development and architecture is more tractable than a similar analysis of the human nervous system. We have demonstrated that in an evolved agent with two modes of environmental interaction that can be viewed as automatic and symbolic, lesions that produce the equivalent of the automatic-voluntary dissociation seen in ideomotor apraxia cause a change in the hierarchical organization of its dynamical neural network controller with a loss of the longer duration time scales corresponding to the loss of symbolic function. Two techniques were utilized to demonstrate these changes in the network dynamics, a Fourier analysis of the activity of all 5 neurons in the network and the calculation of the their Hurst exponents. With Fourier analysis, EAAs that were able to successfully perform both tasks demonstrated a 1/f power spectral distribution. This type of scale free distribution has been encountered in a number of physical and biological systems including heart rate variability, electrical currents and reaction times in human cognition [ 16 , 20 ]. The 1/f distribution has also been associated with self-organized criticality, a theory of the internal interactions of large systems that has been postulated to govern such diverse natural phenomena as the size of avalanches and earthquakes [ 21 ]. This 1/f pattern was altered in the apractic EAAs indicating that this abnormal behavior was associated with a loss of the lower frequency components in the network dynamics. The calculation of the Hurst exponent is based on long range correlations over all time scales in the signal of interest and is a measure of persistence or memory, that is, how long a given fluctuation in a time series will be reflected in future values of the series [ 18 ]. The larger the value of the exponent, the more dominant are low frequencies in the time signal. EAAs that could successfully perform both tasks had a larger value of the Hurst exponent compared to the lesioned, apractic agents again indicating the loss of low frequency components in the network dynamics of the apractic agents. Does the EAA paradigm described capture the essential features necessary to causally explain the automatic-voluntary dissociation seen in ideomotor apraxia? Ideally, to most reliably model this phenomenon, an EAA would need to have language capabilities. Barring that, it is argued that the ability to move to a target site without the continued visual presence of that target captures the minimal requirements necessary to extrapolate to the case of a verbal command. Although controversy exists concerning the conceptual or operational definition of internal representations in evolved agents [ 22 , 23 ], Clark has argued that the notion of representation remains valuable in evolved robotic systems and can be defined well enough to discuss the physical basis of cognition [ 15 ]. He distinguished between what he called weak and strong representations. A weak representation in a robotic system is the dynamics and connectivity of the network which is associated with a behavior requiring ongoing sensory feedback. A movement that is done spontaneously and automatically would be a type of behavior whose underlying network would use a weak representation. A strong representation in a robotic system is the network dynamics and connectivity associated with a behavior that does not require ongoing sensory feedback for its successful execution. He argued that the kind of network organization that supports movement in the absence of visual feedback, such as the movement of an EAA to a target flash location, is the simplest kind of strong representation and is prototypical of representations in robotic systems that could support more sophisticated cognitive processes including off-line reasoning. Since language capability would require the presence of strong representations, it was felt that the movement of an EAA to a target location without continued visual presence of the target was an acceptable surrogate for a verbal command. It was clear while individual EAAs were being observed as they performed the trial in the normal and apractic conditions that the apractic robots often demonstrated minor deficits even in the case in which they were approaching the sustained light target. Usually, this took the form of a mild slowness in movement although other deficits were seen such as altered trajectories toward the sustained light source. It has been assumed that patients with ideomotor apraxia function normally in the environments to which they are accustomed and in which automatic behavior is expected. This has also been suggested as a reason why apraxia is not detected as frequently as its true incidence predicts [ 10 ]. This clear cut dissociation would also be predicted by a computational formalism since distinct and determinate motor programs would lend themselves to inactivation of one with maintenance of the other. Recently, this idea has been disputed and, in fact, careful examination of apractic patients reveals deficits even in movements that are performed automatically [ 10 ]. The results in our EAA simulation corroborates this more recent analysis. Since determinate motor programs or perceptual representation do not exist in the EAAs, it would be expected that a lesion in the network would have some effect on all aspects of the agent's motor performance. The notion that a lesion that causes apraxia has a global effect on the patient that is not restricted to those functions that can be considered symbolic was elaborated in detail by Merleau-Ponty [ 19 ]. He performed an extensive existential analysis of one patient, Schneider by name, who had suffered a shell injury to his brain. Merleau-Ponty suggested that the standard empirical or cognitive explanations of the nature of apraxia are limited because they fail to take into consideration the phenomenology of apraxia. According to Merleau-Ponty, "Beneath the intelligence as an anonymous function or as a categorical process, a personal core has to be recognized, which is the patient's being, his power of existing. It is here that illness has its seat". He argued that it is "ridiculous to think that the shell splinter directly struck symbolic consciousness" but rather "any pathological degeneration should affect the whole of consciousness." This phenomenological insight was mirrored in the observation that lesions that resulted in a behavioral dissociation in the EAAs had an influence on the dynamics of the network even in the tasks that were behaviorly unchanged. By avoiding the assumptions inherent in the computational framework and by evolving hierarchical self-organization, including symbolic thought, from the most basic level of sensorimotor interactions in an embodied agent, EAAs provide the optimal framework by which these phenomenological observations on the nature of apraxia can be modeled. In fact, taking phenomenological accuracy in addition to physiological plausibility as a constraint in the selection of those EAAs that successfully perform a behavioral task will help to discover those networks whose architecture and function most closely parallel those of the human nervous system [ 24 ]. Conclusion In an embodied agent with two modes of environmental interaction which can be viewed as automatic and symbolic, lesions that produce the experimental equivalent of voluntary-automatic dissociation of ideomotor apraxia cause a change in the hierarchical organization of its dynamical neural network controller with a loss of the longer duration time scales corresponding to the loss of symbolic function. The identification of the dynamical changes in an apractic nervous system may provide a causally mechanistic explanation for the syndrome that may complement the higher-order descriptive explanation of the standard computational approach. The methodology of EAAs is just beginning to be recognized and it is hoped that further studies can be performed with this methodology to further understand the physical origin of normal and pathological brain function. This study was funded by a research grant from the Toronto East General Research Foundation. We thank Weyland Cheng and Owen Wong for assistance in the research. Author's contributions DSB conceived the study, participated in its design and the data analysis and drafted the manuscript. FJ carried out all computer simulations, participated in the study design and the data analysis and contributed to the manuscript draft. HCK conceived the study, participated in its design and the data analysis and contributed to the manuscript draft. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544844.xml

423161

A Protein's Role in Progressive Renal Disease

Focal segmental glomerulosclerosis (FSGS) made a brief media splash last year when a kidney transplant forced NBA superstar Alonso Mourning into early retirement. Mourning's condition elicited a flood of calls from fans offering their kidneys, but most people with kidney disease are not so lucky. Some 56,000 patients await transplants; many have waited over five years. FSGS, which underlies about 25% of the 60,000 kidney-related deaths each year, causes inflammation and irregular scarring in the glomeruli, clusters of blood vessels in the kidney that filter toxins from the blood. These lesions, which allow protein and blood to escape into the urine, cause progressive kidney failure. FSGS commonly occurs as an outgrowth of various primary disorders, including obesity, HIV infection, diabetes, and hypertension. Though it's not clear what causes FSGS, this form of renal pathology is becoming more common. By using the genes underlying inherited forms of FSGS as probes, scientists hope to uncover the mechanisms that unleash the disease and to find ways to stem the damage. Mutant α-actinin-4 is mislocalizes and aggregates in renal glomeruli Mutations in the ACTN4 gene, which encodes a protein called α-actinin-4, cause an inherited form of FSGS. The protein normally remodels actin filaments, the primary structural component of muscle and cytoskeleton. Having a single mutated copy of the gene can cause FSGS in humans, though it is unclear how. In this issue of PLoS Biology , Martin Pollak and his colleagues at Brigham and Women's Hospital at Harvard Medical School use a three-pronged approach to figure out how the defective protein wreaks renal havoc and how these physiological changes lead to FSGS. Using biochemical analysis, cell-based studies, and a newly developed “knockin” mouse model, the researchers report that FSGS-related mutations cause α-actinin-4 to engage in various aberrant behaviors that ultimately rob the protein of its function and poison cells. In previous experiments, Pollak's team had discovered that some families with the inherited form of FSGS carried mutations in ACTN4 . In these individuals, the disease appeared to strike podocytes (glomeruli epithelial cells) first. While engineering mice designed to carry mutations in this gene, the researchers created mice that lacked detectable Actn4 expression. These “knockout” mice developed severely damaged podocytes and progressive glomerular disease. In the current experiments, the researchers returned to their “knockin” mice, which carry two copies of the mutation found in the families with inherited FSGS. They also generated “normal” mice and mice harboring one normal and one mutant copy of the gene. In the biochemical experiments, the researchers investigated the mutant protein's binding behavior. Typically, two α-actinin-4 proteins form a twosome without incident, but here the mutants behaved badly, assuming improper structural conformations and forming aggregates rather than pairs. Next, Pollak and colleagues introduced the genes with the Actn4 mutation into podocytes, using a variety of methods, to see where in the cells the expressed proteins turned up. They also introduced fluorescently labeled mutant and normal Actn4 genes into podocytes that were grown from the three mouse types: normal proteins were diffused throughout the cytoplasm in each cell type, but the mutant proteins showed an uneven distribution. Analysis of various tissues taken from the knockin mice revealed normal levels of mRNA transcripts—indicating normal gene transcription—but “markedly reduced” α-actinin-4 protein levels. The mutant proteins, it turns out, were manufactured normally but were degraded far more quickly than normal proteins. Electron microscopy showed that podocytes in the kidneys of the knockin mice had structural defects, while the mice themselves had significantly higher levels of protein in their urine than mice with one or two normal copies of the gene did. The finding that FSGS-associated Actn4 mutations produce α-actinin-4 aggregates with significantly reduced life span, the authors explain, suggests two possible mechanisms of initiating disease: aggregation and the toxic affects of aggregation could injure podocytes, or loss of α-actinin-4 function caused by rapid degradation of the protein could produce injury. Pollak and colleagues argue that both factors likely play a role: Actn4 mutations lead to both reduced α-actinin-4 activity and protein aggregation, and the loss of protein function and the toxic effects of protein aggregation produce glomerular injury. As inherited renal disease typically emerges later in life, it may be that α-actinin-4 aggregation causes incremental but cumulative podocyte damage over time. So what does this mean for patients with progressive renal disease? While these findings may not translate into clinical applications anytime soon, they do suggest that therapies aimed at repairing the structure or expression of these essential cytoskeletal proteins might return the renegade proteins to the fold.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC423161.xml

529464

Treatment outcomes in locally advanced colorectal carcinoma

Background Locally advanced colorectal cancers form a distinct subgroup where contiguous organs could be involved without distant metastases and so may be amenable to curative surgical resection. It was our objective to report our experience in treating six such patients with operable locally advanced colorectal carcinomas. Methods We retrospectively reviewed the case notes of 47 patients who were diagnosed with colorectal cancers at M S Ramaiah Medical Teaching Hospital between the years 1996 – 2001. Six patients were identified with T4 lesions, adjacent organ involvement and with no nodal involvement. The treatments and outcomes for these patients were then reviewed. Results Two of three patients with rectal malignancies who underwent pelvic exenteration succumbed to disease recurrence within the first 18 months. One of the three patients with colonic cancers died of non malignant causes. The other two are disease free till date. Conclusions Aggressive multivisceral resections for locally advanced colonic cancers might be appropriate. Rectal cancers when locally advanced may be considered for pelvic exenteration, but a more guarded prognosis may apply.

Background Locally advanced colorectal tumors constitutes to about 5 – 22% of all colorectal cancers at the time of presentation [ 1 ]. This type of tumor forms a distinct sub class of colorectal tumors characterized by aggressive local behavior in the form of invasion of adjacent organs or structures with somewhat surprisingly no distant metastasis at presentation. The survivals of such cases that undergo multivisceral resections are 58% and 43% for UICC stage II and III respectively. These results are similar to those undergoing conventional resections [ 1 ]. In addition, there is a suggestion of elevated stage related late results to the same level as that associated with tumors where there is no direct invasion of contiguous organs [ 1 ]. Addressing the local disease adequately with multi-visceral resections when necessary could result in favorable out come. However en-bloc surgical resection of the tumor forms a surgical challenge and the risks of complications and death must be weighed against probable survival benefits. The post operative complication rate of multivisceral resection is about 11.5%; 30 day operative mortality is 3.6% and compares favorably with non-multivisceral resections [ 1 ]. It was our objective to report our experience in treating six such patients with operable locally advanced colorectal carcinomas. Methods During the years 1996–2001, 47 patients with colorectal cancers were treated at our institute. All patients underwent a rectal exam, punch biopsy if it were an accessible rectal lesion or colonoscopy and biopsy if it were a colonic lesion. Screening colonoscopy for a synchronous lesion was however done for all patients. Patients in whom a biopsy confirmed a malignant lesion underwent an ultrasound (US) study of the abdomen, chest X-ray and Carcino-embryonic antigen (CEA) estimation. Patients with doubtful involvement of adjacent organs or structures underwent CT scan. Cystoscopy was performed when urinary bladder involvement was suspected on CT scan. Patients were diagnosed as locally advanced when the staging evaluation showed involvement of adjacent organ or structure. Though nodal evaluation is suboptimal with US or CT scan, all patients with 'N0' status on these investigations were included as locally advanced while those with 'N+' on evaluation were excluded. Patients with metastasis to any organ were excluded. The preoperative staging of the tumor was T4, N0, M0. Elective surgery was performed on all cases. No pre-operative therapy was administered. All patients were administered adjuvant radiation and chemotherapy which was instituted immediately after wound healing (between 16 and 29 days post surgery). A total dose of 5000 centi-Gray of external beam radiation in combination with chemotherapy consisting of 5-Fluoro-Uracil (5-FU) and leucovorin was administered. The dose of 5-FU was 425 mg/m 2 and that of leucovorin 20 mg/m 2 . This combination was administered as infusion on day 1 to 5, and six such cycles every 28 days were instituted as part of chemotherapy regimen. The therapy in colonic carcinoma was sequenced as 1 cycle of chemotherapy followed by radiation and the remaining 5 cycles after completion of radiation. In case of rectal cancers, the chemotherapy was administered concurrently with radiation. The administration of leucovorin was with-held during radiation and instead, the dose of 5-FU was increased to 500 mg/m 2 . Follow up was by clinical examination every 3 months for first two years, and six monthly until 5 years. The evaluation included alternate year CT scan, annual US abdomen, chest X-ray, colonoscopy and CEA levels every six months. The two live patients have given consent for publication. Consents have been obtained from the legal heirs of the remaining four patients. Results The study group consisted of six patients (29 to 70 years), three of whom had locally advanced colonic disease and three of whom had locally advanced rectal disease. Pre operative CEA levels were surprisingly within normal limits in all six patients. The pre operative colonoscopic or rectal biopsy had determined adenocarcinoma in all six patients (Grade as shown in Table 1 ). In two of the patients with rectal disease where involvement of the bladder wall was suspected, cystoscopic findings were normal. All the patients were offered radical surgery on elective basis. The radicality included removal of adjacent involved organs which could increase the complication rate of extensive surgery. In addition such extended resections for rectal cancer would result in exenterative pelvic surgery resulting in abdominal stomas. Surgical resection margins were negative in all cases. Table 1 Salient Clinico – pathologic features of locally advanced colorectal carcinomas Site of malignancy Age Sex Pathologic 'T' Status Grade M/SR * ca Adjacent Org † Nodal Status PNS ‡ PN/PV § spread Colon 56 Male T 4 II - + (abdominal wall) N0 - - 58 Male T4 II - + (abdominal wall) N0 - + 67 Female T3 II - - (bladder dome) N0 - - Rectum 45 Male T4 I + + (bladder / prostate) N1 + + 27 Female T4 I + + (uterus and vagina) N2 + + 69 Male T4 II + + (bladder / prostate) N2 + + * Mucinous or signet ring carcinoma † Pathologic involvement of adjacent organ ‡ Perinodal spread § Perineural / perivascular spread Case 1 Patient presented with a 2 month history of abdominal mass. There was no history of weight loss or altered bowel habits. The mass was firm with irregular borders and restricted mobility. CT scan of abdomen revealed a large mass lesion arising from the caecum and ascending colon apparently infiltrating the anterior abdominal wall (Fig 1 ). Patient underwent a radical right hemicolectomy with en-bloc resection of involved abdominal wall (Fig 2 ). The right branch of the middle colic, the right colic and ileocolic vessels were ligated at the origin. Retroperitoneum was not involved. A two layer hand sewn ileocolic anastomosis was performed. The abdominal wall muscles were resected with a 2.5 cm margin and the resultant defect was repaired with a polypropylene mesh. The patient received adjuvant radiation and chemotherapy as described above and is disease free 3 1/2 years after surgery. Figure 1 CT scan of the abdomen showing the involvement of abdominal wall muscles and adhesions to neighboring intestines. Figure 2 Specimen of radical right hemicolectomy with en-bloc resection of abdominal wall muscle (indicated by small arrow) and neighboring small intestines (indicated by dotted arrows). Case 2 Patient presented with a 6 month history of bleeding per rectum. There was no history of weight loss. Patient's hemoglobin was 10.2 gms / dl. Patient had ignored the complaint for 3 months but even at a later date was unfortunately not advised to undergo sigmoidoscopy where he was evaluated. Patient underwent colonoscopic evaluation at our institute after evaluation. CT scan showed possible involvement of abdominal wall with the growth arising from sigmoid colon. Intra-operatively, the mass was found to involve the posterior rectus sheath. Sigmoid colectomy was performed along with resection of the involved posterior rectus sheath and the rectus muscle en-bloc. Splenic flexure was mobilized to obtain colorectal hand sewn anastomosis. Primary closure of the rectus defect was achieved and patient received adjuvant radiation and chemotherapy. The patient is disease free 3 years post surgery. Case 3 Patient was diabetic, asymptomatic; ultrasound performed for evaluation of kidneys for diabetic nephropathy incidentally detected the mass lesion in sigmoid colon. Biopsy specimen was obtained at colonoscopy; no synchronous lesion was detected in the rest of the colon. CT scan revealed close relation of the sigmoid mass lesion to the bladder dome but was unable to comment categorically on wall infiltration. Urinary bladder mucosa was intact and normal on pre-operative cystoscopy. Intra-operatively, the lesion was found adherent to bladder dome. Sigmoid colectomy and part of the bladder wall was resected en-bloc. There was no pathologic evidence of bladder involvement. Patient had ishcaemic heart disease and diabetic nephropathy; she later developed myocardial infarction and renal failure while on adjuvant therapy 7 months post surgery and succumbed to the same. Case 4 Patient had bleeding per rectum and constipation for 18 months. Patient was evaluated and treated as hemorrhoids initially (no surgical intervention). About 6 months prior to presentation at our institute, he was diagnosed as rectal cancer and was advised abdomino-perineal resection. Patient was scared of surgery and waited without any therapy. Rest of the large bowel was normal on colonoscopy. A CT scan showed involvement of the urinary bladder and prostate by a rectal mass while bony pelvis appeared free of tumor involvement. The patient underwent a total pelvic exenteration with formation of a urinary diversion by ileal conduit. After a lower midline laparotomy, abdomen was evaluated for any ascites or liver metastasis. Para-aortic area was palpated and rest of peritoneal cavity was evaluated for any metastatic deposit. Inferior mesenteric artery was ligated beyond the origin of the left colic artery. Bilateral pelvic nodal dissection was completed. Sigmoid colon and rectum was mobilized as in abdominoperineal resection over the sacral hollow. Anteriorly the dissection was carried out in the retropubic space to access the urethra beyond the prostate. Lateral dissection was carried out which included the ligation of the superior vesical vessels. Both the ureters were ligated below the pelvic brim in the true pelvis at least 3 cm proximal to palpable disease. Isolated loop of ileum was mobilized based on two vessels. One end of the loop was closed and both ureters were anastomosed separately to the loop. Intestinal continuity was obtained with ileo-ileal end to end anastomosis. This ileal loop was brought out as a urinary stoma on the right side and the cut colon was brought out as colostomy on the left side. The patient's postoperative recovery was unremarkable except for a uereteric leak which settled by 10 th post operative day. Patient received adjuvant therapy but was irregular on follow-up and refused to be investigated. Patient developed hepatic metastasis and succumbed to the disease after 18 months. Case 5 Patient had bleeding per rectum and constipation for 4 months. Patient had significant weight loss over previous 2 months. Patient had circumferential rectal carcinoma involving uterus and upper part of vagina. Rectal carcinoma was proved by biopsy. Rest of the colon was normal on colonoscopy. Upper vaginal mucosa was intact but indurated underneath. CT scan showed infiltration of lower part of uterus and upper vagina. Patient underwent posterior pelvic exenteration. Uterus, including both the fallopian tubes and ovaries were removed en-bloc along with vagina and rectum. Surgery was followed by adjuvant radiation and chemotherapy. Patient succumbed to the disease 1 year later with multiple metastases in liver, brain and malignant ascites. Case 6 Patient had bleeding per rectum and constipation for 3 months. A preoperative CT scan with rectal and intravenous contrast revealed involvement of prostate and possibly bladder. Colonoscopy did not reveal any other lesion. He underwent a total pelvic exenteration (Fig 3 ) with ileal conduit similar to case 4. Patient developed fever 48 hours after surgery. Evaluation for infective pathology including cultures from catheters and venous access tips did not reveal any source of infection. There was no pocket of collection on repeated abdominal ultrasound. There was raised leukocyte count. A possibility of sepsis from an occult focus was considered and treated. However, on the 5 th post operative day, patient developed vomiting and had raised serum creatinine values. Values of fibrin degradation products were also raised and a diagnosis of disseminated intravascular coagulation and renal failure was entertained. Patient had prolonged prothrombin time but did not have any clinical bleed. He died on 30 th day post operative while on recovery from the same. The post surgical pathology reports are summarized in Table 1 . Figure 3 Specimen of pelvic exenteration with cut open rectum showing the rectal adenocarcinoma infiltrating the prostate. Foley's catheter is also seen. Discussion Surgical resection remains the primary and most effective treatment for advanced colorectal cancers [ 2 ]. The 5-year survival rate of the overall group of patients undergoing multivisceral resection is 42%, that of the subgroup undergoing curative surgery is 51%, and that of the subgroup receiving only palliative resection is 0% [ 1 ]. In the presence of local tumor extension the distinction between inflammatory adherence and malignant invasion is impossible to make intra operatively. Adherence of the neoplasm to surrounding structures demonstrates pathological tumor invasion in approximately 50% of cases [ 1 , 2 ]. The operative intent of the surgeon should be to achieve complete extirpation with adequate margins in the involved structures. Dissection between malignancy and adherent structures, or biopsy and frozen section is not recommended as these procedures may promote tumor dissemination [ 1 , 2 ], which may have a detrimental impact on prognosis [ 3 - 5 ]. The concept of en-bloc resection has significantly reduced the local recurrence rate from 77% to 36% and significantly improved the 5 year over all survival of 43% [ 1 , 4 - 6 ]. Other studies have also shown improved survival of similar staged colorectal cancers with en-bloc resection [ 7 , 8 ]. The survivals reported in various studies are shown in Table 2 [ 1 , 2 , 4 , 9 - 11 ]. The 5-year survivals are comparable ranging from 38% to 52%. Although colon and rectal cancers which required multivisceral resections are shown separately, the survivals are not indicated differently. For similar 'T' status, the corresponding 'N+' status of colon versus rectum has not been studied. Our present study is small but the overall survival is 33% and compares with larger studies. In addition, since a very high percentage of even large T4 tumors have not yet metastasized to the regional lymph nodes, a multivisceral resection offers the chance to radically remove the local disease and affect a cure [ 1 , 12 ]. All patients in our study group underwent en-bloc resection. No attempt was made to separate the adherent structure to confirm or negate the involvement of adjacent organ per operatively. Since colon spans the entire periphery of the abdomen, almost all the abdominal organs have been reported to be involved [ 2 ]. Multivisceral resections have been recommended whenever necessary as it could offer cure or at least significant palliation [ 13 - 15 ]. En-bloc resections have been recommended even when it warrants a pancreatico-duodencetomy [ 16 , 17 ]. In the present report, the en-bloc resections included abdominal wall in two cases and urinary bladder in one. Table 2 Summary of survival data from a few large studies Author Multivisceral Resections Colon Rectum Death Rate 5-year Survival Rate Reiner (1987) 158 53 105 12% 38% Heslov (1988) 58 26 32 5% 38% Montesani (1991) 33 - - - 30% Hermanek (1992) 197 119 78 3% 52% Gebhardt (1999) 173 122 51 4% 51% Lehnert (2002) 201 139 62 7.5% 51% Rectum cancers involve the pelvic genital organs and or urinary bladder anteriorly. Posteriorly it could involve the sacrum. Exenterative pelvic surgeries are warranted for locally advanced rectal cancers [ 18 ]. Pelvic exenteration and sacral resection for primary or recurrent rectal cancer are tolerable procedures with a low mortality rate [ 19 ]. Although they provide a survival benefit if curative resection is possible, the associated morbidity remains high [ 19 - 21 ]. The complications described include sepsis, intra-abdominal abscess, pelvic abscess, enteric fistula, enteric anastomotic leak, wound infection, ileoureteral anastomotic leak, ileoureteral anastomotic stenosis, bowel obstruction, vascular injury, bleed, liver dysfunction and pneumonia. In the present study, both the male patients underwent total pelvic exenteration with ileal conduit for urinary diversion. The female patient underwent posterior pelvic exenteration. The male patient who lived for 18 months did not have any major problem to take care of both the stomas. Similar thoughts that a small reduction in patient's quality of life due to urinary diversion should not be a major contraindication when surgery with urinary diversion is warranted to obtain curative resection have been echoed [ 22 ]. Pre-operative nodal evaluation by abdominal CT scan, MRI or endoscopic ultrasound may be inadequate. Though endosonography can detect perirectal nodes, its inability to reliably predict pathologic involvement is a constraint [ 23 ]. Nodal status assessment is considered adequate when at least 14 nodes are examined [ 24 ]. In some studies, nodal metastasis has insignificantly altered survival [ 4 , 6 , 25 , 26 ]. In contrast, other studies have shown that presence of lymph node involvement is associated with poor prognosis [ 21 ]. Studies have shown that 5 year survival in patients with nodal metastasis was 0% to 11%, significantly lower than the 37% to 76% survival rate in their patients without nodal metastases [ 11 , 27 , 28 ]. These studies have even cast doubts over usefulness of pelvic exenteration in patients with nodal disease though some would still recommend it [ 29 ]. None of the patients in the study were evaluated by endosonography but were evaluated by CT scan. Accepting that it is a poor tool for nodal evaluation, pre-operative involvement was not detected in any patient in the study group. Moreover, lymph node status can be accurately determined by pathologic examination only. Hence the investigations might not contribute to decision making though cases with obvious metastasis could be excluded from major resections. In our series of six cases, all the colonic patients were node negative and had no vascular invasion. With multi visceral resections, T4 colonic cancers had acceptable morbidity and better treatment outcome. However, rectal cancer patients had poor prognostic factors like vascular invasion, lymph nodal involvement and histological type of mucinous / signet ring variety. Another striking feature was that though colonic cancers were grade II tumors, they were pathologically N0. But rectal cancers were pathologically N1, N2 and had perinodal spread despite being grade I tumors. Down staging of locally advanced rectal cancers have been achieved with pre-operative radiation or chemotherapy or both resulting in decreased recurrence and improved disease free survival [ 30 - 35 ]. The results of IORT hold some promise [ 36 , 37 ]. It would be fair to say that there is still no agreement as to the optimal sequencing of chemotherapy, radiation, and definitive surgery in immediately operable patients, and both preoperative and postoperative approaches have vocal proponents [ 38 ]. In the present study, pre-operative chemotherapy or radiation was not administered to any patient. There have been no studies to suggest that pre-operative chemoradiation would decrease the magnitude of surgical resection and hence a patient suitable for exenteration would still require the same after such a therapy. In addition, administration of radiation pre-operatively increases the morbidity after exenteration [ 39 ]. It must also be noted that studies on pre-operative chemoradiation have shown improvement is DFS and not in OS. Till such time there are conclusive results in these areas for pre-operative therapy with chemoradiation, such a therapy would continue to be debated. Although it could be inappropriate to draw conclusions from a small number of cases, certain features require further deliberation. T4 colonic cancers have had better outcomes compared to T4 rectal cancers. Many of the earlier reports have combined results of colonic and rectal cancers. In addition, the reports have combined locally advanced and recurrent cancers. As the tumor biology of these areas is different, the results of colon and rectum have to be viewed separately. In addition, the morbidity and mortality associated with multivisceral resections are different of these areas. Since nodal metastasis were seen in all T4 rectal cancers and in none of the colonic T4 cancers, it would be interesting to evaluate percentage of nodal metastasis separately for colonic and rectal cancers with similar T stages. Another indicator of aggressive biology of rectal cancers compared to colonic cancers is the fact that lower grade rectal cancers had more nodal metastasis, perineural and perivascular invasion compared to higher grade colon cancers. Conclusions Locally advanced colonic cancers are to be evaluated and treated aggressively with multivisceral en-bloc resections. The outcomes are likely to be favorable and hence the results gratifying. Locally advanced rectal cancers require exenterative pelvic surgery which carries higher morbidity and mortality. In addition, rectal cancers are biologically more aggressive. Hence exenterative pelvic surgeries are to be done more selectively with guarded disease out come for T4 rectal tumors. Future improvements in chemotherapeutic agents and radiation techniques could make down staging of these malignancies a reality not only in terms of operability but in improving overall survival. List of abbreviations used US: Ultrasound CEA: Carcinoembryonic antigen CT: Computerized tomogram 5FU: 5-Fluoro-uracil MRI: Magnetic Resonance Imaging IORT: Intraoperative Radiotherapy DFS: Disease free survival OS: Overall survival Competing interests The author(s) declare that they have no competing interests. Authors' contributions KH: Was the principal treating surgeon apart from designing and conceptualizing the article. YVN: Was the assistant in the procedures and made the preliminary draft of the article. SN: Planned and administered adjuvant radiation.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529464.xml

526757

On the emergence of multifocal cancers

Several tumors can exist as multiple lesions within a tissue. The lesions may either arise independently, or they may be monoclonal. The importance of multiple lesions for tumor staging, progression, and treatment is subject to debate. Here we use mathematical models to analyze the emergence of multiple, clonally related lesions within a single tissue. We refer to them as multi-focal cancers. We find that multifocal cancers can arise through a dynamical interplay between tumor promoting and inhibiting factors. This requires that tumor promoters act locally, while tumor inhibitors act over a longer range. An example of such factors may be angiogenesis promoters and inhibitors. The model further suggests that multifocal cancers represent an intermediate stage in cancer progression as the tumor evolves away from inhibition and towards promotion. Different patterns of progression can be distinguished: (i) If tumor inhibition is strong, the initial growth occurs as a unifocal and self contained lesion; progression occurs through bifurcation of the lesion and this gives rise to multiple lesions. As the tumor continues to evolve and pushes the balance between inhibition and promotion further towards promotion, the multiple lesions eventually give rise to a single large mass which can invade the entire tissue. (ii) If tumor inhibition is weaker upon initiation, growth can occur as a single lesion without the occurrence of multiple lesions, until the entire tissue is invaded. The model suggests that the sum of the tumor sizes across all lesions is the best characteristic which correlates with the stage and metastatic potential of the tumor.

1 Introduction The occurrence of multiple lesions is observed in a variety of cancers. That is, not one, but several lesions are observed within a given tissue. Multiple lesions can occur by two basic mechanisms [ 1 - 5 ]. Either they originate independently by separate carcinogenic events, or they are generated by a single transformation event (monoclonal origin). Sometimes, the term "multicen-tric cancers" is used to describe the occurrence of clonally unrelated lesions, while the term "multifocal" refers to a mono-clonal origin [ 6 ]. Clinically, it is important to determine the nature of multiple lesions. The occurrence of multiple lesions can be indicative of a familial cancer, especially if they occur at a relatively young age. Examples are familial adenomatous poliposis (FAP) in the colon, and familial retinoblastoma [ 7 ]. The genetic predisposition of such individuals renders multiple independent carcinogenic events likely. Alternatively, multiple independent lesions can be the result of a large area of tissue which has been altered and is prone to the development of cancer, such as Barrett's esophagus [ 8 ], or by other mechanisms which are not yet under-stood. On the other hand, genetic analysis has indicated that multiple lesions in several cases have a monoclonal origin [ 8 - 18 ]. Examples are mammary carcinoma, gliomas, renal cell carcinoma, hepatocellular carcinoma, and esophageal adenocarcinoma. In this paper we focus on multiple lesions with a monoclonal origin. We will refer to them as "multifocal" cancers. The mechanism by which such multifocal cancers are generated, and their relation to the stage and metastatic potential of the cancer, are not fully understood [ 19 ]. Yet, this understanding is important for decisions regarding treatment and surgery. Here, we report that multifocal cancers can be generated through the dynamical interplay between tumor promoting and inhibiting factors. Mathematical modeling indicates that somatic evolution away from tumor inhibition and towards tumor promotion results in the transition from a small contained tumor, to multi-focal tumors, and finally to a large tumor mass within a tissue. Multifocal tumors therefore represent an intermediate stage in tumor progression. Several studies have identified tumor promoting and inhibiting factors, produced either by the tumor cells themselves, or by surrounding tissue cells. An obvious example is angiogenesis inhibition and promotion, where simple mutations can change the balance away from inhibition and in favor of promotion [ 20 , 21 ]. Other inhibiting factors which are not related to angiogenesis have also been observed, although their exact identity and function remain unknown [ 22 ]. 2 Results We start with a simple model which describes tumor growth in relation to the production of promoters and inhibitors. We then extend this model to describe the local spread of cancer cells across space (tissue), and examine somatic evolution of cells away from tumor inhibition and towards promotion. The basic model We consider a basic mathematical model which describes the growth of a cancer cell population, assuming that the amount of blood supply influences the rate of cell division. The model includes three variables: the population of cancer cells, C ; promoters, P ; and inhibitors, I . It is assumed that both promoters and inhibitors can be produced by cancer cells. In addition, inhibitors may be produced by healthy tissue. The model is given by the following set of differential equations which describe cancer growth as a function of time, The equations are based on a previous study [ 23 ]. The population of cancer cells grows with a rate r . Growth is assumed to be density dependent and saturates if the population of cancer cells becomes large (expressed in the parameter ε ). In addition, the growth rate of the cancer cells depends on the balance between promoters and inhibitors, expressed as P /( I + 1). The higher the level of promoters relative to inhibitors, the faster the growth rate of the cancer cell population. If the level of promoters is zero, or the balance between promoters and inhibitors in heavily in favor of inhibitors, the cancer cells cannot grow and remain dormant [ 24 - 26 ]. Cancer cells are assumed to die at a rate δ . Promoters are produced by cancer cells at a rate a p and decay at a rate b p . Inhibitors are produced by cancer cells at a rate a I and decay at a rate b I . In addition, the model allows for production of inhibitors by normal tissue at a rate ξ . Insights from the model The analysis of the model above is presented in detail in the Materials and Methods section. It suggests the following patterns. There are two outcomes. ( i ) The cancer cells cannot grow and consequently go extinct.That is, C = 0, P = 0 and I = 0. The cancer goes extinct in the model because we only consider cells which require the presence of promoters for division. If the level of promoters is not sufficient, the rate of cell death is larger than the rate of cell division. In reality, however, it is possible that a small population of non-angiogenic tumor cells survives. Here, we omit this for simplicity. ( ii ) The population of cancer cells grows to significant levels, that is, C = . How do the parameter values influence the outcome of cancer growth? The cancer extinction outcome is always stable. The reason is as follows. The cancer cells require promoters to grow. The promoters, however, are produced by the cancer cells themselves. If we start with a relatively low initial number of cancer cells, this small population cannot produce enough promoters to overcome the presence of inhibitors. Consequently, the cancer fails to grow and goes extinct. This outcome is always a possibility, regardless of the parameter values. Significant cancer growth can be observed if the intrinsic growth rate, r , lies above a threshold relative to the death rate of the cells, δ , and degree of tumor cell inhibition ( a p and b p relative to a I and b I , i.e. the production and decay rates of promoters and inhibitors, respectively). The exact condition is given by (9). In this case, the outcome is either failure of cancer growth, or successful growth to large numbers. Which outcome is achieved depends on the initial conditions. Successful growth is only observed if the initial number of cancer cells lies above a threshold. Then, enough promoters are initially produced to overcome inhibition. This provides an important barrier to the successful growth of cancers. It could explain why it is difficult for cancers to escape angiogenesis inhibition, and why autopsies often reveal the existence of multiple small, non-pathogenic tumors which have failed to progress [ 27 ]. Modeling the spread of tumors across space In this section, we introduce space into the above described model. We consider a one-dimensional space along which tumor cells can migrate. The model is formulated as a set of partial differential equations and is written as follows, The model assumes that tumor cells can migrate, and this is described by the diffusion coefficient D c . Inhibitors can also diffuse across space, and this is described by the diffusion coefficient D I . It is generally thought that inhibitors act over a longer range, while promoters act locally [ 21 , 24 ]. Therefore, we make the extreme assumption that promoters do not diffuse. Again, we ignore for simplicity the production of inhibitors by healthy tissue, ξ . As before, numerical simulations indicate that results are not changed qualitatively by this simplification. As mentioned above, the model considers tumor spread across space. It is important to point out that we do not consider long-range metastatic spread. Instead, we consider local spread of a tumor within a tissue, such as the breast, liver, brain, or esophagus. Here we investigate the process of tumor growth and progression in relation to the degree of inhibition and promotion. A mathematical analysis is presented in the Materials and Methods section. Here we present biological insights and results of numerical simulations. Insights from the spatial model We start with a scenario where the degree of inhibition is much larger than the degree of promotion ( a I / b I >> a p / b p ). This corresponds to the early stages when the tumor is generated. We then investigate how tumor growth changes as the degree of inhibition is reduced relative to the level of promotion (i.e. the value of a I / b I is reduced). We consider the following parameter regions (Figure 1 ). Figure 1 Outcome of the spatial model depending on the relative balance of promoters and inhibitors, captured in the variable a i . Parameters were chosen as follows: r = 1; δ = 0.1; a P = 5; b P = 0.1; b I = 0.01; D C = 0.00001; D I = 0.001; L = 2. For (a) a I = 3, (b), a I = 2, (c), a I = 1, (d) a I = 0.1. 1. If the degree of inhibition is strong and lies above a threshold, growth of the cancer cells to higher levels does not occur (not shown). Only a small number of cells which do not require promotion for survival would remain. 2. If the degree of inhibition is weaker, the cancer cells can grow. The spread across space is, however, self-limited (Figure 1a ). The cancer cells migrate across space. The inhibitors produced by the cancer cells also spread across space, while the promoters do not. Therefore, as the cancer cells migrate, they enter regions of the tissue where the balance of inhibitors to promoters is heavily in favor of inhibitors. Consequently, these cells cannot grow within the space. They remain dormant and may eventually die. In biological terms, this corresponds to a single coherent but self-limited lesion ( uni-focal ). Note that this does not mean that it is in principle impossible to generate more lesions. It means that the space between lesions is bigger than the space provided for cancer growth within the tissue. 3. As the production of inhibitors is further reduced, we enter another parameter region. Now fewer inhibitors diffuse across space. We observe that multiple lesions or foci are formed (Figure 1b ). They are separated by tissue space which does not contain any tumor cells. The separate lesions produce some inhibitors, and they diffuse across space. This explains the absence of tumor cells between lesions. Because the production of inhibitors is weakened, however, tumor growth is only inhibited in a certain area around the lesion, and not across the whole space. How many lesions are found within a tissue depends on the parameters in the model, in particular on the relative strength of inhibition and promotion (Figure 1b and 1c ). The stronger the degree of inhibition, the larger the space between lesions, and the fewer lesions we expect. The weaker the degree of inhibition, the smaller the space between lesions, and the larger the expected number of lesions. In biological terms, this corresponds to the occurrence of multi-focal cancers. 4. If the degree of inhibition is further reduced and lies below a threshold, spread of inhibitors is sufficiently diminished such that the tumor cells can invade the entire space and tissue (Figure 1d ). In biological terms, this corresponds to the most extensive tumor growth possible within a tissue. In summary, as the relative degree of inhibition is reduced, the patterns of tumor growth change from absence of significant growth, to a single self-limited tumor, to the occurrence of multiple foci, and to the maximal invasion of the tissue by tumor cells. Multi-focal cancers may arise through the dynamical interplay between long range inhibition and local promotion. The following section will examine this in the light of somatic evolution. 3 Discussion We have shown how the pattern of cancer growth can depend on the relative balance of promoters and inhibitors. Here we consider these results in the context of somatic evolution, and suggest some clinical implications. Somatic cancer evolution and progression At early stages of cancer progression, the balance between inhibitors and promoters is in favor of inhibition. Inhibitors are likely to be produced by healthy cells (e.g. in the context of angiogenesis), and they are more abundant than an initiating population of transformed cells. In the context of angiogenesis, specific mutations have been shown to result in the enhanced production of promoters or reduced production of inhibitors in cancer cells. Our model has shown that such mutants have to be produced at a relatively high frequency, so that a sufficient number of promoting cells are present in order to ensure that enough promoters are produced to overcome the effect of inhibition. Once the promoting cells have succeeded to expand, cancer progression can occur in a variety of ways according to the model. How the cancer progresses depends on how much the balance between promotion and inhibition has been shifted in favor of promotion. We distinguish between three possibilities (Figures 2 , 3 & 4 ). Figure 2 Tumor progression if the initial mutant cell line has only shifted the balance between promoters and inhibitors slightly in favor of promotion. This cell line can only give rise to self limited growth. Further tumor growth requires the generation of further mutants. The new mutant in the simulation is depicted by the dashed line. Parameters were chosen as follows: r = 1; δ = 0.1; a P = 5; b P = 0.1; a I = 3; b I = 0.01; D c = 0.00001; D I = 0.001; L = 2. For mutant: a I = 0.5; a P = 20. Figure 3 Tumor progression if the initial mutant cell line has shifted the balance between promoters and inhibitors more substantially towards promotion. Now, multiple foci can develop without the need for further mutations. The multiple foci develop, however, by first generating a single lesion which subsequently splits to give rise to two lesions during the natural growth process. Parameters were chosen as follows: r = 1; δ = 0.1; a P = 5; b P = 0.1; a I = 1; b I = 0.01; D C = 0.00001; D I = 0.001; L = 2. Figure 4 Tumor progression if the initial cell line has largely escaped inhibition, and promotion is the dominant force. Now the tumor grows in space as a single lesion until the whole tissue is invaded. Parameters were chosen as follows; r = 1; δ = 0.1; a P = 5; b P = 0.1; a I = 0.1; b I = 0.01; D C = 0.00001; D I = 0.001; L = 2. (i) The balance between inhibition and promotion has been shifted only slightly in favor of promotion, such that self-limited growth of the cancer is observed (Figure 2 ). That is, we observe a single lesion which can grow to a certain size but which is limited in the spread through the tissue. In order to progress further towards the occurrence of multiple lesions or towards more extensive invasion of the tissue, further mutants have to be generated which are characterized by enhanced production of promoters or by reduced production of inhibitors. This introduces a new problem: such a mutant will not have a selective advantage, but is selectively neutral relative to the other cells. This is because the promoters and inhibitors secreted from one cell affect the whole population of cells. If the mutant produces more promoters, not only the mutant, but the entire population of tumor cells benefits. This means that a mutant characterized by enhanced production of promoters will not invade the tumor cell population. Instead, we observe genetic drift which is stochastic and not described by the equations considered here. The model does, however, suggest the following (Figure 2 ): if the population of mutant cells remains below a given threshold relative to the rest of the tumor cells, it will not alter the growth pattern. If the population of mutant cells grows beyond a threshold relative to the rest of the tumor cells, it can change the pattern of cancer growth, even if the mutants do not become fixed in the population (Figure 2 ). The change can either be the generation of multiple lesions, or invasion of the whole tissue, depending on the amount by which the level of promotion has been enhanced by the mutant cell population. The chances that the mutant cell population drifts to levels high enough to cause such a change in tumor growth depend on the population size of the lesion. The larger the number of tumor cells, the lower the chance that the relative population size of the mutants can cross this threshold. If this cannot occur, further cancer progression not only requires the generation of a mutation which enhances the level of promotion, but an additional mutation which gives the promoter mutant a selective advantage over the rest of the cell population. That is, in addition to the mutation which shifts the balance in favor of promotion, a mutation is required either in an oncogene or a tumor suppressor gene so that the mutant can grow to sufficiently high numbers or fixation. (ii) The first mutation shifts the balance between promoters and inhibitors to a lager extent which is sufficient to result in the generation of multiple lesions (Figure 3 ). The multiple lesions do not, however, occur immediately. First, the tumor grows as a single and self limited lesion (Figure 3 ). Over time, this lesion bifurcates to give rise to two lesions, or further lesions if the degree of promotion is large enough relative to the degree of inhibition (Figure 3 ). The temporal sequence from a single and self-controlled lesion to the occurrence of multiple lesions is the same as in the previous case. But in contrast to the previous case, no further mutations are required. This is because multiple foci arise from the split and migration of a single lesion. The number of foci that form depends on the exact degree of promotion which was achieved by the initial mutation. The higher the degree of promotion, the larger the number of lesions. Growth beyond this number of lesions (which will eventually result in maximal invasion) then requires higher levels of promotion. This is in turn achieved by further mutational events according to the same principles as described in the previous section. (iii) Finally, assume that the initial mutation shifts the balance so much in favor of promotion that maximal invasion of the tissue is possible (Figure 4 ). Now we observe cancer progression without the generation of multiple foci. Instead, a relatively small single lesion expands in space until all the tissue has been invaded. In summary, the model predicts different modes of cancer progression in relation to the evolution away from tumor inhibition and towards promotion. A single cancer lesion may spread across the tissue without the occurrence of multiple lesions. Alternatively, the cancer can first grow as a single, self-contained lesion. This can then bifurcate to give rise to multiple foci, either as a result of additional mutations, or as a result of the natural pathway by which multiple foci are generated, depending on the degree of tumor promotion conferred by the initial mutation. Further evolutionary events can then induce the multiple foci to become a single, maximally invasive mass. The occurrence of multiple foci therefore represents an intermediate stage in tumor progression towards malignancy. Clinical implications The models discussed here show that multiple foci with a monoclonal origin can develop through a dynamical interplay between tumor promoters and inhibitors. The cancer can only grow to high loads as a single mass if it has largely escaped all inhibitory effects. Otherwise, the cancer is likely to grow via the generation of a relatively small and self limited tumor which then bifurcates into multiple foci until it finally invades the entire tissue. The occurrence of multiple foci is therefore an intermediate stage in cancer progression. The higher the number of foci, the further advanced the stage of cancer progression. A clinically important step in carcinogenesis is the process of metastasis. That is, the spread of tumor cells to the lymph node, entry into the blood supply, and the spread to other tissues. Various studies have investigated the metastatic potential of multi-focal compared to uni-focal cancers [ 19 , 28 , 29 ]. In uni-focal cancers, tumor size has been found to be a predictor of metastatic potential. For staging multi-focal breast carcinomas, it has been suggested to use the diameter of the largest tumor only [ 19 ]. This, however, assumes that the other foci do not significantly contribute to tumor progression. According to our arguments, this would under-stage the cancer. According to the model, the number of foci correlates with the stage of the disease. This has also been concluded in clinical studies, and is supported by data which show reduced patient survival with multi-focal compared to uni-focal cancers [ 19 ]. Moreover, because our model suggests that multi-focality can occur as a result of reduced tumor cell inhibition, successful metastatic growth might be easier to achieve. Although under debate, some data suggest that inhibitors produced by the primary tumor can prevent metastatic cells from growing [ 24 ]. If multi-focality correlates with reduced inhibition, then it could also correlate with an increased chance that metastatic cells grow and do not remain dormant. Further, it is important to note that studies which aim to assess the correlation between multi-focality and metastatic potential should not only concentrate on the number of foci, but also on the size of the foci. As we have shown with the model, cancer progression might start with a small single lesion which can be considered uni-focal. It can then bifurcate to give rise to multiple foci, and finally spread through the entire tissue. When such spread occurs, the multiple foci turn into a big and single mass, and this would again be considered uni-focal. Hence, the cumulative size or volume of the tumor is likely to be the best predictor of malignant progression. 4 Conclusions In conclusion, we suggest that the balance between tumor promoting and inhibiting factors might be an important driving force which determines the pattern on cancer progression, and can account for the occurrence of multi-focal cancers. The best worked out example of such promoter-inhibitor dynamics is angiogenesis. In this context, inhibitors are produced both by healthy tissue cells and by tumor cells. During the course of progression, tumor cells can mutate and evolve to produce less inhibitors and more promoters. The initial establishment of an angiogenic cell line is the most difficult step. Since the promoting factors are produced by angiogenic cells themselves, their initial abundance has to be sufficiently high, such that the balance can be shifted away from inhibition. This enables the population of cancer cells to expand beyond a very small size. This growth can then give rise to a self-limited uni-focal cancer which can bifurcate to give rise to multi-focal cancers. Further evolutionary events can finally lead to maximal tissue invasion. If the initial mutation allows the cells to sufficiently escape from inhibition, cancer progression can occur as a single expanding mass without the occurrence of multi-focality. These arguments not only apply to angiogenesis, but to any tumor promoting and inhibiting factors where inhibitors act over a long range while promoters act locally. Therefore, the therapeutic use of inhibitors should be further explored. This is an active area of research in the context of angiogenesis [ 30 ], and the identification of possible alternative inhibitors might open new avenues of investigation in this context. 5 Materials and Methods Here we present mathematical methodology used to analyze the equations described in the text. Linear stability analysis of the ODEs Here we discuss a linear stability analysis of system (1–3). Let us first simplify the problem by using a quasistationary approach, that is, we will assume that the level of promoters adjusts instantaneously to its steady-state value ( P = Ca P / b P ). It is convenient to denote Now we have a two-dimensional system, For simplicity we ignore the constant input term, ξ , which describes the production of inhibitors by healthy tissue. Numerical simulations have shown that results are not altered qualitatively by this simplification. There can be up to three fixed points in this system, where , and It is obvious that if γ + ε - W < 0, and ( γ + ε - W ) 2 - 4 ε γ > 0, then there are exactly three positive equilibria in the system. If either of these conditions is violated, the (0,0) solution is the only (biologically meaningful) stable point. Stability analysis can be performed by the usual methods. For the (0,0) equilibrium, the Jacobian is that is, this equilibrium is always stable. For the points ( C ± , I ± ), we get the following Jacobian, where we denote for convenience, . It is easy to show that the eigenvalues of this matrix for the solution ( ) are given by and for the solution ( ) we have eigenvalues where Y ± ≡ 2 b I W + δ ( ε - γ - W ± Γ). We can see that solution ( ) is always unstable and we will not consider it any longer. Solution ( ), which we call for simplicity ( ) from now on, is stable as long as Y + > 0     (9) Turing stability analysis Here we present a linear analysis of system (4–6). As before, we are going to assume that promoters adjust instantaneously to their equilibrium level. By replacing P with C defined by , we can rewrite equation (4) as This equation together with equation (6) gives a Turing model. Let us go back to the system of ODEs, (7–8), and assume that solution ( ) is a stable equilibrium. Of course, this solution also satisfies the system of PDEs, (10,6). Let us consider a wave-like deviation from this spatially uniform solution: Here, the amplitudes of the perturbation, A and B , are small compared to the amplitude of the spatially uniform solution, and we assume an infinitely large space. The equation for the new eigenvalue, λ is where we define Equation (11) can be written as λ 2 + λ ( b I - α + ( D C + D I ) ω 2 ) + a I β - ( b I + D I ω 2 )( α - D C ω 2 ) = 0.     (12) This is the dispersion relation which connects the growth-rate, λ , with the spatial frequency of the perturbation, ω . The stability conditions now are given by b I - α + ( D C + D I ) ω 2 > 0,     (13) a I β - ( b I + D I ω 2 )( α - D C ω 2 ) > 0.     (14) Note that the stability conditions for solution ( ) of the system of ODEs, (7–8), are obtained automatically from the conditions above by setting ω = 0: b I - α > 0,     (15) a I β - b I α > 0.     (16) Inequality (13) is always satisfied because of inequality (15). Let us derive conditions under which the spatially uniform solution is unstable. This requires that condition (14) is reversed. This can be expressed as follows: F ( ω ) ≡ D I D C ω 4 - ω 2 γ 1 + γ 2 < 0.     (17) where we denoted for simplicity, γ 1 = α D I - b I D C , γ 2 = α I β - α b I > 0. This is a fourth order polynomial, symmetrical with respect to the line ω = 0, with a positive leading term. The points, ±| ω |, satisfying correspond to the two minima of the left hand side of inequality (17). Let us call these values of ω , ± ω c . The condition F ( ω c ) < 0 defines that the uniform solution ( ) is unstable. Let us plot the function F ( ω ) for different values of a I , see Figure 5 . For small values of a I , F ( ω ) is strictly positive, and the spatially uniform solution is stable. As a I increases, the function F ( ω ) crosses the line F = 0. The critical value of a I , a I,c , for which F ( ω c ) = 0, is determined from Figure 5 Emergence of Turing instability. As a I increases through its critical value, the function F ( ω ) (equation (17)) crosses zero. Negative regions of F ( ω ) correspond to unstable wave-numbers. The wave-number which becomes unstable first is denoted by ω c . The parameters are as follows: r = 1; δ = 0.1; a P = 5; b P = 0.1; b I = 0.01; D C = 0.00001; D I = 0.001. ( α D I - b I D C ) 2 = 4 D I D C ( a I β - α b I ), where α and β both depend on a I . We solved this equation numerically to find the critical value of a I , c , see Figure 5 . The applicability of the above analysis depends on the parameters of the system. First of all, we need conditions (15–16) to be satisfied. They mean that without diffusion, a positive, spatially uniform solution is stable. Next, we need to be in a weakly nonlinear regime , where the function F ( ω ) has only very narrow regions of ω corresponding to negative values. More precisely, Δ ω ~ L -1 , where L is the spatial dimension of the system. In terms of parameter a I , we require that it is sufficiently close to a I , c . Then, we can calculate the "most unstable" wavenumber, that is, ω c defined by equation (18), with a I , c . This value will determine the spatial period of the solution, Stationary periodic solutions In numerical simulations described in this paper, we used the following (Neumann) boundary conditions: The simulation results are presented in the main body of the paper. Here we discuss the behavior of the system in the light of the analysis presented above. Let us assume that the value a I is below the critical, a I < a I , c . The system exhibits bistability. If we start in the vicinity of a (0,0) solution, then cancer will not grow and decay to zero. If we start from a point ( C , I ) in the domain of attraction of the solution ( ), then the system will develop towards this positive spatially homogeneous stationary solution. Next, let us suppose we have a I > a I , c , but make sure that it is sufficiently close to a I , c (the exact meaning of "close" is specified in the analysis above). Again, if the initial conditions are close to the zero solution, then the zero state will be the state that the system will attain. However, if we start in the vicinity of the ( ) state, we will observe interesting behavior. Solution ( ) is now unstable, and we will see "ripples" developing on top of this solution. This is Turing instability. The spatial period of the ripple was calculated in the previous section. Long-time evolution of this state is of course not in the realm of linear stability analysis, but we can predict that the spatial scale of the resulting solution will be given by (19). Finally, let us assume that a I is much higher than critical. Now, solution ( ) is unstable even in the system of ODEs. However, a periodic solution will develop, unless the initial condition is in the domain of attraction of the zero solution. The spatial scale of the periodic solution is determined intrinsically by the parameters of the system, and it grows with a I . Intuitively this is easy to understand, because higher values of a I correspond to higher levels of inhibition, so the distance between regions of large C will become larger. Note that the exact period of the periodic solution is adjusted to fit the boundary conditions of the system. For instance, with the Neumann boundary conditions, the boundary points are forced to be troughs of the wave-like pattern. In other words, the period of the solution must be an integer fraction of L .

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526757.xml

555949

ISSR markers show differentiation among Italian populations of Asparagus acutifolius L

Background Asparagus acutifolius L. is a dioecious and native plant species, widely distributed in the Mediterranean Basin. It is known for its fine flavour and could represent an important resource for cultivation programs in desert areas. Few molecular studies have been performed on this species. In the present paper, the ISSR technique was employed to study genetic diversity in Italian A. acutifolius . Results Twenty-three primers produced a total of 228 polymorphic fragments used to evaluate genetic variation. F ST (0.4561) and Theta B (0.4776) values indicate a wide genetic variation among the samples examined. The distance UPGMA tree grouped together the genotypes strictly according to their geographical origin, showing that each sample is genetically structured and can be considered a distinct population. AMOVA analysis further confirmed genetic structuring of the populations. Population-specific fragments were also detected. Conclusion The results suggest that ISSR markers are useful in distinguishing the populations of A. acutifolius according to geographical origin, and confirm the importance of genetic studies for designing germplasm conservation strategies.

Background The availability of a variety of DNA markers, such as restriction fragment length polymorphism (RFLP), amplified fragment length polymorphism (AFLP), random amplified polymorphic DNA (RAPD), simple sequence repeat (SSR) and intersimple sequence repeat (ISSR) has enabled researchers to investigate genetic diversity among various plant species across natural populations [ 1 - 5 ]. Among these, PCR-based techniques of random multilocus analysis (RAPD, AFLP, ISSR) have been successfully used in genotyping, genome mapping and phylogenetic studies in horticultural crops such as strawberry [ 6 ], soybean [ 7 ], and potato [ 8 ]. Local populations of traditional cultivars provide a valuable resource for plant breeding as well as for the preservation of genetic diversity [ 9 ]. The exploration, evaluation, and conservation in situ and ex situ of genetic diversity in natural populations is imperative to guarantee sustainable development [ 10 ]. Asparagus acutifolius L. (Liliaceae) is a native, perennial plant species widely distributed throughout the Mediterranean areas, whose flowers are classified as dioecious and are mainly bee-pollinated; it generally does not reproduce by self-pollination. It grows in bushy and semi-dry places, sunny or semi-shade, mainly on limestone. This species is known for its strong taste compared to the cultivated A. officinalis and does not require rich soils for cultivation; for these reasons, it could be an economically important resource for the recovery of arid rural areas where controlled introduced programs could be achieved. To date, there is little information available on the genetic variability of this species. At present, the most widely studied species is A. officinalis , for which many molecular markers have been characterized (RAPD, RFLP, AFLP) [ 11 ]. The few molecular data regarding A. acutifolius are drawn from RAPD analyses [ 12 ] and the identification of microsatellite loci [ 13 ]. The ISSR technique is similar to that for RAPD, except that ISSR primers consist of a di- or trinucleotide simple sequence repeat with a 5' or 3' anchoring sequence of 1–3 nucleotides. Compared with RAPD primers, the ISSR primers sequence is usually larger, allowing for a higher primer annealing temperature which results in greater band reproducibiliy than RAPD markers [ 14 ]. They have been successfully used to assess genetic variation in plants such as citrus [ 15 ], Viola pubescens [ 14 ], potato [ 8 ], and Oryza [ 16 ]. In this study we used ISSR markers to analyse the genetic diversity of Italian A. acutifolius collecting samples in eight different scattered rural areas: six continental and one each from the Italian islands of Sardinia and Sicily. Results Figure 1 and Table 1 show the eight different Italian sites where A. acutifolius was collected and their characteristics. Figure 1 Collection sites of A. acutifolius . Table 1 Populations of A. acutifolius . Site Abbr Characteristics N Hab Alt Exp Ped Assn Caserta Vecchia (Caserta) CAS Clay hill 400 m South, sunny Acidic ground Rubus 15 Colli al Volturno (Isernia) ISE Limestone hill 380 m Semi-shadow Alkaline ground Quercus, Genista 16 Lotrine (Livorno) LIV Hedges 100 m South-west, sunny Alkaline ground Quercus, Rubus 14 Recco (Genova) REC Torrent levee 5–10 m West, shadow Not recorded Rubus 6 Sassari (Sardinia) SAS Coast 10–15 m North, shadow Neutral ground Pinus 18 Spoleto (Perugia) SPO Cultivated fields 600 m South- west, sunny Alkaline ground Quercus, Rubus, Genista 14 Vignacastrisi (Lecce) LEC Dry walls 50–100 m North-west, shadow Neutral ground Quercus, Rubus 11 Vizzini (Catania, Sicily) CAT Ground cover 500 m North, shadow Neutral ground Quercus, Rubus 15 Abbr, abbreviation; N, number of individuals; Hab, habitat; Alt, altitude; Exp, exposure; Ped, pedology; Assn, ecological association. Among the 42 primers tested, 23 proved useful to characterize the samples (Table 2 ), whereas 19 were excluded due to absence of amplification (9 primers) or to amplification of the same single fragment in all samples (10 primers). The 23 useful primers gave a total of 228 polymorphic fragments, ranging from 150 to 1100 bp, with 100% repeatability. Fragments of the same molecular weight were considered as the same locus [ 17 ]. The validity of ISSRs in assessing genetic variability in the eight samples of Italian A. acutifolius is summarized in Table 3 . Table 2 ISSR primers useful for the amplification of the eight populations of A. acutifolius . Primer Name Sequence 5'-3' AT (°C) 3 (CA) 8 AT 50 4 (CA) 8 AC 51.7 5 (CA) 8 GT 51.7 8 (CA) 8 GAC 54.7 15 GGTC(AC) 7 56 16 CGTC(AC) 7 56 17 CAGC(AC) 7 56 21 CAGC(TC) 7 56 23 GAG(TC) 8 56 8082 (CT) 9 G 57 8564 (CAC) 7 T 58 8565 GT(CAC) 7 58 BEC (CA) 7 YC 54 CHR (CA) 7 YG 54 DAT (GA) 7 RC 54 HAD CT(CCT) 3 CRC 54 MAN CA(CCA) 3 CRC 54 OH (GAG) 7 RG 66.7 TE GT(GGT) 3 GRC 54 W7 (CT) 8 RG 52.8 W814 (CT) 8 TG 52 W844 (CT) 8 RC 52.8 W902 (GT) 6 AY 39 AT, annealing temperature. Table 3 Genetic variability among the populations of A. acutifolius . a) CAS CAT ISE LEC LIV REC SAS SPO 3 0.1554 0.2111 0.1018 0.1811 0.2157 0.0428 0.3172 0.0727 4 0.0599 0.3073 0.1664 0.1685 0.0000 0.0000 0.2753 0.1102 5 0.0581 0.1603 0.0823 0.1032 0.1059 0.1502 0.1037 0.0518 8 0.0698 0.1182 0.0667 0.1209 0.0621 0.0000 0.0563 0.1454 15 0.2086 0.2980 0.1330 0.2445 0.2612 0.1695 0.1734 0.1865 16 0.1661 0.1646 0.2129 0.1870 0.0470 0.1616 0.1820 0.1560 17 0.0846 0.1034 0.2370 0.0938 0.1660 0.2356 0.1616 0.1346 21 0.2490 0.2258 0.1345 0.1251 0.2058 0.1145 0.1185 0.1551 23 0.0711 0.1478 0.2607 0.0930 0.2663 0.0749 0.2746 0.2650 8082 0.1558 0.1163 0.1106 0.0127 0.0672 0.1474 0.1398 0.1395 8564 0.2033 0.0000 0.1453 0.2546 0.1567 0.0145 0.0825 0.1299 8565 0.2316 0.0924 0.1752 0.0783 0.1579 0.0856 0.1337 0.1001 BEC 0.0907 0.0860 0.1107 0.0475 0.1847 0.1471 0.1855 0.1250 CHR 0.2614 0.1785 0.0198 0.1980 0.0156 0.0530 0.0240 0.1161 DAT 0.1202 0.0676 0.1308 0.1706 0.1434 0.1699 0.1472 0.0905 HAD 0.1248 0.2192 0.1576 0.1725 0.1492 0.2171 0.2668 0.0529 MAN 0.2177 0.1119 0.2266 0.2135 0.0584 0.1394 0.0617 0.1455 OH 0.2527 0.2122 0.2099 0.2512 0.2150 0.1953 0.2615 0.1010 TE 0.0233 0.0559 0.0555 0.2077 0.0923 0.0525 0.1646 0.1333 W7 0.1613 0.1383 0.1502 0.1638 0.0738 0.1324 0.1987 0.1577 W814 0.1575 0.3503 0.2574 0.2106 0.0000 0.0000 0.3189 0.2179 W844 0.1063 0.2670 0.0962 0.0000 0.0000 0.2641 0.1253 0.0687 W902 0.2554 0.2413 0.2115 0.3393 0.1764 0.1636 0.1896 0.0895 Mean ± SD 0.1512 ± 0.1887 0.1539 ± 0.1940 0.1489 ± 0.1875 0.1625 ± 0.1964 0.1173 ± 0.1744 0.1180 ± 0.1771 0.1618 ± 0.1903 0.1259 ± 0.1842 P 50.44% 45.61% 46.49% 46.93% 39.91% 35.09% 51.75% 39.91% Number of specific bands 3 M 9 P 4 P 4 P 2 P 2 P 3 P 1 M 3 P b) H T H S D ST F ST Theta-B Mean ± SD 0.2618 ± 0.0240 0.2859 ± 0.0155 0.1424 ± 0.0084 0.1619 ± 0.0026 0.1194 0.4561 0.4766 ± 0.0173 a) Gene diversity for each primer set and population and over-all populations, and percentage of polymorphic loci (P) per population; the first column indicates the primer name; the last row indicates the number of population specific fragments (M = monomorphic, P = polymorphic);b) Mean values ± SD of total heterozygosity (H T ), intrapopulation heterozygosity (H S ) (Left column, POPGENE result; right column, Hickory result), diversity among population (D ST ), fixation index (F ST ), Theta-B. A high level of genetic variation was observed using ISSR markers, with 100% polymorphic loci at the species level. The highest number of polymorphic loci (51.75%) was exhibited in the Sassari and the lowest (35.09%) in the Recco samples. Genetic structuring was evident due to the detection of specific bands in each sample examined. Spoleto and Caserta samples showed one and three fixed specific fragments, respectively, found to be statistically significant (P < 0.0001). For the other samples, 27 ISSR specific polymorphic fragments were detected, with a varying degree of statistical significance ranging from P < 0.0400 to P < 0.0001. Genetic distances [ 18 ] were examined for all pairwise comparisons between the sub-populations (Table 4 ). The mean distance for all comparisons was 0.1680, ranging from 0.0916 (between Isernia and Lecce) and 0.2865 (between Recco and Caserta). The Mantel test showed no correlation between the genetic and geographic data (-0.220). Table 4 Genetic (below diagonal) and geographic (above diagonal) distances among the eight populations of A. acutifolius . LIV SPO LEC ISE CAT SAS REC CAS LIV *** 1.89 7.39 3.87 7.84 3.46 1.46 4.30 SPO 0.1354 *** 5.53 2.12 6.53 4.19 3.27 2.65 LEC 0.1687 0.1134 *** 3.58 4.17 8.13 8.80 3.33 ISE 0.1498 0.1257 0.0916 *** 4.62 4.84 5.31 0.60 CAT 0.1180 0.1288 0.1321 0.1326 *** 6.67 9.25 4.02 SAS 0.1265 0.1528 0.1404 0.1255 0.1096 *** 4.12 4.86 REC 0.1388 0.1702 0.1821 0.1759 0.1267 0.1257 *** 5.76 CAS 0.2591 0.2751 0.2591 0.2573 0.2421 0.2549 0.2865 *** Samples collected at different geographic site grouped together, as shown in the UPGMA tree (Fig. 2 ), and the AMOVA analysis revealed significant genetic structuring (p = 0.001). Figure 2 UPGMA tree of the 109 A. acutifolius samples used in the ISSR analysis . The numbers indicate the bootstrap values. The values of gene diversity are summarized in Table 3a . For some primers, the value was 0.0000, and the highest value (0.3503) was found for the primer W814 in the Catania sample. This explains the high standard deviation values observed. As summarized in Table 3b , the total variation (H T ) was 0.2618 ± 0.0240 and the average variation within samples (average H S ) was 0.1424 ± 0.0084. The mean diversity among the samples (D ST ) was 0.1194. The fixation index F ST = (H T -H S )/H T was 0.4561, indicating a reduction of genetic diversity of about 45%. The Theta-B value obtained by Hickory analysis is an estimate of F ST under a random-effects model of population sampling. Its mean value is 0.4766 ± 0.0173; the H T and H S values are, respectively, 0.2859 ± 0.0115 and 0.1619 ± 0.0026 showing that there is a general agreement between the results obtained using the two different approaches. Discussion ISSR markers can be used in population genetic studies of plant species as they effectively detect very low levels of genetic variation [ 19 ]. They also may have potential for analysing biogeographic patterns among populations of a single plant species. In this study, we have shown that these markers revealed genetic variation among geographically separated samples of A. acutifolius in an Italian population. ISSRs also revealed diversity within each sub-population. The results obtained are in accordance with the principle that the number of individuals used to estimate average heterozygosity can be very small if a large number of loci is studied [ 18 ]. The gene diversity values (Table 3a ) ranged from 0.0525 (TE-Recco) to 0.3503 (W814-Catania). As expected, some primer-sample combinations showed no diversity (0.0) for two reasons: i) the combination primer-sample produced the same amplification pattern in all the samples (primer 4, Livorno and Recco; primer 8, Recco; primer W814, Livorno and Recco); ii) the combination primer-sample produced no measurable fragments (primer 8564, Catania; primer W844, Lecce and Livorno). These primers were not excluded from the analysis because in some cases they produced sample-specific fragments (e.g.: primer 8564 in Caserta and Lecce; W844 in Catania and Recco; primer 8 in Caserta, Catania, Isernia, Sassari, and Spoleto). The fixation index is 0.4561, which indicates a substantial reduction of genetic diversity (about 45%), probably due to the high genetic isolation of samples analysed. The F ST and the Theta-B value (0.4766) demonstrated a very great genetic differentiation among samples, possibly caused by random genetic drift. Further statistical support to the genetic structuring of the samples examined comes from the AMOVA analysis. Despite the continuous distribution of A. acutifolius , the eight samples represent genetically differentiated populations. In each population it was possible to identify ISSR specific fragments. As shown in Table 3a , the Caserta and Spoleto populations had specific fixed fragments that distinguished them from all others. Although the Isernia specific fragments are not statistically significant (P > 0.05), population-specific fragments were detected for all the other populations, with varying levels of intra-population polymorphism, ranging from 0.147 to 0.764. Thus, we have identified reproducible markers that distinguish the geographical origin of the A. acutifolius populations. The high degree of genetic differentiation is confirmed by the UPGMA tree topology, in which all accessions from the same population grouped together (Fig. 2 ). The populations showed genetic distances ranging from 0.0916 and 0.1821 with the exception of the Caserta population that is more distant from the others (from 0.2421 with Catania to 0.2865 with Recco, Table 4 ). This result is probably due to the high number of specific ISSR fragments found in the Caserta population (12, including monomorphic and polymorphic) and can be attributed to genetic drift. In particular, the high distance between Caserta and Isernia (0.2573) is unexpected because of the short geographical distance separating the regions, and could explain the absence of correlation between genetic and geographical data matrices obtained with the Mantel test. The high genetic structuring of the eight populations shows that despite the continuous distribution of A. acutifolius throughout the Italian peninsula, there is poor gene flow through the isolates. The high genetic differentiation of the A. acutifolius populations examined might be attributed to the kind of pollinators (mainly bees) that can act at short distances, preventing the gene flow, and to the effects of anthropogenic habitat fragmentation. The results obtained using ISSR markers are in agreement with the RAPD analysis that also identified population-specific fragments in different Italian A. acutifolius populations [ 12 ]. Conclusion Information about the spatial organization of genetic variability is essential for the conservation of genetic resources [ 20 ]. Our results provide an important contribution toward confirming that A. acutifolius has well-differentiated populations, despite their morphological low variability. These results show that to maintain genetic diversity within A. acutifolius it is necessary to conserve many populations. Methods Plant materials A total of 109 samples of A. acutifolius collected in the eight different locations in Italy (listed in Table 1 and showed in Fig. 1 ) were used for the analysis. Although they are only a tiny fraction of the A. acutifolius Mediterranean distribution, they are representative of the Italian population. ISSR amplification DNA was extracted from silica gel dried cladodes following the Doyle and Doyle protocol [ 21 ]. A total of 42 ISSR primers were tested on the eight populations of A. acutifolius . The polymerase chain reaction was conducted in a 9600 Perkin Elmer Thermal Cycler using the following reaction conditions: 2–5 ng DNA, 1.5 mM MgCl 2 , 0.2 mM dNTPs, 0.6 μM primer, 1.5 UE Taq polymerase (BIOLINE) and 1X Taq DNA polymerase buffer, in a total volume of 25 μL. The amplification programme was 1.5 min at 94°C; 35 × 40 s at 94°C, 45 s at the primer annealing optimal temperature (see Table 2 ), 1.5 min at 72°C; 45 s at 94°C, 45 s at the annealing temperature, 5 min at 72°C. Following PCR, the samples were loaded onto a 1.5% agarose gel in TAE 1X buffer, stained with ethidium bromide. Additionally, 100 bp ladder (Promega) and negative and positive controls were loaded and run at constant voltage (150 V) for 2 hours. After running, the gels were UV visualised and recorded using a Kodak Digital Science dS1D DC40/DC120 Camera. To verify the repeatability of the results, each DNA extraction, PCR amplification, and gel running was repeated twice. Data analysis Unequivocally scorable and consistently reproducible amplified DNA fragments were transformed into binary character matrix (1 = presence, 0 = absence). Genetic variation within and among sub-populations was analysed on the basis of the banding profile using various parameters such as percentage polymorphism (P), total heterozygosity (H T ), heterozygosity within population (H S ), diversity among populations (D ST ), fixation index (F ST ), and genetic distance [ 18 , 22 - 24 ], using POPGENE software [ 25 ]. Since ISSR are dominant markers, data were also analysed using Hickory software [ 26 ] based on a Bayesian method that does not require prior decisions about the breeding system and Hardy-Weinberg equilibrium; the analysis was conducted under the f-free model. AMOVA analysis, implemented in Arlequin [ 27 ], was conducted to document the degree of genetic structure among sub-populations. The Mantel test of genetic and geographic distances was carried out to evaluate the correlation between the two data matrices. The UPGMA tree was generated using the PAUP*4.0 software [ 28 ], and Bootstrap analysis was conducted using 1000 replicates. Authors' contributions MS carried out part of the sample collection, designed ISSR primers and carried out ISSR work; GG did part of DNA extraction; SM carried out part of the sample collection and DNA extraction; LG participated in the manuscript preparation and revision; SA conceived the study, carried out data analysis, co-ordination and interpretation of the results.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555949.xml

529302

Risk factors for acute chemical releases with public health consequences: Hazardous Substances Emergency Events Surveillance in the U.S., 1996–2001

Background Releases of hazardous materials can cause substantial morbidity and mortality. To reduce and prevent the public health consequences (victims or evacuations) from uncontrolled or illegally released hazardous substances, a more comprehensive analysis is needed to determine risk factors for hazardous materials incidents. Methods Hazardous Substances Emergency Events Surveillance (HSEES) data from 1996 through 2001 were analyzed using bivariate and multiple logistic regression. Fixed-facility and transportation-related events were analyzed separately. Results For fixed-facility events, 2,327 (8%) resulted in at least one victim and 2,844 (10%) involved ordered evacuations. For transportation-related events, 759 (8%) resulted in at least one victim, and 405 (4%) caused evacuation orders. Fire and/or explosion were the strongest risk factors for events involving either victims or evacuations. Stratified analysis of fixed-facility events involving victims showed a strong association for acid releases in the agriculture, forestry, and fisheries industry. Chlorine releases in fixed-facility events resulted in victims and evacuations in more industry categories than any other substance. Conclusions Outreach efforts should focus on preventing and preparing for fires and explosions, acid releases in the agricultural industry, and chlorine releases in fixed facilities.

Background Recent high-profile hazardous materials incidents, such as an explosion at a pharmaceutical supply plant in Kinston, North Carolina [ 1 ] and an explosion at a manufacturing plant that makes automotive insulation products in Corbin, Kentucky [ 2 ], highlight the need to determine risk factors for hazardous materials incidents to reduce the public health consequences from uncontrolled or illegally released hazardous substances. Previous analyses of risk factors for chemical releases have been restricted to one state, one chemical, or only a few years of data. An analysis of hazardous substance releases in Wisconsin found that ammonia releases occurred more frequently in the food processing, manufacturing, and agricultural industries; and ammonia releases were more likely than releases of all other chemicals to result in evacuation and injury [ 3 ]. Burgess et al. showed that chemical releases in Washington state inside buildings and releases involving three to five victims were more likely to require evacuation or sheltering in place [ 4 ]. An analysis of hazardous substance releases during 1990–1992 showed that ammonia, chlorine, and acids were frequently released and were more likely than other substances to result in injuries or evacuations [ 5 ]. A more comprehensive analysis is needed to determine which risk factors are associated with releases of hazardous substances that result in victims or evacuations. The Hazardous Substances Emergency Events Surveillance (HSEES) system, maintained by the Agency for Toxic Substances and Disease Registry (ATSDR), is an active, multistate surveillance system designed to collect and analyze information about acute releases of hazardous substances. Data elements captured in HSEES include time, date, and day of the week when the event occurred; event type (fixed-facility or transportation-related event); geographic information; factors contributing to the release; type of chemicals released; information about injured persons (victims); industry responsible for the event; and information about decontaminations and orders to evacuate. Data from HSEES were analyzed to identify potential risk factors associated with chemical releases involving victims or evacuations during 1996–2001. Methods Surveillance HSEES events are defined as sudden, uncontrolled, or illegal releases of at least one hazardous substance that had to be removed, cleaned up, or neutralized according to federal, state, or local law. A substance is considered hazardous if it might reasonably be expected to cause adverse human health outcomes. Threatened releases are also included in HSEES, if the amount threatened to be released would have required removal, cleanup, or neutralization under federal, state, or local law and the threat led to an action to protect the public health (e.g., rerouting traffic, closing a road, or ordering an evacuation). Events involving only petroleum are excluded from HSEES. The purpose of HSEES is to reduce morbidity and mortality associated with acute releases of hazardous substances. The goals of the system are to: 1) describe the distribution and characteristics of hazardous substances emergencies, 2) describe morbidity and mortality of employees, emergency responders, and the general public resulting from hazardous substance releases, 3) identify risk factors for morbidity and mortality, and 4) identify strategies that might reduce future morbidity and mortality resulting from the release of hazardous substances. The pilot phase of the surveillance system was from January 1, 1990 through December 31, 1992. Data for 1996–2001, the most recent period for which complete data are available, were analyzed. Thirteen states participated in HSEES during the entire period analyzed: Alabama, Colorado, Iowa, Minnesota, Missouri, Mississippi, New York, North Carolina, Oregon, Rhode Island, Texas, Washington, and Wisconsin. An additional four states participated during portions of that period: Louisiana (2001), New Hampshire (1996), New Jersey (2000–2001), and Utah (2000–2001). State health department personnel used a variety of sources (e.g., records and oral reports of state environmental agencies, police and fire departments, and hospitals) to collect information about the hazardous events. Before January 1, 2000, data were entered into a computerized data-entry system designed by ATSDR and were transmitted quarterly to ATSDR for quality control checks and analysis. Beginning on January 1, 2000, data were entered into a web-based application that enabled ATSDR to instantly access the data. A standardized data-collection instrument was used to obtain information on each event. ATSDR provided the states with a training manual to ensure uniformity. A victim is defined as a person experiencing at least one documented adverse health effect (such as respiratory irritation or chemical burns) that probably resulted from the event and occurred within 24 hours after the release. The HSEES system does not identify the immediate cause of the adverse health effect other than the event itself. Official evacuations were recorded in the system if they were ordered by on-scene coordinators such as a fire or police chief, a member of a HazMat team, or some other type of official. Reasons for the evacuation are not captured. Industry codes for the type of industry responsible for each HSEES event were assigned according to the 1990 Industrial Classification System of the United States Census Bureau [ 6 ]. The industry classification system consists of 243 codes. From this, 18 major industry categories were defined as follows: agriculture, forestry, and fisheries; construction; mining; manufacturing chemical and allied products; manufacturing petroleum and coal products; manufacturing (excluding chemical and allied products and petroleum and coal products); transportation; communications; utilities and sanitary services; wholesale trade; retail trade; finance, insurance and real estate; business and repair services; personal services (comprising industries such as private households, hotels/motels, dry cleaners, and beauty parlors); entertainment and recreation services; professional and related services (which includes hospitals and schools); public administration; and military. Individual chemicals that were released or threatened to be released were assigned to one of 11 substance categories (acids, ammonia, bases, chlorine, other inorganic substances, paints and dyes, pesticides, polychlorinated biphenyls, volatile organic compounds, mixtures of substances from different categories mixed before release, and other). Events that involved chemicals from more than one substance category were categorized as "multiple substances." Analysis Data were analyzed using SAS [ 7 ] logistic regression to identify potential risk factors associated with chemical releases involving victims or evacuations during 1996–2001. Events that involved only a threatened release were excluded from analyses that examined risk factors for events with victims. Fixed-facility and transportation-related events were analyzed separately. Potential risk factors included time of day (0000–0559, 0600–1159, 1200–1759, and 1800–2359), day of week (weekday vs. weekend), and season (December-February, March-May, June-August, and September-November) when the event occurred; type of general land use of the surrounding area in which the event occurred (industrial, commercial, residential, agricultural, other); release type; and substance category. For each chemical reported, up to two entries for type of release could be selected from six choices (spill, air, fire, explosion, threatened release, other). To describe an event, release type was assigned in a hierarchic arrangement (fire and explosion, explosion, fire, air, spill, or threatened) according to all release types reported for an event. When an event had multiple release types, the highest rank was used to assign the release type. For example, an event that had a chemical released as both an air emission and a fire (or one chemical released by air and one chemical released by fire) would be classified as fire. Threatened releases were analyzed only for events with evacuations. The area of evacuation recorded if the evacuation zone was downwind/downstream of the release, a circular area, the building(s) or affected part of the building(s), both a circular area and downwind/downstream, or if no criteria were used. For fixed-facility events, industry category and factor contributing to the release (improper mixing or filling; equipment failure; human error; system problem; beyond human control, including power failures and bad weather; illegal dumping or deliberate damage; other) were also included. Although up to two factors could have been selected for each event, we analyzed only the first factor selected, because only 13% of events had a second factor selected and specific combinations of factors were too sparse to provide meaningful analysis. We examined the likelihood of an event resulting in a victim or evacuation for specific substance and industry categories separately. Substance categories were coded as yes/no; the referent group was all other substances. Similarly, industry categories were coded as yes/no; the referent group was all other industries. Because most transportation-related events were coded into the transportation industry (79%), the association between industry type and an event with a victim (or evacuation) was examined only for fixed-facility events. Mode of transportation, categorized as ground (i.e., truck), rail, water, air, and other, was also included in the analysis for transportation-related events. To further specify risk factors, multivariate and stratified analyses were performed to examine whether certain combinations of variables significant as univariate predictors were associated with a greater likelihood of an event with a victim or evacuation. A multiple logistic regression model was constructed to examine the potential interactions among time of day, weekday, and season. For fixed-facility events, analysis stratified by industry category examined the association between specific substances and an event with a victim or evacuation within industries. Results During 1996–2001, participating state health departments reported 39,766 events to the HSEES system. The analysis of risk factors for events with at least one victim excluded 568 events that involved only a threatened release, and thus focused on 39,198 events (29,974 (76%) fixed-facility events and 9,224 (24%) transportation-related events). At least one victim was reported for 2,327 (8%) of the fixed-facility events with the number of victims per event ranging from 1 to 259 (mean = 4 and median = 1). Similarly, at least one victim was reported for 759 (8%) of the transportation events with the number of victims per event ranging from 1 to 65 (mean = 2 and median = 1). The analysis of risk factors for evacuations focused on 39,622 events (30,190 (76%) fixed-facility events and 9,432 (24%) transportation-related events) because information about whether an evacuation was ordered was missing for 144 events. Evacuations were ordered for 2,961 (10%) fixed-facility events with the number of persons evacuated per event ranging from 0 (10 events) to 11,000 (mean = 112 and median = 20). For transportation events, 453 (5%) events involved an ordered evacuation; the number of persons evacuated per event ranged from 0 (3 events) to 8,700 (mean = 163 and median = 20). For fixed-facility events, the area of evacuation was primarily the building(s) or affected part of the building(s) (2,222 events, 75%), while for transportation events, the evacuation zone was primarily a circular area (199, 44%). Events involving victims Emergency chemical releases at fixed facilities involving victims were more likely to occur on weekdays than weekends (prevalence odds ratio [POR] = 1.16, 95% CI = 1.03–1.29) (Table 1 ). Furthermore, events at fixed facilities with victims were most likely to occur during 1200–1759 (POR = 1.90, 95% CI = 1.61–2.25) compared with the hours of 0000–0559. Transportation-related events were less likely to occur during specific hours or on weekdays. Seasonality was not associated with a higher likelihood of chemical releases with victims for either fixed-facility or transportation-related events. Table 1 Univariate analyses for acute chemical releases involving victims, by type of event. Variable Fixed Facilities Transportation No. % with victims POR (95% CI) No. % with victims POR (95% CI) Time of day 0000–0559 4117 4.3 referent 1205 9.8 referent 0600–1159 10104 7.4 1.78 (1.51–2.11) 3011 9.2 0.94 (0.75–1.18) 1200–1759 9289 7.9 1.90 (1.61–2.25) 2539 8.7 0.87 (0.69–1.11) 1800–2359 5069 7.4 1.78 (1.48–2.14) 1171 8.8 0.89 (0.67–1.17) Day of week Weekend 5765 7.0 referent 1226 7.0 referent Weekday 24209 8.0 1.16 (1.03–1.29) 7998 8.4 1.22 (0.97–1.54) Season Dec.-Feb. 6643 7.3 referent 1631 8.3 referent Mar.-May 7988 7.9 1.08 (0.96–1.23) 2812 7.4 0.88 (0.70–1.10) Jun.-Aug. 8480 8.1 1.11 (0.99–1.26) 2794 8.9 1.08 (0.87–1.35) Sep.-Nov. 6863 7.7 1.07 (0.94–1.21) 1987 8.5 1.02 (0.81–1.30) Area* Industrial 18936 5.6 0.59 (0.45–0.77) 2547 12.7 0.20 (0.15–0.27) Commercial 5336 3.4 3.49 (2.67–4.57) 3385 2.9 0.46 (0.36–0.59) Residential 2097 17.3 1.68 (1.53–1.85) 612 6.3 1.08 (0.98–1.19) Agricultural 2253 22.1 1.70 (1.27–2.29) 1628 15.5 1.29 (1.02–1.64) Other 1064 9.2 referent 903 15.9 referent Release Type Spill 11165 6.2 referent 8020 6.5 referent Air 17280 7.1 1.16 (1.06–1.28) 945 18.3 3.25 (2.69–3.91) Fire 878 22.8 4.47 (3.75–5.33) 134 35.8 8.08 (5.62–11.64) Explosion 244 53.7 17.57 (13.51–22.86) 11 45.5 12.07 (3.67–39.68) Fire and explosion 98 57.1 20.21 (13.45–30.38) 5 40.0 9.66 (1.61–57.91) Substances †,‡ Acid 1832 14.8 2.20 (1.92–2.53) NA Ammonia 2150 12.6 1.81 (1.58–2.07) 280 17.5 2.46     (1.79–3.38) Chlorine 597 28.0 4.90 (4.07–5.89) 18 33.3 5.62   (2.10–15.01) Other inorganics § NA 968 10.7 1.40     (1.23–1.74) Pesticides 810 14.8 2.13 (1.74–2.59) NA Multiple substances ¶ 1142 24.0 4.12 (3.57–4.75) 352 36.4 7.46     (5.92–9.41) *Type of general land use of the surrounding area in which the event occurred. † Only substance categories with a 95% CI greater than 1.0 are presented. ‡ Substance categories were coded as yes/no. The referent group is all other substances. §Excludes acids, bases, ammonia, and chlorine. ¶ Events with more than one hazardous substance released from different chemical categories. Fixed-facility events with victims were most likely to occur in areas described as commercial, than in the referent group "other," while transportation-related events were more likely to occur in agricultural areas. Compared with spills, fire and explosion was the release type with the greatest likelihood of victims for fixed-facility events (POR = 20.21, 95% CI = 13.45–30.38). Releases of chlorine (POR = 4.90, 95% CI = 4.07–5.89) were most likely to result in fixed-facility events with victims compared with releases from all other chemical categories while releases of multiple substances from different categories (POR = 7.46, 95% CI = 5.92–9.41) were most likely to result in transportation events with victims. Type of industry was examined as a potential risk factor for victims in fixed-facility events only. Industries classified as personal services were most likely to result in events with victims than were releases not involving these industries (POR = 8.24, 95% CI = 7.20–9.44) (Table 2 ). Factors identified as illegal dumping or deliberate damage were most likely to contribute to a fixed-facility event with at least one victim (POR = 15.93, 95% CI = 10.41–24.36) compared with "other" factors. Factors classified as "other" included maintenance, vehicular accident, fire, explosion, and factors that could not be classified into an existing category. Table 2 Univariate analyses for acute chemical releases in fixed facilities, by events with victims and evacuations. Variable Victims Evacuations No. % with victims POR (95% CI) No. % with evac. POR (95% CI) Industry*, † Agriculture ‡ 578 17.8 2.75 (2.21–3.42) NA Business and repair services 265 23.0 3.76 (2.81–5.02) 268 15.7 1.74 (1.25–2.42) Communications 33 21.2 3.33 (1.45–7.68) NA Construction 319 15.4 2.26 (1.67–3.08) NA Entertainment and recreation services 196 29.6 5.29 (3.88–7.21) 200 29.5 3.96 (2.91–5.38) Finance, insurance & real Estate 108 37.0 7.36 (4.97–10.91) 107 41.1 6.59 (4.47–9.70) Manufacturing § 4594 9.6 1.39 (1.24–1.55) 4598 18.5 2.59 (2.38–2.83) Personal services 1025 36.3 8.24 (7.20–9.44) 1064 23.7 3.03 (2.62–3.51) Professional and related services 953 25.6 4.64 (3.99–5.41) 978 42.3 7.82 (6.85–8.94) Public administration 273 22.7 3.70 (2.77–4.92) 277 22.4 2.72 (2.05–3.62) Retail trade 451 33.0 6.45 (5.27–7.88) 456 36.6 5.65 (4.65–6.86) Wholesale trade 1061 10.7 1.49 (1.22–1.82) NA Factors Beyond human control ¶ 1076 2.2 0.58 (0.37–0.92) 1084 8.9 1.56 (1.19–2.05) Equipment failure 15193 4.5 1.21 (0.97–1.51) 15186 7.1 1.23 (1.02–1.47) Human error 5900 14.3 4.28 (3.43–5.35) 5930 15.6 2.97 (2.47–3.57) Illegal dumping or deliberate damage 1152 26.7 15.93 (10.41–24.36) 1295 16.6 2.05 (1.59–2.65) Improper mixing or filling 771 22.6 7.46 (5.69–9.77) 775 27.0 5.94 (4.71–7.49) System problem 1838 0.9 0.23 (0.13–0.38) 1833 1.4 0.23 (.015–0.35) Other 2393 3.8 referent 2425 5.9 referent *Only industry categories with 95% CI greater than 1.0 are presented. † Industry categories were coded as yes/no. The referent group is all other industries. ‡ Includes forestry and fisheries. § Excludes chemical and allied products and petroleum and coal products, which were analyzed as separate categories. ¶ Includes power failures and bad weather. Events involving evacuations Emergency chemical releases at fixed facilities involving evacuations were more likely to occur on weekdays (POR = 1.26, 95% CI = 1.14–1.40) (Table 3 ). Furthermore, events at fixed facilities with evacuations were most likely to occur during 1800–2359 (POR = 1.69, 95% CI = 1.45–1.97) compared with the hours of 0000–0559. Fixed-facility events that resulted in evacuation orders were slightly more likely to occur during June through August than during December through February (POR = 1.12, 95% CI = 1.01–1.25). Transportation-related events were not more likely to occur during specific hours, on weekdays, or during a particular season. Table 3 Univariate analyses for acute chemical releases involving evacuations, by type of event. Variable Fixed Facilities Transportation No. % with evac. POR (95% CI) No. % with evac. POR (95% CI) Time of day 0000–0559 4130 6.5 referent 1237 4.9 referent 0600–1159 10188 10.4 1.67 (1.45–1.92) 3085 5.1 1.04 (0.77–1.41) 1200–1759 9363 10.2 1.64 (1.42–1.88) 2591 5.4 1.11 (0.82–1.51) 1800–2359 5100 10.5 1.69 (1.45–1.97) 1210 6.1 1.26 (0.89–1.78) Day of week Weekend 5790 5.0 referent 1262 7.7 referent Weekday 24400 4.8 1.26 (1.14–1.40) 8170 8.8 0.95 (0.73–1.25) Season Dec.-Feb. 6683 9.2 referent 1682 5.4 referent Mar.-May 8058 9.6 1.04 (0.93–1.17) 2860 4.6 0.86 (0.65–1.13) Jun.-Aug. 8559 10.3 1.12 (1.01–1.25) 2849 4.7 0.87 (0.66–1.14) Sep.-Nov 6890 10.1 1.10 (0.98–1.23) 2041 4.8 0.89 (0.67–1.20) Area* Industrial 18958 8.5 0.60 (0.48–0.75) 2579 5.7 0.61 (0.43–0.85) Commercial 5438 5.2 3.11 (2.49–3.88) 3436 3.5 0.87 (0.63–1.19) Residential 2158 22.3 1.50 (1.38–1.62) 637 5.0 1.12 (0.98–1.28) Agricultural 2269 23.6 0.68 (0.51–0.89) 1681 7.9 0.86 (0.60–1.22) Other 1089 5.9 referent 947 4.9 referent Release Type Spill 11129 6.9 referent 7999 3.1 referent Air 17222 9.3 1.38 (1.27–1.51) 945 12.7 4.55 (3.61–5.72) Fire 869 35.7 7.48 (6.40–8.75) 134 20.2 7.89 (5.08–12.25) Explosion 242 36.8 7.85 (5.98–10.29) 10 20.0 7.81 (1.65–36.99) Fire & explosion 96 46.9 11.90 (7.92–17.89) 4 25.0 10.42 (1.08–100.5) Threatened 335 34.9 7.24 (5.72–9.17) 231 20.8 8.20 (5.82–11.54) Substances †,‡ Acid 1850 9.6 1.46 (1.27–1.67) NA Ammonia 2144 24.6 3.44 (3.09–3.83) 290 18.3 4.89 (3.57–6.69) Chlorine 599 33.1 4.79 (4.03–5.71) 22 27.3 7.52 (2.93–19.31) Multiple substances § 1213 25.6 3.41 (2.98–3.90) 386 14.8 3.79 (2.81–5.10) *Type of general land use of the surrounding area in which the event occurred. † Only substance categories with a 95% CI greater than 1.0 are presented. ‡ Substance categories were coded as yes/no. The referent group is all other substances. § Events with more than one hazardous substance released from different chemical categories. Fixed-facility events with evacuations were most likely to occur in areas described as commercial compared with the referent group "other." Area type was not associated with ordered evacuations for transportation-related events. The release type fire and explosion was most likely to result in fixed-facility events with evacuations compared with spills (POR = 11.90, 95% CI = 7.92–17.89). Releases of chlorine were most likely to result in events with evacuations compared with releases from all other chemical categories for both fixed-facility and transportation-related events (POR = 4.79, 95% CI = 4.03–5.71 and POR = 7.52, 95% CI = 2.93–19.31, respectively). We examined type of industry as a potential risk factor for evacuations in fixed-facility events only. Industries classified as professional services were most likely to result in events with evacuations compared with releases not involving these industries (POR = 7.82, 95% CI = 6.85–8.94) (Table 2 ). Factors identified as improper mixing or filling were most likely to contribute to releases with evacuations in fixed facilities (POR = 5.94, 95% CI = 4.71–7.49) compared with "other" factors. Multivariate and stratified analyses of fixed-facility events Results from the multiple regression analysis of temporal variables indicated an association of borderline significance between fixed-facility events with victims and the three-way interaction indicated by time of day (0600–1159), day of week (weekday), and season (March through May) (p = 0.06); no interaction among temporal variables was indicated for evacuations. The likelihood of an event with at least one victim or evacuation in fixed facilities was examined for substances within industry category (Table 4 ). The strongest association was observed for acid releases in the agriculture, forestry, and fisheries industry for fixed-facility events involving victims (POR = 7.28, 95% CI = 2.02, 26.30). Releases of acids, chlorine, and multiple substances in the manufacturing industry had an elevated likelihood of both an event with at least one victim and an evacuation. There were fewer associations between substances and industry categories for events with evacuations compared with events with at least one victim. Table 4 Multivariate analyses* for acute chemical releases in fixed facilities, by industry and substance category. Industry Acids Ammonia Chlorine Pesticides Multiple Substances POR (95% CI) POR (95% CI) POR (95% CI) POR (95% CI) POR (95% CI) Victims Agriculture † 7.28 (2.02–26.30) 2.65 (1.68–4.17) Construction 5.92 (1.43–24.50) Entertainment & recreation services 5.78 (2.80–11.94) Manufacturing ‡ 1.71 (1.27–2.29) 4.03 (2.69–6.05) 2.74 (1.90–3.96) Personal services 3.21 (2.30–4.47) Professional and related services 2.43 (1.49–3.99) 2.99 (1.23–7.27) Public administration 3.99 (1.47–10.85) Retail trade 3.37 (1.08–10.49) Wholesale trade 4.19 (2.32–7.59) 2.35 (1.57–3.54) Evacuations Business & repair services 3.84 (1.55–9.48) Entertainment & recreation services 3.60 (1.77–7.31) Manufacturing ‡ 4.31 (3.64–5.10) 3.32 (2.23–4.67) 2.64 (1.94–3.60) Personal services 3.07 (1.72–5.45) Retail trade 4.06 (1.23–13.38) *Analysis stratified by industry category examined associations between specific substances and events with victims or evacuations within industries for categories that were significant in the univariate analyses. † Includes forestry and fisheries. ‡ Excludes chemical and allied products and petroleum and coal products, which were analyzed as separate categories. Discussion Almost 40,000 chemical release events were reported to HSEES during the 6-year period included in this analysis. Approximately 8% of the events resulted in almost 12,000 victims, and approximately 9% involved evacuation of more than 325,000 people. Because the public health consequences of chemical releases can be serious and affect large numbers of people, identifying the risk factors more likely to be associated with hazardous substance releases that result in victims or evacuations is important to target appropriate prevention activities. For fixed-facility events, fire and explosion was the strongest single risk factor for events involving either victims or evacuations. Illegal dumping or deliberate damage also was strongly associated with fixed-facility events resulting in victims. For transportation-related events, explosion was the strongest single risk factor for events involving victims, and fire and explosion was the strongest for events involving evacuations. The observed associations between time of day (0600–2359) and weekday occurrence are likely to be influenced by production intensity, however information on production intensity is not available in HSEES to examine this relationship further. When temporal variables were modeled together, fixed-facility events with victims were more likely to occur on weekday mornings in the spring, which may coincide with the planting season. These results are consistent with a previous analysis that showed that acute hazardous substance releases with victims were more likely during the planting season in Midwestern states [ 8 ]. In the stratified analysis of substances within industry category, the strongest association was for acid releases in the agriculture, forestry, and fisheries industry for fixed-facility events involving victims. Previous analyses have shown that acute chemical releases are common in the agricultural industry, nitric acid and sulfuric acid are frequently released in agriculture events, and releases of acids result in significantly higher proportions of releases involving victims [ 5 , 8 , 9 ]. Chlorine releases in fixed facilities resulted in victims and evacuations in more industry categories than any other substance. This is consistent with an analysis that found chlorine releases had the greatest significant risk of having events with victims and were almost five times more likely than nonchlorine events to involve evacuations [ 10 ]. This is not surprising given that chlorine is a strong corrosive agent. Acute health effects of exposure to low levels of chlorine include sore throat, coughing, and eye and skin irritation; exposure to higher levels may cause severe burning of the eyes and skin, rapid breathing, narrowing of the bronchi, wheezing, bluish discoloration of the skin, and lung collapse [ 11 ]. Additionally, chlorine is one of the most frequently produced chemical substances in the United States, with more than 30 million tons produced annually [ 12 ]. Chlorine is widely used in a variety of processes, such as synthesizing other chemicals and making bleaches and disinfectants. The HSEES system collected data in only 17 states during 1996–2001. Each state has different reporting requirements for the amount of hazardous substances released that has to be removed, cleaned up, or neutralized; and reporting of events to participating state health departments is not mandatory. Therefore, the total number of events, events with victims, and events with evacuations may be underestimated. However, HSEES is the only federal hazardous substances release database designed specifically to assess and record the public health consequences of hazardous substances emergency events. The HSEES system captures more events than other federal reporting systems, such as the United States Environmental Protection Agency's (EPA) Risk Management Program (RMP) [ 13 ]. EPA's RMP requires all companies that use certain flammable and toxic substances to develop procedures for hazard assessment, prevention activities, and emergency response [ 14 ]. Because company-specific information (e.g., name and street address) is not available to HSEES at the federal level, RMP accident history data provides important insight regarding the hazardousness and regulatory practices associated with accidental releases of chemicals at United States manufacturing facilities [ 15 ]. These results indicate that prevention activities should focus on ways to reduce fire and/or explosion hazards in facilities that store, use, or manufacture hazardous substances such as distributing material safety data sheets to employees, keeping work areas dust free, and storing chemicals away from ignition sources. Previous analyses of HSEES data have shown that human error and equipment failure are the most frequent causes of acute chemical releases [ 10 , 16 - 18 ]; releases due to these causes may result in fires and explosions. Facilities should also have chemical inventories and risk management plans available for responders to enhance their response capabilities [ 19 ]. Additionally, security measures (e.g., surveillance cameras, chain-link fences, patrol guards, alarms) should be put in place to reduce deliberate chemical releases, such as theft [ 20 ]. Prevention activities also should target agricultural industries that use or store acids and industries that store, use, or manufacture chlorine. Several state health departments participating in HSEES already have developed strategies aimed at reducing releases and injuries associated with chlorine. These activities include distribution of fact sheets to county emergency management agencies, fire departments, other first responders, and industries; and presentations to municipal water directors, engineers, and hotel/motel swimming pool owners and operators. The predictors of public health consequences associated with acute chemical releases presented in this paper are broad categories of characteristics available in the national data. Future analyses of HSEES data may explore industry-specific root causes of events and risk of events with victims or evacuations for specific industries. Conclusion The results of this analysis should help guide prevention activities aimed at reducing emergency chemical releases and their associated victims and evacuations. Attention should focus on preventing fires and explosions, releases from illegal dumping and deliberate damage, and acid releases in the agricultural industry. Efforts should continue to educate industry, first responders and other users about the potential hazards of chlorine. List of Abbreviations ATSDR, Agency for Toxic Substances and Disease Registry CI, confidence interval EPA, Environmental Protection Agency HSEES, Hazardous Substances Emergency Events Surveillance POR, prevalence odds ratio RMP, Risk Management Program Competing Interests The author(s) declare that they have no competing interests. Authors' contributions PZR led the writing of the manuscript and assisted with the statistical analysis. WAW led the statistical analysis and assisted with the writing of the manuscript. WEK conceived of the manuscript and participated in its writing and interpretation. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529302.xml

543468

Student responses to the introduction of case-based learning and practical activities into a theoretical obstetrics and gynaecology teaching programme

Background The fourth-year Obstetrics and Gynaecology course at our institution had previously been taught using theory classes alone. A new teaching model was introduced to provide a better link with professional practice. We wished to evaluate the impact of the introduction of case discussions and other practical activities upon students' perceptions of the learning process. Methods Small-group discussions of cases and practical activities were introduced for the teaching of a fourth-year class in 2003 (Group II; 113 students). Comparisons were made with the fourth-year class of 2002 (Group I; 108 students), from before the new programme was introduced. Students were asked to rate their satisfaction with various elements of the teaching programme. Statistical differences in their ratings were analysed using the chi-square and Bonferroni tests. Results Group II gave higher ratings to the clarity of theory classes and lecturers' teaching abilities (p < 0.05) and lecturers' punctuality (p < 0.001) than did Group I. Group II had greater belief that the knowledge assessment tests were useful (p < 0.001) and that their understanding of the subject was good (p < 0.001) than did Group I. Group II gave a higher overall rating to the course (p < 0.05) than did Group I. However, there was no difference in the groups' assessments of the use made of the timetabled hours available for the subject or lecturers' concern for students' learning. Conclusions Students were very receptive to the new teaching model.

Background In Brazil, medical school courses last for six years and demand full-time study. In our school, the curriculum follows the traditional model, and it is divided into the basic cycle (first and second years), clinical cycle (third and fourth years) and pre-intern cycle (fifth and sixth years; full-time outpatient and hospital practice). The practice of institutional self-evaluation, especially for educational institutions, has become part of the country's recent culture [ 1 ]. It has come about as a result of the democratic transition that Brazil went through from 1986 onwards and with the introduction of quality control principles in the past decade. In our school, we introduced an annual subject evaluation programme (SEP) in 2000, for the purposes of self-evaluation. Every year, in the middle of the second semester, all students in each year-group fill out a standard questionnaire that aims to assess each subject according to the following variables: teaching ability, teaching quality, lecturer's punctuality, student's improvement and commitment to each subject, test evaluation, stimulus given to discussion and clinical reasoning, guidance on practical activities, emphasis on the doctor-patient relationship, clinical correlation between the subject taught, and general impression. Each of these items is rated by the students as very weak , weak , regular , good or very good . The annual report consists of the evaluations on each subject, for all the above-mentioned variables, and is handed in to each lecturer in charge and the department representative [ 2 ]. The teaching of medicine centred on diseases and hospital care and, consequently, centred on the less prevalent disorders, has arisen as a consequence of the current curricular model. This has proven to be inadequate, inefficient and onerous for Brazil's health care sector. Students' participation in the health system is practically non-existent in the basic cycle of the current medical curriculum. In this light, over the last 15 years or so, there has been a series of movements among institutions, aiming towards changing the Brazilian medical school system [ 3 ]. In addition to the implementation of curricular directives of greater efficacy, other measures may stimulate medical students and better prepare them for professional practice. Such invigorating measures may include the linking of basic sciences with all the phases of the professional cycle, scientific initiation programmes that are accessible to all students, support programmes and academic guidance [ 4 ]. To adapt our medical sciences course to these new concepts within the Brazilian setting of health care and medical teaching, a committee for discussing and drawing up a new teaching system was set up. The process began in the second semester of 2000 and has gradually involved more and more of the lecturers, students, members of the board of directors and the school's sponsoring foundation. Thus, in the new model that was proposed, problem-solving techniques would be the main teaching tool [ 5 ]. The model was also grounded in the basic principles of adult education: adults have a profound need for self-motivation [ 6 ] and must therefore take on an active role in the learning process. Adults are motivated much more to learn because of their own inner needs, such as their drive to succeed and satisfaction in learning in order to reach specific personal objectives, than because of outside factors [ 7 ]. Consequently, the content of the fourth-year programme was restructured for the 2003 academic year, with the aim of providing the new teaching tool of clinical case discussions alongside the learning of theory. Students' responses to these changes were assessed in comparison with the 2002 course. Our working hypothesis was that we would be promoting four positive actions: a) integration of theoretical and practical learning from the beginning of the students' contact with the speciality; b) greater consolidation of knowledge of the speciality; c) optimisation of the time that is made available for the speciality in the fourth year of the course; and d) preparation of the staff for the curricular model that would be implemented in the coming years. Methods Each year-group of the medical course at Centro Universitário Lusíada (UNILUS) consists of 120 students. The Obstetrics and Gynaecology course is given in the fourth year (4 hours per week, giving a total of 120 hours) and the fifth and sixth years (a total of 925 hours for the pre-intern cycle). Up to and including 2002, the fourth-year course was taught by means of theory classes only, which aimed to cover all normal aspects of Obstetrics and Gynaecology and principal diseases encountered. Subsequently, during the pre-intern cycle, other relevant topics concerning disorders within the speciality were dealt with through visits to patients and seminars prepared by the students under staff supervision. In 2003, small-group discussions of cases and practical activities based on the normal aspects and diagnosis methods of the speciality were introduced for the fourth year. These activities replaced 50% of the theory content (the content relating to obstetrical and gynaecological pathology). The theory content taken from the fourth-year curriculum would be taught during the pre-intern cycle, together with the practical learning. A large proportion of the staff was mobilised. Each week, during the four hours of teaching, students were divided into two groups. Sixty students attended two theory classes (one topic within Obstetrics and the other within Gynaecology), while the other sixty students were divided into six groups of ten students for discussions of clinical cases or practical activities. The two groups of sixty students alternated throughout the year. It should be stressed that the theory classes were taught by the same lecturers, using the same teaching material, during the two years of the present study. The study began with a standardised SEP questionnaire that was administered to all fourth-year students in the second semester of the teaching year. This was done on the day of their assessment test, to ensure full attendance. The questionnaire bore the institution's official stamp and consisted of several questions, as described in the introduction, above. The questions used in the present study sought ratings for the use made of the timetabled hours available for the subject, lecturers' concern for students' learning, clarity of theory classes and lecturers' teaching abilities, lecturers' punctuality, quality of the assessment tests, students' learning of the subject and general evaluation of the subject. In the 2003 questionnaire, the variable new methodology – activities in small groups was introduced. Each of these items was rated by the students as very weak , weak , regular , good or very good . Group I consisted of 108 fourth-year students on the medical sciences course who answered the questionnaire in 2002 and group II consisted of 113 fourth-year students on the course in 2003. The findings were tabulated using Microsoft ® Excel 2002 for later evaluation. For analysis purposes, positive evaluations were considered to be the sum of the good and very good ratings, and negative evaluations the sum of the weak and very weak ratings. For statistical analysis, the variables were represented by absolute (n) and relative (%) frequency, and the difference between them was analysed using the chi-square test (χ 2 ). The significance level adopted was 0.05 (α = 5%), and descriptive levels (p) that were less than this value were considered significant and marked by an asterisk (*). Significant values were also submitted to the Bonferroni test to ratify their statistical value. Results Group I consisted of 66 female (61%) and 42 male students (39%), whose average age was 23.1 years, while group II consisted of 67 female (59%) and 46 male students (41%), whose average age was 23.6 years. There was no significant difference between their ages. In group I, 95 students (88%) gave a positive rating for the use made of the timetabled hours available for the subject and 13 (12%) gave a regular rating for it, whereas 101 students in group II (89%) gave a positive rating and 12 (11%) a regular rating. There was no significant difference in this evaluation between the two groups (Table 1 – item 1). Eighty-five students (79%) in group I considered that the lecturers had great concern for students' learning, and 98 students (87%) in group II also believed this. Thus, there was no statistically significant difference between the groups (Table 1 – item 2). Table 1 Distribution of course evaluation. Question Rating Group I 2002 n (%) Group II 2003 n (%) χ 2 p 1. Use made of timetabled hours available for the subject 1 95 (88%) 101 (89%) 2 13 (12%) 12 (11%) 0.1106079 > 0.05 3 0 0 2. Lecturers' concern for students' 1 85 (79%) 98 (87%) learning 2 14 (13%) 9 (8%) 2.4986106 > 0.05 3 9 (8%) 6 (5%) 3. Clarity of theory classes and lecturers' teaching abilities 1 72 (67%) 98 (87%) 2 31 (29%) 12 (11%) 12.765231 < 0.05 * 3 5 (4%) 3 (2 %) 4. Lecturers' punctuality 1 108 (100%) 81 (72%) 2 0 32 (28%) 32.620532 < 0.001 * 3 0 0 5. Quality of knowledge assessment tests 1 38 (35%) 73 (65%) 2 49 (45%) 34 (30%) 21.978341 < 0.001 * 3 21 (20%) 6 (5 %) 6. Students' learning of the subject 1 60 (56%) 97 (86%) 2 46 (42%) 13 (12%) 24.619225 < 0.001 * 3 2 (2 %) 3 (2%) 7. General evaluation of the subject 1 80 (74%) 100 (89%) 2 25 (23%) 10 (9%) 7.6007955 < 0.05 * 3 3 (3 %) 3 (2%) Ratings: 1 – Good and very good 2 – Regular 3 – Weak and very weak The clarity of theory classes and lecturers' teaching abilities received a positive evaluation from 72 students in group I (67%) and from 98 students (87%) in group II. This was a statistically significant difference (Table 1 – item 3). Lecturers' punctuality received a positive evaluation from all 108 students in 2002, but only from 72% (81 students) in 2003, and this was statistically significant (Table 1 – item 4). Only 35% of group I (38 students) gave a positive rating for the quality of the knowledge assessment test, whereas 65% of group II (73 students) gave this a positive rating, which was a significant difference (Table 1 – item 5). In 2002, 60 students (56%) rated their learning of the subject as good or very good, while in 2003, 97 students (86%) rated it as positive, which was a significant difference (Table 1 – item 6). Finally, the general evaluation of the subject was rated as good or very good by 80 students (74%) in 2002 and by 100 students (89%) in 2003, which was a statistically significant difference (Table 1 – item 7). The new methodology adopted in 2003 for the Obstetrics and Gynaecology course was considered to be good or very good by 89% of the students, regular by 8% and weak or very weak by 3%. Discussion The traditional curriculum model was developed with reference to the Flexner report of 1910 [ 8 ]. In this, medical education was considered to be a process of initiation in a science. The teachers' role was to establish what students must learn, to transmit information that was considered relevant, and to evaluate students' capacities to retain and reproduce the information presented. Theory would be dealt with before practice, with the aim of preparing students for the use of theory during students' internship and subsequent professional lives. In this model, medical practice is detached from scientific practice, thereby promoting fragmentation of knowledge and neglect of the psychosocial and cultural aspects of medical activities [ 9 ]. This teaching approach has been criticised for the excessive value given to content and for its low efficacy, which brings about the subsequent need for re-qualification. We believe that this "banking concept of education" that Freire [ 10 ] refers to is conclusively condemned to history. On the other hand, the teaching concept of meaningful learning calls for linkage between the roles of universities, health care administrators and social services. It suggests that there should be co-operation in the selection of content, production of knowledge and development of professional competence. In meaningful learning, the teacher is no longer the main source of information, but the facilitator of the teaching-learning process. The teacher's aim is to stimulate the learner to take on an active, critical and reflective attitude in the knowledge building process. The content dealt with must have the potential to be meaningful (functionality and relevance for professional practice), giving value to matters that are pertinent and correlatable with students' cognitive structure. However, the absorption by students of knowledge of the so-called basic subjects in this context presents a great challenge [ 11 ]. The curricular directives for medical courses (Report 583/01, of August 7, 2001) from the Brazilian National Education Council (part of the Ministry of Education) give guidance on the changes to be made to the teaching model for courses. They indicate that courses must involve students in practical activities from the outset and promote active integration between health care service users and professionals from the beginning of their instruction, using methodology which reinforces students' active participation in knowledge-building, thereby bridging the gap between academic medical learning and the social needs of Brazilian health care. It is evident that the new curricular directives have used the concepts and logic of problem-based learning as their reference point. They have been based on various American and European curricula that, over the past decade, have been giving emphasis to free time for self-study instead of traditional lectures [ 12 - 14 ]. Thus, more than half of the medical schools in the United States are at present undergoing a process of curricular reform [ 15 ], as are a large proportion of the medical schools in the United Kingdom [ 16 , 17 ]. In the "problematization" methodology based on Maguerez's Arch, as presented by Bordenave [ 18 ], five phases develop from reality: observation, key points, formulation of theory, putting forward of solution and application to reality (Figure 1 ). This is an alternative methodology that is appropriate to higher education. It differs significantly from problem-based learning in some points that are summarised in Table 2 (adapted from Berbel, 1998 [ 19 ]). Figure 1 Maguerez's Arch. Table 2 Main differences between " problematization " and problem-based learning . "Problematization" Problem-based learning Observation of reality Problems constructed by the lecturers of subjects in which this methodology is used (subject option) Construction of problems by the lecturers, with complete vertical and horizontal integration (institutional option) Key points Not defined Defined in the curriculum Formulation of theory Investigation-guided study Investigation-guided study Putting forward of solution Done after study Done by students before study, on the basis of previous knowledge Application to reality (practice) Results must intervene in reality as much as possible Intervention in the social environment is considered to be fundamental In problem-based learning, the cognitive objectives are all previously established, while in "problematization", total control over the resultant knowledge does not exist. The essence of problem-based learning is that the problems define objective concepts to be learned and non-objective concepts that can be excluded from the learning because they are not relevant to the study in question [ 16 ]. Although it may be difficult and scientifically dangerous to compare results from conventional curricula (lecture-based learning) and models like problem-based learning or "problematization" [ 20 - 23 ], this was not our intention. Our only objective was to evaluate a teaching tool that is already well known and make a contribution towards discussions on curricular reform. The present study does not prove that the "modernised" curriculum is better than the previous one, but it emphasises that the strengths of the "new" curriculum are worthy of more exploration. In our opinion, the perception that a qualitative improvement in students' learning has taken place during the course is the first step towards a more substantial and effective change in the teaching-learning process. In the present study, the intention was to transform a totally theoretical course into a more stimulating and efficient course. In this, concepts acquired during classes would be applied clinically to real cases obtained by the students themselves in the wards. A recent study at Manchester University [ 16 ] has shown that changing a conventional course into a new integrated course, using problem-based learning throughout, has significantly improved recently graduated students' perceptions of their preparedness for entering the professional market. There was no significant difference in students' evaluations of the use made of the time available for the subject between the two groups, because there was already a positive assessment among the 2002 group (Table 1 – item 1). Likewise, students gave positive evaluations regarding their perception of lecturers' concern for their learning. Although there was no significant difference between the groups in relation to this question, there was a mild tendency towards increased positive evaluation among the 2003 group (Table 1 – item 2). An improvement in the assessment of the course can be seen from item 3 of Table 1 onwards. From 2002 to 2003, there was a significant increase in the positive rating given to clarity and teaching abilities in the classes taught. At first, this seemed odd to us, considering that the teaching material used and the staff who taught the theory classes were identical for the two groups. We concluded that the insertion of clinical cases and practical classes into the traditionally theoretical course was the decisive factor in students' perception that the 2003 lessons had improved. Although the fact that the questionnaire was administered at the time of the final assessment test may have had an influence on the data, the questionnaire was administered on the same occasion for each of the two year-groups. The decrease in the rating of lecturers' punctuality can be easily explained by the fact that the theory classes were always predictably held in the same place in 2002 (group I), while group II used various locations that were specially booked for them. On some occasions in 2003, unexpected events occurred at the beginning of the activities (Table 1 – item 4). Assessment tests for Obstetrics and Gynaecology are traditionally considered to be difficult. There was a perception in our school that they did not reflect the overall knowledge of the subject that is required. The tests consist of forty to fifty multiple-choice questions (each with five alternatives presented) and five essay-type questions. The former perception can be seen among the 2002 year-group in item 5 of Table 1 , alongside the significant improvement among the 2003 group. This indicates to us that the 2003 year-group studied with greater satisfaction and interest, stimulated by the new process, and that this group consequently made the interpretation that there was greater coherence in the preparation of tests. Nonetheless, the tests did not undergo any substantial change from 2002 to 2003. Despite this improvement in the rating, we are still far from achieving the desired positive evaluation rate for the quality of our tests, and the present study shows us that the tests need to be improved. One of the most important objectives in a change in the teaching system is to obtain greater course efficiency and increased student learning. Items 6 and 7 of Table 1 show us that, at least with regard to student perceptions, this aim has been achieved. Our assessment is that the change in the teaching system was very stimulating for the development of students' study routines. The holistic concept of modern education directs us towards integrating knowledge, understanding and practice for learners. In this, learning is taken to be an ongoing part of life and not just a preparation for it [ 24 ]. In keeping with this view, the medical curriculum needs to drum into students the ethos of self-evaluation [ 7 ]. Students responded well to the new method, as shown by the positive rating of 89% given by the 2003 year-group. This provides us with the basis for further advances in this subject in the years to come. It gives the staff the confidence to institute significant changes in the curricular reform that has been under discussion for three years. Although the staff's level of satisfaction was not an objective of our study, initial observation of this indicates great commitment to the course and, probably, better performance. However, it will only be through future longitudinal studies that we will know whether there has really been greater consolidation of knowledge and course efficiency. Conclusions Students were very receptive to the new teaching model in this study. An active role in their learning process seems to be more pleasant and productive than usual method. Thus, active learning methodology should be stimulated on the medical courses throughout the world. Competing interests The author(s) declare that they have no competing interests. Authors' contributions All authors participated in the study design and application. JCM was the study coordinator. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC543468.xml

516441

Disparities in lipid management for African Americans and Caucasians with coronary artery disease: A national cross-sectional study

Background Individuals with coronary artery disease are at high risk for adverse health outcomes. This risk can be diminished by aggressive lipid management, but adherence to lipid management guidelines is far from ideal and substantial racial disparities in care have been reported. Lipid treatment and goal attainment information is not readily available for large patient populations seen in the fee-for-service setting. As a result, national programs to improve lipid management in this setting may focus on lipid testing as an indicator of lipid management. We describe the detection, treatment, and control of dyslipdemia for African Americans and Caucasians with coronary artery disease to evaluate whether public health programs focusing on lipid testing can eliminate racial disparities in lipid management. Methods Physicians and medical practices with high numbers of prescriptions for coronary artery disease medications were invited to participate in the Quality Assurance Program. Medical records were reviewed from a random sample of patients with coronary artery disease seen from 1995 through 1998. Data related to the detection, treatment, and control of dyslipidemia were abstracted from the medical record and evaluated in cross-sectional stratified and logistic regression analyses using generalized estimation equations. Results Data from the medical records of 1,046 African Americans and 22,077 Caucasians seen in outpatient medical practices in 23 states were analyzed. African-American patients were younger, more likely to be women and to have diabetes, heart failure, and hypertension. The low density lipoprotein cholesterol (LDL-C) testing rate for Caucasian men was over 1.4 times higher than that for African-American women and about 1.3 times higher than that for African-American men. Almost 60% of tested Caucasian men and less than half of tested African Americans were prescribed lipid-lowering drugs. Tested and treated Caucasian men had the highest LDL-C goal attainment (35%) and African-American men the lowest (21%). Conclusions Although increased lipid testing is clearly needed for African Americans, improvements in treatment and control are also necessary to eliminate racial disparities in lipid management. Disparities in treatment and goal attainment must be better understood and reflected in policy to improve the health of underserved populations.

Background Individuals with coronary disease (CAD) are at high risk for subsequent cardiovascular disease events and mortality[ 1 ]. Clinical trials have shown that this risk can be substantially reduced though the detection, treatment, and control of dyslipidemia[ 2 , 3 ]. To that end, clinical guidelines have been established for the lipid management of CAD patients[ 4 , 5 ]. Adherence to these guidelines is far from ideal[ 6 , 7 ]. Substantial racial disparities in the diagnosis and management of dyslipidemia have been reported in the general population and among CAD patients [ 8 - 16 ]. Improved lipid management through the diagnosis of dyslipidemia has been a focus of quality improvement programs in the outpatient fee-for-service setting such as Medicare's Health Care Quality Improvement Program[ 17 ]. Lipid testing is used as an indicator of lipid management in the fee-for-service setting because testing is a service identifiable in insurance claims data. In contrast, assessment of treatment and goal attainment in the fee-for-service setting requires resource-intensive medical record review which is generally not performed for large national patient populations. This study describes outpatient lipid management for African Americans and Caucasians with CAD seen in medical practices throughout the United States. Data from medical records were examined for indicators of lipid management including lipid testing, lipid-lowering drug prescription, and goal attainment. Our objectives were to characterize lipid management across race-sex groups and evaluate the extent of disparities for the three components of lipid management: detection, treatment, and control. We then discuss the implications of our findings with respect to possible underlying causes and health policies for closing the gap in race-sex lipid management disparities. Methods The Quality Assurance Program The Quality Assurance Program (QAP) is a national program sponsored by Merck & Company conducted during the late 1990's to identify physician practice patterns and to promote evidence-based best practices for the medical management of patients with cardiovascular disease seen in the outpatient setting [ 18 ]. The QAP database provides abstracted medical record data collected in two distinct time periods and study populations nationwide. Our analyses were limited to the study population identified in the most recent period of data collection (QAP-II). Patients from QAP-II included in these analyses were seen at participating medical practices from January, 1995 through March, 1998. Medical records were reviewed by Access Medical Ltd (Arlington, VA) using a standardized electronic abstraction tool developed specifically for QAP-II. The QAP database contains data obtained from the medical record including race, sex, date of birth, medical history, and medical procedures. The most recent serum lipid testing results and the most recently recorded prescriptions for lipid-lowering drugs were also determined from the medical record. The medical record of each patient was reviewed only once. Patients were not followed over time. Patient and physician identifying information were not included in the QAP-II database to ensure confidentiality. QAP participant selection Physicians and medical practices throughout the United States with high numbers of prescriptions for medications used in the treatment of cardiovascular disease were invited to participate in the QAP. The specialties of participating physicians included cardiology, internal medicine, family medicine, and endocrinology. Patients with cardiovascular disease were randomly selected within each participating medical practice. Inclusion and exclusion criteria Patients included in the QAP study were at least 21 years of age with CAD and/or heart failure and were seen at least twice in two years by the participating physician. Patients were excluded if the medical record indicated a terminal illness, history of a transplant or awaiting transplant, or deceased. Patients without medical record documentation of CAD were excluded from analysis. The presence of CAD was ascertained during analysis from abstracted medical record data based on medical history, International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes (410–414), and cardiac procedures consistent with CAD (i.e., coronary artery bypass graft, angioplasty, and stent). Only patients with medical record documentation of African-American or Caucasian race were included. To reduce the influence of between-state variation in lipid management from states that contribute little information about African American populations, medical practices were excluded if they were located in states with fewer than 10 African-American patients in QAP-II Indicators of lipid management Measures of detection (lipid testing), treatment (lipid-lowering drug prescription), and control (goal attainment) were the indicators of lipid management considered in this study. Low density lipoprotein cholesterol (LDL-C) testing was measured as the percentage of patients with at least one serum LDL-C value documented in the medical record. The use of lipid-lowering drugs (i.e., "treated" patients) was measured as the percentage of patients with medical record documentation of at least one prescription for a statin (3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitor) or non-statin lipid drug (e.g., gemfibrozil). Goal attainment among those with documented LDL-C values was based on recommended guidelines for patients with coronary artery disease (LDL-C < 100 mg/dL)[ 5 ]. Analysis We conducted a cross-sectional analysis of abstracted medical record data obtained from QAP-II. Indicators of lipid management (i.e., LDL-C testing, lipid-lowering drug prescription, and LDL-C goal attainment rates) and potential confounding or explanatory variables were examined within strata of race and sex. Co-morbid conditions including diabetes mellitus, myocardial infarction, heart failure, and hypertension were identified from medical history and diagnosis codes. Logistic regression analyses were performed to evaluate the associations of race and sex with each of three dichotomous lipid management indicators while controlling for multiple confounding and explanatory variables. We accounted for correlations within medical practices using the generalized estimation equation (GEE) method[ 19 ]. This method was implemented in generalized linear models with PROC GENMOD of SAS Version 9 (SAS, Inc, Cary, North Carolina) for logistic regression with correlated data[ 20 ]. Separate models were run for each indicator of lipid management as the dependent variable. The entire study population was included for logistic regression analyses of LDL-C testing as the dependent variable. Regression analyses with lipid-lowering drug prescription as the dependent variable included only patients with LDL-C tests. Regression analyses for LDL-C goal attainment were limited to patients who received at least one LDL-C test and had a documented prescription for a lipid-lowering drug. The independent variables included in the regression models in addition to race and sex were age, medical history (diabetes mellitus, myocardial infarction, heart failure, hypertension), and geographic region of medical practice. Logistic models predicting lipid-lowering drug prescriptions included a term for serum LDL-C concentration in addition to the above variables. LDL-C concentration was included to control for the severity of hyperlipidemia as a factor in the physician's decision to treat with lipid-lowering drugs. Results Population characteristics A total of 23,123 CAD patients with documented race and sex seen by 1,171 physicians at 238 medical practices in 23 states were included in the study. Of these patients, 1,046 were African American and 22,077 were Caucasian. African-American compared to Caucasian patients were more likely to be women (53% versus 36%). The average age for the study population was 69 years and ranged from 22 to 97 years. Within each race group, women were older than men on average and within each sex group Caucasians were older than African Americans (Table 1 ). The prevalence of co-morbid conditions was high for all race-sex groups. Despite their younger ages, African Americans were more likely than Caucasians to have diabetes, heart failure, and hypertension. Almost half of African-American women and a third of African-American men had diabetes compared to about a quarter of the Caucasian population. About half of African-American men had heart failure and over three quarters of African-American women had hypertension. Consistent with the geographic distribution of race-specific populations, African-American patients were more likely seen in southern medical practices than elsewhere. Table 1 Characteristics of patients by race and sex.* Characteristic African Americans Caucasians Women (n = 558) Men (n = 488) Women (n = 8,038) Men (n = 14,039) Age (mean ± SEM) 66.8 ± 0.5 62.7 ± 0.6 72.1 ± 0.1 67.1 ± 0.1 Medical History Diabetes Mellitus 43 33 25 21 Myocardial Infarction 38 47 40 48 Heart Failure 42 48 39 34 Hypertension 78 70 63 52 Region Northeast 27 25 34 31 Midwest 24 30 31 33 South 42 36 22 23 West 8 8 13 13 Serum Lipid Levels LDL-C (mean ± SEM mg/dl) † 131.3 ± 2.5 133.0 ± 2.5 124.8 ± 0.6 117.2 ± 0.3 LDL-C at goal (%) † 19 18 25 31 *Percent of race- and sex-specific total unless otherwise specified. † For patients with documented tests having valid values Lipid testing Within sex strata, the percent of patients with LDL-C tests was lower for African-Americans compared to Caucasians (Figure 1 ). Within race strata the percent of patients with LDL-C tests was lower for women than for men. The LDL-C testing rate for Caucasian men was over 1.4 times higher than that for African-American women and about 1.3 times higher than that for African-American men. Figure 1 Lipid Management among CAD patients. Lipid testing, treatment, and goal attainment rates for African-American and Caucasian women (W) and men (M). Lipid treatment Almost 60% of Caucasian men with LDL-C testing were prescribed lipid-lowering drugs (Figure 1 ). The proportion of tested Caucasian women receiving these drugs (55%) was similar to, but slightly lower than that for Caucasian men. Less than half of tested African-American men (47%) and women (46%) were prescribed lipid-lowering drugs. Goal attainment Including those prescribed and not prescribed lipid-lowering drugs (Table 1 ), a quarter of Caucasian women and less than 20% of African-American men and women achieved the recommended LDL-C goal. But a higher proportion (31%) of Caucasian men achieved goal. The mean serum LDL-C concentration for Caucasian men (117 mg/dL) was lower than that for Caucasian women (125 mg/dL) and for African-Americans. Mean LDL-C concentrations were similar for African-American men (133 mg/dL) and women (131 mg/dL). Among patients tested and treated with lipid-lowering drugs about two-thirds or more failed to achieve goal (Figure 1 ). Tested and treated Caucasian men had the best goal attainment rates (35%). Tested and treated Caucasian women had lower levels of goal attainment than Caucasian men. Goal attainment for Caucasian men and for women of either race exceeded that of African-American men. Only about 1 of 5 African-American men prescribed lipid-lowering drugs achieved LDL-C goal. Logistic regression Results from GEE logistic regression analyses controlling for age, co-morbid conditions, and geographic region (Table 2 ) were consistent with results from the stratified analyses described above. Relative to Caucasians, African-American men and women were under-tested, under-treated, and less likely to be at goal. Relative to men, regardless of race, women were less likely to be tested. But if tested, women were as likely as men of the same race to receive prescriptions for lipid-lowering drugs. African-American women were the least likely to be tested (Odds Ratio, OR = 0.49), and if tested, African-American men were the least likely to be prescribed lipid drugs (OR = 0.59). Among those tested and treated, African-American men were least likely to be at goal (OR = 0.47). That is, among tested patients who were prescribed lipid-lowering drugs, the odds for goal attainment for Caucasian men were more than twice the odds for African-American men. Table 2 Lipid testing, pharmacologic treatment, and goal attainment among African-American men and women and Caucasian women relative to Caucasian men from logistic regression analyses.* African American Caucasian Women Odds Ratio (95% CI)* Men Odds Ratio (95% CI)* Women Odds Ratio (95% CI)* Men Reference Group LDL-C tested (n = 23,104) 0.49 (0.37,0.64) 0.60 (0.47,0.77) 0.80 (0.74,0.86) 1.00 Lipid drug prescribed (n = 14,499) † 0.62 (0.46,0.83) 0.59 (0.45,0.78) 1.04 (0.95,1.13) 1.00 LDL-C goal attainment (n = 8,336) ‡ 0.55 (0.36,0.82) 0.47 (0.30,0.74) 0.76 (0.68,0.85) 1.00 *Odds ratio and 95% confidence intervals (95% CI) from logistic regression models accounting for within-practice correlations with GEE and controlling for race, sex, age, medical history (diabetes mellitus, myocardial infarction, heart failure, hypertension), and geographic region of medical practice. † Regression model includes serum LDL-C concentration. ‡ LDL-C goal attainment (<100 mg/dL) among those with documented LDL-C values and treated with lipid-lowering drugs. Discussion Consistent with previous reports, our findings demonstrate that outpatient lipid management for CAD patients in the late 1990's had much room for improvement and that substantial race-sex disparities existed[ 8 , 12 , 16 , 21 - 26 ]. African Americans experienced markedly lower levels of LDL-C testing than Caucasians and, as a result, they may benefit more than Caucasians from interventions to improve testing. Among those tested, African Americans were less likely to be treated and, if treated, they were less likely to be at goal compared to Caucasians. Marked lipid testing disparities by sex suggest a need for more aggressive testing in women. Much of the information needed to assess lipid management can be found only in the medical record. This information is generally not available to national and local public health programs attempting to implement policies promoting quality improvement in the many and diverse medical practices treating large patient populations. The most readily available data for patients seen in the fee-for-service setting is derived from administrative insurance claims for the reimbursement of costs associated with drugs and services. Insurance coverage for lipid-lowering drugs varies across plans. In contrast, lipid testing for CAD patients is a widely covered service that can be identified using electronic billing data without the need to review patient records in medical practices. For this reason, lipid testing is a focus of national efforts in the Medicare population to improve outpatient lipid management in the fee-for-service setting[ 17 ]. Our findings demonstrate that substantial disparities in treatment and goal attainment exist among CAD patients with lipid tests. This implies that current public health programs and policies designed to increase lipid testing alone will have limited impact on lipid management disparities. Substantial disparities in lipid treatment and control will likely persist in the absence of disparities in testing. Underlying causes of inadequate lipid management among CAD patients are multiple and likely vary by race and process of care (i.e., detection, treatment, and control). Factors that limit patient-physician encounters and continuity of care may partially account for racial disparities in lipid management. For example, African-American Medicare consumers with diabetes were more likely to receive outpatient care from emergency departments and had fewer physician visits per year than their Caucasian counterparts[ 12 , 27 ]. The QAP data provide insufficient information to evaluate whether health care access and continuity explain lipid management disparities. A report regarding racial disparities in the use of prescription drugs from the Center for Studying Health System Change provides evidence of other factors explaining racial disparities in lipid management[ 11 ]. In this report, Medicare consumers 65 years of age and older were surveyed regarding their ability to obtain prescription drugs. African Americans were more than twice as likely as Caucasians to have not filled a prescription because they could not afford it. More than 16% of Medicare insured African Americans reported that they could not afford to fill at least one prescription in 2001. One-fifth of African Americans and 13% of Caucasians with low income could not afford to fill at least one prescription. During the time period of QAP-II, Medicare did not cover the cost of lipid-lowering drugs. Many Medicare consumers have supplemental insurance that assists with drug costs. In the Medicare population, African Americans were less likely than Caucasians to have supplemental insurance and more likely to be of low income[ 11 ]. Barriers to lipid management due to affordability may result in racial disparities if affordability differs by race. Patients who can not afford treatment may be less likely to aggressively pursue it with their physicians and may be less likely to comply with physician recommendations for testing and treatment. Secondary prevention of cardiovascular diseases among CAD patients with pharmacologic agents such as statins has been shown to be cost effective[ 28 ]. But drug costs that may exceed $2,000 annually can be beyond the reach of low income and underinsured patients[ 29 ]. These costs are more likely a barrier for African Americans than for Caucasians and may contribute to lipid management disparities. The fact that African Americans are more likely to be of low income has greater implications than simply the inability to afford medications. Income is one of several indicators of socioeconomic status correlated with factors related to health including education[ 30 ]. Low education has been identified as a factor limiting a patient's personal involvement in lipid management[ 31 ]. There is evidence that African Americans are less knowledgeable about cholesterol compared to Caucasians[ 24 ]. African Americans may be less aware of the need for lipid management and less likely to pursue it with their physician. Achieving LDL-C goal is challenging for all races, but African Americans may require especially aggressive lipid management accompanied by an enhanced understanding of reasons for failure to achieve goal. A recent report concerning patients with CAD and/or diabetes seen at a Veterans Affairs Medical Center found that African Americans were less likely to achieve lipid goal than Caucasians when prescribed identical doses of the same lipid-lowering drug even though African Americans had more clinic visits and lipid tests[ 10 ]. The authors speculate that racial disparities in goal attainment may have occurred due to differences in compliance, lifestyle, and baseline LDL-C (higher for African Americans). In a study of LDL-C lowering with pravastatin in an African-American population, only 13% of patients with a LDL-C goal of 100 mg/dL actually achieved it. Incorrect drug regimen, inadequate lipid monitoring, and compliance problems were thought to have contributed to these goal attainment failures[ 32 ]. Additional studies are needed to investigate the underlying causes of lower goal attainment for African Americans receiving treatment for dyslipidemia. Physicians have indicated that oversight is a common reason for failure to adhere to lipid testing guidelines[ 33 ]. Whether oversight contributes to racial disparities in lipid management is not known, but oversight can be alleviated by system changes in the medical practice[ 34 ]. Awareness is growing that implementation of electronic health records is a necessary component of efforts to improve healthcare quality and prevent medical errors[ 35 , 36 ]. The impact of electronic systems on disparities is an interesting area for future research. Lipid management is also influenced by physician attitudes about guidelines and drug effectiveness[ 37 ]. A better understanding is needed of physician attitudes and their relations with healthcare disparities. Physician-patient interactions are influenced by race and cultural factors related to race. African-American and Caucasian patients may differ with respect to cultural perceptions of health and disease and the ability of patients to influence or control their health outcomes[ 38 ]. African Americans may be less likely than Caucasians to trust their physicians and racial groups may view their relationship with physicians differently[ 39 ]. A survey of adults seen in a managed care setting revealed that African Americans viewed their visits with physicians as less participatory than did Caucasians, but they felt more participatory when seeing a physician of their own race[ 40 ]. The relatively poorer lipid management for African Americans compared to Caucasians may be partially explained by racial differences in the prevalence of co-morbid conditions. African Americans in the QAP were more likely than Caucasians to suffer from multiple chronic conditions including diabetes and heart failure. It has been shown that patients with diabetes compared to those without diabetes receive poorer lipid management and are less likely to be at goal[ 41 ]. An earlier report from the QAP has shown that patients with CAD and heart failure were less likely to receive lipid testing and cholesterol-lowering drugs than those without heart failure[ 42 ]. Co-morbid conditions may hinder lipid management in a number ways. These patients may be more likely to have contra-indications to lipid treatment and to present with acute life-threatening conditions that divert attention away from lipid management. In addition, it has been reported that patients with multiple chronic conditions are less likely to afford medications than those with one or fewer chronic conditions[ 11 ]. But even after controlling for several highly prevalent co-morbid conditions in logistic regression analyses, we find that significant lipid management disparities persist. There are a number of limitations in the QAP study design and its population that potentially pose a threat to internal and external validity. These limitations suggest that our findings may not reflect the experiences of the general population. The medical practices included in the QAP were restricted to those writing large numbers of prescriptions for cardiovascular disease drugs. It is difficult to evaluate the impact of this selection bias, but we suspect that medical practices included in the QAP represent the larger and more sophisticated providers of care. Thus, lipid management among QAP participants may be better than that found in the general population. Another limitation to our study is the lack of information characterizing medical practices, physicians, and patients with respect to factors related to race, sex, and lipid management. Because of the unavailability of this information, we could not identify underlying factors potentially explaining our findings. Lipid management data for the QAP patients seeing multiple physicians were unavailable from physicians not participating in QAP and this may have influenced results in an unpredictable manner. Considering the relatively small proportion of the study population identified as African American (<5%), it is likely that African Americans in this study do not represent the national population. We have no specific information about QAP medical practices or their patient populations, but we speculate that African Americans in the QAP were likely receiving better care on average than their counterparts in the general population, especially those with little or no access to outpatient care. If this is true, then racial disparities in lipid management in the general population may be even greater that those suggested by our findings. The time frame for this study includes years 1995 through 1998. Despite changes in guidelines and therapies since that time, our findings remain relevant to current practices. They provide historic context and baseline measures of lipid management disparities necessary to evaluate trends. In addition, they direct attention to the persistent need to provide aggressive treatment to high risk populations with CAD, especially African Americans and women who continue to be underserved. Finally, they highlight the growing view that public health strategies in lipid management must shift focus from testing to treatment and goal attainment[ 43 ]. Clearly, a multi-pronged approach including all three elements of the process of care (i.e., detection, treatment, and control) is needed to improve lipid management for all race-sex groups, and particularly for African Americans. Although beyond the scope of this report, the application of conceptual models to the process of care may be useful in understanding how racial disparities arise at each step of the process. Furthermore, policies addressing health promotion and non-medical determinants of health such as socioeconomic status, community environment, and lifestyle choices need also be considered in confronting these disparities[ 44 , 45 ]. Successful lipid management likely depends on a variety of processes that determine the provision of medical services including their availability, accessibility, and acceptability. The substantial racial disparities in lipid management among patients in contact with medical providers suggest that the effectiveness of medical services and patient characteristics play a prominent role in lipid management. Policies promoting appropriate delivery of care through system change as well as those ensuring equal access to care are required to eliminate lipid management disparities in the population of high-risk CAD patients[ 34 , 46 ]. Conclusions Our results suggest that policies and programs focusing solely on the elimination of lipid testing disparities can have only limited benefit in reducing the major disparity in lipid management. The elimination of lipid management disparities will require policies that view untested, untreated, and under-treated individuals as separate populations with unique challenges and solutions[ 43 ]. Disparities in testing are just one element in explaining overall disparities in lipid management and ultimately, in cardiovascular outcomes. Future research should address patient, physician, and health system factors that lead to lower rates of testing, treatment and goal attainment for African Americans. Disparities in treatment and goal attainment must be better understood and reflected in policy in order to improve the health of underserved populations through optimal lipid management. List of abbreviations CAD Coronary Artery Disease LDL-C Low-density lipoprotein cholesterol QAP Quality Assurance Program GEE Generalized Estimation Equation Competing interests Analyses were funded by an unrestricted grant from Merck & Co., Inc. No other competing interests are declared. Authors' contributions MM conceived of the study, provided analytic and statistical support, and was the leading contributing author. KF participated in the design of the study and provided analytic support and interpretation of findings. LCE provided critical technical review and major contributions to the discussion section. CS provided conceptual guidance, assistance with QAP project database, and interpretation of analytic findings. CA provided critical review, QAP project experience, and contributions to the presentation and interpretation of findings. RS provided critical review, QAP project experience, and contributions to discussion and interpretation of findings. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516441.xml

545596

A scale of functional divergence for yeast duplicated genes revealed from analysis of the protein-protein interaction network

Protein-protein interaction networks were used to analyze the functional evolution of duplicated genes in yeast. Pairs of paralogs can be grouped into 3 classes, which likely form part of a continuous scale of diversity.

Background Complete genome analysis showed the tremendous extent to which gene and genome duplication events have shaped genomes over time. Remarkably, 30% of the Saccharomyces cerevisae genome, 40% that of Drosophila melanogaster , 50% that of Caenorhabditis elegans , and 38% of the human genome are composed of duplicated genes [ 1 , 2 ]. According to Ohno's theory [ 3 ], such duplication events should have provided genetic raw material, a source of evolutionary novelties, that could have led to the emergence of new genes and functions through mutations followed by natural selection. But despite the recent increase in genomic knowledge, the patterns by which gene duplications might give rise to new gene functions over the course of evolution remain poorly understood. This is mainly explained by the fact that there are very few ways of experimentally investigating the evolution of function of duplicated genes. Studying the function of duplicated genes usually means estimating the extent of the conservation/divergence between duplicates from comparison of actual sequences. For this purpose, the sequence divergence, the divergence time and the selective constraints on gene pairs are usually calculated (as in [ 4 ]). Given that these calculations are only valid on a relatively short timescale [ 4 , 5 ], they exclude de facto the study of ancient duplication events (such as the complete duplication of the yeast genome [ 6 - 8 ]), even though remnants of such events are still present in the genomes [ 9 ]. Enlarging the timescale on which we are able to work is thus a desirable goal, which may be reached by using other means to evaluate the functional conservation/divergence between duplicates. In addition, sequence analysis generally only reveal the possible molecular (biochemical) function(s) of proteins and even this only applies when domains of known function are identified in the sequences. As discussed previously [ 10 ], the function of a gene or protein can be defined at several integrated levels of complexity (molecular, cellular, tissue, organismal) As far as genome evolution is concerned, consideration of the functional evolution of genes and proteins not only at the basal molecular level, but also at upper, more integrated, levels is particularly important. In this respect, it is essential to consider the cellular function of genes/proteins - that is, the biological processes they are involved in. One can easily imagine, for instance, that the evolution of a duplicated pair of protein kinases, having the same molecular function, could potentially result in the emergence of a new signaling pathway involved in a different cellular function. Being able to study the evolutionary fate of duplicated genes at the level of cellular function using bioinformatics methods, something that was quite difficult until now, may thus provide new insights into the field. To do so, one needs to be able to easily compare the functions of many proteins at once and to estimate their functional similarities at the cellular level. Function comparison was one of our aims while developing PRODISTIN, a computational method that we recently proposed [ 11 ]. This method permits the functional classification of proteins solely on the basis of protein-protein interaction data, independently of sequence data. It clusters proteins with respect to their common interactors and defines classes of proteins found to be involved in the same cellular functions. In the work presented here, we addressed the question of the cellular functional fate of duplicated genes in the yeast S. cerevisiae , focusing on the 899 duplicated genes which represent remnants of an ancient whole-genome duplication (WGD) [ 6 - 8 ]. This event took place 100-150 million years ago in the Saccharomyces lineage, after the divergence from Kluyveromyces waltii , and was probably followed by a gene-loss event leading to the current S. cerevisiae genome [ 8 ]. Overall, these duplicated genes form 460 pairs of paralogs, accounting for 16% of the current genome [ 6 ]. After applying the PRODISTIN method to the yeast interactome, we established and analyzed the functional classification of the duplicated yeast genes originating from the WGD. This analysis allowed us to compare the cellular function(s) of 41 paralog pairs for which enough interaction data was available. Three different behaviors of the pairs of paralogs in respect of the PRODISTIN classification were identified from this analysis, allowing us to establish a scale of functional divergence for the duplicated genes based on the protein-protein network analysis. This work validates the use of interaction data and the analysis of interaction networks as a new means of investigating evolutionary processes at the level of the cellular function. Results GO annotations do not functionally distinguish between duplicated pairs from the ancient genome duplication To obtain a first estimation of the functional conservation/divergence of the yeast duplicated genes, we analyzed available textual information relative to the actual functions of the 460 pairs of paralogs from the WGD. For this purpose, we used the Gene Ontology (GO) annotations. The Gene Ontology consortium [ 12 ] develops structured controlled vocabularies describing three aspects of gene function: 'Molecular Function' describes the biochemical function of proteins (their molecular activity); 'Biological Process' describes their cellular function (the "broad biological goals that are accomplished by ordered assemblies of molecular functions"); and 'Cellular Component' describes their subcellular localization. These structured vocabularies, or ontologies, are not organized as hierarchies but as directed acyclic graphs (DAGs), in which child terms (the more specialized terms) can have several parent terms (less specialized terms). These functional annotations thus provide a means of comparing gene functions as long as one is able to take into account the structure of the ontology in the comparison process. We performed a pairwise comparison of the functions of the 460 pairs of duplicates by processing their functional GO annotations with GOproxy [ 13 ]. This tool calculates a functional distance between genes based on the shared and specific GO annotations. The calculation is made separately for the three ontologies, and for each gene the complete hierarchy of GO terms, from the root term to the leaf term of the DAG, is considered in the comparison process without differentiating the two parent-child relationships existing in GO (the 'is-a' and the 'has-a' relations) (for details see Materials and methods). Two genes that do not share any GO terms would have a maximum distance value (equal to 1), whereas two genes sharing exactly the same set of GO terms would have a minimum distance value (equal to 0). The distributions of the calculated distance values are showed in Figure 1 . First, as expected, paralog pairs are globally closer in term of functional distance based on the annotations (Figure 1a ) than pairs of proteins chosen randomly from the proteome (Figure 1a , inset). Indeed, the distribution of the distances peaks at the minimum distance value for the paralogs while it peaks at the maximum distance value for the randomly selected pairs. Second, the vast majority of the duplicated pairs do not differ significantly when Molecular Function terms are compared: 74.5% of the pairs have a zero distance based on annotations (Figure 1a , purple bars). This could be explained by the fact that on one hand, a tight relationship exists between protein sequence similarity and molecular function(s) similarity, and on the other the majority of the paralogs share a percentage sequence identity above the 'twilight zone' (20-35%) [ 14 ], usually considered as a threshold for molecular function similarity. Given that paralogs with the same molecular function may potentially be involved in different cellular functions, we also considered the Biological Process annotations of gene products. Interestingly, the majority of the paralogs also display a zero distance value, suggesting that a majority of duplicated genes from the ancient duplication do not significantly differ when considering the cellular function annotations. However, although the distribution of the distances between the duplicates for the Biological Process annotations displays the same overall shape, only 56.5% of the pairs show a zero value (Figure 1a , blue bars) as compared to 74.5% for the Molecular Function annotations. The fact that, on average, the molecular functions of duplicated pairs are more conserved than their corresponding cellular functions may reflect the fact that changes in function that occurred during evolution are more measurable and discernible at the cellular level than at the molecular level at the present time. This is corroborated by the fact that paralog pairs are found to be globally closer according to the Molecular Function annotation compared to the Biological Process annotation when the expectation values are calculated for each distribution, whereas the converse is encountered for randomly selected pairs (see Additional data file 1). Similarly, changes in subcellular localization (Cellular Component annotations, Figure 1a , yellow bars) also appear to be more apparent than changes in Molecular Function (see Additional data file 1). PRODISTIN interaction network analysis: three classification behaviors Immediately after a genome-duplication event, the two duplicated proteins will have the same interactors. As time goes by and mutations occur, these proteins may gain or lose interactors; that is, the number of interactors for each protein of the pair may change as well as their identity. Taking account of the fact that protein action is seldom isolated but rather is exerted in concert with other proteins, studying duplicates according to the interactors they still share and the ones they have lost or acquired since the duplication event may give a hint about how their cellular functions have evolved. We thus applied the PRODISTIN method [ 11 ] to 4,143 selected binary protein-protein interactions involving 2,643 yeast proteins. Briefly, the PRODISTIN method consists of three different steps: first, a functional distance is calculated between all possible pairs of proteins in the interaction network with regard to the number of interactors they share (proteins must have at least three interactors to be considered further); second, all distance values are clustered, leading to a classification tree; third, the tree is visualized and subdivided into formal classes. A PRODISTIN class is defined as the largest possible sub-tree composed of at least three proteins sharing the same functional annotation and representing at least 50% of the individual class members for which a functional annotation is available. Classes of proteins are then analyzed for their biological relevance and tested for their statistical robustness (see Materials and methods and [ 11 ] for a detailed explanation). The relevance of the method has been assessed biologically and statistically in a previous study (its first application to a smaller interaction dataset led to the prediction of the cellular function of 42 uncharacterized yeast proteins with a success rate of 67% [ 11 ]). In the present work, 890 proteins were classified (Figure 2 ). Among them, 154 correspond to products of duplicated genes from the ancient duplication and 82/154 form 41 pairs of paralogs. These 41 pairs thus correspond to the only pairs from the ancient duplication for which more than three interaction partners per protein are presently known. Then, following the PRODISTIN procedure, the clustering of the proteins was analyzed, defining classes of proteins involved in the same cellular function(s) according to the GO Biological Process ontology (for details, see Materials and methods). In total, 123 classes corresponding to 53 different cellular functions were identified in the tree (see Additional data file 2) and evaluated statistically (data not shown), allowing the classification of 38/41 pairs of duplicated genes (Table 1 ). We then investigated the details of the distribution of the duplicates in the tree by analyzing the PRODISTIN classes. Interestingly enough, three different situations were encountered (Figure 2 , Table 1 ). First, for 26 pairs both gene products were found in the same class. This means that their list of interactors is very similar and that these proteins should thus be involved in the same biological process. This is illustrated by Tif4631 and Tif4632 (Figure 2 ), which are subunits of the translation initiation complex that binds the cap on the 5' end of mRNAs [ 15 ]. In our analysis they both belong to a class devoted to 'Protein biosynthesis'. Interestingly, they are clustered with other actors of the initiation of translation (Cdc33, Pab1), as well as with proteins involved in cell-wall biogenesis (Kre6, Pkc1, Stt3), thus reinforcing the recent proposal of the existence of a functional link between these two biological processes [ 16 ]. Second, three other pairs of duplicates were recovered in different PRODISTIN classes, relatively far away when considering the tree topology (they therefore no longer share the majority of their interactors), but interestingly, the classes containing the duplicates were dedicated to the same biological process. This is reminiscent of a previous observation we made while studying in detail the rationale sustaining the PRODISTIN clustering [ 11 ]: classes distant in the tree but corresponding to the grouping of proteins involved in the same biological process often correspond to different aspects of the same biological process. This is the case for the pair composed of Tub1 and Tub4 (Figure 2 ), which are classified in different PRODISTIN classes both annotated 'cytoplasm organization and biogenesis' and 'cell cycle' (PRODISTIN classes may be annotated with several cellular functions [ 11 ]). These two proteins are structural components of the cytoskeleton that are implicated in microtubule organization. But strikingly, these two paralogous genes have different roles relative to microtubules. Tub1 is an alpha-tubulin and thus a component of the microtubule itself, whereas Tub4 is a gamma-tubulin involved in the nucleation of the microtubules on both the nuclear and the cytoplasmic sides of the spindle-pole body [ 17 ]. Consequently, the class containing Tub1 is more structural and mainly composed of proteins implicated in microtubule formation, orientation and catabolism (Kar9, Bim1, Pre4), whereas the class containing Tub4 includes actors of the nuclear processes in which the microtubules are involved: chromosome segregation, spindle orientation and nuclear migration (Spc72, Spc97, Spc98, Spc110, Mcm16, Yfr008w, Far3, Vps64, Ylr238w, Ynl127w). Thus, it appears that the PRODISTIN classification of these two paralogous proteins reflects their functions in two different aspects of the same biological process. Finally, nine pairs of duplicated genes were found in different classes devoted to different biological processes. This is exemplified by the case of Ace2 and Swi5 (Figure 2 ), which are two transcription factors regulating the expression of cell-cycle-specific genes. Although they regulate a shared set of genes in vivo , they display different specificities in some cases. Swi5 specifically promotes transcription of the HO gene whereas Ace2 localizes to daughter cell nuclei after cytokinesis, regulates the expression of daughter-specific genes and delays the G1 progression in daughters [ 18 - 20 ]. The PRODISTIN classification was successful in pointing towards these differences as Swi5 and Ace2 localize in different classes annotated for 'transcription' and 'cell cycle', respectively. Indeed, Swi5 is found with Pho2, a transcription factor acting in a combinatorial manner, with which it interacts to regulate HO transcription [ 21 ]. Other Pho2 partners populate the rest of the class. On the other hand, Ace2 partitioned with Mob2 and Cbk1, which form a kinase complex regulating the localization of Ace2 in the daughter cell [ 20 ]. Overall, this analysis shows that the duplicated gene pairs from the ancient duplication present in the tree display three different behaviors in respect of the PRODISTIN classification (Table 2 ). The three groups are populated differently: 63% of the protein pairs are located in the same class, and are therefore involved in the same biological process (behavior I); 7.5% of the duplicated pairs are located in different classes with the same function, therefore suggesting that they are involved in different aspects of the same biological process (behavior II); and, finally, the remaining 22% are implicated in different cellular functions because they are located in different classes devoted to different biological processes (behavior III). We propose considering the three behaviors identified by the PRODISTIN classification as a scale of functional divergence for duplicated pairs. First, the duplicated pairs found in the same class and which essentially have identical interactors would compose the basic level of the scale. This level represents paralogous genes for which cellular function is identical or highly conserved. Higher in the functional scale of divergence are found the duplicates that have different interactors. They are found either in different classes of the same cellular function, thus defining the intermediate level of the functional scale of divergence, or in different classes of different function. This latter case populates the higher level of the scale and represents paralogs for which the cellular function has diverged. The relationship between the functional distance based on annotation and the classification behavior based on protein-protein interactions As noted above, most of the 460 duplicated gene pairs from the ancient duplication were not distinguishable when considering either the functional annotations for Molecular Function or Biological Process as their functional distances based on annotations were mainly equal or close to zero. We have also shown (Figure 1b ) that the subset of 41 paralogous pairs characterized in the PRODISTIN analysis exhibits the same distribution of distance values based on annotations as the 460 pairs. Because the PRODISTIN method allowed us to distinguish three categories of duplicated gene pairs with different types of functional similarities, we wondered if and how the results of the annotation and interaction clustering were correlated. To investigate this, we reported the PRODISTIN behaviours of the paralogs on the distribution of their functional distance based on the Biological Process annotations (Figure 3 ). Among the duplicated pairs that are similarly annotated, we were able not only to distinguish gene pairs found in the same class, as expected for a correlation between the results of the two approaches (behavior I, blue), but also gene pairs involved in different aspects of the same biological process (behavior II, pink) as well as gene pairs not implicated in the same biological processes (behavior III, gray). The last two cases reveal that whereas annotations do not allow us to differentiate certain paralogs from each other functionally, interactions do unveil subtle functional differences. Conversely, paralogous genes may be grouped in the same PRODISTIN class even though their annotations are not completely similar (up to an annotation-based functional distance equal to 0.6). Interestingly, pairs of duplicated genes partitioning into different classes with different functions are encountered independently of the functional distance based on annotation range. This again underlines the fact that the classification based on interactions identifies functional details that are not discernible at the level of annotation only. Therefore, the protein-protein interactions processed by PRODISTIN bring supplementary functional information about the function of the duplicated genes. Sequence evolution versus functional evolution of duplicated genes The availability of 41 yeast paralog pairs for which a pairwise functional comparison can be proposed, offers for the first time the possibility of studying the relationship (if any) between sequence conservation/divergence and evolution of cellular function. Because we have proposed here a three-level scale of possible functional divergence between paralog pairs, what can be said about the sequence-identity patterns shown by protein pairs within and between these three groups? To answer this question, 41 binary sequence comparison analyses were performed (one for each paralogue pair) and the results are displayed according to the classification behavior of the pair identified in the PRODISTIN analysis (Figure 4 ). If paralogs displaying behaviors I, II and III are compared, three observations can be made: first, all gene pairs that show more than 55% sequence identity display behavior I, with one noticeable exception. It is clear, however, that despite the fact that all the protein pairs of this class have been classified by the PRODISTIN analysis as essentially having a conserved function, their degree of sequence identity covers, in a nearly uniform manner, a wide range comprising 16 to 95% sequence identity. Second, and conversely, gene pairs with between 15 and 55% sequence identity are found in all three classes, clearly indicating that neither cellular functional similarity nor divergence can confidently be deduced for paralog pairs with sequence identity falling in this range. Third and strikingly, no clear distinction can be made on the basis of sequence identity between paralogs found in different classes with (behavior II) or without (behavior III) identical functions. In summary, as suggested by a preliminary study [ 22 ], a simple relationship cannot be established between sequence identity and the cellular functional similarity revealed by the interaction-network analysis. So, as previously shown for the annotations, the functional classification based on interactions is able to underline properties of the duplicates that are not discernible when only sequences are compared. Discussion Bioinformatic study of the interaction network as a tool to investigate the function of the duplicated genes We have shown here that studying the cellular interactome using bioinformatics methods leads to a proposal of a functional scale of divergence for yeast duplicated genes. As our work makes use of functional gene annotations and interaction lists, it is important to examine how the quality of these two types of data could potentially affect the conclusions that can be drawn from our studies. Gene annotations provided by the GO consortium [ 12 ] are the result of collaborative work by experts, and all annotations are supported by at least one type of experimental evidence. This, together with the use of a controlled vocabulary consistently applied for all annotations, is in principle a good guarantee of annotation quality. However, several potential problems should be taken into account when using annotations. First, all gene products are not annotated. This is the case for 30% of the pairs of duplicated genes, for which at least one gene is not annotated. Second, annotation errors can propagate in the databases, due to the transfer of annotations from gene to gene based only on sequence or structural similarities. In GO, some functional annotations are "inferred from sequence or structural similarity" (ISS), meaning that the annotation assignment is not supported by experimental evidence per se . It can then can be argued that paralog pairs may be more prone to such annotation transfers than other genes because of their sequence identity. In such a case, our measure of functional distance according to annotations would be largely meaningless. We thus estimated the amount of genes for which GO annotations are solely 'inferred from sequence or structural similarity'. Interestingly enough, they account, at the level of the complete genome, for only 10.3% and 4.95% of the Molecular Function and the Biological Process annotations, respectively. Similar low values are encountered for the 460 pairs of paralogs (11.2% and 4.5%), allowing us to neglect the weight of such inferred annotations in our distance calculation. As far as the quality of interactions is concerned, two main problems result from erroneous (false-positive) interactions and missing (false-negative) interactions. Taking into account that the PRODISTIN method was largely statistically assessed for robustness against the presence of false interactions in our previous study [ 11 ], we can anticipate that the classification behaviors found in the present analysis will be confirmed, or only slightly modified, in the near future when new interactions are discovered. The ancestral yeast genome duplication as a case study for functional evolution of paralogs In the present analysis, we worked solely on pairs of paralogs that supposedly originated from the ancient WGD [ 6 , 7 ]. This choice was made for several reasons. First, after the yeast WGD hypothesis, we can consider that all genes, remnants from this event, have duplicated simultaneously. This sets a 'time 0' for the duplication event and therefore enables us to avoid the problem of determining the age of the duplication events, a problem inherent in all genome-wide analyses of paralogs. Second, after a WGD, polyploidization preserves the necessary stoechiometric relationships between gene products, while the duplication of a single gene does not: duplicates are then out of balance with their interacting partners. This is an important parameter to consider when one wants to study the evolution of the duplicated genes through the analysis of interactions, as we did in this work. Third, studying the remnants of a WGD after more than 100 million years [ 7 , 23 ] allows one to estimate how the sequence, function and interactors of the paralog gene products have evolved since their origin, when their sequence, function(s) and interactor(s) were identical. An important issue for the interpretation of our results is the validity of the hypothesis of the existence of a WGD in S. cerevisiae . Initially proposed by Wolfe and Shields [ 7 ], the WGD model has been controversial and alternative models of local duplications have been proposed [ 24 - 27 ]. Very recently, a novel proof of WGD was provided [ 8 ]. Among the 460 paralog pairs we studied, 362 were shown by this new analysis to arise from the WGD. Revisiting our results to take into account the new dataset of duplicated genes did not change them drastically. The distribution of the duplicated pairs becomes 68, 4.5 and 18% for the three different categories of classification behaviors (I, II, III), respectively, compared to 63, 7.5 and 22% for the dataset we used (Table 2 ). The evolution of cellular function: from the scale of functional divergence to the evolutionary fates of the duplicated genes Our study was driven by the idea that investigating the cellular rather than the molecular function of the duplicated genes might provide new information about the extent of their actual divergence and, consequently, might help us to envisage how their cellular function has evolved since the duplication event. Indeed, the first important outcome of our study, based on the comparison of annotations for duplicated pairs, is that although both the molecular and cellular functions of the majority of protein pairs have been conserved since the date of the WGD, cellular functions have evolved more rapidly than molecular functions. Although this finding could seem rather intuitive, it is, to the best of our knowledge, the first time that evidence has been proposed in its favor. Conservation of the same molecular function for two duplicated proteins while allowing the diversification of their cellular functions may represent a simple and economical way of introducing functional diversity and complexity in a controlled manner during evolution. This may be the result of a change in interaction partners and/or subcellular localization. The second important result of our study is that since the date of the ancient WGD, cellular functions have evolved at variable rates, since a scale of functional divergence can be detected. In this respect, we propose to interpret this functional scale of divergence in the light of different theoritical evolutionary scenarios for cellular function. First, the first level of the functional scale (behavior I) may contain duplicates which have been conserved as such, because keeping two copies may confer an evolutionary benefit on the cell (for instance, Rps26A/Rps26B; Table 1 ). Second, we propose that the majority of the paralog pairs populating the two first levels of the functional scale of divergence based on interactions (behaviors I and II) evolved functionally according to the duplication-degeneration-complementation (DDC) or subfunctionalization model proposed by Force et al . [ 28 ]. This predicts that duplicated genes are preserved by the partitioning of the function(s) of the ancestral gene between the two duplicates. This may happen, for instance, by the complementary loss of regulatory elements or the modification of the coding regions. Even though our analysis does not pretend to reveal the molecular mechanisms by which the subfunctionalization of the duplicated pairs has occurred, several lines of evidence sustain our proposal. First, the first level of the functional scale is populated by paralog pairs, which have kept their interactors identical or still share common interactors. This is in good agreement with a situation in which duplicates have slightly diverged by subfunctionalization to form two subunits of a same complex (for example, Tif4631/Tif4632, Rfc3/Rfc4, Yck1/Yck2; Table 1 ) or to increase the complexity of a signaling pathway (for instance, Mkk1/Mkk2; Table 1 ). Second, the intermediate level of the functional scale of divergence (behavior II) contains paralog pairs that do not have the same interactors but have still conserved their cellular function(s) since the duplication event. They may represent paralog pairs involved in different aspects of the same biological process (see Results and [ 11 ]) and/or pairs for which the spatio-temporal regulation has evolved by subfunctionalization, therefore implying a new cast of interactors. Finally, the third level of the functional scale (behavior III) may correspond to duplicates that have evolved by neofunctionalization, as not only their interactors are different but they are also involved in different cellular processes (for instance, Swi5/Ace2). These genes may illustrate Ohno's theory [ 3 ] of the emergence of new functions from gene duplication events. Even though we have shown here that there is no simple relationship between sequence identity and cellular function, it is interesting to note that data newly generated by Kellis et al . [ 8 ] strengthen our proposal. Indeed, the frequency of pairs showing accelerated protein evolution is almost twice as high among the paralog pairs displaying behavior III (37.5% (3/8) of the pairs common to both studies) than among pairs with the same function (20% (5/25) of the pairs common to both studies with behaviors I and II). Overall, these results corroborated our proposal. Conclusions Most network analyses carried out up to now either emphasized the prediction of function for uncharacterized proteins [ 29 , 30 ] or, in the frame of evolutionary studies, estimated the rate of evolution of proteins according to their number of interactors [ 31 ] and addressed the issue of the link between protein dispensability and rate of protein evolution [ 32 , 33 ]. As far as we know, this work constitutes the first attempt to address the functional evolutionary fate of duplicated genes using a bioinformatic analysis of the protein-protein interaction network in which the products of these genes are involved, and to provide detailed protein function comparisons based on interaction data. Our approach might thus provide a new way to analyze the evolution of the function of duplicated genes in different organisms. A limitation of this type of analysis is the present knowledge of interaction networks. Even in a well-studied organism such as Saccharomyces cerevisiae , less than 10% of the gene pairs, remnants of the WGD, are amenable to such a detailed analysis. As our knowledge on interaction networks is increasing and as more interactions become available, we can expect to improve both the coverage of duplicated pairs of interactors and the relevance of the functional clusters found by the PRODISTIN method. Finally, it should be emphasized that the study of evolutionary processes greatly benefits by being approached using different tools not only at the sequence level, as is usual, but also directly at the functional level. In the case of the study of the 41 paralog pairs reported here, functional conclusions inferred from the sequence level would have been incomplete and even erroneous in several instances. Materials and methods Functional distance based on GO annotations GOproxy [ 13 ], a tool that calculates the Czekanowski-Dice distance between gene annotations was used to compare the GO annotations [ 12 ] of the duplicated gene products as well as that of five datasets of 460 pairs of proteins randomly selected from the yeast genome. The Molecular Function, Biological Process and Cellular Component ontologies were processed separately. The Czekanowski-Dice distance formula used in the algorithm is: Dist( i,j ) = number of (Terms( i ) ΔTerms( j ))/ [number of (Terms( i ) ∪ Terms( j )) + number of (Terms( i ) ∩ Terms( j ))], in which, i and j denote two genes, Terms( i ) and Terms( j ) are the lists of their GO terms and Δ is the symmetrical difference between the two sets. This distance formula increases the weight of the shared GO terms by giving more weight to similarities than to differences. The GOToolBox website can be accessed at [ 13 ]. Protein-protein interaction dataset The protein-protein interaction dataset we investigated contains a total of 4,143 selected interactions involving 2,643 proteins. We updated our former dataset [ 11 ] with 1,244 new interactions taken from the Munich Information Center for Protein Sequences (MIPS) [ 34 ] and from the literature. As previously, only direct binary interactions were selected according to the method used for their identification (two-hybrid experiments, in vitro binding, far western, gel retardation and biochemical experiments). PRODISTIN analysis PRODISTIN, a computational method we recently proposed [ 11 ], was used to analyze the protein-protein interaction dataset. Starting with a binary list of interactions, only proteins involved in at least three binary interactions were selected for further classification (because poorly connected proteins have a higher chance of being involved in false-positive interactions). A graph in which vertices are proteins and edges correspond to the relation 'interact with and/or share at least one common interactor' was computed and the Czekanowski-Dice distance was calculated between all possible pairs of proteins belonging to the connected component of this graph (using the formula above and applying it to the list of protein interactors instead of the list of GO terms). The distance matrix was then clustered using BioNJ [ 35 ] and the tree was visualized using TreeDyn [ 36 ]. PRODISTIN classes corresponding to the largest possible subtree composed of at least three proteins sharing the same functional annotation and representing at least 50% of the individual class members for which a functional annotation is available were detected in the tree. GO annotations corresponding to the Biological Process ontology were used for this purpose. Given that GO is organized as a DAG, proteins may be annotated at different levels of the ontology. Our goal was to analyze subtrees regarding to the proteins commonly annotated as participating in them, so we considered annotations for all proteins at a specific level of the ontology. We chose to work at level 4 because we estimated, on previous experience using the Yeast Proteome Database [ 37 ] system of annotation, that this particular level provides a good representation of the complexity of cellular functions. For this, we used GODiet, a tool enabling us to restrict the list of GO terms to a given depth in the ontology [ 13 ]. Sequence analysis Pairwise sequence alignments were carried out on the set of 460 pairs of duplicated protein sequences using the Needleman-Wunsch (global alignment) algorithm. The program used is available at [ 38 ]. The chosen alignment matrix was BLOSUM50, and the gap-opening and gap-extension penalties were set to 12 and 2, respectively. The resulting 460 alignments have been processed to calculate the percent identity for each protein pair. Additional data files The following additional data are available with the online version of this paper. Additional data file 1 contains the expectation values for the distribution of functional distances based on the GO annotations. Additional data file 2 contains details of the 123 PRODISTIN classes contained in the classification tree. Supplementary Material Additional data file 1 The expectation values for the distribution of functional distances based on the GO annotations Click here for additional data file Additional data file 2 Details of the 123 PRODISTIN classes contained in the classification tree Click here for additional data file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545596.xml

554114

The Bosnian version of the international self-report measure of posttraumatic stress disorder, the Posttraumatic Stress Diagnostic Scale, is reliable and valid in a variety of different adult samples affected by war

Background The aim of the present study was to assess the internal consistency and discriminant and convergent validity of the Bosnian version of a self-report measure of posttraumatic stress disorder (PTSD), the Posttraumatic Stress Diagnostic Scale (PTDS). The PTDS yields both a PTSD diagnosis according to the Diagnostic and Statistical Manual of Mental Disorders 4 th edition (DSM-IV) and a measure of symptom severity. Methods 812 people living in Sarajevo or in Banja Luka in Bosnia-Herzegovina, of whom the majority had experienced a high number of traumatic war events, were administered the PTDS and other measures of trauma-related psychopathology. The psychometric properties of the instrument were assessed using Cronbach's alpha and principal components analysis, and its construct validity was assessed via Spearman correlation coefficients with the other instruments. Results The PTDS and its subscales demonstrated high internal consistency. The principal components revealed by an exploratory analysis are broadly consistent with the DSM-IV subscales except that they reproduce some previously reported difficulties with the "numbing" items from the avoidance subscale. The construct validity of the PTDS was supported by appropriate correlations with other relevant measures of trauma related psychopathology. Conclusion The Bosnian version of the PTDS thus appears to be a time-economic and psychometrically sound measure for screening and assessing current PTSD. This self-report measure awaits further validation by interview methods.

Background To obtain a diagnosis of PTSD and an estimation of PTSD severity a wide range of measures either relying on interviews or self-report exist in many languages. However, most of the relevant validation studies for these instruments were carried out for English-language versions [ 1 ]. For many languages, validated instruments do not exist. A standard approach in this situation is to translate one of those English-language instruments which are well validated, to carry out a validation study for the translation and to compare the results of the validation study with the studies for the original. Self-report instruments have several advantages as compared to interview measures. They are relatively economic in terms of administration and demand minimal clinician time. If clinicians are not familiar with psychiatric diagnostic procedures and especially the clinical diagnosis of PTSD, it is more advisable to use a psychometrically sound self-report measure which is less prone to mistakes than interview measures. A good self-report measure for PTSD should allow a diagnosis of PTSD as well as an estimation of PTSD severity and should conform to the DSM-IV criteria for PTSD [ 2 ]. The English version of the Posttraumatic Stress Diagnostic Scale [PTDS; [ 3 ]] fulfils these criteria and has been shown to have adequate psychometric properties. The PTDS has been translated into a German version which also has adequate psychometric properties [ 4 ]. These two different language versions of the PTDS have been used in numerous studies [e.g. [ 4 - 9 ]]. Table 1 provides an overview over the internal consistency and the test-retest-reliability of the PTDS as published in the literature. Table 1 Internal consistency and test-retest reliability of the PTDS and its DSM-III-R precursor the PSS-SR Authors Samples Scales Cronbach's alpha Test-retest reliability Foa, Riggs, Dancu, & Rothbaum (1993) [8] 44 women (rape and other non-sexual attack) Total score .91 .74 after one month (N = 29) 5 – 6 weeks after the event Reexperiencing .78 .66 Avoidance .80 .56 Hyperarousal .82 .71 Engelhard et al (2001) [7] 113 women after miscarriage Total score .87 Stieglitz, Frommberger, Foa, & Berger (2001) [9] 152 persons: Total score .85 and .86 .60 after six months 1. time point: a few days after accident Reexperiencing .75 and .82 .39 Avoidance .56 and .74 .53 2 time point: 6 months later Hyperarousal .75 and .64 .47 Foa, Cashman, Jaycox, & Perry (1997) [3] 284 victims of various traumatic experiences Total score .92 .83 (approx 2 weeks later) Reexperiencing .78 .77 Avoidance .84 .81 Hyperarousal .84 .85 Kappa = .74 for PTSD diagnosis In terms of convergent validity, Foa, Riggs, Dancu, and Rothbaum [ 8 ] compared Posttraumatic Symptom Scale scores (PSS-SR; the DSM-III-R version of the PTDS) with the diagnosis obtained by administering the Structured Clinical Interview for the DSM-III-R [SCID;[ 10 ]]. 86 % of the participants with a PTSD diagnosis according to DSM-III-R criteria were correctly identified with the self-report instrument. The sensitivity was 62% and the specificity 100%. The DSM-IV version of the PTDS achieved a sensitivity of .89 and a specificity of .75. Percentage agreement between SCID and PTDS diagnosis was 82 % and kappa was .65. Overall, the criterion validity of the PTDS with respect to SCID was encouraging. Table 2 provides an overview of convergent and divergent validity for the PTDS and some other self-report measures of trauma related psychopathology. Table 2 Convergent and divergent validity of the PTDS and its DSM-III-R precursor PSS-SR Authors PSS/PTDS Scales IES Total score IES Intrusion IES Avoidance BDI Foa et al. (1993) [8] PSS Total score .81 .53 .80 Reexperiencing .81 .47 .66 Avoidance .71 .52 .73 Hyperarousal .70 .45 .75 Stieglitz et al. (2001) [9] PSS Total score .67 & .65 .61 & .57 .61 only at first measurement (a few days after the accident) Reexperiencing .63 & .59 .53 & .47 .45 Avoidance .56 & .55 .50 & .51 .50 Hyperarousal .52 & .49 .47 & .45 .60 Foa et al. (1997) [3] Total score .78 .80 .66 .79 Reexperiencing .68 .77 .51 .67 Avoidance .75 .72 .69 .77 Hyperarousal .70 .74 .58 .73 The symptom items of the PTDS, which reflect more or less verbatim the corresponding items in the DSM-IV criteria, in empirical studies do not necessarily fall into the three groups explicit in DSM-III-R and DSM-IV. The results of a number of factor-analytic studies suggest that the avoidance symptoms load on two separate factors [ 11 - 13 ]. One factor captures wilful and effortful avoidance and the other factor captures involuntary strategies of "shutting down" the emotional system when effortful strategies fail, which thus may load together on the same factor as hyperarousal symptoms. This issue is to be borne in mind when examining the structure of instruments intended to measure PTSD symptoms according to DSM-IV. Because of the many advantages of the PTDS we decided to use it for estimating rates of PTSD in a series of studies in different samples of war-traumatized inhabitants of Sarajevo and Banja Luka, Bosnia and Herzegovina. The results of these studies have been published elsewhere or are still in the process of being published [ 14 - 16 ]. The PTDS had to our knowledge never been used before in the area of former Yugoslavia; instead, many studies have used similar but more or less ad-hoc constructed checklist versions of the DSM-IV criteria. The introduction of the PTDS would therefore mean providing clinicians and researchers with a sound Bosnian version of an internationally accepted PTSD self-rating instrument. The goal of this paper is to report first results of the psychometric evaluation of the Bosnian PTDS. Methods Diagnostic assessment Although all applied measures are questionnaires, not all subjects proved literate enough to complete them on their own. Therefore in some cases the interviewers had to read some of the questions to them and sometimes to reread or reformulate the questions. Thus the administration deviated slightly from the standard procedures. The instrument under assessment was the Posttraumatic Stress Diagnostic Scale [ 3 , 17 ] which allows, as mentioned before, a diagnosis of PTSD as well as an estimation of symptom severity. The PTDS consists of four parts. Part 1 has 12 items in the original and asks about possible traumatic events (A1 criterion of DSM-IV). In part 2 the time of occurrence of the "most upsetting" event, together with the respondent's assessment of whether the event was life-threatening and whether it was accompanied by feelings of helplessness and intense fear are all evaluated (A2-criterion). Part 3 asks about symptoms of reexperiencing (5 items; criterion B), avoidance (7 items, criterion C), and arousal (5 items, criterion D). Part 4 explores the duration of the disturbance (criterion E) and the consequences of the symptomatology for important areas of functioning (criterion F). Since the original PTDS was designed for a civilian population in times of peace we replaced part 1 with a checklist of traumatic events specific to the war in Bosnia and Herzegovina 1992–5, the Checklist of War Related Experiences, CWE, the items of which are reproduced in Appendix 1. (The checklist also included other significant life events relevant to life in post-war Bosnia-Herzegovina. As these items are not relevant to this study, they are not discussed here.) To obtain a Bosnian version we applied the procedures suggested by Vijver and Hambleton for the translations of psychological assessment measures [ 18 ]. That is, we performed an alternating procedure of translations and back-translations until no significant differences could be detected. In a second step we field-tested the resulting pilot versions to further check the appropriateness of the wording to the Bosnian language and the cultural context. The resulting modifications were then back-translated again. The Impact of Event Scale [IES; [ 19 ]] is a questionnaire which assesses the frequency of intrusion and avoidance phenomena as a consequence of experiencing a particular event. In the more than 20 years since its publication it has very frequently been used to diagnose PTSD; however, that is neither the intended nor an appropriate use for it. The IES consists of 15 items each to be answered on a four-point scale assessing the frequency of the occurrence of stress reactions in the preceding week (0 = not at all; 1 = occasionally; 3 = sometimes; 5 = frequently). This means that total scores for the IES range between 0 and 75, with higher scores indicating more frequent intrusion and avoidance reactions. The IES has been applied in nearly every kind of traumatisation [for an overview, see [ 20 ]] and has been translated into many languages. The IES is one the most frequently used traumatic stress questionnaires internationally. The version used in the present study was almost identical to one which has been used in other studies in the region during and after the war and which has since been subject to a validation study [ 21 ] and found to have satisfactory factor structure and reliability. The Symptom Checklist-90-R [SCL-90-R; [ 22 ]] is a 90 item self report questionnaire for measuring subjective psychological and somatic stress in the preceding seven days. Like the IES, the SCL-90-R is used widely internationally and has been used in a large number of research projects in a very wide variety of applications [for an overview, see [ 23 ]]. The SCL-90-R consists of nine scales and three global indices, of which the GSI, the Global Severity Index, is the most widely used. Beck Depression Inventory (BDI) The Beck Depression Inventory [BDI; [ 24 ]] is probably the best documented self-report method of measuring the intensity of depression [ 25 , 26 ]. By 1998 more than 2000 studies had been published using the BDI [ 27 ]. The current, revised, version consists of 21 items whose scores vary between 0 and 3 [ 24 ]. Zero indicates that the symptom is not present whereas three indicates the most extreme level of symptoms. Clients are instructed to report on how they felt in the preceding seven days. Samples The following data was collected between February 1998 and October 1999 in Sarajevo, Banja Luka and Prijedor, which are all in Bosnia-Herzegovina. Sarajevo is in the Federation of Bosnia and Herzegovina, namely that part of Bosnia and Herzegovina which has a predominantly Muslim and Catholic population, and Banja Luka and Prijedor are in the other part, the Republika Srpska, which is predominantly Serbian Orthodox. The samples were stratified by age and sex. The number of years of schooling was also recorded. All subjects participated voluntarily and gave fully informed consent. Table 3 shows sampling procedures, region, and numbers for each sub-sample included in the following analysis. Table 4 provides a description of the demographics. Table 3 Overview of samples used Sample Region Sampling procedure N A 1998 Sarajevo randomised via maps of Sarajevo area 98 B 1998 Sarajevo admission to psychological treatment 114 C 1998 Sarajevo admission to medical treatment 99 D 1999 Sarajevo randomly selected repatriates to B&H from lists held by local councils 103 E 1999 Sarajevo randomly selected displaced or formerly displaced persons from lists held by local councils 97 F 1999 Banja Luka randomly selected subjects who stayed in the Banja Luka throughout the war, selected via maps of area 100 G 1999 Banja Luka randomly selected returned displaced persons, selected from lists of residents 100 H 1999 Prijedor randomly selected from lists of residents in collective centres 100 Table 4 Sample description N Minimum Maximum Mean Std. Deviation years of education 809 8.00 16.00 11.72 2.50439 age 812 16.00 68.00 37.89 13.78230 N % Missing sex female 426 52.5 % male 386 47.5 % Total 812 100.0 % 0 employment status unemployed or waiting list 178 21.9 % other (housewife, student) 360 44.3 % employed 274 33.7 % Total 812 100.0 % 0 family status single 363 44.8 % married or long-term relationship 447 55.2 % Total 810 100.0% 2 Other 70 8.7 % religion Islam 383 47.3 % Catholicism 45 5.6 % Orthodox 311 38.4 % Total 809 100.0 % 3 In total 812 persons participated. Inclusion criteria for all were a) age between 16 and 65, b) not suffering from a psychotic disorder and c) literate enough to answer the questionnaires with help. All subjects completed the PTDS and the SCL-90-R; therefore correlations for these subscales are based on the data of all the subjects. However for reasons of economy, in 1999 the full package of questionnaires including the BDI and IES were only administered to a random selection of participants in only the two Sarajevo sub-samples. All other participants in 1999 only answered a smaller package of questionnaires including the PTDS. Correlations between the PTDS and BDI and IES are therefore based on a smaller dataset. In 20 cases an entire instrument was missing, as detailed in table 5 . In the remaining cases, the number of individual missing values for individual items was small (much less than 5%), so it was deemed acceptable to form the total scores for the scales simply by multiplying the mean item score for each individual, allowing for any missing items, by the total number of items on each scale. So in the case of the inter-scale correlations the Ns are merely reduced by the number of completely missing questionnaires. In the case of the reliability analyses for the subscales of the PTDS, instruments with any missing items on the scale in question were excluded from the analyses, in each case slightly reducing the Ns. Table 5 Details of which instruments were given to which sub-samples BDI IES PTDS SCL not given missing available not given missing available missing available missing available total 1998 samples, Sarajevo non-displaced random sample 98 2 96 1 97 1 97 98 non-displaced medical treatment 1 98 4 95 5 94 1 98 99 non-displaced psychological treatment 114 1 113 114 114 114 1999 samples, Sarajevo returnees from outside Former Yugoslavia 40 64 40 62 1 103 2 102 104 displaced or former displaced 21 76 21 1 75 97 97 97 1999 samples, Banja Luka and Prijedor Banja Luka displaced or former displaced 100 100 100 100 100 Banja Luka non-displaced 100 100 100 100 100 Prijedor displaced in camps 100 100 100 100 100 Table Total 361 1 450 361 8 441 7 805 4 808 812 Interviewers The medical and psychological samples were assessed through a total of 15 experienced counsellors/therapists, who were working at a variety of clinics and counselling centres in Sarajevo. All other samples were assessed by pairs of final year and third year students of Psychology at Sarajevo University and Banja Luka University. All interviewers were trained in the use of the questionnaires. Two pilot studies were performed to insure the appropriate use of the assessment. During the studies constant supervision for all interviewers was provided. Statistical analysis To obtain an estimation of internal consistency Cronbach's alpha was calculated for the total scores and the subscales of the PTDS. Convergent and divergent validity were estimated by using Spearman correlations between the scales. Spearman correlations were used because most of the distributions were not normal. For the principal components analysis, oblimin oblique rotation was used. Results and discussion The standardised Cronbach's alphas for the Bosnian PTDS were .93 for the total symptom score, .89 for the reexperiencing subscale, .84 for the avoidance subscale and .84 for the arousal subscale. The results correspond well with other published results. The Spearman's correlations between the total scale and the subscales were all quite high at .89, .93 and .87 for re-experiencing, avoidance and hyperarousal respectively; re-experiencing correlated .74 and .67 with avoidance and hyperarousal; and the correlation between avoidance and hyperarousal was .72. The item characteristics for the symptom items and subscale totals are shown in table 6 . The characteristics are acceptable, with the lowest standard deviation being .77 for the item about not being able to remember details, which also had the lowest mean (.36 on a scale of 0 to 4). Table 6 Item characteristics of the PTSD symptom items of the Bosnian PTDS sex female male total Mean Standard Deviation Mean Standard Deviation Mean Standard Deviation B1 intrusions 1.00 1.12 .76 1.00 .89 1.07 B2 bad dreams .73 1.02 .57 .92 .65 .98 B3 reexperiencing .70 .99 .53 .86 .62 .94 B4 upset after remembering 1.14 1.07 .89 1.01 1.02 1.04 B5 physical reaction after remembering .95 1.09 .62 .92 .79 1.02 C1 attempt not to think about it 1.14 1.16 .83 1.06 .99 1.12 C2 avoiding places people .86 1.13 .65 1.03 .76 1.09 C3 not being able to remember details .40 .80 .33 .74 .36 .77 C4 less interest in activities .66 .96 .51 .89 .59 .93 C5 detachment estrangement .65 1.00 .50 .93 .58 .97 C6 restricted affect .81 1.07 .50 .87 .66 .99 C7 foreshortened future .79 1.08 .58 .95 .69 1.02 D1 difficulty falling or staying asleep .92 1.11 .63 .96 .78 1.05 D2 irritability .70 .94 .55 .88 .62 .92 D3 difficulty concentrating .92 1.02 .68 .92 .80 .98 D4 hypervigilance .55 .88 .40 .77 .48 .83 D5 exaggerated startle response .75 .99 .42 .81 .60 .93 total score on subscale b (reexperiencing) 4.53 4.43 3.36 3.94 3.97 4.24 total score on subscale c (avoidance) 5.32 5.02 3.89 4.79 4.64 4.96 total score on subscale d (arousal) 3.82 3.77 2.67 3.44 3.28 3.66 total score on all symptom subscales 13.66 11.73 9.93 10.84 11.88 11.46 The items were scored on a scale of 0 (not at all or once a month) to 4 (5 or more times a week /almost always). The items from the symptom subscales were submitted to a principal components analysis with oblimin oblique rotation. Factors with eigenvalues greater than 1 were retained. Items were considered as belonging to a factor if their loadings on that factor were above 0.4. (see table 7 ). The first solution had three factors explaining a total of 61.41% of the variance and was deemed to be satisfactory, so that no further solutions were sought. The first factor, which explains 47.64% of the variance, was labelled Arousal / Numbing. It contains all the items from the DSM-IV arousal scale and three DSM-IV avoidance items, two of which (detachment/estrangement and restricted affect) are also associated with numbing [ 11 ]. The second factor, explaining 7.85% of the variance, was labelled Intrusion and includes all the items from the DSM-IV intrusion scale together with one item ("attempting not to think about it") from the DSM-IV avoidance scale. The third factor, which explains 5.92% of the variance, was labelled Avoidance. It contains all the items from the DSM-IV avoidance scale except for two items which load on Arousal/Numbing. Every item loaded on at least one factor and only two items loaded on more than one factor (the item "attempt not to think about it" loaded on the Intrusion and Avoidance factors, and the item "detachment, estrangement" loaded on the Arousal/Numbing and Avoidance factors). Table 7 Rotated factor pattern of the PTSD symptom items of the Bosnian PTDS Loadings Symptom Factor 1: Arousal / Numbing Factor 2: Intrusion Factor 3: Avoidance b1 intrusions .031 -.824 .050 b2 bad dreams .127 -.779 -.049 b3 reexperiencing .189 -.704 -.047 b4 upset after remembering -.042 -.830 .079 b5 physical reaction after remembering .084 -.729 .074 c1 attempt not to think about it -.163 -.480 .544 c2 avoiding places people -.080 -.256 .666 c3 not being able to remember details .103 .014 .649 c4 less interest in activities .209 -.105 .511 c5 detachment, estrangement .567 .089 .438 c6 restricted affect .523 -.002 .397 c7 foreshortened future .596 .056 .326 d1 difficulty falling or staying asleep .460 -.361 .063 d2 irritability .652 -.199 -.031 d3 difficulty concentrating .732 -.123 -.017 d4 hypervigilance .753 -.072 -.122 d5 exaggerated startle response .746 -.065 -.003 Factor loadings greater than 0.40 are shown in bold underline. In short, the three DSM-IV scales can be broadly identified, except that three DSM-IV avoidance items including two of the somewhat contentious numbing items load on the arousal scale, which replicates well the findings reported above [ 11 - 13 ]. Table 8 provides the correlations between the various other measures of psychopathology and the Bosnian PTDS. With samples of this size, correlations even as small as approximately .1 are significant, so all the correlations are highly significant and thus the significances are not reported here. Table 8 Convergent and divergent validity of the Bosnian PTDS IES Total IES Intrusion IES Avoidance BDI SCL GSI PTDS Total Spearman's rho .709 .703 .619 .622 .568 N 439 438 439 444 802 Reexperiencing Spearman's rho .634 .687 .491 .485 .437 N 439 438 439 444 802 Avoidance Spearman's rho .651 .603 .610 .581 .515 N 439 438 439 444 802 Hyperarousal Spearman's rho .573 .574 .493 .596 .595 N 439 438 439 445 803 The correlations between the PTDS and the IES are somewhat lower than in the two American publications, closer to those in the German article. Re-experiencing on the PTDS correlates higher with intrusion than with avoidance on the IES, and avoidance on the PTDS correlates higher with avoidance on the IES than with intrusion on the IES, all of which are desirable results in that they support construct validity. The correlations between the re-experiencing and avoidance scales of the IES and the avoidance scale of the PTDS are quite similar, possibly indicating weak specificity of the latter, which was however also the case for all except the oldest of the three previous studies. The correlation between the BDI total and the PTDS/PSS total is high, as reported in the literature. In fact the Bosnian version seems to differentiate a little better between PTSD and depression than do the American and German versions; nevertheless the specificity is still quite weak. In the same way there are also quite high correlations with the SCL-90-R. Although the Bosnian version of the BDI and SCL have also not been adequately validated before, validating one new instrument against other instruments which are also not validated is not a meaningless affair but on the contrary the only possible procedure in a situation such as the one we (and our local and international researcher colleagues) found ourselves in, namely that very few world-standard instruments existed. If one does find, as we did, inter-instrument correlations similar to those for the corresponding instruments in other languages then that provides at least some provisional evidence for the psychometric quality and construct validity of all of those instruments. One of the main uses of the PTDS is to provide a PTSD diagnosis in an economical way. As the PTDS assesses in questionnaire form all the information necessary for the diagnosis according to DSM-IV, the PTDS prevalences can be easily calculated and are in fact 24.72% for the whole sample, 31.37% for women and 17.40% for men. The most important factor which restricts the interpretation of these results is that the PTDS was not compared with clinical interview, which would have been standard procedure in this kind of study. However, when we began the study there was no suitable validated interview available in the Bosnian language, which meant that we would have had to translate and extensively validate such an interview ourselves, and again we would have run into the problem of validating the interview against instruments which had also not been validated at that time. It also should be stressed that this study says very little about the cultural or contextual validity of the instrument or the construct PTSD which it is intended to measure. On the other hand, the samples are quite large and taken together quite heterogeneous, and the selection methodologies in each case provided a reasonable approximation to randomness, so that all in all the data can be considered to be of good quality. Conclusion In conclusion it can be said that the psychometric properties of the Bosnian version of the PTDS are as good as those published for other languages. The internal consistencies are at least as good and the Bosnian version appears even to distinguish a little better than the American and German versions between PTSD as measured by the IES and depression as measured by the BDI. The principal components revealed by an exploratory analysis are broadly consistent with the DSM-IV subscales except that they reproduce some previously reported difficulties with the "numbing" items from the avoidance subscale; this issue might explain the poor specificity of the avoidance scale with respect to the IES subscales. None of the analyses revealed anything unusual or indicated problems either with the translation or with the application of the concepts inherent in the instrument to the post-war Bosnian population, all of which indicates that the Bosnian PTDS can be given the green light for further application in the future. Yet our results are only a necessary first step in the validation of the applied measures; a comparison with a validated translation of a Bosnian interview measure for PTSD still needs to be done. Competing interests The author(s) declare that they have no competing interests. Authors' contributions RR participated in the design of the study, and drafted the manuscript. SP carried out the actual study and performed the statistical analysis. Both authors worked on and approved the final manuscript. Appendix 1 The war traumatic event items of the Checklist of War Events (which replaces the standard traumatic event checklist in the PTDS) group 0: injury to self Were you severely injured during the war? group 1: sexual violence to self Were you raped or sexually assaulted during the war? During the war, were you sexually assaulted by a member of your close family who had been forced to do that? During the war, were you sexually assaulted by a member of your close family who was not forced to do that? group 2: torture to self Were you tortured during the war? group 3: other threat to self During the war, were you in a situation in which you strongly believed you would be severely injured or killed? During the war, did a bullet come so close to you that you could have been severely injured or killed? During the war, did a bomb or grenade explode so close to you that you could have been severely injured or killed? During the war, did anyone threaten to kill you or severely injure you? Were you captured or held in a detention camp during the war? During the war, were you without food or water for so long that you strongly believed you would die? During the war, were you so cold that you strongly believed you would die? During the war, did you stay in a cellar longer than 3 weeks without a break? During the war, were you assaulted in a non-sexual way by a member of your close family who had been forced to do that? During the war, were you assaulted in a non-sexual way by a member of your close family who had not been forced to do? Were you in the army during the war? During the war, were you seriously ill because of the war (e.g. heart attack) group 4: witnessed: loved ones Did you eyewitness a loved one being killed during the war? Did you see dead body of a loved one who had been killed in the war? (excluding funerals) Did you see a loved one being tortured or physically assaulted during the war? Did you see a loved one being sexually assaulted during the war? Did you touch a loved one who had been killed or wounded in the war? During the war, did you see a loved one who was severely injured before he/she received medical help? group 5: witnessed: others Did you eyewitness somebody being killed (not a loved one) in the war? Did you see the body of a person (but not a loved one) who had been killed in the war? (excluding funerals) Did you see someone being tortured or physically assaulted during the war (but not a loved one)? Did you see someone being sexually assaulted during the war (but not a loved one)? Did you touch someone (but not a loved one) who had been killed or wounded in the war? During the war, did you see a severely injured person (not a loved one) before they received medical help? group 6: losses, nuclear family Was your father killed in the war? Was your mother killed in the war? Was your spouse killed in the war? Was a child of yours killed in the war? Was a brother or sister of yours killed in the war? group 7: losses, other loved ones Was a close relative of yours killed in the war? Was a close friend of yours killed in the war? group 8: threat, violence, injury to loved ones Was a loved one in the army during the war? Was a loved one severely injured in the war? Was a loved one raped or sexually assaulted in the war? Was a loved one tortured in the war? Was a loved one captured or held in a concentration camp during the war? During the war, was a loved one seriously ill (e.g. cancer or heart attack) or had some chronic health problem? group 9: other war events Other traumatic event since 1991 due to war: 1 Other traumatic event since 1991 due to war: 2 Other traumatic event since 1991 due to war: 3 group 10: other events since 1991 not related to war Did a loved one die in the war for reasons unrelated with the war? (Other stressful and traumatic events since 1991 and unrelated to the war) Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC554114.xml

514524

Characterisation of the Escherichia coli membrane structure and function during fedbatch cultivation

Background Important parameters during recombinant protein production in Escherichia coli, such as productivity and protein activity, are affected by the growth rate. This includes the translocation of protein over the membrane to gain better folding capacity or reduced proteolysis. To vary the growth rate two techniques are available: fedbatch and continuous cultivation, both controlled by the ingoing feed rate. Results During fedbatch cultivation, E. coli contains phosphatidylethanolamine, phosphatidylglycerol, cardiolipin and saturated fatty acids in amounts which are stable with growth rate. However, the levels of cardiolipin are very high compared to continuous cultivation. The reason for fedbatch triggering of this metabolism is not known but hypothesised to result from an additional need for carbon and energy. The reason could be the dynamic and sometimes rapid changes in growth rate to which the fedbatch cell has at all times to adjust. The membrane flexibility, essential for translocation of various components, is however to some degree sustained by production of increased amounts of unsaturated fatty acids in phosphatidylglycerol. The result is a functionally stiff membrane which generally promotes low cell lysis and is constant with respect to protein leakage to the medium. At comparatively high growth rates, when the further stabilising effect of cyclic fatty acids is gone, the high level of unsaturated fatty acids results in a pronounced effect upon sonication. This is very much in contrast to the membrane function in continuous cultivation which shows very specific characteristics as a function of growth rate. Conclusions The stiff and unchanging fedbatch membrane should promote a stable behaviour during downstream processing and is less dependent on the time of harvest. However, optimisation of protein leakage can only be achieved in the continuously cultivated cell where leakage is twice as high compared to the constant leakage level in fedbatch. If leakage is undesired, continuous cultivation is also preferred since it can be designed to lead to the lowest values detected. Induction at low growth rate (<0.2 h -1 ) should be avoided with respect to productivity, in any system, since the specific and total protein production shows their lowest values at this point.

Background For several recombinant proteins produced in Esherichia coli there seems to be a strong dependence of the productivity and product quality on the limiting substrate feed rate at induction [ 1 - 8 ]. Different feed profiles in fedbatch cultivation have thus been used to affect the growth rate and thus the production of the desired product. A common strategy to increase cell productivity and product quality is furthermore to export products to the periplasm. This reduces proteolysis since the periplasm contains less proteases and the more oxidising environment is favourable for correct folding of proteins containing disulphide bridges. Furthermore, the translocation to the periplasm can result in a primary purification provided that the outer membrane can selectively be removed which requires a stable inner membrane. Of particular importance for protein translocation processes are the anionic phospholipids but also unsaturated fatty acids. It was reported a great number of times that increased amounts of phosphatidylglycerol could facilitate protein translocation [ 9 , 13 - 15 ]. In continuous cultivation data has shown that there is clear growth rate coupling of the membrane phospholipid/fatty acid structure [ 9 , 10 ] but also a coupling between the structure and function [ 9 ]. It is also known that the accumulated amount of protein translocators are dependent on for example the medium composition but are also influenced by metabolic events like catabolite repression [ 11 , 12 ]. The latter process is known to be a function of the growth rate [ 12 ]. In continuous cultivation it was observed that conditions for establishment of an optimum in phosphatidylglycerol and unsaturated fatty acids could be achieved. An optimum in the accumulation of these compounds occurred at a growth rate of 0.3 h -1 simultaneously to an optimum in β -lactamase leakage over the outer membrane [ 9 ]. It was further shown that at already at growth rates lower than 0.3 h -1 unsaturated fatty acids where replaced by cyclic fatty acids [ 9 ] which was earlier thought to take place first in stationary phase, as shown in batch cultivation experiments [ 26 , 27 ]. If the formation of cyclic fatty acids is allowed to take place this influences some important functions of the membrane due to the higher rigidity caused by this rapid accumulation. This is of great concern since this generally constitutes the point of induction in protein producing processes but also since the cell organelles have a particular influence on the unit operations of the cell harvest and further in the downstream processing performance. There is thus a strong incentive to select and operate processes which allow the establishment of controlled growth conditions and to understand the specific growth characterises established with respect to the membrane at particular feed rates. Two main techniques are at present available: fedbatch and continuous cultivation. We have previously established data from continuous cultivation [ 9 ] which we here will compare to the present data from fedbatch cultivation. The fedbatch technique is used when a high cell density is the primary goal. A variety of feeding strategies can be applied for different research purposes where exponential feed can be used to create conditions that resemble a steady state in all points except for the continuous cell mass accumulation. This technique can however only be used for a few number of generations since the oxygen transfer rate rapidly becomes limiting. In order to reach a high cell density, i.e. cell densities of 50 g, l -1 and above, the traditional fedbatch concept established early in the former century based on bakers yeast and later on for antibiotics production, is generally used. The feed profile, characterised by this general concept, is initially exponential until the oxygen or heat transfer capacity of the reactor is reached and thereafter constant. This implies that the substrate concentration and the specific growth rate decline with cultivation time after the constant feed profile is adopted. This constitutes the advantage: the declining growth rate allows the cell mass productivity to increase at constant oxygen consumption/heat production. A further advantage of this cultivation mode, in comparison to a standard batch process with substrate overflow and to fedbatch with high exponential feed, is the possibility to avoid inhibitory by-products such as acetate. At a preset time during the feeding process, induction is performed. The drawback is that when the cell mass is very high the substrate consumption is very low and this point might not represent the optimal condition for high product formation, achievement of high product quality or facilitated transport of the protein. In continuous cultivation the feed/dilution rate is also used to control the growth rate where there is a limit of the chemostat to cultivation at the maximum or near maximum rates when the pH-auxostat is the preferred method [ 16 ]. Continuous cultivation relies on a theoretical steady state in all variables and the production can thus take place under constant growth conditions. Fedbatch cultivation for high cell density production, on the other hand, is a dynamic operation, as described above, where the cell at all times has to adapt to a new environment. Although the processes can be designed for induction at the same growth rate, the overall cell response might still severely differ due to the difference in process conditions preceding the induction, sometimes referred to as a result of the microbial "memory" of the recent past. One reason is that the time of adaptation of important cellular processes related to production and product quality by far exceeds those of the changes in substrate uptake. Thus, after a rapid change in the feed rate it might very well take some time before seen in a product variable. In spite of all benefits and drawbacks characterising both techniques, the fedbatch technique, using the concept described above, is the most common production technique in industry for recombinant production of protein pharmaceuticals in Escherichia coli . The goal of this work is to show how growth rate controls the performance in fedbatch cultivation with respect to the structure of the membrane and to compare these changes to the function of some selected process variables and methods. These data will, in turn, be compared to earlier data derived from continuous cultivation [ 9 ]. From this, conclusions could be drawn as to which are the bottlenecks of either method at a particular event which will facilitate the choice of production method and enhance the design of the growth rate profile of the chosen method to achieve an increased cultivation but also harvest and downstream processing performance. Results Escherichia coli W3110, a K-12 derived strain, was grown in fedbatch cultivation. The fedbatch concept was characterised by an initial exponential feed, started at time zero, of the growth limiting component, glucose, leading to the establishment of a specific growth rate of 0.6 h -1 . This was followed by a constant feed phase where the growth rate declined asymptotically to a value of approximately 0.05 h -1 . During this phase the cell mass increases linearly to a cell density of approximately 27 g, l -1 , as expected (Figure 1 ). The characteristic and rapid declination in the growth rate, following the shift from exponential to constant feed, is also shown in the figure where the growth rate thus decreases by half in a little more than two hours. This feed process was designed to cover the larger part of the possible growth rates achievable with this strain of E. coli to permit comparison with earlier data from continuous cultivation [ 9 ]. However, the profile follows the general industrial feed profile concept except for the levels, which are generally kept below the limit for acetic acid production [ 9 ]. During the exponential feed phase acetic acid is thus formed however only in minor amounts due to the high glucose influx which induces overflow metabolism. This amount of acetic acid is consumed concomitantly to glucose during the rapid declination of the growth rate from 0.6 h -1 to 0.4 h -1 . The phospholipid content of the E.coli membranes is shown as a function of time from constant feed start in Figure 2A . The amount of phosphatidylethanolamine (PE) is always predominant and constitutes 76–77% of the phospholipids in the cell membranes at all growth rates. The content of phosphatidylglycerol (PG) and cardiolipin (CL) is kept at a level slightly below 12 % throughout the cultivation and does not vary to a great extent. At a growth rate of approximately 0.3 h -1 a minimum of CL of 10.9 % is indicated which is also the time of the maximum PG level. At lower growth rates there is less PG than CL which is directly opposite to the amounts of these compounds at high growth rates. Apart from these small changes the phospholipid values are constant throughout the cultivation. The fatty acids are shown as a function of the time from constant feed start in Figure 2B . The fatty acids predominant in the E.coli cytoplasmic and inner part of the outer membrane, are saturated (SFA: C14:0, C16:0), unsaturated (UFA: C16:1, C18:1), and cyclic (CFA: C17:cyc). From this figure is evident that the amount of SFA's is the most stable during the declination of growth. There is only a slight reduction of the amount at a growth rate of approximately 0.3 h -1 , which is accompanied by a maximum in the amount of unsaturated fatty acids. The unsaturated fatty acids are markedly decreased as the growth rate goes below 0.3 h -1 to 56/70 % of their maximum values, respectively. These maximal values, which coincide with the above mentioned minimum in CL/maximum in PG, thus constitutes a switch-point in the fatty acid and phospholipid metabolism during fedbatch cultivation. At this point there is also a minimum of cyclic fatty acids. The levels of the cyclic fatty acids are then rapidly increasing the lower the growth rate and from its minimum the cyclic fatty acids are dramatically increased by almost 100 % in less than two hours. This occurs although there is a supply of nutrients at all times which is in contrast to earlier literature where it is stated that it is the lack of carbon and energy in the stationary phase which results in cyclic fatty acid formation. During the continuing declination of the growth rate, the cyclic fatty acids continues to increase to a final of three times their minimum value. It should be noted that since the fatty acids of the lipopolysaccharide (LPS) were removed during the sample extraction, the results are only reflecting the inner part of the outer membrane and the cytoplasmic membrane. The three phospholipids were furthermore examined for their individual fatty acid content. The results are shown in Figure 3 . Since the dominating phospholipid is PE the pattern of fatty acids in PE consequently mirrors the overall results seen in Figure 2B . The amount of saturated fatty acids does not change to a high degree in any phospholipid, the differences lie more in the changes of unsaturated and cyclic fatty acids. In PG the same general pattern as in PE is observable but the increase in C18:1 with growth rate is more pronounced and this increase is made at the expense of C17:cyc. In CL, it is obvious that the levels of the fatty acids do not change very much at all except for the well-documented increase in cyclic fatty acids towards low growth rate. Since the metabolic pathway depicts that the formation of CL is a condensation of two molecules of PG and since the fatty acid pattern in CL is not mirrored by the one in PG it is thus evident that there is a modification also of the fatty acids during the condensation. This is in favour of considerably less unsaturated fatty acids in CL specifically with respect to C18:1 and also less cyclic fatty acids compared to PG. The difference is however less at lower growth rates. The effects of these membrane changes were investigated on basis of the function of some selected methods and process variables: the mechanical strength of the membrane to ultrasound, the effect of osmotic chock, the shredding of endotoxin to the medium, the leakage to the medium and the amount of cell lysis. The mechanical strength of the membrane during different growth rates was quantified as the release of the periplasmic marker, β -lactamase, at different time of sonication using the same amount of cell and sonication effect. Figure 4 illustrates the release compared to French press data. A full release of the amount of the specific protein is achieved in all cases by a two-minute sonication except for the samples at the lowest growth rate. There is no evident change in the pronounced effect of sonication at growth rates down to 0.3 h -1 but from this point on there is continuously increased membrane stability. A comparison to the pattern of unsaturated fatty acid accumulation in Figure 2B shows the close resemblance between structure and performance and where the high amounts of the double bond might be the reason to the effect of the ultrasound. Osmotic shock is a frequently used method to release protein in the small scale. The methodology chosen as a model system gives a step-wise permeabilisation of the membrane, which is followed by breakage of the cell wall murein saccculus and formation of sphaeroplasts when lysozyme enters the cell [ 17 ]. Several conditions must be met before lysozyme can efficiently penetrate the outer membrane which was shown for several methods in the publication by Witholt et al [ 17 ]. The employed method relies six individual steps, as further described in material and methods. Figure 5 shows the drop in absorbance caused by each addition as a function of sampling at selected growth rates. At the top is shown the initial values and each following line in the diagram refers to the effect compared to the preceding addition. Consequently a large distance between two lines is a large effect and vice versa . The figure shows that the sucrose addition and the late addition of water results in the major effect on permeabilisation and, for unknown reason, also the initial dilution by Tris-HCl. The addition of lyzosyme, in itself, has no effect which is a result of the fact that it can only penetrate the membrane by the addition of water [ 17 ]. The effect, read as a change in absorbance is due to the change in the shape of the cells which was also clearly visible in the microscope. The growth rate dependent changes were evident between comparison of the first and second measuring points at high growth rate, specifically the result of the addition of water, but the effect stays thereafter more or less constant shown by the constant distance between the lines. The release of lipopolysaccharide, LPS, to the medium was measured by means of detection of the presence of a specific inner part of this structure, lipid A, which has properties of an endotoxin. This compound has to be removed in downstream processing of pharmaceutical proteins since it gives a toxic response in patients and must always be proven absent in the final product. It was considered likely that the changes in the underlying phospholipid and fatty acid structure might have large consequences for the release of this component to the medium which was to some extent confirmed by literature studies [ 12 ]. In Figure 6A the volumetric accumulation of endotoxin in the medium is seen as a function of time compared to cell mass accumulation. In this graph is also included the measurements of the cell metabolite, cyclic adenosine monophosphate (cAMP), which is important for carbon utilisation and which was proposed to constitute the signal for changes leading to membrane reorganisation during statis and stress [ 18 ]. The figure shows that all metabolites are accumulated with time. In Figure 6B are plotted the specific accumulation of the parameters of figure 6A and it can be concluded that endotoxin and cAMP accumulation are constant and follows the cell mass accumulation. A major concern of this work was the effect of membrane structural changes on protein leakage from the periplasm to the medium. Since it is clear that a periplasmic protein can be present in the medium both by cell lysis and by true leakage from the periplasm, cell lysis was registered as a function of cultivation time. This is shown in Figure 7A . The leakage to the medium due to cell lysis was detected both by total protein analysis and by presence of DNA in the medium (the latter results not shown). Both variables showed the same pattern. At low growth rates there seems to be a marked change of the amount of total protein in the medium but it should be noted that the value is exceedingly low. The conclusion is that during the present fedbatch operation cell lysis is at all times very low and does not exceed 1 %. Earlier data from continuous cultivation [ 9 ] has shown that both total and specific protein synthesis can vary at different growth rates. To understand if this takes place also in fedbatch cultivation the total protein was measured at different growth rates. In Figure 7A is seen that total protein accumulation in each cell is constant down to a growth rate of 0.2 h -1 where after it slowly declines from 0.50 to a final value of 0.42 g protein, g cells -1 . In Figure 7B is plotted the time course of the formation of the specific protein of interest i.e. periplamsic protein β -lactamase, which is constitutively produced from its own promoter. The total β -lactamase production is shown as well as the specific production values in the medium per gram of cells. It is clear that during fedbatch cultivation the total amount of this specific protein follows the total protein accumulation pattern and is constant independent if calculated per cell mass or per total cell protein except for the values at very low growth rate. The β -lactamase value per gram of cells in the medium is thus slightly higher during the first part of the cultivation down to μ = 0.3 h -1 where after the value declines. At very low growth rate there is yet another period of increased specific accumulation. It is realised that this latter period of slowly increased accumulation in the medium coincides with an increased however small lysis in the end of the cultivation. Since the leakage is not increased during this period lysis seems most likely cause for the accumulation. The level of β -lactamase in the medium amounts to approximately 10% of the total β -lactamase production throughout the cultivation as seen from the quotient of β -lactamase in the medium in relation to the total value. Discussion The conditions during fedbatch leads to a structure, which is approximately constant with respect to phospholipids and saturated fatty acids compared to growth rate. The changes in lipid components arrive from changes in unsaturated fatty acids where the increased values above the maximum at 0.3 h -1 is formed in expense of cyclic fatty acids which are accumulating at low growth rate. Compared to literature data the amount of cardiolipin is approximately 70 % higher than from earlier observations [ 9 ]. The resulting effects on the membrane function are several but are generally and overall characterised by their stability as a function of growth rate. Cell lysis is low at all times i.e. the membrane is not easily breaking-up and this is due to the high content of cardiolipin at all times which is further emphasised by high levels of cyclic fatty acids specifically at growth rates below 0.3 h -1 . The high content of cardiolipin is to some extent balanced by high level of unsaturated fatty acids, which compensates the need for flexibility, permeability and movement of compounds within the membrane which growth rate. The growth dependent accumulation and the generally high level of unsaturated fatty acids influence the results of sonication. The high levels above μ = 0.3 h -1 thus supports a weaker membrane, in this respect, while the substitution of unsaturated to cyclic fatty acids reduces the effect at low growth rate. The transport of protein over the membrane is however only slightly higher at higher amounts of unsaturated fatty acids which is implicated by the larger effect of enzyme/water penetration during osmotic chock and the leakage of the model product, β -lactamase, at high growth rate. In total, β -lactamase is however always present in the medium at a level of approximately 10 % of the total production due to that the membrane flexibility caused by unsaturated fatty acid is balanced by the high and constant amount of cardiolipin. There is a very slight increase in cell lysis at the end of cultivation. The reduced capacity of this cell for total and specific protein production at low growth rate, which might be due to a reduced amount of ribosomes [ 1 ], low supply of carbon and energy and the comparatively high maintenance demand, makes it likely that also the protein content in the membranes is reduced with increased susceptibility to lysis as a result. The changes at low growth rate often results in a change in the shape of the cells which are known to become smaller and more round [ 19 ] which might influence the performance. Flow cytometric studies could however not reveal any major changes of these cells during the time of the process (data not shown). Continuous cultivation gives theoretically always the highest productivity due to the reduced shut-down periods for cleaning and refilling. However this technique is rarely used in industrial production of recombinant protein production due to problems of separation of growth and product formation. The pharmaceutical industry will furthermore need a validation protocol, which must be derived for this technique and where the concept of batch reproducibility needs some clarification. There is also a tendency to use techniques which are since long established and to which the personnel are educated. However, sometimes reasons are not well developed and a problem is the lack of data for a scientific choice of production technique. In table 1 and 2 continuous cultivation and fedbatch data are summarised as a function of growth rate in a process based on the same cell, the same vector, the same medium and the same product. Table 1 shows selected values of membrane components and data has been collected as to represent either a relatively high (0.6 h -1 ) or a low (0.05 h- 1 ) growth rate. The table furthermore indicate if curves, with respect to growth rate, show a trend which reveals the existence of maximum/minimum values of the specific compound. Table 2 shows the comparative effects on a selection of functional parameters at the same growth rates. One of the most significant structural differences between the cultivation techniques is the high content of CL in fedbatch cells in expense of both PG and PE. The formation of large amounts of CL is an obvious cell strategy to save energy and carbon by the release of glycerol to further metabolisation. Since the only difference between the two cultivation techniques is the constant conditions in continuous cultivation and the dynamically changing ones in fedbatch the increased need for carbon/energy must lie in the rapidly changing growth environment. In fedbatch there are also more unsaturated fatty acids accumulated, which develop in expense of saturated fatty acids, which are higher in continuous cultivation. The fatty acid level is also more or less constant in CL, which promotes stability. The conclusion is that cells in fedbatch cultivation interpret the environmental signals in a way that leads to a membrane structure with several features that generally only develop during nutrient exhaust in stationary phase. This is not the case for cells growth in continuous cultivation which develop distinct characteristics at each stable growth rate. This leads to a fedbatch membrane that is rigid compared to cells in continuous cultivation. The overall result is fewer changes in the membrane function due to growth rate. This is represented by constant leakage, constant lysis and an almost constant effect of permeabilisation. The fedbatch membrane shreds also less endotoxin although levels are low for both cultivation techniques. However, the fedbatch membrane is sensitive to sonication, specifically at high growth rate due to the comparatively high amount of unsaturated fatty acids and thus releases more β -lactamase due to the growth dependent changes of this fatty acid. From the comparative data of table 1 and 2 , from both process techniques it is evident that there is a growth dependent switch-point in the membrane regulation at approximately 0.3 h -1 . However, this is not allowed to develop in fedbatch to the extent seen in continuous cultivation where a specific growth rate is kept much longer of the process time. The result is that no optimum in the leakage to the medium is clearly established in fedbatch cultivation and the fedbatch process cannot be designed for this since a constant growth rate of 0.3 h -1 cannot be kept while at the same time leading to a high cell accumulation. Protein transport is generally considered to benefit from a high content of PG and unsaturated fatty acids. This accumulation leads to high leakage which is verified by the continuous cultivation leakage optimum [ 9 ] and which leads to a comparatively high overall level in fedbatch (present data). The high fedbatch content of unsaturated fatty acids is however counteracted by signals leading to reduce levels of PG in favour of CL. The control of cellular activities at the switch-point leading to the establishment of the membrane compound pattern cannot be clarified at this point. However, it is clear that it is not the signals relating to the accumulation of cAMP. This compound is indeed a function of the growth rate, but not of the accumulation of the structural components. The overall conclusion is thus that if leakage is undesired, fedbatch should not be used but the process could well be run in continuous cultivation at low growth rate. If an increased leakage is preferred this can be optimised in the continuous operation mode principally by change of growth rate but probably also by mutants accumulating a combination of phosphatidylglycerol and unsaturated fatty acids if these are stable enough for prolonged cultivation. Conclusion The stability of the fedbatch membrane, which is not growth dependent, is the strength of this cultivation technique. The time of cell harvest is thus not critical since the membrane strength and function does not rapidly change which allows for flexibility in operation. The stability of the fedbatch derived cell membrane might thus also by the same reasoning be an advantage in further downstream processing however depending on the chosen unit operations. The merits of the cell characteristics in continuous cultivation are due to the possibility to run the cultivation at a feed rate which is optimal for membrane leakage independent if this is to be small or maximised. Total and specific protein production is lowered at low growth rate and the data points to that recombinant protein production should not be performed below a growth rate of 0.2 h -1 and induction should thus take place well above this value. This is a considerable drawback in fedbatch cultivation, which relies on high cell density accumulation leading to low growth rates before induction but easily controlled in continuous cultivation. The coupling of the productivity and even product quality to a specific growth rate further strengthens the implication of using continuous cultivation. The conclusion is that the drawbacks of industrial operation of continuous cultivation in the biopharmaceutical industry earlier mentioned should be preferably be solved to explore the benefits that are obviously present with this technique. Methods Strain and medium Escherichia coli W3110 (F - , λ - , IN (rrnD-rrnE)) [ 20 ], a K12 derivative with a maximum growth rate, in the selected medium and temperature, of 0.65 h -1 , was used. The plasmid pBR322 (Pharmacia Biotech, gene Bank Accession number V0119) was transferred to the host cell. This plasmid carries the genes bla and tet where β -lactamase (TEM-1) was used as the model product. The maximum levels were approximately 1 % of the total protein. The medium for the bioreactor and the feed was a mineral salts medium composed of (g, l -1 ): (NH 4 ) 2 SO 2 , 7.0; KH 2 PO 4 , 1.6; Na 2 HPO 4 *2H 2 O, 6.6; (NH 4 ) 2 -H-citrate, 0.5. Glucose was autoclaved separately and added to bioreactor together with a sterile filtered trace element solution and 1.0 M MgSO 4 (1 ml, l -1 , respectively). A further addition of magnesium, of the same amount, was done at eight hours of feeding. The glucose concentration in the feed was 300 g, l -1 . The trace elements solution was composed of (per litre): CaCl 2 *2H 2 O, 0.5g; FeCl 3 *6H 2 O, 16.7 g; ZnSO 4 *7H 2 O, 0.18 g; CuSO 4 *5H 2 O, 0.16g; MnSO 4 *4H 2 O, 0.15 g; CoCl 2 *6H 2 O, 0.18 g; Na-EDTA, 20.0 g. Cultivation Conditions Four independent fedbatch cultivations were performed in the following manner: inoculum was prepared from a shake flask culture grown overnight at 37°C at 150 rpm. This was transferred into the bioreactor when an optical density at 600 nm (OD 600 ) of 2–3 was reached from which point the fedbatch cultivation was performed. The experiments were carried out in a 14-l bioreactor with a working volume of 8.0 l. The bioreactor was equipped with an air sparger and the air-flow was 0.5–1.0 vvm. By varying of the stirrer speed up to 1000 rpm the dissolved oxygen concentration did never go below (30 ± 2) % air saturation. A pH of 7.0 was kept by addition of 28% (w/w) ammonia solution. Temperature was controlled at 37°C. The cultivations were started as batch cultures with an initial glucose concentration of 2 g, l -1 . The feed was started when glucose was almost exhausted. The feed profile consists of two different phases where an exponential feed was followed by a constant feed phase. The exponential feed profile was calculated from the following equation: F(t) = F 0 * e - μ t where F is the feed rate (L/h), μ the desired specific growth rate (1/h) and t is the process time (h). The calculation of the initial feed rate, F 0 , is based on a mass balance at the time of the feed i.e. after the batch growth assuming: pseudo steady state growth during the feed phase, a theoretical yield coefficient (Y x/s ) of 0,5 g, g -1 and a negligable amount of the limiting substrate in the reactor. F 0 =( μ *V*X)/(S i *Y x/s ) Where Y x/s is the cell yield on glucose (g/g), X is the cell concentration at F 0 (g/L), V is the culture volume (L) and S i is the glucose concentration in the feed (g/L). The continuous cultivation conditions and data, compared to in the conclusions section, were reported elsewhere [ 9 ]. These were designed for steady states at similar growth rates to those reported from the fedbtach cultivations of this paper. Steady state was kept for approximately eight generations and the same strains, vectors and media were used. Analyses Cell mass, substrate and metabolites The cell growth was followed by optical density at 600 nm. Biomass concentration was determined as cell dry weight (CDW, g, l -1 ) by 10 min centrifugation (4500 rpm, 2250 g) of 3*5 ml of cell suspension in preweighed test tubes and drying of the pellet overnight at 105°C before weighing. The supernatant from the CDW samples was sterile filtered and used for acetic acid and cyclic adenosine monophosphate, cAMP, analysis. Acetic acid was measured with the enzymatic method from Boehringer Mannheim GmbH (Cat. No 148 261) and cAMP was analysed by an enzyme immunoassay system from Amersham Pharmacia Biotech AB (code RPN 225). Samples for glucose analysis were taken as described earlier [ 21 ]. The β -lactamase assay was performed as described in reference [ 22 ]. Inductively Coupled Plasma (ICP) analysis was used for magnesium determination. Total protein and DNA analysis were used for cell lysis detection. Total-protein was measured according to the method by [ 23 ] using bovine serum albumin as standard (Sigma Chemical Co, USA). DNA was measured by a fluorescence assay and quantified with Hoechst 33258 (Sigma Chemical Co, USA) according to the Dyna Quant 200 application method (Pharmacia, Sweden). The amounts of endotoxin i.e. lipid A, the inner part of the lipopolysaccharide (LPS) structure, was measured by the chromogenic Limulus method (Chromogenix, LAL, U.S. License No. 1197). Mechanical strength of the cell membrane The mechanical strength was estimated by sonication (Vibra-Cell, Sonics & Materials, USA) at constant frequency, cell dry mass, sample volume and acoustic power. Samples were placed in a water bath at +4°C during cell disruption to prevent overheating. The release of β -lactamase protein was determined at the time events of 0.5, 1.0, 1.5 and 2.0 minutes and compared to cell that was disintegrated in a high-pressure homogeniser (French press FA-073) at 55 bar. Data from French press was set as the 100% disruption level and all other values were taken as a percentage of this value. Chemical strength of the cell membrane Chemical strength was estimated by permeabilisation as described by Witholt et al [ 17 ]. The following steps were thus undertaken: (1) t = 0 min, 0.5 g, l -1 cultures were harvested and suspended in 0.3 ml, 200 mM Tris-HCL (pH 8.0), (2) t = 1 min, 0.5 μ l, 100 mM EDTA (pH 7.6) was added, (3) t = 2 min, 0.5 ml, 1 M sucrose (pH 8.0) was added thus giving excess of EDTA over Mg 2+ ions, (4) t = 3.5 min, egg white lysozyme (EC 3.2.1.17, C. F. Boehringer und Soehne GmbH, Mannheim, Germany) was added to a final concentration of 60 μ g/ml, (5) t = 4 min, the cell suspension was exposed to a mild osmotic shock by two-fold dilution in water to trigger lysozyme penetration of the outer membrane, and finally (6) t = 8 min, the cell suspension was diluted 11-fold in 10 mM EDTA (pH 7.6). Permeabilisation was recorded by optical density at 450 nm. The samples taken at different growth rate were diluted to the same cell concentration (0.5 g, l -1 ) and the same volume (0.2 ml) was used at the sampling point. Phospholipids and fatty acids analysis Phospholipids and fatty acids were determined by the method of Arneborg [ 24 ] where the total and individual phospholipid content was quantified by an inorganic phosphorous assay after extraction and thin liquid chromatography. The presented value is the mol% of individual to the total phospholipids. The standard deviation was +/- 0.03 %. Fatty acids were released from extracted phospholipids by transmethylation with KOH in methanol. The methyl esters are thus representing the actual fatty acid. Chromatography was performed with C13:0 as an internal standard as described by [ 25 ]. The presented value is the mol% of individual to the total fatty acids (shortened for the number of carbon atoms according to: C14:0, C16:0, C16:1, C18:1 and 17:cyc). The methodology is further described in reference [ 9 ]. Author's contributions AS made the fermentations, analyses and data treatments and evaluation. The project was conceived by GL who further participated in the project planning, design and data evaluation. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514524.xml

548267

Rivastigmine: an open-label, observational study of safety and effectiveness in treating patients with Alzheimer's disease for up to 5 years

Background Rivastigmine, a butyl- and acetylcholinesterase inhibitor, is approved for symptomatic treatment of Alzheimer's disease (AD). Data supporting the safety and efficacy of second-generation cholinesterase inhibitors, such as rivastigmine, are available for treatment up to 1 year, with limited data up to 2 1/2 years. The purpose of this report is to present safety and effectiveness data for rivastigmine therapy in patients with mild to moderately severe AD receiving treatment for up to 5 years. Methods An observational approach was used to study 37 patients with originally mild to moderate AD receiving rivastigmine as a therapy for AD in an open-label extension (ENA713, B352 Study Group, 1998). Results The initial trial demonstrated rivastigmine was well-tolerated and effective in terms of cognition, global functioning and activities of daily living. In this open label extension, high-dose rivastigmine therapy was safe and well tolerated over a 5-year period. Two thirds of the participants still enrolled at week 234 were in the original high-dose rivastigmine group during the double-blind phase, suggesting that early therapy may confer some benefit in delaying long-term progression of symptoms. Conclusions Long-term cholinesterase inhibition therapy with rivastigmine was well tolerated, with no dropouts due to adverse effects past the initial titration period. Early initiation of treatment, with titration to high-dose therapy, may have an advantage in delaying progression of the illness.

Background Alzheimer's disease (AD) is the most common form of dementia affecting elderly people in the United States. Prevalence is 1% to 2% at age 65 years, but increases markedly to 35% or greater by age 85. Because of a demographic shift toward a more aged population, the percentage of affected individuals is rapidly increasing. This trend is expected to continue for the foreseeable future. Therefore, accurate and timely diagnosis and effective treatments are critical to optimal outcomes over the 8- to 10-year course of the illness [ 1 ]. Traditionally, a probable diagnosis of AD was accomplished by history, clinical examination, neuroimaging, and neuropsychological and laboratory testing to rule out treatable causes for the patient's symptoms and to differentiate AD from other possible causes of dementia [ 2 , 3 ]. Much effort has gone into defining risk factors for the development and progression of Alzheimer's dementia, as well as to identify biological markers for the disease. Clinical-demographic variables that are consistently associated with AD in prior studies include family history of AD, age, and Down's syndrome [ 1 , 3 ]. None of these variables has been demonstrated to affect the rate of disease progression or show any utility in defining subgroups that may be more amenable to therapy. Currently, predominant symptoms of dementia are treated primarily with second-generation cholinesterase (ChE) inhibitors. These drugs have demonstrated efficacy, as measured by cognitive, behavioral, and functional outcomes, in randomized, placebo-controlled clinical trials, the majority of which have been of 6 months' duration [ 4 - 6 ]. In an open-label extension study of the cholinesterase inhibitor donepezil, Doody et al [ 7 ] concluded that donepezil was safe and effective for treating the symptoms of mild to moderate AD for up to 2 1/2 years. Cognitive, behavioral, and functional outcomes in patients treated with ChE inhibitors over the longer term are of great interest given the substantial social and economic implications of AD, which has a course that averages 8 to 10 years. Due to their relatively recent approval, however, longer-term data on the clinical benefits and/or limitations of ChE inhibitor therapy in AD patients is virtually nonexistent [ 8 ]. Rivastigmine's approval by the FDA in 2000 was supported by several pivotal trials, including a randomized US trial (ENA 713 B352)[ 5 ]. In this pivotal trial, 699 patients with mild to moderately severe AD were randomized to high dose rivastigmine (6–12 mg/day), low dose (1–4 mg/day) or placebo with a 7 week fixed dose-titration phase followed by a flexible dosing phase during weeks 8–26. Results of the 26-week open-label extension of this study found that at 52 weeks, patients originally treated with 6–12 mg/day rivastigmine had significantly better cognitive function than patients originally treated with placebo [ 9 ]. In this paper the authors present descriptive findings for a cohort of 37 patients who participated in the long-term open-label extension of the ENA713B352 rivastigmine trial. Much work remains to be done to more definitively answer questions about when to start therapy, which patients are most likely to benefit, what constitutes clinically relevant beneficial effects over the longer term, and when these drugs are no longer clinically effective. Consideration should also be given to withdrawal of therapy. Findings presented in this article will add to the current limited dataset for long-term efficacy and outcomes with cholinesterase inhibitor therapy for persons with probable AD. This report describes our experience in following the cohort of patients at our center with AD treated with the ChE inhibitor rivastigmine (a medication that inhibits both butyl- and acetylcholinesterase) as part of the ENA 713 B352 pivotal trial for a period up to 5 years. Methods Data in this analysis came from a subgroup of 37 patients with originally mild- to moderate-stage (defined by a Mini-Mental State Examination [MMSE] score of 10 to 26) AD followed at a large Mid-Western university site in a 26-week, prospective, randomized, double-blind, placebo-controlled, parallel-group study of rivastigmine as therapy for AD conducted at 22 research sites across the United States (ENA713, B352 Study Group, 1998) [ 5 ]. Patients were enrolled according to previously described inclusion criteria [ 5 ]. Of note, this study allowed rather broad inclusion of AD patients with other comorbid illnesses; presumably this would allow this cohort to more closely mirror real-world populations. The study was conducted in accordance with ethical standards of the institutional committee on human experimentation and with the Helsinki Declaration of 1975, revised 1983. The initial study protocol was therefore reviewed in our center by the institutional review board and all patients or caregivers consented to inclusion based on appropriate informed consent. Additional consent was obtained for the open-label extension of the study. Cholinesterase inhibitor treatment Immediately following the double-blind phase of the study, open-label rivastigmine was flexibly titrated over a 12-week time period to a maximum tolerated dosage of up to 6 mg BID. By the end of the 12 week titration 25 participants were on 4–6 mg. of rivastigmine and 11 participants had dropped from the study. One participant remained on a 2 mg. dose and dropped from the study between weeks 52 and 78. Assessment of treatment response Outcomes measures included the Alzheimer's Disease Assessment Scale-cognitive subscale (ADAS-cog) and the Clinician's Interview-Based Impression of Change, with caregiver input (CIBIC-Plus). Ability to carry out activities of daily living (ADL) was assessed by the Progressive Deterioration Scale (PDS). Disease-staging measures included the Geriatric Deterioration Scale (GDS) and the MMSE. In the open-label phase of the study, efficacy evaluations were performed every 6 to 8 weeks for titration and early maintenance, and every 26 weeks for the next 5 to 6 years. Statistical analysis These data are predominantly descriptive, with analyses including Kaplan- Meier survival plots when appropriate. All statistical analyses on our single center extension were performed at our center using SPSS. Subgroup analyses by initial treatment randomization were also performed. Results Demographics and population Twenty-one Caucasian women and 11 Caucasian men participated in the open label extension of this pivotal study. Twenty-two participants reported a family history of AD. Selected demographic and baseline characteristics of the subsample are presented in Table I . Of the 32 patients, 11 were originally randomized to the high dose (6–12 mg/day) group, 10 to the low dose group, and 11 to placebo. Five of the patients in the cohort chose not to participate in the extension. Table 1 Selected baseline and demographic characteristics (N = 32). Characteristic Mean (SD) Age (y) 71.8 (7.5) Length of symptoms (mo)* 29.6 (18.3) Baseline MMSE 21.3 (4.1) Baseline GDS 3.7 (0.78) MMSE = Mini-Mental Status Examination; GDS = Geriatric Deterioration Scale. *Length of symptoms prior to enrollment in the double-blind study. A total of 25 patients during the 5-year term eventually withdrew from the study. Reasons for termination, stratified by original group during the double-blind phase, are summarized in Figure 1 . The most frequent reason for termination of participants initially randomized to the high-dose group was the lack of availability of free rivastigmine following FDA approval, which had been provided at pre-launch at no charge as part of the clinical study. It is of interest that no deaths occurred in the group initially randomized to high-dose rivastigmine during the initial double-blind, placebo-controlled trial. Furthermore, disease severity, as measured by the GDS, was greater in the original high-dose group (mean = 4.0; median = 4.0) as compared with the low-dose (mean = 3.7; median = 3.5) and placebo groups (mean = 3.4, median = 3.0). A total of 8 terminations were due to withdrawal of consent (n = 3) or caregiver discontinuation (n = 5). Five of the 8 were in the original placebo group. Seven patients withdrew because of adverse events; 2 each in the original low-dose and placebo groups (57%) and 3 in the original high-dose group (43%). Treatment failure was cited as the reason for termination of one 69-year-old man in the original placebo group. His baseline MMSE score was 21, and his final MMSE score at week 26 was 17. No other patients withdrew as a result of treatment failure. Figure 1 Reason for termination stratified by original group. Of the 4 patients who withdrew from the study due to disease progression, 3 were men between the ages of 57 and 64 years. The 57-year-old man was in the original high-dose group, with a baseline MMSE score of 16 and an MMSE score of 0 at the 234-week data collection. The other two male participants were aged 62 and 64 years with baseline MMSE scores of 22 and 20, respectively. The 62-year-old man in the original low-dose group withdrew at week 104 with an MMSE score of 9; the 64-year old man in the original high-dose group withdrew at week 156 with an MMSE score of 11. Interestingly, the 76-year-old woman in the original low-dose group, with a baseline MMSE score of 22, withdrew after the 26-week data collection (MMSE = 21). Quantitative (objective) analysis The Kaplan-Meier survival analysis, shown in Figure 2 , illustrates time to dropout from week 26 to week 234. The survival curve reveals a relatively steep decline in participation in the first 9 months of open-label extension, mostly related to adverse events (AEs), since almost all AEs causing discontinuation occurred during this phase, followed by a "flattening" of the curve. Fourteen participants were still taking high-dose rivastigmine at 2 years, 12 participants at 3 years, and 10 participants at 4 years. Figure 2 Kaplan-Meier analysis of time to dropout from week 26 to week 234 Of the subjects starting open-label rivastigmine, 25% were still participating at week 234. Of the 8 subjects still participating, 5 were female and only 1 reported a family history of AD. Interestingly, this group was characterized by a broad age range (57–85 years), a broad range of baseline MMSE scores (15–26), and by the fact that 5 of the 8 remaining participants at week 234 were in the original high-dose rivastigmine group. For the end point defined as a 5-point drop from the baseline MMSE score, the group mean was 94 weeks. However, an age-related observation was noted such that the mean time to a 5-point drop from baseline MMSE for the ≤ 70 age group was 67 weeks as compared with 122 weeks for the >70 years age group. A similar trend was observed for scores on the ADAS-cog. For the group as a whole, mean time to a 4-point deterioration on the ADAS-cog was 84 weeks; with a mean of 70 weeks for the younger group and 104 weeks for the older group. [In the original study, mean age across the 22 research sites was 74.5 years; mean age for participants in this report was 71.8 years] Figures 4 through 8 summarize mean scores for the ADAS-cog, CIBIC, GDS, PDS, and MMSE from week 26 through week 234. As expected, mean values for cognitive and functional status decline over time; however, significant within subject variability is present. Figure 3 Mean scores on the Alzheimer's Disease Assessment Scale-cognitive subscale (ADAS-cog) for weeks 26–234. Qualitative analysis When evaluating qualitative data, it is important to examine particular patients in terms of the data collected and treatment response at end of study. Data for the youngest man and the elderly woman, considered side-by-side, seem counterintuitive. Part of the explanation, however, may rest with the caregivers' experiences with and beliefs about the patients. This information can be accessed by reviewing the "symptoms most troubling" item from the CIBIC-Plus. This item asks, "With respect to the above symptoms, which are the biggest problems for you and/or other caregivers?" The spouse of the 57-year-old man reported that the "symptoms most troubling" for her were the patient's "selective difficulties and his attitude problem" (week 104, MMSE = 11) and that the patient "doesn't try" (week 130, MMSE = 7). These comments suggest a lack of acceptance of the patient's diagnosis and a lack of belief regarding the patient's documented decline that could explain continued clinic visits until, at week 234, the patient's MMSE score was 0. In contrast, the caregiver for the 76-year-old woman cited the fact that the patient "doesn't want to leave the house" (week 12) and the patient's anxiety about coming to clinic appointments (week 26) as the "most troubling" symptoms. These comments suggest that the caregiver was encountering patient resistance and observing patient distress, which were exacerbated by efforts to participate in the study. Therefore, further clinic visits were declined. Retrospectively, it is impossible to determine how the caregiver's beliefs and experience with the patient affect perceptions about disease status, treatment response, and decisions to continue or terminate treatment. These are interesting hypothesis-generating observations that should be explored in future investigations. Discussion Limited data are available on the tolerability and effectiveness of cholinesterase inhibitor therapy for periods up to 3 years [ 4 - 7 ], but nothing has been reported in the literature concerning the percentages of patients remaining on therapy and its effect beyond this time point. This study adds to the limited long term data on patients treated with high-dose rivastigmine therapy by reporting descriptive data for up to 5 years. Findings presented in this report indicate that, in this sample of patients, high-dose rivastigmine therapy was well tolerated over a 5-year period. Of note, approximately two thirds of the participants still enrolled at week 234 were in the original high-dose rivastigmine group during the double-blind phase; this finding suggests that early therapy with rivastigmine may confer some benefit in delaying long-term progression of symptoms, as has been previously suggested by analysis of the combined 26 weeks of double-blind and first 26 weeks of open-label data from the B352 US trial [ 9 ]. Throughout the initial 26-week double-blind portion, patients receiving placebo steadily deteriorated, while those treated with high-dose rivastigmine were able to maintain their baseline level of performance on the ADAS-Cog [ 5 ]. This approximated a delayed-start design for the open-label portion, which demonstrated that patients who started rivastigmine late never "caught up" with patients who had been on high-dose rivastigmine from the beginning of the trial. This suggests a disease-progression-delaying effect of the drug, which may allow this population to maintain their autonomy for a longer period of time. However, it is important to emphasize the limitations of this data; the analysis was retrospective, the sample was small, there were significant numbers of drop-outs, and the availability of free rivastigmine ceased with FDA approval which occurred near to the end of the study Conclusions In summary, long-term therapy with the ChE inhibitor rivastigmine was well tolerated with no dropouts due to adverse effects past the first 9 months of the open-label extension. In contrast to the generally reported community experience of relatively brief duration of therapy with ChE inhibitors in AD, 25% of patients in this study were still taking rivastigmine by the end of 5 years. Of interest is the multifactorial nature of reasons for treatment discontinuation. Only 1 of these patients withdrew from the study because of perceived treatment failure over the 5-year period. Disease progression accounted for the withdrawal of 4 patients. These results are informative both for the duration of treatment, and because the majority of the patients who continued treatment during this 5 year period were from the group originally titrated up to a high dose during the initial double-blind phase. Given that the sample size prohibits significance testing, these data suggest that rivastigmine therapy may be sustained over long periods of time in a significant percentage of patients with AD, and that high-dose therapy, particularly when begun early, may have an advantage in delaying the progression of the illness. Competing interests MRF has received grant support and has served as consultant and received honoraria from Novartis Pharmaceuticals Corporation. MLL has no financial interest to declare. Authors' contributions MF carried out the original clinical trial, conceived of the design for the observational study, and participated in the drafting and revisions of the manuscript. ML participated in the design of the observational study, performed statistical analysis, and participated in the drafting and revisions of the manuscript. Both authors read and approved the final manuscript. Figure 4 Change in Clinician's Interview-Based Impression of Change for weeks 26–56. Figure 5 Mean scores on Geriatric Deterioration Scale for weeks 26–234. Figure 6 Mean scores on Progressive Deterioration Scale for weeks 26–156. Figure 7 Mean scores on Mini-Mental Status Examination for weeks 26–234. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548267.xml

555578

Mechanical properties of femoral trabecular bone in dogs

Background Studying mechanical properties of canine trabecular bone is important for a better understanding of fracture mechanics or bone disorders and is also needed for numerical simulation of canine femora. No detailed data about elastic moduli and degrees of anisotropy of canine femoral trabecular bone has been published so far, hence the purpose of this study was to measure the elastic modulus of trabecular bone in canine femoral heads by ultrasound testing and to assess whether assuming isotropy of the cancellous bone in femoral heads in dogs is a valid simplification. Methods From 8 euthanized dogs, both femora were obtained and cubic specimens were cut from the centre of the femoral head which were oriented along the main pressure and tension trajectories. The specimens were tested using a 100 MHz ultrasound transducer in all three orthogonal directions. The directional elastic moduli of trabecular bone tissue and degrees of anisotropy were calculated. Results The elastic modulus along principal bone trajectories was found to be 11.2 GPa ± 0.4, 10.5 ± 2.1 GPa and 10.5 ± 1.8 GPa, respectively. The mean density of the specimens was 1.40 ± 0.09 g/cm 3 . The degrees of anisotropy revealed a significant inverse relationship with specimen densities. No significant differences were found between the elastic moduli in x, y and z directions, suggesting an effective isotropy of trabecular bone tissue in canine femoral heads. Discussion This study presents detailed data about elastic moduli of trabecular bone tissue obtained from canine femoral heads. Limitations of the study are the relatively small number of animals investigated and the measurement of whole specimen densities instead of trabecular bone densities which might lead to an underestimation of Young's moduli. Publications on elastic moduli of trabecular bone tissue present results that are similar to our data. Conclusion This study provides data about directional elastic moduli and degrees of anisotropy of canine femoral head trabecular bone and might be useful for biomechanical modeling of proximal canine femora.

Background The mechanical properties of canine trabecular bone in the femoral head are important for a better understanding of normal biomechanics of the bone and are needed for assessing changes occurring under pathological conditions like osteoarthritis of the hip, osteonecrosis or Legg-Calvé-Perthes disease of the femoral head. Particularly the elastic moduli are important for finite element modeling of the proximal femur. Several methods have been used in the literature for identifying the elastic modulus of bone, such as mechanical testing [ 1 - 4 ], combinations of micro-computed tomography and finite element modeling [ 4 - 11 ], and ultrasonography[ 12 ] including acoustic microscopy [ 13 ]. Some studies have investigated canine bone [ 13 - 19 ], but to our knowledge no publication has presented any details about directional elastic moduli of canine femoral heads including degrees of anisotropy. Neither is it clear whether assuming isotropy on the tissue level is a justified simplification for trabecular bone in canine femoral heads as Kabel et al.[ 9 ] reported in a study of whale bone specimens. Therefore, in this study we determined Young's moduli of trabecular bone obtained from healthy canine femoral heads by ultrasonography. We then calculated degrees of anisotropy and used statistical testing in order to estimate whether assuming isotropy of the trabecular tissue might be a valid simplification. Methods Eight dogs (weight 30–63 kg) were selected that had been euthanized for several medical reasons. From each dog both femora were obtained and were examined by a veterinarian for signs of metastatic malignant disease, Legg-Calvé-Perthes disease, osteoarthritis of the hip or bone necrosis. An x-ray of the whole femur was obtained with the femoral head and the intertrochanteric region placed directly on the film, and the main pressure and tensile trajectories were marked on the image. The bones were kept moist, wrapped in plastic bags and stored at -21°C. Each bone was placed on the x-ray image and the direction of tension and pressure trajectories were marked on the bone according to the x-ray template. An orthogonal coordinate system was defined (Fig. 2 ). The positive x-axis was oriented along the main pressure trajectories and the y-axis was aligned with the main tension trajectories. One cubic specimen of 10 × 10 × 10 mm was cut (Fig. 1 ) from each frozen femoral head using a precision bone saw (Exakt Makro 310 CP, EXAKT Apparatebau, Norderstedt, Germany). The edges of the cubes were cut parallel to the x, y and z axes of the coordinate system. The cubic specimens were weighed using a laboratory scale (Acculab ALC-110.4, Acculab Europe, Göttingen, Germany). Figure 2 Schematic representation of femoral head and neck. The main tensile and compressive trajectories and the orientation of the cubic specimen and coordinate system are shown. Figure 1 Cubic specimen cut from one canine femoral head. For sonographic testing, a specially designed device with a custom-made ultrasound transducer (Institute of Materials Science, University of Hannover) was used. An ultrasound frequency of over 2 MHz was chosen for measuring the material properties of trabecular bone tissue[ 20 ]. The ultrasound frequency was adjusted so that a clear signal could be detected by the ultrasound receiver. A frequency of 100 MHz was chosen because ultrasound signals remained undetectable when using lower frequencies even with maximum power. The cubic specimens were placed in a container after thawing and immersed in standard Ringer's solution at room temperature. An ultrasound receiver was placed at the surface of the cube opposite the transducer which was also directly touching the specimen surface, and the runtime through the bone material of each bone cube (n = 16) was recorded ten times in all three orthogonal directions. The edge lengths of each cube were measured using the digital image analysis system IMAGE C ® (IMTRONIC GmbH, Berlin, Germany), and the specimen volumes were calculated. Specimen densities were determined by equation (1): where ρ is specimen density, m is specimen mass and V is the specimen volume The ultrasound wave runtimes were processed by excluding the minimum and maximum results of the ten subsequent measurements, and the average of the remaining results was calculated. The transmission velocity was calculated by equation (2) and the elastic modulus was determined using equation (3). where c long is the transmission velocity, s denotes the edge length of the specimen and equals the distance between ultrasound transmitter and receiver which are placed in direct contact with the opposing specimen surfaces, t 1 and t 0 the time at reception and sending of the ultrasound wave, respectively. where E x,y,z is Young's modulus along x, y and z axes and ρ is the specimen density calculated from equation (1). C long is the ultrasound velocity calculated from specimen edge length and transmission time (see equation (2)) Mean density and mean Young's moduli of the specimens and standard deviations were determined using the statistical software package SPSS 12.0 (SPSS, Chicago, USA). Strong correlations between density and elastic moduli could be expected from equation (3), nonetheless Pearson correlations were calculated for confirmation. The directional elastic moduli were checked for significant differences using one-way ANOVA (analysis of variance). The degrees of anisotropy [ 4 , 21 ](E x /E z; E x /E y ; E z /E y ) were calculated and subsequently checked for a significant relationship with specimen density using Pearson correlations. Results The edge lengths of the bone specimens varied by ± 1 % (± 0.1 mm). Elastic moduli in the bone specimens ranging from 6.3 to 14.3 GPa were found. The elastic moduli in X, Y and Z direction were 11.2 ± 0.4 GPa, 10.5 ± 2.1 GPa and 10.5 ± 1.8 GPa, respectively. Minimum, maximum, mean values and standard deviations of bone sample density and directional Young's moduli are listed in Table 1 . Pearson correlations between density and directional Young's moduli (E x , E y and E z ) were significant as could be expected from equation (3) (p < 0.005). The degrees of anisotropy ranged from 0.82 to 1.59 and were significantly correlated with specimen density (Table 2 , Figure 3 ). Figure 3 Degrees of anisotropy in dependence of specimen density. Correlation coefficients are listed in the symbol legend. Table 1 Minimum, maximum and mean values for sample densities and directional Young's moduli Moduli are arranged according to testing direction (along x, y and z axes, see fig. 2). Minimum Maximum Mean Standard deviation Density [g/cm 3 ] 1.19 1.51 1.40 0.09 E X [MPa] 10600 11760 11217 376 E Y [MPa] 6283 14285 10459 2071 E Z [MPa] 6832 13097 10506 1839 Table 2 Degrees of anisotropy The minimum and maximum degrees of anisotropy (range), the mean values, standard deviations (SD) and the correlation coefficients (r) indicating correlation of the respective degree of anisotropy with specimen density ρ. p designates the significance probability (p-value). Range Mean SD r p E x /E z 0.89–1.59 1.10 0.18 -0,93 0.000 E z /E y 0.92–1.11 1.01 0.05 -0,58 0.009 E x /E y 0.82–1.73 1.11 0.22 -0,96 0.000 Discussion This study presents detailed data about mechanical properties of canine femoral trabecular bone tissue and degrees of anisotropy. Despite the strengths of our work, some limitations have to be noted. The correct calculation of the sample volume depends on exactly cubic specimens, but deviations might occur due to errors in the sawing technique. We found a maximum variation of edge length of ± 0.1 mm (1%) in our specimens so this error appears to be negligible. The apparent densities of the specimens were calculated by weighing whole samples and measuring sample volumes instead of calculating the densities of ashed samples[ 2 , 19 ] or cleaning the bone marrow out of the specimens using water jets prior to measurement. According to Rho [ 20 ] ultrasonic waves at frequencies of >2 MHz travel along the trabecular material and allow calculation of the elasticity of the trabecular bone material rather than the elasticity of whole specimens which is investigated by conventional compression testing. Considering this statement, in our study using a 100 MHz ultrasound transducer the application of apparent densities rather than the densities of the trabecular bone material might lead to an conspicuous underestimation of the elastic modulus (equation (3)) because it is expected that the density of the trabecular bone tissue is higher than the density of whole specimens. However, Kang et al. [ 22 ] measured densities of cylindrical trabecular bone specimens from canine femoral heads and reported a mean density of 1.17 ± 17 g/cm 3 for whole bone cylinders after cleaning and 0.65 ± 0.09 g/cm 3 for ashed samples which is much lower than our results. The lower densities might be caused by the specimen volume that was used by the authors: they geometrically measured the volume of the cylindrical bone specimens so that a considerable intertrabecular volume is included and bone tissue densities are underestimated according to equation (1). Hence the results of Kang's publication can not be used for comparison or correction of our density data. It is noticeable that in our study we found a small relative standard deviation (SD/mean) in apparent densities and Young's modulus in x direction (6.6 % and 3.3 %) although we used femora from a heterogeneous selection of different breeds. However, the relative SD in y and z directions was computed to be 19.5 % and 17.5 %, respectively. It is not clear whether the broader range of elastic moduli in these directions reflects real differences or whether it might be caused by a less accurate positioning of the saw when cutting the specimens from the femoral heads. The significant relationship between the specimen density ρ and directional elastic moduli found in our study was to be expected because Young's modulus was calculated from ρ. This significant relationship was also described in a study using compression testing of bone cubes from human donors (0.74 <= r <= 0.84; p < 0.001) [ 2 ]. No significant differences were found between directional elastic moduli (p = 0.34). This result could support the concept of an "effective" isotropic elastic tissue modulus as described by Kabel et al.[ 9 ]. Several works have been published concerning elastic moduli of canine or human bone. Studies that investigated the apparent elastic modulus of human bone specimens using mechanical testing and finite element models found much lower elastic moduli[ 4 , 21 ] than we did in our work. Kang et al. [ 22 ] reported elastic moduli of trabecular specimens from canine femoral heads of 428 ± 237 MPa which is also much lower than what we observed in our study; their results were obtained by conventional compression testing. Several other studies have been carried out investigating elastic moduli of canine trabecular bone specimens[ 14 , 16 , 19 ]. Those results are not comparable with our data because the authors measured elastic moduli of whole specimens rather than Young's moduli on the tissue level. Additionally, Odgaard et al. reported that conventional compression testing underestimates Young's modulus by about 20%[ 23 ]. Keaveny et al. [ 24 ] found a percentage difference in modulus when using platens compression testing of up to 86%. They recommend using the endcap technique for obtaining more accurate data, which however restricts testing to one direction. Jacobs et al. [ 25 ] investigated porous samples made from bone cement. Using finite element modeling and mechanical testing, they found that the mean error when using parallel platen compression testing was 8 or 15% depending on FE mesh size and was reduced to 2 or 0.5% with the endcap technique. Only one study investigating elastic moduli of canine trabecular bone tissue is available to our knowledge: Jorgensen and Kundu [ 13 ] used a 1 GHz acoustic microscope for examining a trabecular strut obtained from a canine distal femur. They computed a mean Young's modulus of 19.9 ± 2.5 GPa which is higher than our results; this could be caused by a different trabecular structure and higher bone volume fraction in the distal femur. The authors state that anisotropy was clearly detected at micrometer level, but no further quantification is given. More studies are available reporting mechanical properties of human trabecular bone. Rho [ 20 ] found an elastic modulus of 14.9 ± 1.7 GPa in trabecular bone specimens from human tibiae using a 2.25 MHz transducer. This modulus is higher than in our data but within the same order of magnitude. Ashman and Rho [ 26 ] measured elastic moduli from three human trabecular bone specimens using an ultrasonic technique and found a mean elastic modulus of 13.0 GPa which is about 30% higher than in our study of canine bone. This observation is supported by Kuhn's [ 19 ] assertion that elastic moduli are higher in human trabecular bone than in canine bone. Our results are further supported by Zysset et al. [ 27 ] who used a nanoindentation technique and found average elastic moduli in trabecular lamellae of human femoral necks of 11.4 ± 5.6 GPa. Conclusion Our study provides detailed data about elastic moduli and degrees of anisotropy of canine femoral bone tissue. No significant differences between directional Young's moduli were found indicating that the concept of an effective isotropy of trabecular bone in canine femoral heads might be a justified simplification. The degrees of anisotropy were highly correlated with specimen densities. The results of elastic moduli are comparable to similar studies of canine and human trabecular bone tissue [ 13 , 20 , 26 , 27 ]. Authors' contributions TP carried out the statistical analysis of the results, took part in the detailed design of the study and prepared the manuscript. AB designed the study, prepared the bone specimens, conducted the ultrasound measurements and calculated the elastic moduli. UV and AML obtained the canine femora, examined the bone for exclusion criteria and prepared the bone specimens. BAB, IN and HW designed the study concept from a technical and medical perspective, respectively, and corrected the manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555578.xml

554100

Comparative genome analysis of cortactin and HS1: the significance of the F-actin binding repeat domain

Background In human carcinomas, overexpression of cortactin correlates with poor prognosis. Cortactin is an F-actin-binding protein involved in cytoskeletal rearrangements and cell migration by promoting actin-related protein (Arp)2/3 mediated actin polymerization. It shares a high amino acid sequence and structural similarity to hematopoietic lineage cell-specific protein 1 (HS1) although their functions differ considerable. In this manuscript we describe the genomic organization of these two genes in a variety of species by a combination of cloning and database searches. Based on our analysis, we predict the genesis of the actin-binding repeat domain during evolution. Results Cortactin homologues exist in sponges, worms, shrimps, insects, urochordates, fishes, amphibians, birds and mammalians, whereas HS1 exists in vertebrates only, suggesting that both genes have been derived from an ancestor cortactin gene by duplication. In agreement with this, comparative genome analysis revealed very similar exon-intron structures and sequence homologies, especially over the regions that encode the characteristic highly conserved F-actin-binding repeat domain. Cortactin splice variants affecting this F-actin-binding domain were identified not only in mammalians, but also in amphibians, fishes and birds. In mammalians, cortactin is ubiquitously expressed except in hematopoietic cells, whereas HS1 is mainly expressed in hematopoietic cells. In accordance with their distinct tissue specificity, the putative promoter region of cortactin is different from HS1. Conclusions Comparative analysis of the genomic organization and amino acid sequences of cortactin and HS1 provides inside into their origin and evolution. Our analysis shows that both genes originated from a gene duplication event and subsequently HS1 lost two repeats, whereas cortactin gained one repeat. Our analysis genetically underscores the significance of the F-actin binding domain in cytoskeletal remodeling, which is of importance for the major role of HS1 in apoptosis and for cortactin in cell migration.

Background Cortactin (also designated EMS1 , CTTN, cttn, Amplaxin, see Genecard [ 1 ]) was initially identified as one of the most prominent tyrosine phosphorylated proteins in v-Src infected chicken embryo fibroblasts [ 2 ]. Cortactin was independently isolated from mouse NIH3T3 cells [ 3 ] and human tumor cell lines [ 4 ]. Human cortactin is encoded by the EMS1 gene, which is located on chromosome 11q13 [ 4 , 5 ]. Gene amplification of 11q13 region and concomitant overexpression of cortactin frequently occurs in several human carcinomas [ 4 , 6 - 8 ] and correlates with lymph node metastasis and increased mortality [ 9 - 11 ]. Elevated expression of cortactin increases cell motility, invasion [ 12 - 14 ] and metastasis [ 15 ]. The deduced amino acid sequence of cortactin revealed three main distinguishable domains: the N-terminal acidic domain containing a DDW-Arp2/3 binding motif followed by a six and one-half 37-amino acid F-actin binding repeat domain, a central region and an SH3 domain at the very C-terminal. The DDW-Arp2/3 binding site and the actin-binding domain together regulate F-actin polymerization and dynamics by activating the Arp2/3 complex [ 16 ] and both are necessary for translocation of cortactin to sites of actin polymerization [ 17 ]. Recently, we reported the identification of two alternative splice variants of human cortactin lacking either 6 th or the 5 th /6 th repeat, present in normal tissues as well as squamous cell carcinomas cell lines [ 14 ]. These splice variants differ significantly in their ability to (i) bind F-actin, (ii) cross-link F-actin (iii) activate Arp2/3 mediated actin polymerization and (iv) induce cell migration in vitro [ 14 ]. This indicates that also the number of repeats determines the affinity for F-actin and ability to regulate cell migration. Similar cortactin splice variants were also reported in the mouse [ 18 ], rat [ 19 ] and frog [ 20 ]. The SH3 domain is a conserved protein module found in various signal proteins and mediates the interaction with various proteins such as N-WASP involved in actin polymerization, dynamin-2 in endocytosis, ZO-1 in cell-cell interactions and SHANK-2 in neuronal growth cones (reviewed in [ 21 ]). The central part of the protein between the F-actin repeat domain and the SH3 domain contains an alpha-helix sequence and a proline-rich region with three c-Src tyrosine phosphorylation sites [ 22 , 23 ] and three serine/threonine phosphorylation sites [ 24 ]. Cortactin tyrosine phosphorylation occurs in response to growth factor treatment, integrin cross-linking, bacterial invasion and cell shrinkage (reviewed in [ 21 ]). Tyrosine phosphorylation of cortactin reduces its F-actin cross-linking activity and is required for its ability to stimulate cell migration [ 13 ]. Since cortactin operates mainly in cytoskeletal rearrangements, it may link other proteins via its SH3 domain to sites of actin polymerization. Alternatively, serine phosphorylation of cortactin by Erk enhances, whereas Src phosphorylation inhibits the activation of N-WASP by cortactin [ 25 ] and as a result affects actin polymerization. This suggests that cortactin at first instance may be directed to the site of actin polymerization by other proteins. Thus, changes in protein expression level, phosphorylation state, the relative expression of splice variants and interactions with other proteins can all influence cell migration. Cortactin shows the highest similarity to the hematopoietic lineage cell-specific protein 1 (HS1). Human HS1 (also designated HCLS1 , see Genecard [ 26 ]) was originally isolated by its homology to the adenovirus E1A gene [ 27 ]. HS1 overall similarity to cortactin at the amino acid level is 51% but is highest at the SH3 domain (86%) and the 37-amino-acids repeat domain (86%), except that HS1 carries only three and one-half repeats. Despite this high homology, the function of HS1 differs considerable from cortactin. First, HS1 is mainly expressed in hematopoietic cells [ 27 ], whereas cortactin is widely expressed in all cell types except most hematopoietic cells [ 28 ]. Only in platelets and in megakaryocytes both genes are expressed [ 29 , 30 ]. Second, in concordance with this tissue distribution, HS1 is tyrosine phosphorylated after receptor cross-linking in B-cells [ 31 ], T-cells [ 32 ], mast cells [ 33 ] and erythroid cells [ 34 ], but at different residues compared to the functional phosphorylation residues in cortactin [ 13 , 23 ]. Third, HS1 is, like cortactin, a cytoplasmic protein, but after tyrosine phosphorylation HS1 translocates to the nucleus [ 35 ], whereas cortactin is never found in the nucleus. This is because HS1, but not cortactin, contains a nuclear localization signal (NLS) [ 36 , 37 ]. Fourth, HS1 plays an important role in the receptor-mediated apoptosis and proliferative responses as demonstrated by the analysis of HS1 deficient mice [ 38 ] and WEH1-231 B lymphoma cells [ 37 , 39 ]. An HS1 tyrosine mutant that could not translocate to the nucleus, also failed to induce apoptosis [ 37 ]. Consistent with its role in apoptosis, HS1 is able to bind to the mitochondrial protein HAX-1, a Bcl2 like protein [ 40 ]. Finally, the SH3 domain of HS1 at the C-terminus binds to other proteins (Ste20 related kinase HPK1 [ 41 ] and HS1-BP3 [ 42 ]) than those binding to cortactin, despite the very high amino acid sequence similarity of both SH3 domains (86%). This most probably reflects the different tissue-specific expression pattern. Cortactin and HS1 share also remarkable similarities. First, HS1 binds with its DDW-motif directly to Arp2/3 and is involved in Arp2/3 mediated actin polymerization in vitro , although less efficient than cortactin [ 43 ]. Second, HS1 binds to F-actin with its 37-amino-acid repeat domain [ 36 ], however, it contains only three and one-half repeat in contrast to cortactin. Third, also HS1-splice variants have been detected such as a variant lacking the 3 rd repeat of the F-actin binding domain in a systemic lupus erythematosus (SLE) patient resulting in increased apoptosis after B-cell receptor (BCR) stimulation [ 44 ]. Fourth, HS1 is sequentially phosphorylated on three tyrosine residues by various Src family tyrosine kinases [ 31 , 45 ] and two serine/threonine residues [ 30 ], although at different residues than cortactin [ 25 ]. Finally, both cortactin and HS1 can accumulate into podosomes, structures found in osteoclasts [ 46 ] and marcrophages [ 47 ], but also in RSV transformed cells [ 48 ] and carcinoma cells [ 49 ]. Although cortactin and HS1 share a high amino acid sequence and structural similarity, their functions differ considerable. In this paper, we compare their genomic organization in order to provide more insight into their evolution, which may form the basis towards understanding specific functions of both genes. We describe the genomic organization and the exon-intron boundaries for human cortactin. Both the genomic cDNA and deduced amino acid sequences of human cortactin were compared to cortactin and HS1 genes from other species. Genomic comparisons revealed the evolution and underscore the significance of the conserved F-actin binding repeat domain for HS1 and cortactin and the importance of alternative splicing for cortactin function. Results and discussion The genomic organization of cortactin homologues We have previously described the isolation and sequencing of the EMS1 cDNA [ 28 , 49 ] (DDBJ/EMBL/GenBank Accession No. M98343) coding for the human cortactin protein. To evaluate the genomic structure, we determined the exon/intron-boundaries. Nucleotide sequence comparisons with human EMS1 cDNA sequence revealed homology with two human genomic clones (DDBJ/EMBL/GenBank Accession No. AP000487 and AP000405) (Table 1 ). The genomic structure of the EMS1 /cortactin gene was determined by performing BLASTn comparisons of EMS1 cDNA against the genomic clones (Figure 1A ). By amplifying the intron sequences (smaller than 2 Kb) using primers on adjacent exons followed by end-sequencing of these products, we confirmed the intron/exon boundaries of the human EMS1 /cortactin gene. The EMS1 gene contains 18 exons spanning over about 38 Kb of genomic DNA. The length of the individual exons ranges from 55 to 178 bp, except the last exon (1564 bp). The splice donor and acceptor sequences, the sizes of the introns and exons of the human EMS1/ cortactin gene are provided in the supplementary materials [see Additional file 1 ]. The ATG is at position 169, at the first nucleotide of exon 3, indicating that the first two exons encode the 5' untranslated region (UTR). The F-actin-binding repeat domain is encoded by exon 5 to exon 12 with 5 exons of 111 nucleotides in length (exons 6, 8, 9, 10 and 11) (Figure 1A and [see Additional file 1 ]). The sequence encoding the DDW Arp2/3 binding site is located within exon 3 and the SH3 domain is encoded by exon 17 and 18. The 3' UTR is 1420 nucleotides in length with a polyadenylation signal AATAAA at position 3225. Table 1 Accession numbers of cortactin and HS1 sequences Gene mRNA/EST Protein Genomic DNA Chromosome COMPLETE CORTACTIN AND HS1 SEQUENCES Human (Homo sapiens, Hs) wt-cortactin M98343 a AAA58455 AP000487 11q13 b AP000405 SV1-cortactin c BC008799 AAH08799 BC033889 AAH33889 NM_138565 NP_612632 HS1 d X16663 CAA34651 NT_005612 3q13 BC016758 AAH16758 Chimpanzee (Pan troglodytes, Pt) wt-cortactin AADA01305241 e 9 HS1 AADA01307895 e 2 Mouse (Mus musculus, Mm) wt-cortactin U03184 AAA19689 NT_00336 7F5 SV1-cortactin f BC011434 AAH11434 XM_144788 XP_144788 AK084249 BAC39148 HS1 X84797 CAA59265 NW_006107 16B BC007469 AAH07469 D42120 Rat (Rattus Norvegicus, Rn) wt-cortactin NW_043405 1q41 SV1-cortactin (isoform B) AF054619 AAC08425 SV2-cortactin (isoform C) AF054618 AAC08424 HS1 XM_221421 XP_221421 NW_042728 11q11 Chicken (Gallus gallus, Gg), wt-cortactin M73705 AAA49031 AADN01110316 g 5 SV1-cortactin BU109838 g HS1 ENSGALG00000009778 e,p Un Frog (Xenopus laevis, Xl), wt-cortactin AB027611 h BAB79435 Frog (Xenopus tropicalis, Xt), wt-cortactin scaffold_32906 I Zebrafish (Danio rerio, Dr), wt-cortactin AF527956 i AAQ09010 HS1 Finished_845 o 4 Pufferfish (Takifugu rubripes, Tr), wt-cortactin SINFRUG00000156355 e scaffold_853 e HS1 SINFRUG00000124755 e scaffold_1329 e Pufferfish (Tetraodon nigroviridis, Tn), HS1 CAG04186 scaf14731 19 Fruit fly (Drosophila melanogaster, Dm) NM_079702 NP_524426 AE003733 3R AB009998 BAA34397 AB030177 BAB01490 Mosquito (Anopheles gambiae, Ag) XM_315193 XP_315193 AAAB01008952 i 2R Sea urchin (Strongylocentrotus purpuratus, Sp) NM_214617 NP_999782 scaffold_101 e AF064260 i AAD08655 Sponge (Suberites domuncula, Sd) Y18027 CAC38778 Y18860 CAC80140 INCOMPLETE CORTACTIN AND HS1 SEQUENCES Cattle (Bos taurus, Bt), SV1-cortactin TC154749 j,k B222447 k Pig (Sus scrofa), wt-cortactin TC48123 j,l Frog (Xenopus laevis, Xl) HS1 BC060434 AAH60434 Sea squirt (Ciona intestinalis, Ci) TC32922 j,m scaffold_101 i White shrimp (Litopenaeus setiferus, Ls) BE846976 White shrimp (Litopenaeus vannamei, Lv) BE188605 Root knot worm (Meloidogyne incognita, Mi) BE188583 BQ613692 n BQ625292 n Root knot worm (Meloidogyne chitwood, Mc) CB856307 BQ613692 Root knot worm (Meloidogyne javanica, Mj) BE578389 a All accession numbers except as noted below, may be found in the mRNA, EST, protein of genomic databases of NCBI [65]. b Chromosomal locations were obtained from UniGene, NCBI [65]. c SV, splice variant. Accession numbers from EST's BE714795, BE717740, BE717751, BE717765, BE717811, E717819, BE717829, BE717871, BE274120, BE728099 [65]. d HS1= Haematopoietic lineage cell-specific protein 1 = HCLS1= hematopoietic cell-specific Lyn substrate 1 e Accession number was obtained from the EnsEMBL [67]. f SV, splice variant. Accession numbers from EST's BE290787, BF321856, BG519413, BG174188, AA762862, AI099054, BF135250 [65]. g Accession numbers were obtained from the U.S. Poultry Gene Mapping Project [75]. h From 2500 bp untill 3578 bp of this mRNA is mRNA from another gene. Genomic DNA are pieces of sequences. i Accession numbers were obtained from the DNA Data Bank of Japan [70]. j Accession numbers from EST's from the TIGR) [69]. k Homologue to actin binding domain of human SV1-cortactin. l Homologue to C-terminal part of human cortactin incuding the SH3 domain. m Homologue to repeat 1 to 5 of the actin binding domain of human cortactin. n Accession numbers were obtained from the European Bioinformatics Institute. Homologue from 5' untills repeat 3 of the actin binding domain of human cortactin [73]. o The deduced cDNA and protein sequence from the genomic zebrafish Finished_845 sequence is more related to human HS1, while the zebrafish mRNA/protein sequence (AF527956, AAQ09010) showed more homology to human cortactin. p The deduced cDNA and protein sequence from the genomic chicken ENSGALG00000009778 showed more homology to human HS1[67]. Figure 1 Exon map of cortactin and HS1 from different species . Exon/intron boundaries found in the genomic databases by performing BLAST searches with the cortactin cDNA of different species to their genomic DNA, are indicated as vertical boxes in different colors. A lack of boxes means that the boundaries were not found. The genomic organization of some species could not be fully elucidated, because cDNA/genomic sequences were not completely available. The actin binding repeat domain of the cortactin protein is represented by red boxes and the SH3 domain by the purple box. The vertical green stripe indicates the sequence coding for the Arp2/3 binding domain. Pro = proline rich region. The y in the proline rich region represents tyrosine phosphorylation sites. Hs, human; Pt, chimpanzee; Mm, mouse; Rn, rat. Other cortactin homologues have been reported in mouse [ 3 ], rat [ 19 ], chicken [ 50 ], fruit fly ( Drosophila melanogaster ) [ 51 ], and frog ( Xenopus laevis ) [ 20 ]. We searched in numerous databases for all known cortactin genes in other species (listed in Table 1 ). The identification is based on overall amino acid sequence and overall structural homology with human cortactin. Cortactin homologues exist in mammalians (human, chimpanzee, cattle, pig, mouse, rat), birds (chicken), amphibians (frog), fishes (zebrafish, pufferfish), urochordates (sea squirt), invertebrates (sea urchin), insects (fruit fly, mosquito), shrimps, worms and sponges. To date, there is no evidence for the existence of cortactin in unicellular species, nor in plants. Thus, cortactin seems to be restricted to metazoans. For several species, both cDNA and genomic sequences (total or partial) are available and therefore we were able to reveal their genomic organization using BLASTn. The exon/intron-boundaries were determined and compared to human cortactin [see Additional file 1 ]. As schematically presented in Figure 1 , the genomic organization and the lengths of the exons as well as the locations of the exon/intron boundaries are highly conserved from urochordates to mammalians. Pufferfishes have the shortest known genome of all vertebrate species due to much shorter introns, nevertheless most exon/intron boundaries were conserved and similar to mammalian cortactin. Intriguingly, the number of repeats in the actin-binding domain differs between species (Figure 1A–G ). The number of exons and the location of the intron/exon borders of insect cortactin ( Drosophila and mosquito) differ considerably with mammalian cortactin, despite the proteins sequences are very similar. Drosophila and mosquito carry 4 repeats in the actin-binding domain. In both species, repeat 1-to-3 and 4 are on separate exons with in mosquito the 4 th repeat of the actin binding domain to be encoded by a single 111 bp large exon 2 (Figure 1F,G ). Both, sponge (the lowest metazoan) and sea squirt (urochordate) cortactin protein carry 5 repeats. During evolution, after creation of sponges and worms, the coelomata divided into insects and urochordates (that evolved later into vertebrates). The genomic organization of ancestors of the coelomata should reveal the roots of cortactin evolution. However, complete cDNA and/or genomic DNA of cortactin homologues in these species are not yet available. The genomic organization HS1 homologues Both nucleotide and amino acid sequence comparisons with cortactin revealed the highest similarity with the hematopoietic lineage cell-specific protein 1 (HS1). So far, HS1 homologues have been reported in human [ 27 ], mouse [ 33 ], rat and chimpanzee (NCBI database), suggesting that HS1 exists in mammalians only. We determined the intron/exon boundaries of mammalian HS1 genes by aligning the cDNA with the genomic DNA using BLASTn (Figure 1H and [see Additional file 2 ]). The number and lengths of the exons and the locations of the exon/intron boundaries were very similar to cortactin, especially in the exons that encode the actin-binding domain (compare [see Additional file 1 ] and [see Additional file 2 ]). The exons 10–13 of HS1 encoding the centre region between the actin-binding domain and the SH3 domain are longer (633 bp versus 489bp in cortactin) and more divergent compared to corresponding exons of cortactin. In addition to a single cortactin homologue in all other species, nucleotide sequences comparisons using the mammalian HS1 mRNA and genomic DNA sequences revealed (incomplete) genomic sequences in chicken, pufferfish, zebrafish and frog (Table 1 and Figure 1I–M ) that were more related to the HS1 protein (Figure 3 and [see Additional file 3 ]). Because no HS1 homologues for these species were present in the mRNA/dbEST database (except for X. laevis HS1), the cDNA (and corresponding protein) sequences were deduced from the genomic DNA with BLASTn or were predicted by Ensemble program. In these lower species, two cortactin related proteins exist. To distinguish between cortactin and HS1 variants, only the most conserved N-terminal part of cortactin and HS1 protein variants, including repeat 3 (corresponding to amino acid 1–190 of human cortactin) was used in BLASTp analysis. In each species, one protein variant turned out to be more homologous to human cortactin, and was called cortactin, whereas the other protein variant appeared to be more related to HS1 and was called HS1. This analysis unveiled HS1 proteins with more than 3 repeats in chicken and pufferfish Tetraodon nigroviridis (containing 4 1/2 repeats), pufferfish Takifugi rubripes and Xenopus laevis HS1 (5 1/2 repeats) and zebrafish HS1 (6 1/2 repeats) (Figure 1 I-M). Moreover, alignments of the exon/intron boundaries of these HS1 genes to the mammalian HS1 genes [see Additional file 2 ] revealed that exon 7 (repeat 3) of HS1 was most similar to exon 10 (repeat 5) of cortactin suggesting that in mammalians exon 8 and 9 (repeat 3 and 4) of HS1 were lost during evolution. This is supported by the presence of at least one sequence of 111 nucleotides in the 5670 bp intron 6 of human HS1 (location 3271–3381) that is predicted by the program HMMER when performing alignments using a consensus sequence of the 37 amino acid repeats. However, this sequence is not functional because it does not represent an exon based on the consensus sequence of exon-intron junctions ('gt ... ag' rule of intron sequences) and no human transcripts or ESTs of HS1 including this sequence are present in the NCBI databases. In summary, HS1 is not restricted to mammalians only, but exist also in fishes, amphibians and birds and its genomic structure is very similar to that of cortactin. Different promoter regions explain distinct tissue specificity of cortactin and HS1 Cortactin is widely expressed in most cell types suggesting to be important for vital functions, while HS1 expression is restricted to hematopoietic cells suggesting to be tailored later in evolution to serve a specific function in these cells. In concordance with their tissue-specific expression pattern, we suppose that their expression might be differently regulated. Therefore, we compared the upstream promoter regions of several cortactin and HS1 genes (Figure 2 ). The mammalian cortactin gene is very GC rich and contains putative SP-1 transcriptional factor binding sites that are common to many TATA-less promoters and typical for promoter regions in 'widely-expressed housekeeping genes'. Ets family transcription factors, found in the HS1 promoters, are specific for hematopoietic cells and involved in controlling the expression of many B cell- and macrophage-specific genes [ 52 ] and are critical for development of lymphoid and myeloid cell lineages. The promoter region of Drosophila and mosquito cortactin shares putative transcription factors found both in mammalian cortactin and HS1. Thus at least in mammalians, the nature of the promoters seemed to determine the broad distribution of cortactin expression in various tissues except most hematopoietic cells and the limited expression of HS1 to hematopoetic cells. Figure 2 A schematic view over 800 bp of the proximal promoters . Distribution of putative binding sites where represented for the transcription factors SP1 (red), GATA1 or GATA2 (green), AP-1 (dark blue), E2F (yellow), cEts (purple), C/EBPa or C/EBPb (light blue) and the TATAA box (gray) and CCAAT box (white) in the promoter regions of cortactin, human (HsCort), chimpanzee (PtCort), mouse (MmCort), mosquito (AgCort), Drosophila (Dmcort), and HS1, human (HsHS1), chimpanzee (PtHS1), mouse (MmHS1) and rat (RnHS1). The mRNA starting point (assigned +1) is indicated by an arrow. Figure 3 Phylogenetic relationship of cortactin and HS1 genes . Evolutionary comparison of the N-terminal of cortactin and HS1 proteins including repeat 3 (corresponding to nucleotide 1–190 of human cortactin), represented in a phylogenetic tree based on a cluster alogorithmic alignment generated using GeneBee ClustalW 1.83 program. The number of repeats in the full length actin binding domain for the indicated species are depicted between brackets. Hs, human; Pt, chimpanzee; Mm, mouse; Rn, rat; Gg, chicken; Xl, frog Xenopus laevis ; Dr, zebrafish; Tr, pufferfish Takifugu rubripes ; Tn, pufferfish Tetraodon nigroviridis ; Dm, fruit fly Drosophila ; Ag, mosquito; Sp, sea urchin; Sd, sponge. The significance of the actin binding repeat domain in cortactin and HS1 We recently reported the identification of two alternative splice variants of human cortactin; SV1-cortactin lacking the 6 th repeat and SV2 lacking the 5 th and 6 th repeat resulting in a different F-actin binding properties and decreased cell migration [ 14 ]. As shown in Table 1 , cortactin splice variants exist in other mammalians as well as in chicken and frog. So far, splice variants in other species have not been identified, suggesting that alternative splicing of cortactin seems to be restricted to higher metazoans. All intron sequences of cortactin bordering the splice site junctions follow the general GT/AG rule [ 53 ] except for intron 11 (GC/AG) [see Additional file 1 ]. As has been shown for other genes, a GT-to-GC transition might be responsible for the generation of an alternatively mRNA transcript [ 54 ]. However, in frog ( Xenopus laevis ), the SV1-cortactin variant exists despite the splice donor of intron 11 begins with a GT [ 20 ]. Thus, concerning the genome of these different species, alternative splicing of the actin-binding domain of cortactin seems to be facilitated during evolution by modulating the splicing machinery by a GT-to-GC transition to create cortactin related variants that influences cellular properties [ 14 ]. The relative expression of cortactin splice variants by tissue origin [ 14 ] suggested that splice variants might have tissue-specific functions such as fine-tuning the organization of the F-actin cytoskeleton and consequently regulating cell adhesion and migration. Alternative splicing also occurs in human HS1. Recently a splice variant lacking the 3 rd repeat (exon 7) has been found in an SLE patient [ 44 ], resulting in enhanced BCR-mediated cell death. This alternative splicing event was due to a germ line mutation. In contrast, the splice donor of HS1 intron 6 begins with a GC [see Additional file 2 ]. With respect to the similarities between cortactin and HS1, it might be of interest to investigate the occurrence of splicing of HS1 exon 6 and possible biological consequences. The 3 rd repeat and its NLS links HS1 to a role in apoptosis, while such a role has not been described for cortactin lacking a NLS. Since the cytoskeleton architecture in hematopoietic lineage cells is very different from that in adherent cells, it is likely that HS1 plays an important role in the construction of tissue-type specific actin networks. Other types of actin cytoskeleton factors, such as the Arp2/3 complex activators of the WASP family have been reported to have distinct tissue specific expression profiles as well. Thus, the apparent role of HS1 in apoptosis is likely due to its actin remodeling related function. Additionally, our genomic comparisons revealed that the 3 rd repeat of HS1 corresponds with the 5 th repeat of cortactin, and therefore it might be of interest to investigate whether cortactin SV2 variant (lacking the 5 th and 6 th repeat) might be involved in apoptosis. The 4 th repeat of cortactin has been suggested to be required for F-actin-binding [ 17 ]. Genomic comparisons revealed that HS1 lacks this 4 th repeat. Nonetheless, HS1 does bind to F-actin and activate the Arp2/3 complex, although at a lower efficiency than cortactin [ 43 ]. This suggests that not only a single repeat but the number of repeats is crucial for the F-actin-binding affinity [ 14 , 18 ]. In addition, HS1 contains a PIP 2 binding site in each of its 3 repeats, whereas cortactin has only one in the 4 th repeat. PIP 2 reduces F-actin cross-linking by cortactin, probably due to competition for the same binding site. Due to its higher affinity for PIP 2 [ 36 ], HS1 restores this cortactin/F-actin cross-linking process by trapping PIP 2 . This might be of importance in platelets and megakaryocytes where both, cortactin and HS1 are expressed. Taken together, the composition of the repeat domain is also involved in diverting the functions of both genes. An elegant way to study the function of a protein is to perform loss-of-function experiments. So far, cortactin knock-out models have not yet been generated successfully, because deletion of one allele of cortactin leads to premature differentiation of embryonic stem cells (personal communication in [ 55 ]). However, complete loss-of-function mutants of the Drosophila cortactin gene were viable and fertile, except impaired border cell migration during oogenesis [ 56 ]. Down-regulation of cortactin by RNA interference, revealed an essential role for cortactin in dendritic spine morphogenesis [ 57 ] and in E-cadherin mediated contact formation in epithelial cells [ 58 ]. Mice lacking HS1, showed normal development of the lymphoid system [ 38 ], however, the antigen-receptor induced clonal expansion and deletion of B and T lymphocytes were impaired. Thus, loss of function studies underscores the divergent functions of HS1 and cortactin in different cell systems. Cortactin and HS1 are derived from an ancestral vertebrate cortactin-gene by gene duplication To examine the genesis of the cortactin family, we studied the relationship between the cortactin and HS1 homologues by generating a phylogenetic tree based on a multi-sequence alignment with the ClustalW 1,83 program [see Additional file 3 ]. We compared the N-terminal regions including repeat 3 (corresponding to nucleotide 1 to 190 of human cortactin), because this is the best-conserved region among all homologues (Figure 3 ). One cluster contains all known HS1 proteins and appeared to be closest related to a cluster composed by insects (Mosquito (Ag), Drosophila (Dm)), urochordate (sea urchin, (Sp)) and sponge (Sd) cortactin. In this last cluster all the species with only one gene (with the highest similarity with cortactin) are present. This suggests that with the appearance of the vertebrates, an ancestral gene became duplicated to create two genes, which later evolved into cortactin and HS1. This hypothesis is supported by the fact that many genes duplicated at this stage in the evolution, the overall amino-acid sequence in both genes is very similar and the introns are located at the same amino acid position. Furthermore, gene duplication often correlates with a tissue specific expression pattern of the duplicated genes, which is true for mammalian cortactin and HS1. Figure 4 displays a hypothetical model for the origin of the cortactin and HS1 genes during evolution. The oldest ancestor is the sponge that, like sea squirt (urochordate), carries one cortactin protein with 5 c1/2 repeats. Insects have also one cortactin gene and evolved to 4 1/2 repeats. During evolution, after the creation of the sponge and the worms, the coelomata divided into insects and urochordates (that evolved later into vertebrates). This suggests that during the evolution, the number of repeats decreased in the insects. Unfortunately, no genomic sequences of ancestors of the coelomata that could reveal the roots of cortactin evolution are available yet to perform more detailed genomic analysis. Figure 4 Model for the origin of cortactin and HS1 during evolution . Exon/intron boundaries from the exons encoding the actin binding repeat domain are represented in yellow. The actin binding repeat domain of the cortactin protein is represented by red boxes. The genome of pufferfish Takifugu rubripes contains two cortactin-related genomic sequences both including 5 1/2 repeats. Most likely, an ancestor vertebrate cortactin gene underwent gene duplication. From this moment on during evolution, two cortactin/HS1-releated genes are present in all higher species. One gene evolved to mammalian HS1 with a specific function in apoptosis in hematopoietic cells. For its function, exon 8 and 9 (encoding repeat 3 and 4) were not useful and lost during evolution. However, the HS1 protein in pufferfish Takifugu rubripes and frog Xenopus laevis contains 5 1/2 repeats, while chicken and pufferfish Tetraodon nigroviridis HS1 carries 4 1/2 repeats. It might be of interest to investigate the function of these HS1 proteins and their functional differences to mammalian HS1. The other gene evolved to a ubiquitously expressed mammalian cortactin protein with a vital function in the organization of the cytoskeleton and cell migration. The 6 th repeat of cortactin most likely originated from a duplication event of the 5 th repeat, since the 6 th repeat is most similar to the 5 th repeat in all species with 6 1/2 repeats. We recently demonstrated that 6 1/2 repeats are necessary for optimal F-actin cross-linking activity and cell migration, while the splice variant lacking both the 5th and 6th repeats (SV2) was less efficient [ 14 ]. Thus, the number of repeats in the F-actin binding domain of cortactin fine-tunes its function in cytoskeletal remodeling. For that reason, in higher metazoans, alternative splicing of the F-actin binding domain is most likely facilitated by a GT-GC transition in the splice donor. Alternatively, we can not exclude that gene duplication might have taken place after duplicated of the 5 th repeat (dotted arrows), since both zebrafish cortactin and HS1 contain 6 1/2 repeats. Conclusions We report the genomic organization of cortactin and HS1 genes of several species. These genes display a conserved genomic organization as the coding regions have almost identical exon/intron structure. Comparison of 5' sequences allows possible regulatory elements that stress their specific tissue distribution. Comparative analysis of the genomic organization and amino acid sequences of cortactin and HS1 provides insight into the evolution of the conserved actin-binding repeat domain, which forms the basis towards understanding specific functions of both genes. Most likely, both genes originated from a gene duplication event and subsequently HS1 lost two repeats, whereas cortactin gained one repeat. Our analysis genetically underscores the significance of the F-actin binding domain in cytoskeletal remodeling, which is of importance for the major role of HS1 in apoptosis and for cortactin in cell migration. Methods The genomic structure of human cortactin To determine the genomic structure of the human cortactin gene, an algorithm was applied based on the consensus sequence of exon-intron junctions ('gt ... ag' rule of intronic sequence) as well as on the codon usage within ORF. Nucleotide sequence comparisons with human cortactin sequences (NCBI, GenBank accession no. M98343) using BLASTn [ 59 ] revealed homology with two genomic clones (GenBank accession no. AP000487 and AP000405). With these clones, we determined all exon/intron boundaries and size of all introns and exons (Table 2A) of the human cortactin gene by (1) performing BLAST comparisons with the cDNA against the genomic DNA and (2) using the GeneFinder program [ 60 ] based on the consensus sequence of exon-intron junctions ('gt ... ag' rule of intronic sequence) as well as on the codon usage within ORF [ 61 ]. To confirm the predicted genomic structure, we determined the intron/exon boundaries using a cloning procedure as described [ 62 ]. Genomic DNA of two cosmid clones COS-7.12 and COS-3.72 covering the cortactin gene as determined by the full-length cDNA [ 5 ], was amplified with randomly selected primers from the cDNA sequence (GeneBank accession no. M98343). All PCR products that were larger than the cDNA control sample were considered to be caused by intron sequences and compared to genomic sequence (accession number AP000487 and AP000405) using BLASTn [ 59 ]. The size of intron 1, 5, 8, 12 and 13 was too large to obtain a reliable sequence. Because no overlapping genomic sequences immediately 5' of the first exon were present in the database, we performed sequence analysis of a 2.7-kb HincII-HincII fragment representing the first exon and its 5'-flanking sequences from cosmid COS-7.12 cloned into pUC18 (p5'EMS_3135). In addition, we sequenced a 5-kb PCR product using a 5'-primer in the vector (within the TET gene) and 3'-primer (p3135p601: 5'-ccgggtcggccctggattcc-3') within exon 1, subcloned in pUC18 (p5'EMS_4911). Nucleotide sequences of both products were compared with the genomic clones representing the cortactin gene present in the NCBI database (Accession number AP000487 (GI 8118774 and GI 6277297) and AP000405 (GI 8118742)) and used to define the 7.4 kb 5'-flanking region. The PROSCAN program [ 63 ] from BIMAS was used to define the 316 bp promoter region preceding exon 1. Putative transcription factor binding sites where determined by the TFSEARCH program [ 64 ] and graphically represented in figure 2 . Sequences from human cortactin were submitted to NCBI GenBank [ 65 ] as accession No. M98343 (cDNA) and AJ288897 (promoter). Database searching The (deduced) protein and genomic sequences of all cortactin and HS1 genes were retrieved from various WEB-sites and their available sequence data are summarized in Table 1 . In addition, partial cortactin sequences (ESTs and/or genomic) of various organisms were identified based on amino acid sequence homology with existing cortactin proteins. The genomic organization of the sea squirt and Takifugu rubripes could not be completely elucidated, because cDNA/genomic sequences were only partially available. All data were compiled using BLAST searches of the following databases: National Center for Biotechnology Information (NCBI) (Bethesda, MD, USA) [ 65 ]; The Wellcome Trust Sanger Institute (Cambridge, UK) [ 66 ]; EnsEMBL of The Wellcome Trust Sanger Institute (Cambridge, UK) [ 67 ]; DOE Joint Genome Institute (Walnut Creek, CA, USA) [ 68 ]; TIGR: The Institute for Genomic Research (Rockville, MD, USA) [ 69 ]; DNA Data Bank of Japan (Mishima, Shizuoka, Japan) [ 70 ]; Nematode.net Genome Sequencing Center (St. Louis, MO, USA) [ 71 ]; Wormbase (NY, USA) [ 72 ]; European Bioinformatics Institute (EBI) (Cambridge, UK) [ 73 ]; Genoscope National Sequencing Center (Evry, France) [ 74 ]; The U.S. Poultry Gene Mapping Project (MI, USA) [ 75 ] and UCSC Genome Bioinformatics (Santa Cruz, CA, USA) [ 76 ]. To determine the exon/intron boundaries of all cortactin and HS1 genes, available genomic sequences were subjected to sequence alignments of each species-specific cDNA sequence using the BLAST program of NCBI. Using the same algorithms, as described for human cortactin, the exon/intron-boundaries could be predicted. The complete genomic sequences of the 5' flanking region of cortactin of human, chimpanzee, mouse, rat, fruit fly, and mosquito were determined using the various accession numbers of genomic DNA in Table 1 . Putative transcription factor binding sites of 800 bp of the 5' flanking regions where determined by the TFSEARCH program (Figure 2 ). The predicted exon in intron 6 of HS1 was predicted by the bio-informatics program HMMER [ 77 ]) The human cortactin 6 1/2 repeats of the actin-binding domain were aligned, resulting in a consensus sequence: (kfGvqkdrvDksAvGfdyqekvekhesqkDysk). With HMMER this consensus sequence was 'tBLASTn' to intron 6 of human HS1. With an acceptable probability (E-value 0.095), the program predicted an exon in this intron 6 (at location 3271–3381). Amino acid sequence comparisons Sequence alignments were carried out using the BLAST program of NCBI. The multiple sequence alignments of various cortactin proteins were constructed using Basic GeneBee ClustalW 1.83 [ 78 ]. The genome, cDNA or protein was completed for all cortactin homologues and the number of repeats differs across species and between HS1 and cortactin. Only the N-terminal of cortactin and HS1 proteins including repeat 3 (corresponding to amino acid 1–190 of human cortactin) was used to generate a phylogenetic tree, because this is the most conserved part. Predicted nuclear localization signals sequences were obtained using Predict NLS program [ 79 ]. List of abbreviations aa, amino acid(s); bp, base pair(s); BCR, B-cell receptor; EST, expressed sequence tag; HS1, hematopoietic lineage cell-specific protein 1; NLS, nuclear localization signal; RT-PCR, reverse transcriptase polymerase chain reaction; SH3, Src homology; UTR, untranslated region. Authors' contributions AGSHvR designed the study on comparative genome analysis, performed database searches, sequence alignments and gene structure prediction and drafted the manuscript. ESS designed, conducted and analyzed the cloning and sequencing of the promoter of human cortactin. VvBvS conducted and analyzed the PCR and sequencing experiments of the exon-intron boundaries of human cortactin and its splice variants. PMK read the manuscript and provided comments. ES helped with writing the paper, provided overall technical guidance and coordination. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Splice donor and acceptor sequences of cortactin in different species. Click here for file Additional File 2 Splice donor and acceptor sequences of HS1 in different species. Click here for file Additional File 3 Multiple amino acid sequence alignment of cortactin and HS1 homologues. Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC554100.xml

524168

Regulation of signaling genes by TGFβ during entry into dauer diapause in C. elegans

Background When resources are scant, C. elegans larvae arrest as long-lived dauers under the control of insulin/IGF- and TGFβ-related signaling pathways. However, critical questions remain regarding the regulation of this developmental event. How do three dozen insulin-like proteins regulate one tyrosine kinase receptor to control complex events in dauer, metabolism and aging? How are signals from the TGFβ and insulin/IGF pathways integrated? What gene expression programs do these pathways regulate, and how do they control complex downstream events? Results We have identified genes that show different levels of expression in a comparison of wild-type L2 or L3 larvae (non-dauer) to TGFβ mutants at similar developmental stages undergoing dauer formation. Many insulin/IGF pathway and other known dauer regulatory genes have changes in expression that suggest strong positive feedback by the TGFβ pathway. In addition, many insulin-like ligand and novel genes with similarity to the extracellular domain of insulin/IGF receptors have altered expression. We have identified a large group of regulated genes with putative binding sites for the FOXO transcription factor, DAF-16. Genes with DAF-16 sites upstream of the transcription start site tend to be upregulated, whereas genes with DAF-16 sites downstream of the coding region tend to be downregulated. Finally, we also see strong regulation of many novel hedgehog- and patched-related genes, hormone biosynthetic genes, cell cycle genes, and other regulatory genes. Conclusions The feedback regulation of insulin/IGF pathway and other dauer genes that we observe would be predicted to amplify signals from the TGFβ pathway; this amplification may serve to ensure a decisive choice between "dauer" and "non-dauer", even if environmental cues are ambiguous. Up and down regulation of insulin-like ligands and novel genes with similarity to the extracellular domain of insulin/IGF receptors suggests opposing roles for several members of these large gene families. Unlike in adults, most genes with putative DAF-16 binding sites are upregulated during dauer entry, suggesting that DAF-16 has different activity in dauer versus adult metabolism and aging. However, our observation that the position of putative DAF-16 binding sites is correlated with the direction of regulation suggests a novel method of achieving gene-specific regulation from a single pathway. We see evidence of TGFβ-mediated regulation of several other classes of regulatory genes, and we discuss possible functions of these genes in dauer formation.

Background Changes in environmental conditions alter the physiology of all organisms. Evolution and experience create a signaling architecture that assesses current conditions and makes changes in physiology to attain the most appropriate state for a predicted future condition. The dauer larva of C. elegans forms when sensory inputs suggest that food resources will be inadequate for successful reproduction [ 1 ]. Food availability, competition for available food resources (population density, as measured by the concentration of a constitutively secreted pheromone), and temperature are assessed by identified chemosensory and thermosensory neurons, and signals are transduced from these neurons to affect the physiology and structure of most cell types in the body. Development is arrested after the second larval molt, and the third-stage larva that is formed is structurally and behaviorally specialized for dispersal and long-term survival. Figure 1 shows two pathways, related to TGFβ and insulin/IGF pathways in vertebrates, which repress dauer formation under non-inducing conditions [ 1 - 4 ]. Studies of these signaling pathways have led to some understanding of the basic role of these pathways in regulating dauer formation. Environmental cues received by chemosensory neurons regulate the transcription of genes that encode ligands in each of these pathways: daf-7 in the TGFβ pathway and daf-28 and perhaps other insulin-like ligands in the insulin/IGF pathway [ 5 - 7 ]. daf-7 and daf-28 are expressed in chemosensory neurons in the amphid sensillae, which senses environmental cues that regulate dauer formation. Experiments in which components of the TGFβ and insulin/IGF pathways were expressed from tissue-specific promoters suggest that these pathways function predominantly or entirely in the nervous system to control dauer formation [ 8 - 10 ]. Little is known about how transcription factors regulated by these pathways control dauer entry, except that these signaling pathways probably influence a hormonal signal, because most of the cells altered in dauer are not innervated. Two components downstream of these pathways are DAF-12, a nuclear hormone receptor [ 11 ], and DAF-9, a cytochrome p450 that is a putative hormone biosynthetic enzyme [ 12 , 13 ]; these two genes suggest that the secondary signal may be a ligand derived from a lipid. Figure 1 Signal transduction pathways that regulate dauer. Relationships between genes are based on mutant phenotypes and genetic interactions, gene expression in mutants, and homology to pathways in other organisms. Dauers are long-lived [ 1 ], and mutations in many genes, particularly insulin/IGF pathway genes, affect C. elegans lifespan [ 14 ]. The role of the insulin/IGF pathway in aging appears to be shared by related pathways in Drosophila and mice [ 15 ]. Unlike mutations in insulin/IGF pathway genes, mutations in TGFβ pathway genes do not extend adult lifespan. Many genes have been shown to be regulated by dauer formation and by the insulin/IGF pathway. However, no previous studies have focused on regulation of the decision to enter dauer. In the present study, we have used microarray-based analysis of gene regulation in C. elegans TGFβ mutants to identify genes that are directly and indirectly regulated by TGFβ signaling during the dauer formation process. Using a stringent definition for significance, we identified over 1200 genes that are upregulated and downregulated in dauer entry. In this study, we focus on the expression and regulation of genes that encode proteins involved in signal transduction and gene regulation. Analysis of these genes has helped suggest mechanisms by which TGFβ signaling might regulate a secondary hormonal cue and how signaling from TGFβ, insulin/IGF, and other dauer regulators is integrated. We have identified genes that show different levels of expression in a comparison between wild type L2 or L3 larvae (non dauer), and several TGFβ mutants undergoing dauer larva formation (L2d-early dauer). We identify a large number of genes encoding regulatory proteins that are regulated by TGFβ signaling; of special interest are a large group of putative hormone biosynthetic enzymes and hormone receptors. We see strong regulation of many genes related to Hedgehog and its receptor, Patched; our analysis of genes coregulated with these signaling molecules suggests that the dauer-regulated members of this family function in cuticle synthesis or other hypodermal functions. Genetic evidence shows that inactivation of either the TGFβ pathway or the insulin/IGF pathway is sufficient to induce dauer formation. These results can be explained in on of two ways. First, inactivation of either pathway, even with continued activity of the parallel pathway, causes dauer formation. Second, inactivation of one pathway may lead to an inactivation of the other. We show that, when the TGFβ pathway is inactive, expression of several genes is regulated in such a way as to decrease signaling in the insulin/IGF pathway. Also, expression of the daf-12 nuclear hormone receptor gene is strongly upregulated and expression of the daf-9 cytochrome p450 gene is downregulated. These results suggest that feedback inactivation of the insulin/IGF pathway and feedback regulation of other genes may be required in order for TGFβ pathway mutants to cause dauer formation. We suggest that this feedback promotes a clear on/off decision for dauer, even under moderate dauer-inducing conditions, and discuss the relevance of feedback for control of aging. We find altered expression of more than 70 genes with putative DAF-16 regulatory sites, and find that whether these genes are upregulated or downregulated depends on the location of the putative DAF-16 site. Finally, we see regulation of many insulin-related genes as well as genes that encode proteins similar to the extracellular domain of insulin/IGF receptors; this variety of signaling molecules may allow insulin/IGF signaling to have a dramatically different outcome in dauers and adults. Results Array-based hybridization experiments and analysis of expression profiles We compared gene expression in C. elegans wild type to that of animals mutant for the TGFβ pathway genes daf-7 (ligand; [ 5 , 6 ]), daf-8 and daf-14 (Smad transcription factors; [ 8 , 16 ]). At 25°, these daf-c mutants form an L2d instead of the L2 larva, and begin dauer morphogenesis after the L2d to dauer molt. Completion of morphogenesis takes about 12 hours from the molt [ 1 ]. For all experiments, we examined RNA from animals close to the molt between the L2 (or L2d) and L3 (or dauer). We compared wild-type animals that had just completed the lethargus associated with the L2/L3 molt to daf-7 animals that had just completed the lethargus associated with the L2d/dauer molt (four independent experiments). For comparisons of daf-8 or daf-14 to wild type, we used a slightly earlier stage, and compared wild type animals near the beginning of the L2/L3 lethargus to animals near the beginning of the L2d/dauer lethargus (three independent experiments each). We found that many genes showed significant regulation in at least one genotype (t-test p < 0.05; Table 1 ; full data in additional file 1: Table S1 ). More than 90% of the genes showed consistent regulation; if a gene had significantly altered expression in one mutANT, the gene was regulated in the same direction in the other two mutants (Fig. 2 ). This consistency is not surprising, because all three daf-c genes are in the same signal transduction pathway and have similar phenotypes. Because the results from the three daf-c genotypes were broadly similar, we analyzed combined data from all three genotypes. By repeating our statistical analysis on the combined data set, we identified 3248 (out of genes that were significantly up- or down-regulated (P < 0.01). We identified over 1200 genes that were strongly regulated (>2.1 fold, p < 0.01; see additional file 2: Table S2 ). Table 1 Significantly regulated early dauer genes in three TGFβ pathway mutants. Genotype Harvest stage Number of experiments # genes up-regulated p < 0.05 a # genes down-regulated p < 0.05 daf-7 vs. wild type early dauer/L3 4 351 1069 daf-8 vs. wild type late L2d/L2 3 180 679 daf-14 vs. wild type late L2d/L2 3 552 1276 daf-c vs. wild type b 10 2540 3076 a Upregulated genes are expressed at a higher level in daf-c mutants than in wild type, and downregulated are the reverse. b This row is not a summary of the rows above, rather, it is a reanalysis of the data with all ten experiments considered together. Figure 2 Cluster diagram of gene expression responses. Each row represents an average of 3–4 independent experiments. Each gene on the array is represented as a line in each column. The color of the line represents the log2 of the expression ratio, as indicated by the scale bar. We compared our results to published analysis of gene expression in dauers. Many studies have reported the identification of "dauer-regulated" genes, but this term covers a vast range of very different experiments, and we carefully selected data for validation (see Methods). We found that our data was in good agreement with published data from experiments of similar design (see Methods and Table 2 ); 7 of 8 genes examined were similarly regulated in our data and published experiments. Table 2 Comparison of microarray data to published experiments. Gene Method Published difference Fold change, this study P value C27H5.5/ col-36 RT-PCR L2d/dauer >> L2/L3 a [64] 6.8 <0.008 C47G2.1/ cut-1 Northern L2d/dauer >> L2/L3 a [65] 5.7 <0.003 C08A9.1/ sod-3 Northern present in dauer, absent in non-dauer [66] 7.8 <0.00001 T01B7.7/ rol-6 Northern and slotblot L2/L3 >> L2d/dauer a [61] -7.4 <0.00002 B0491.2/ sqt-1 Northern and slotblot L2/L3 >> L2d/dauer a [61] -7.0 <0.0002 T23G5.1/ rnr-1 GFP fusion not seen at L2d/dauer molt, seen in cells in S phase at L2/L3 molt [67] -1.8 <0.05 a W01B6.7/ col-2 Northern and slotblot L2d/dauer >> L2/L3 a [61] 2.2 <0.02 T13B5.4/ col-40 RT-PCR L2d/dauer >> L2/L3 a [64] 1.1 <0.8 a L2d/dauer refers to animals near the molt between these two larval stages; L2/L3 also refers to animals near the molt. b The P value for rnr-1 was calculated using only the local standard deviation because the global standard deviation was not available. Many strongly regulated early dauer genes have DAF-16 regulatory sites We examined the regulated genes in our data set to see if we could identify possible cis-acting regulatory sites. Three Smads (DAF-8, DAF-14, and DAF-3) act downstream of the receptors. Two of the Smads do not have an identifiable DNA binding domain [ 4 , 5 , 16 ], and may serve to negatively regulate the DAF-3 Smad [ 4 , 17 ]. Unfortunately, the optimal binding site for DAF-3 is only 5 bp [ 18 ]. Most genes (>80%) have a DAF-3 binding site within 2000 bp upstream of the ATG (data not shown). Therefore, we focused on the insulin/IGF pathway that regulates DAF-16, which has an eight bp optimal binding site in experiments performed in vitro, TTGTTTAC [ 19 ]. We expect that regulated genes in our data set include targets of DAF-16 for three reasons. First, signaling from the insulin/IGF pathway is decreased in the TGFβ pathway mutants (see below). DAF-16 is negatively regulated by DAF-2 signaling, so we would expect that DAF-16 would be more active in our mutant worms. Second, DAF-16 relocalizes to the nucleus when it is active, and daf-7 mutants accumulate DAF-16 in the nucleus during dauer entry [ 20 ]. Third, DAF-16 and DAF-3 both promote dauer formation, so might have targets in common. We see strong evidence that genes with DAF-16 sites upstream of transcription are likely to be upregulated. 2.7% of genes without DAF-16 sites are strongly upregulated, but genes with DAF-16 sites in all intervals from -1 to -1500 have significantly more upregulated genes (Fig. 3A , additional file 3: Table S3 ). 82 genes with a DAF-16 site within 1500 bp upstream of the translation start are strongly upregulated against a random expectation of 40. We also examined genes that did not meet our most stringent criteria for regulation: these genes were at least two-fold upregulated, but not included in the analysis in Fig. 3A . We found 50 genes with a DAF-16 binding site within 700 bp upstream of the translation start against a random expectation of 24 (Fig. 3B ; additional file 3: Table S3 ). We examined 734 strongly downregulated genes, and see no increased likelihood of upstream DAF-16 sites (58 genes expected, 58 observed). When we examined genes with DAF-16 sites within 200 bp downstream of the stop codon, we see no correlation with upregulation (Fig. 3C ). However, we see 15 downregulated genes with DAF-16 binding sites against a random expectation of 6 (Fig. 3D , additional file 3: Table S3 ). Figure 3 Regulation of genes with DAF-16 binding sites. In all panels, the bars show groups defined by the location of DAF-16 sites, as shown on the X-axis. We used overlapping intervals to allow robust statistical analysis. Negative numbers are bp upstream of the initiation codon, and positive numbers are bp downstream of the stop codon. Asterisks indicate statistical significance (exact hypergeometric probability) of the proportion of strongly regulated genes in the group compared to all genes:*** is p < 0.001, ** is p < 0.01, * is p < 0.05. A. Strongly upregulated (>2.1 fold, p < 0.01) genes. B. Two fold upregulated genes, excluding those genes in panel A. C. Strongly upregulated genes with downstream DAF-16 sites. D. Strongly downregulated (<2.1 fold, p < 0.01) genes. Insulin/IGF pathway signaling is regulated by feedback from TGFβ signaling Insulin/IGF receptor kinase and other downstream signaling molecules Because we found significant regulation of genes with DAF-16 binding sites, we examined regulation of genes in the insulin/IGF pathway, which acts to antagonize DAF-16 signaling. Several genes in the insulin/IGF-like pathway that regulates dauer formation (see Fig. 1 ) are strongly regulated. The most critical of these is daf-2 , which encodes the insulin/IGF receptor, because, downstream of the receptor, the pathway bifurcates, and mutants in downstream genes have weaker phenotypes [ 21 - 23 ]. Downregulation of daf-2 in TGFβ pathway mutants suggests that signaling from the TGFβ pathway positively regulates signaling from the insulin/IGF pathway. Published genetic analysis shows that a nuclear hormone receptor, DAF-12, is antagonized by signaling from the TGFβ receptors [ 3 ]. One means of antagonism may be to regulate a hormone that controls DAF-12 activity. A cytochrome p450 gene, daf-9 , has a partial dauer constitutive mutant phenotype, and acts upstream of daf-12 and downstream of the TGFβ pathway to repress dauer formation [ 12 , 13 ]. In dauers, a daf- 9:: GFP reporter is downregulated in hypodermis and upregulated in a pair of cells in the head [ 12 ]. We see strong downregulation of daf-9 in our data; these results are consistent, given that the hypodermis is at least two orders of magnitude larger than the head cells. We also find that daf-12 itself is strongly upregulated in our data; this is a form of regulation that had been previously unidentified. Upregulation of daf-12 and downregulation of daf-2 and daf-9 would all be predicted to induce dauer formation. Other genes in the insulin/IGF pathway are regulated, but the regulation is complex, with both induction and repression. AKT-1, ATK-2 and PDK-1 are kinases that negatively regulate DAF-16, a FOXO transcription factor. A loss of function mutation in pdk-1 causes a dauer constitutive phenotype that is much weaker than null mutations in daf-2 [ 23 ]. Simultaneous knockdown of akt-1 and akt-2 by RNAi also gives a dauer constitutive phenotype that also is weaker than a daf-2 null. Single knockdown of akt-1 or akt-2 by RNAi does not produce a dauer constitutive phenotype [ 24 ]. The weaker phenotype of akt-1 , akt-2 and pdk-1 is at least partly due to redundancy in the pathway; a second output from DAF-2 is transduced by IST-1 [ 21 ]. The phenotypes may also be weaker because the genes were not completely knocked out by RNAi. We see upregulation of pdk-1 and one isoform of akt-1 , and downregulation of akt-2 , one isoform of akt-1 , and some, but not all, isoforms of daf-16 . The overall effect of this mix of regulation is hard to predict, but suggests variable function for different genes and isoforms. For example, akt-1a , akt-1b and akt-2 may be differently regulated in different tissues, with the result that the output of signal transduction is subtly different. Regulation of insulin/IGF and novel genes with similarity to the extracellular domain of insulin/IGF C. elegans has one insulin/IGF receptor kinase, but dozens of insulin-related genes [ 24 ]. The reason for the abundance of ligands is unclear, but perhaps multiple insulin-like ligandss allow for different temporal and spatial patterns of receptor activation, or produce receptor-ligand complexes with different activities. The structure of the insulin-like genes is quite diverse, with three major groups defined by the pattern of disulfide bonds, and subgroups defined by differences in gene structure [ 24 ]. Diversity in function of insulin like genes has been demonstrated; INS-4, INS-6, INS-7 and DAF-28 have been shown by genetic or biochemical analysis to be receptor agonists [ 7 , 25 , 26 ], and INS-1 and INS-18 act as antagonists [ 24 ]. We see strong upregulation and downregulation of a subset of these insulin-like genes. Insulin-like genes in C. elegans have been divided into three types on the basis of disulphide bonding patterns [ 24 ]. These three groups can be further subdivided on the basis of other sequence features. One subgroup has 9 genes, ins-2 through ins-9 , and daf-28 . Six of these are present on the array, and four ( ins-4 , ins-5 , ins-6 and ins-7 ) show significant downregulation (Table 3 ). The other two insulin-like genes in this family that are present on the array ( ins-2 and ins-3 ) are not significantly changed. INS-6 has been shown to be a receptor agonist in vitro [ 25 ]. Overexpression of ins-4 and ins-6 can partially or completely rescue a daf-28 mutant [ 7 ], and ins-7 RNAi enhances a weak daf-2 mutant [ 26 ], suggesting that these insulin-like ligands act as DAF-2 agonists. Three insulin-like genes are upregulated (Table 3 ); only one of the three, ins-18 , has an identified function, and it acts as a receptor antagonist [ 24 ]. Increased expression of this ligand, like reduction in expression of daf-2 , ins-6 , etc., would be expected to promote dauer formation. Table 3 Dauer and aging genes regulated by TGFβ signaling. Gene Induces or represses dauer? a Fold regulation b p value insulin pathway daf-2 (receptor) strongly represses -3.1 <0.0001 akt-2 represses -3.0 <0.0001 akt-1 all c represses 1.0 ns akt-1a c unknown -1.8 <0.003 akt-1b c unknown 1.4 <0.03 pdk-1 represses 1.8 <0.0001 daf-16a d unknown 1.0 ns daf-16 all d promotes -1.7 <0.0004 Insulin-like ligands ins-4 represses e -2.1 <0.002 ins-5 unknown -2.6 <0.02 ins-6 represses e -2.6 <0.02 ins-7 represses -2.1 <0.005 ins-18 promotes e 1.8 <0.002 ins-33 unknown 3.9 <0.0003 ins-35 unknown 6.5 <0.0001 insulin receptor-like F56A4.C unknown -3.5 <0.0001 Y19D10A.7 unknown -2.8 <0.0003 F14D2.6 unknown 1.8 <0.006 F15E11.11 unknown 1.8 <0.008 other dauer and aging regulators daf-12 strongly induces 2.2 <0.0007 daf-5 induces -2.0 <0.0003 daf-9 represses -4.2 <0.02 scl-1 unknown 5.2 <0.006 ns, not significant This table shows all insulin-like ligands, insulin-receptor-like genes, TGFβ, and insulin/IGF pathway genes with greater than 1.8-fold regulation and a p value less than 0.05. Genes checked for regulation that showed <1.8 fold regulation or p > 0.05 or both: clk-1 , clk-2 , sir-2.1 , daf-11 , daf-7 , daf-1 , daf-4 , daf-14 , daf-8 , daf-3 , daf-19 , tax-4 , npc-1 , daf-18 , age-1 , old-1 , mev-1 , hsf-1 , daf-21 , gpa-2 , gpa-3 , kin-29 , kin-8 , unc-3 and tph-1 , as well as all genes annotated as insulins present on the microarray ( ins-2 , ins-3 , ins-11 , ins-17 , ins-22 , ins-23 , ins-24 , ins-26 , ins-30 , ins-32 , ins-34 , ins-37 ) and nearly all genes identified as putative insulin receptors in Dlakic [27]. a Induction or repression of dauer is defined by mutant or RNAi loss-of-function phenotypes unless otherwise noted. daf-9 mutants have characteristics similar to both types, and are thus marked "mixed". c The fold change between mutant dauers and wild-type. A positive number indicates higher expression in dauers, a negative number indicates higher expression in N2. c The table entry " akt-1 all" is for a probe that recognizes both isoforms. The other two akt-1 entries each cover an exon unique to the indicated isoform. d The entry " daf-16a isoform" is for a probe that recognizes:R13H8.1c ( daf-16a1 ), R13H8.1b ( daf-16a2 ), and R13H8.1d, but not R13H8.1a ( daf-16b ) or R13H8.1e. The entry " daf-16 all" is for a probe that recognizes all five isoforms of daf-16 . e Overexpression of ins-4 or ins-6 represses dauer in a daf-28 mutant [7]. ins-6 has also been shown biochemically to act as a receptor agonist [25]. Overexpression of ins-18 promotes dauer [24]. A family of genes with modest but significant similarity to the ligand binding domain of insulin/IGF receptor has been identified [ 27 ], and these may contribute to insulin/IGF signaling and signaling diversity; however, because of weak similarity to insulin/IGF receptor and similarity to other receptor tyrosine kinase ligand binding domains, a function in insulin/IGF signaling is speculative. As with the insulin-like genes, we see both upregulation and downregulation of insulin-receptor like molecules. Regulation of regulatory genes Genetic and molecular analysis has allowed us to gain a basic understanding of the key regulatory pathways that affect dauer formation by acting in the nervous system, but we know little about regulatory events downstream of these pathways, i.e. the regulators that actually control dauer morphogenesis and physiology. Below, we discuss a selected set of regulated genes; a complete list of regulatory genes can be found in additional file 4: Table S4 . Regulators of aging Because dauers are long lived, and because many genes that control lifespan also control dauer entry, we examined the expression of genes with mutant phenotypes of long or short lifespan. Other than the insulin/IGF pathway genes mentioned above, scl-1 was the only aging regulatory gene we identified as strongly regulated (Table 3 ). SCL-1 has an SCP domain, which defines a family of putative signaling molecules with unknown biochemical function [ 28 , 29 ]. The scl-1 gene is required for extension of lifespan in a daf-2 mutant, and scl-1 expression is upregulated in long-lived genotypes [ 28 ]. scl-1 is strongly upregulated in our data as well, and we suggest scl-1 is part of the mechanism of lifespan extension in dauers. Several other genes in the scl-1 family are upregulated in old animals and in mature dauers, but not in our data [ 29 , 30 , 52 ]. Cytochrome p450s Cytochrome p450s are versatile enzymes that oxygenate a wide variety of compounds [ 31 ]. These reactions are involved in protection from toxins, hormone biosynthesis and other functions. Cytochrome p450s are very prominent among our regulated genes ( additional file 4: Table S4 ); 43 of 75 are regulated >1.8 fold (p < 0.05). One putative hormone biosynthetic cytochrome p450, daf-9 , is known to function in dauer formation, and these others may also participate in hormone metabolism in dauer. Hedgehog/Patched In Drosophila and vertebrates, Patched and Smoothened form a receptor complex that binds Hedgehog ligands. C. elegans has two functional Patched orthologs and a family of proteins similar to Patched, called Patched-related. One, ptc-1 , has an RNAi phenotype of defective cytokinesis in the germ line [ 32 ]. However, none of the patched/hedgehog family genes shown in Table 4 has an RNAi phenotype similar to ptc-1 [ 51 , 33 ]. A group of 10 C. elegans proteins have a carboxyl terminus that is related to the carboxyl terminus of Hedgehog; however, no protein with substantial similarity to the N-terminal signaling domain of Hedgehog has been found in C. elegans . Instead the C. elegans proteins have two novel families of N terminal sequences, called Wart and Ground domains [ 34 ]. Genes that encode both the N-terminal domain and the Hedgehog-related C terminal domain are called warthog and groundhog . Some proteins have only Wart or Ground domains. Finally, a large family of proteins has a domain that is similar to the Ground domain, and these are called Ground-like. Careful examination of the sequence indicates that Wart, Ground, and Ground-like domains and the N terminal domain of Hedgehog, despite low sequence identity, share motifs that indicate descent from a common ancestor [ 34 ]. Table 4 Regulatory genes regulated by TGFβ signaling. Hedgehog and Patched Gene Type of gene Fold regulation p value Mountain grd-2 groundhog -2.4 0.0005 14 grd-7 ground domain only 3.8 0.0003 13 grd-6 ground domain only -3.1 0.0002 16 grd-14 ground domain only -8.9 0.0002 16 wrt-7 warthog 6.7 0.01 17 wrt-1 warthog -6.8 0.0001 14 wrt-4 warthog -2.9 0.0003 14 wrt-6 warthog -2.7 0.0005 14 wrt-8 warthog -3.8 0.0005 14 wrt-2 wart domain only -1.8 0.008 14 C56A3.1 ground-like 4.8 0.0001 17 K03B8.7 ground-like 18.9 0.0003 17 ZC487.4 ground-like 3.4 0.009 17 C24G6.7 ground-like -3.6 0.0006 14 F42C5.7 ground-like -2.4 0.0007 1 T02E9.2 ground-like -7.7 0.0001 14 Y75B8A.20 ground-like -13.3 0.0001 14 ZC168.5 ground-like -4.6 0.0001 16 ptc-3 patched -2.1 0.0001 1 ptr-6 patched related -2.3 0.0008 14 ptr-11 patched related -2.6 0.0002 1 ptr-16 patched related -2.7 0.01 14 ptr-18 patched related -3.3 0.0002 16 ptr-23 patched related -2.5 0.0003 6 npc-2 patched family -2.5 0.0004 16 Dauer growth arrest Gene Type of gene Fold regulation p value dbl-1 TGFβ ligand, regulates growth and body size -2.4 0.001 cdk-4 cyclin dependent kinase -1.9 0.006 nhr-73 member of a family of NHR genes expressed exclusively in lateral hypodermis (seam cells) -4.4 0.00002 nhr-74 member of a family of NHR genes expressed exclusively in lateral hypodermis (seam cells) -5.5 0.00005 nhr-25 similar to NHR in Drosophila ecdysone regulatory cascade, required for embryogenesis and molting -2.4 0.001 G-protein-coupled receptors and olfaction gene type of gene fold regulation p value srh-75 chemoreceptor 2.6 0.007 srh-195 chemoreceptor 2.6 0.0009 srj-32 chemoreceptor 5.8 0.002 sru-21 chemoreceptor 6.3 0.0007 32 genes chemoreceptor 1.8 to 6.3 0.05 or less 15 genes GPCR, not chemoreceptor type 1.8 to 3.6 0.05 or less C30F12.6 thyrotropin-releasing hormone receptor ortholog 1.8 0.002 npp-10 GPCR, nucleoporin -1.8 0.001 6 genes chemoreceptors -1.8 to -3.8 0.05 or less gpa-10 G protein alpha subunit 2.0 0.003 Y71H2B.7 G protein alpha subunit -1.8 0.0001 F45B8.2 regulator of G-protein signaling domain 2.4 0.03 gcy-22 receptor guanylate cyclase 2.7 0.03 gcy-31 soluble guanylate cyclase 1.8 0.01 gcy-34 soluble guanylate cyclase 1.8 0.03 tax-2 cyclic nucleotide gated channel beta subunit 3.0 0.02 lim-6 homeobox protein, functions to allow chemosensory neurons to sense different molecules 2.0 0.003 Notch pathway gene type of gene fold regulation p value W02C12.1 notch family 3.9 0.0007 F47C12.1 notch family 7.4 0.0003 apx-1 ligand for GLP-1, notch receptor 1.8 0.02 R03D7.5 shaggy/GSK3 kinase 2.2 0.002 lag-1 ortholog of CBF1 and Suppressor of Hairless 1.6 0.0003 Transcriptional regulators gene type of gene fold regulation p value lin-28 cold-shock domain 1.8 0.0001 lin-29 zinc finger transcription factor -1.9 0.0002 A complete list of significantly regulated regulatory genes can be found in additional file 4: Table S4. We see abundant regulation of these gene families in our data (Table 4 ). Both upregulated and downregulated genes are observed, but downregulated genes are much more abundant. Olfaction and other G-protein-coupled receptor signaling Two of the three well-characterized cues for dauer entry are food and pheromone, which are sensed by chemosensory neurons. Chemosensation in C. elegans is mediated by G protein coupled receptors (GPCRs), which regulate cyclic GMP levels [ 35 ]. We see many upregulated G protein coupled receptors, accessory proteins, and other proteins related to olfaction, but few downregulated genes of these types (Table 4 ). Cell cycle and growth arrest Many cell types undergo cell cycle arrest in dauers. We see downregulation of cdk-4 (homologous to human Cdk-4/Cdk-6 ), which is well suited to control cell cycle arrest in dauers. The cyd-1 (cyclin D) and cdk-4 genes function in larval development to promote exit from the G1 stage of the cell cycle [ 36 ]. Animals with loss of function of either gene fail to carry out postembryonic divisions. We see a significant reduction in expression of cdk-4 (Table 4 ), which would be predicted to cause cells that would normally divide in reproductive L3 to arrest in dauers. However, loss of cdk-4 would not be expected to lead to an arrest of all growth in dauers. Some cells in C. elegans have endoreduplication without cell division, producing large hyperploid cells. DBL-1 is a ligand in a TGFβ pathway that controls larval growth [ 4 , 37 ]. dbl-1 mutants grow less than wild type, and have lesser DNA content in hyperploid hypodermal cells [ 38 ]. We see downregulation of dbl-1 in dauers (Table 4 ), which may block endoreduplication, and thereby arrest growth, in cell types such as intestine and hypodermis. Thus the microarray analysis suggests a model for the arrest of both cell division and endoreduplication. Notch Notch pathways have not been previously implicated in dauer formation, but we see strong evidence for altered activity of these pathways in dauer entry. Two of nine genes annotated as Notch receptors in C. elegans show strong upregulation in our data, as does one ligand, apx-1 . We also see regulation of two genes that act downstream of the receptors: a shaggy/GSK3 kinase and lag-1 , a transcription factor [ 20 ]. The regulation of these various Notch pathway components suggests that Notch signaling has an important unidentified function in dauer formation. lin-28 and lin-29 These two genes are part of a signal transduction cascade that regulates timing of developmental events during larval development [ 39 ]. During larval growth, abundance of the cold shock domain containing LIN-28 protein is post transcriptionally downregulated. This downregulation allows the zinc-finger transcription factor LIN-29 to accumulate and promote events appropriate to the third larval stage. In our data, lin-28 is upregulated, and lin-29 downregulated (as expected, since lin-28 negatively regulates lin-29 ). This regulation is consistent with some old observations of the role of these genes in dauer formation [ 40 ]. lin-28 loss-of-function mutants enter dauer, but have defects in dauer morphogenesis. These defects are suppressed by lin-29 mutations. The regulation we see is consistent with the prediction that downregulation of lin-29 by lin-28 is required for normal dauer morphogenesis [ 40 ]. Discussion Feedback model for crosstalk between TGFβ pathway and other dauer regulatory genes We find that mutations in TGFβ signaling genes affects the expression of numerous genes that are known regulators of dauer formation. For example, loss of TGFβ signaling causes reduction of expression of daf-2 , would be expected to reduce signaling through the insulin/IGF pathway. In addition, the TGFβ pathway mutants have increased expression of daf-12 and reduced expression of daf-9 , a cytochrome p450 that is predicted to be involved in the biosynthesis of a daf-12 antagonist. All of these changes in gene expression would be predicted to promote dauer formation, and suggest that one way that the wild-type TGFβ pathway promotes reproductive growth is by feedback regulation of other dauer pathway genes. We propose that this regulation is part of a feedback mechanism that operates at many levels to insure a "clean" dauer/non-dauer decision. Under conditions that induce dauer formation to different extents, different percentages of the population form dauers, but individual animals either undergo morphologically complete dauer, or have no dauer morphogenesis; therefore, a mechanism must exist to convert ambiguous signals into a clear decision. If reduced TGFβ signaling causes a reduction in insulin/IGF signaling and vice versa, then weak signals can be strengthened to make a clear, organism-wide decision. Negative regulation of daf-12 and upregulation of daf-9 by TGFβ and insulin/IGF signaling would likewise amplify signals. Interestingly, daf-2 and daf-12 both function as "nodes" for signal transduction in dauer entry. Many genes function in TGFβ and insulin/IGF pathways in dauer formation, but all of them except daf-2 show genetic redundancy [ 1 , 3 , 8 , 21 - 23 , 41 ]. Regulation of genes other than daf-2 would produce a weaker effect on dauer formation. Similarly, daf-12 is the gene that TGFβ and insulin/IGF signaling converge on, and regulation of daf-12 would be expected to have a uniquely powerful effect on dauer formation. The feedback regulation to the insulin/IGF pathway explains published observations that link the TGFβ pathway to the insulin/IGF pathway. First, daf-16 mutants partially suppress daf-7 and other mutants in the TGFβ pathway [ 3 ]. We suggest that feedback from the TGFβ pathway to the insulin/IGF pathway is essential for complete dauer formation. In this model, loss of daf-16 function does not have a direct effect on TGFβ signaling, rather, daf-16 mutants suppress the inactivation of insulin/IGF signaling that occurs in the TGFβ mutants. Second, DAF-16 is localized to the nucleus when insulin/IGF signaling is weak, for example in daf-2 mutants of all ages. DAF-16 is localized to the nucleus in daf-7 mutants that are entering dauer, but not at other stages [ 20 ]. We suggest that this localization is a consequence of feedback regulation from the TGFβ pathway to the insulin/IGF pathway. Third, insulin/IGF pathway mutant adults have a variety of stress-resistance phenotypes, including slowed aging, which are shared with dauers, but not with TGFβ pathway mutant adults. We suggest that the stress resistance seen in daf-7 dauers is mostly or entirely caused by feedback regulation of the insulin/IGF pathway, and that this feedback occurs in daf-7 mutant dauers, but not at other stages. Regulation of insulin-like ligands and novel genes with similarity to insulin/IGF receptors suggests functions for these genes in dauer formation We find several insulin-related genes and genes with similarity to insulin receptor that are regulated by TGFβ signaling. Four insulin-like genes are downregulated; these are members of a subfamily in C. elegans that encodes several agonists of insulin/IGF signaling. Published data suggest that three of the downregulated insulin-like ligands act as DAF-2 (receptor kinase) agonists [ 7 , 25 , 26 ]. Thus, downregulation of these genes and perhaps the closely related gene ins-5 would be predicted to promote dauer entry, and may be part of the feedback mechanism proposed above. Published data suggest that ins-18, which is upregulated in our data, acts as a DAF-2 antagonist; [ 24 ] upregulation of this gene would be predicted to promote dauer entry, and this gene is another candidate for participating in the feedback mechanism proposed above. Upregulation of ins-33 and ins-35 suggests that these genes, like ins-18 , encode receptor antagonists. The ins-7 and ins-18 genes, but not the other insulin-like genes in Table 3 , are regulated by insulin/IGF signaling in adults. Conversely, several insulin-like genes that are not regulated in dauer entry are regulated by insulin/IGF signaling in adults [ 26 , 42 ]. The gene programs induced by in dauers have similarities to those induced in long-lived adults; for example, stress resistance genes are common to both. However, many gene regulatory programs are unique to dauer, for example growth arrest. The different spectrum of insulin-like genes expressed in dauer entry versus adults may explain at least part of the difference in events regulated by insulin/IGF signaling. As with the insulin-like genes, we see both upregulation and downregulation of insulin-receptor like molecules; this regulation identifies these genes as good targets for studies to determine if this gene family is involved in insulin/IGF pathway signal transduction. These results suggest that diversity in insulin-related ligands, and perhaps in receptors, contributes the ability of insulin/IGF signaling to produce dramatically different phenotypes in dauers and adults Many strongly regulated early dauer genes have DAF-16 regulatory sites We observed that genes with putative daf-16 binding sites show a tendency to be upregulated in our data, with the exception of a small number of genes with daf-16 sites downstream of the transcription start. These results are important in several ways. Most significantly, our data identify genes that may be directly regulated by DAF-16 to control dauer formation or to provide dauer-specific functions such as protection from environmental insults. Second, genetics predicts that DAF-16 might upregulate genes that promote dauer entry or downregulate genes that promote reproductive growth. Our results suggest that DAF-16 can do both, but upregulated genes are much more prominent in our data set. Third, we see that the direction of regulation is correlated with the position of the DAF-16 site. Our results show interesting similarities and differences with recent reports about DAF-16 target genes [ 53 , 54 , 42 , 43 ]. A significant number of genes that are upregulated in our data are also upregulated in these other experiments, but some genes are differently regulated. These papers all examined gene expression in adults, and in insulin/IGF pathway mutants, whereas our experiments have perturbed but not mutated insulin/IGF signaling (see below), so we would expect that only a subset of genes would be in common. The most dramatic difference between our data and these published reports is that we see evidence for upregulation of genes by DAF-16 sites upstream of the transcription start, but not downregulation, whereas the other reports see as many upregulated genes as downregulated genes. These differences suggest that the activity of DAF-16 is fundamentally different in dauers and adults, presumably because of availability of different cofactors. Other regulatory genes Cytochrome p450s A very large number of cytochrome p450s show strong regulation in our data. One putative hormone biosynthetic cytochrome p450, daf-9, is described above, and these others may also participate in hormone metabolism. Other microarray experiments have identified cytochrome p450s as regulated genes in dauer or under the control of insulin/IGF signaling in C. elegans [ 25 , 42 , 52 ]. Because of the diversity of functions of cytochrome p450s, and especially because of the prominent roles of cytochrome p450 in stress resistance, sorting out the function of these genes in defense and hormone biosynthesis will require detailed functional analysis. Hedgehog/Patched The function of these genes is uncertain, as most have not been studied. Based on the expression pattern of several Ground and Wart domain proteins, Aspock et al .[ 34 ] suggested a function for these genes in hypodermis. A comparison of regulated Hog and Patched genes to groups ("mountains") of putatively coregulated genes in C. elegans [ 44 ] shows that three mountains are highly enriched for these genes. Mount 14 and Mount 16, are enriched for downregulated genes, and Mount 17 is enriched for upregulated genes (Table 4 ). These mountains are even more highly enriched for hog/patched genes that are strongly regulated in our experiments. These three mountains have less than 5% of the genes in C. elegans , but 48% of hog/patched genes and 80% of hog/patched genes that are strongly regulated in our data. All three mountains are also enriched for cuticular collagens; thus, we suggest that the hog and patched genes function in the production of the cuticle by the hypodermis. The fact that we see substantial numbers of both upregulated and downregulated genes supports this hypothesis. Dauers and non-dauers each form a cuticle, but with dramatically different structures [ 1 ]; therefore, we would expect to see the same types of genes in each program, just as we do when examining expression of cuticle structural genes such as collagens and cuticlins (Table 2 and data not shown). Olfaction and other G-protein-coupled receptor signaling We see regulation of many genes that are candidate olfactory and gustatory signal transduction molecules. This regulation may be involved in changes in response to odor and taste cues in dauers. We see many more upregulated genes in this class than downregulated genes. However, both dauer and non-dauer larvae are responsive to chemical cues, and published results of GPCRs with dauer-regulated changes in expression of reporters are not noticeably biased toward upregulation [ 45 , 46 ]. Perhaps anatomical changes can explain the bias toward upregulation. The key sensillum for regulation of dauer formation is the amphid. In dauers, the amphid pore is filled with an unknown material. In addition, ASI and ASG, which are members of the subset of amphid neurons with a known role in dauer formation, have shortened nerve endings that are more distant from the pore [ 1 ]. These changes insulate the neurons from the environment, to an extent. For example, lipophilic dyes such as DiI are taken up by amphid neurons in all stages of C. elegans development, except in dauers [ 45 , 46 ]. However, dauers respond readily to chemical cues that regulate dauer formation, which are thought to be sensed by amphid neurons. Perhaps genes that are required for chemosensation are more strongly induced in dauers in order to compensate for reduced signals due to the reduced exposure of the sensory endings. Conclusions In summary, perhaps the most striking conclusions from study are that, in TGFβ pathway mutants, we see altered expression of many genes that are known regulators of dauer formation, and that regulation of DAF-16 target genes depends on the location of the DAF-16 binding site. We see that loss of TGFβ signaling causes changes in gene expression that would be expected to reduce signaling through the insulin pathway. In addition, in TGFβ mutants, we see increased expression of daf-12 and reduced expression of daf-9 , a cytochrome p450 that is predicted to be involved in the biosynthesis of a daf-12 antagonist. We also see changes in expression of several insulin-related genes and putative non-kinase insulin receptors. This result suggests that diversity in insulin-related ligands, and perhaps in receptors, contributes the ability of insulin signaling to produce dramatically different phenotypes in dauers and adults. We see evidence that DAF-16 binding sites upstream of the coding region can promote upregulation, but not downregulation of gene expression. We identify many other regulatory genes that have altered expression in TGFβ mutants. We see several genes likely to function in Notch pathway signaling; this is the first implication of Notch signaling in dauer formation. We find that many divergent C. elegans homologs of hunchback and patched are regulated by TGFβ. Because these genes are coregulated with genes that encode structural components of the cuticle, we suggest that these genes regulate the production of the dramatically different cuticles of dauer and reproductive L3. We see coordinate regulation of genes that are predicted to arrest cell division and endoreduplication, which together would be expected to promote growth arrest. We identify many genes that are candidate hormone biosynthetic enzymes that might help transduce signals from the nervous system to other tissues. Figure 4 is a graphic summary of the main conclusions of this work, with new regulatory relationships suggested by our data shown in red. Figure 4 Model for gene regulatory events under the control of TGFβ signaling. Transcriptional regulatory events suggested by expression data in this paper are indicated in red. The TGFβ pathway is shown in green; DAF-3 and DAF-5 form a transcription factor complex that function in neurons to control dauer entry [9]. DAF-3 and DAF-5 are shown regulating DAF-9 and DAF-12 indirectly because these genes are likely to be regulated in non-neuronal tissues. Insulin genes are shown as directly regulated by DAF-3 and DAF-5, but it is equally likely that these genes are regulated by feedback from the insulin pathway or by DAF-12. Methods RNA Isolation, cDNA synthesis, and microarray hybridization Wild type is C. elegans variety Bristol, Strain N2; Daf-c genotypes are daf-7(e1372) , daf-8(e1393) , and daf-14(m77) . Temperature-sensitive strains were maintained at 15°. Eggs were collected by hypochlorite treatment and incubated overnight at 15°. L1 larvae were grown on plates at 25°. During C. elegans larval development, each molt is accompanied by a brief period of lethargus when the animal stops pharyngeal pumping, and sheds its old cuticle. Cessation of pumping was scored according to [ 47 ] and used to judge the stage of worms for harvesting. For comparing daf-7 to N2, worms were harvested at early L3 stage for N2, when >95% of the worm had entered the L3 stage and daf-7 at early dauer stage when > 85% of the worm had passed the L2d to dauer molt. Synchronization of development in C. elegans populations is imperfect. For daf-8 /N2 and daf-14 /N2 experiments, a small sample plate was started 2 hours ahead of the harvest plates. The worms were harvested at late L2 stages for N2 when the worm on sample plate showed maximal cessation of pumping and the very first animals on the harvesting plates entered lethargus. daf-8 and daf-14 animals were harvested at late L2d stage when 50% worm on the sample plates stopped pumping and the very first animals on the harvesting plates entered lethargus. These animals were not perfectly synchronous, but our monitoring of pumping suggests that the vast majority of worms fall within a four-hour developmental window. Worms were harvested quickly to avoid changes in mRNA levels caused by the collection procedure; no more than 15 minutes elapsed between the start of harvest and addition of Trizol. Each experiment compared RNA samples from the wild-type N2 to Daf-c genotype worms. For each Daf-c genotype, we did three or four independent experiments. RNA was prepared as described previously [ 48 ]. Labeled cDNA probe for DNA microarray hybridization was made from 5 μg of poly (A) + as described [ 49 ]. The two cDNA probes were simultaneously hybridized to a single DNA microarray. Imaging and data analysis Arrays were scanned using Axon scanner as described previously [ 49 ], collecting measurements for Cy5 and Cy3 separately. The average level of regulation for each gene for each genotype was calculated as the mean of the Cy5/Cy3 ratios. Genes with significant regulation were identified by using the standard deviation from the ten experiments presented in this paper ("local standard deviation") and the standard deviation derived from a several hundred microarray experiments of various types ("global standard deviation"), as described [ 50 ]. In most cases, genes were considered significantly regulated only if they were significant in a Student's t test using both measurements of standard deviation. However, for a few hundred genes, global standard deviation was unavailable, and significance was measured using the local standard deviation. Our processed, annotated data are available in additional file 1: Table S1 . Raw data compliant with the "Minimum information about a microarray experiment" (MIAME) standard is available at: . Analysis of duplicated and reannotated spots All spots on the array were generated by PCR of genomic DNA, as described [ 49 ]. Some PCR reactions were spotted in two separate locations on the array. In this case, both data points were used to calculate the average regulation. The PCR primers were designed a few years ago, and in some cases, EST data or other information has revealed that predictions upon which the primers were based were in error [ 51 ]. Three types of error account for the vast majority: 1) PCR products that are no longer believed to amplify sequence from a gene. These spots are excluded from the analysis. 2) Cases in which two PCR products that were thought to represent two different genes actually represent different parts of a single gene. In these cases, each spot was initially analyzed independently, but, when we count the number of genes with a particular property, the genes are not double counted. 3) Cases in which a single PCR product is now believed to overlap two genes. Data from these spots were excluded from the analysis. Fortunately, this type of error was uncommon. Cross hybridization Many genes in C. elegans have enough mRNA sequence similarity that our labeled cDNA probes might hybridize to homologous genes, obscuring true patterns of gene regulation. We examined our data and published data to evaluate how widespread this problem is. As a crude measure, we compared pairs of homologous genes and asked whether the genes ever show opposite regulation. Our expectation is that, if cross-hybridization is strong, the two genes will show similar regulation. That is, if one member of a pair shows upregulation in dauers and the other member shows down-regulation, then cross-hybridization is not at a high enough level to obscure true regulation. Of course, the converse is not necessarily true: if two homologous genes show similar regulation, that does not imply that cross hybridization is occurring. We examined 8 groups of genes using blast to identify the most similar genes. In this way, each of the eight genes was matched with a small group of similar genes (6 were collagens, one was cuticlin, and one was superoxide dismutase). In seven groups, the match was approximately 80% identity over 200–700 bp, and several groups had short stretches (<100 bp) of 90% identical sequence. In each group, significantly different regulation was seen in the data collected for this paper. The last group had matches of >90% over 500 bases or more, and this group showed strong correlation in our experiments. This analysis defines the limits of specificity. For genes with identities of <80%, and genes with short stretches of 90% identity, cross hybridization is not strong enough to obscure differences between genes. When similarity goes above 90% for more than 100 bp, genes show strongly correlated expression, and these results should be interpreted with caution. Our examination of the collagen, cytochrome p450, glutathione S-transferase, heat shock, and peroxidase, insulin ligand, and insulin receptor families of C. elegans indicates that no more than 10% of genes in families have similarity great enough to be an issue. Of course, this number applies to genes in families, and the bulk of C. elegans genes will have less similarity. Validation of the DNA microarray results We examined our data to see if it was consistent with published data on dauer gene regulation obtained by Northern blot and reporter gene constructs. The term "dauer-regulated" has been used in many published studies to describe results from experiments that are only superficially similar. It would be inappropriate and misleading to compare our data to most published data on "dauer-regulated" genes. For example, Wang, et al. [ 52 ] identified dauer-regulated genes, but they were examining wild-type animals over time (rather than comparing mutants to wild type) and studying the process of exiting dauer (these animals were several days older than the animals we were studying. If we were to use this data to validate our data, we would be making the assumption that the set of genes that are regulated during dauer entry are similar to the set of genes that are regulated in dauer exit. There is no data to support this assumption. Comparison of our data set to the Wang et al. [ 52 ] would be appropriate to identify interesting similarities and differences between dauer entry and exit, but to use one data set to validate the other would be an error. Most experiments that we consider inappropriate for use in validation have one or both of the following characteristics: 1) In our experiments and published experiments summarized in Table 2 , animals from which RNA is collected are of similar age, either dauers near the L2d/dauer molt or non-dauers near the L2/L3 molt. Many experiments compare dauer RNA to that of mixed stage worms or adult worms [ 53 - 55 ]. which is not useful for our goal of comparing the L3 to the dauer alternative. For example, using mixed stage RNA, a gene that is expressed at a high level in adults but is otherwise off, would be expected to show expression in a mixed-stage RNA prep, but not in dauer RNA. Yet a gene of this sort does not have a meaningful relationship to the dauer/non-dauer decision or dauer morphogenesis. Similarly, a gene that is expressed equally in dauers and in non-dauer L3 animals would be diluted by RNA from other stages, and therefore, less abundant, in the mixed stage RNA than in the dauer RNA, but the gene is not in fact more abundant in dauers than the non-dauer alternative. 2) Many studies use dauers ranging from about 15 hours to 7 days past the L2d/dauer molt [ 52 , 54 - 59 ]. Some of these studies also compare starvation-induced dauers to well-fed animals. These studies are often aimed toward understanding issues related to survival of dauers and aging. We would not expect that these studies would necessarily find the same genes to be regulated, because of differences in timing and differences in physiology due to starvation. In particular, we are looking at the very beginning of dauer morphogenesis, so would expect to see genes that are necessary for dauer formation, while the other studies are looking after the completion of dauer morphogenesis, so would expect to see genes necessary for dauer maintenance and survival. We identified the small subset of experiments that are appropriate for validation of our data set, by identifying experiments that: compare animals close to the L2/L3 and L2d/dauer molts. Table 2 shows that our data agrees very well with published experiments of similar design. The first five genes show strong regulation in the published experiments, and 7–8-fold regulation in our experiments. The next two genes, rnr-1 , and col-2 , are also regulated similarly in our experiment and published data, although the magnitude and statistical significance are not as great as for the first 5 genes. The final gene, col-40 , shows strong upregulation at the L2d-dauer molt in published experiments, but in our data does not show regulation. One possible explanation for this discrepancy is that this gene has been shown to have very rapid induction and repression near the time of molting [ 60 , 61 ]. Thus, seeing regulation of col-40 may depend on precisely when the RNA is collected. This gene showed very high variation in mutant/wildtype ratio from experiment to experiment, higher than 99% of the genes on the array. This unusual variation is consistent with the idea that timing is critical for measuring regulation of this gene. A second possible explanation is that the regulation of these genes is obscured by cross-hybridization to other collagen genes; however, the similarity of col-2 and col-40 to other genes at the level of coding DNA is low enough that we do not expect cross-hybridization to be a serious problem (see previous section of Methods). Overall, seven of eight genes show strong correlation with published data. Identification of DAF-3 and DAF-16 binding sites Binding sites for transcription factors were identified using RSA tools [ 62 , 63 ]. Authors's contributions TL participated in the design of the study, performed wet lab experiments, analyzed data, and co-wrote the manuscript. KKZ participated in the growth and collection of samples for RNA. GIP conceived of the study, participated in the design and data analysis, and co-wrote the manuscript. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Full data set. Table S1 may be found online as an Excel file at: Please feel free to use any annotation in these tables, provided that the original source of the data is cited, and this collection of annotations is cited. Table S1 has the data for all of the genes on the microarray. Raw data in unprocessed, MIAME compliant format is available at: . Column A, "SMD gene name" is the name given to the probe on the microarray. The probe is a PCR product from a pair of primers made to amplify the ORF named in this column (Reinke, V., Smith, H. E., Nance, J., Wang, J., Van Doren, C., Begley, R., Jones, S. J. M., Davis, E., Scherer, S., Ward, S., & Kim, S. K. (2000). Mol. Cell 6 , 605–616). Many of these ORF predictions have changed, and the current match is in column B. Column B, "wormbase gene name" gives the current gene prediction that is complementary to the probe on the array. Columns C-K are annotations downloaded from Wormbase in 2003, from April to June. The column headings are defined in Wormbase. Columns L and M are optimal DAF-16 binding sites, which are the sequence TTGTTTAC. Negative numbers are sites upstream of the start of translation, and positive numbers are downstream of the stop codon. These sites were predicted in Fall 2002 using data and software from (van Helden, J., André, B., & Collado-Vides, J. (2000). Yeast 16 ,177–187.). All sites within 2000 bp upstream and 300 bp downstream are shown, except for a few (<2%) that we were not able to match because of different annotation in our database and the rsat database. Columns N-O are our annotation of function based on sequence similarity, annotation from wormbase, and published reports. For the following groups, we have annotated all genes in the group, to the best of our knowledge: G protein coupled receptors, glutathione S transferases, cytochrome p450s, heat shock proteins, peroxidases, UGTs, epoxide hydrolases, collagens, cuticlins, NRF6 related, scl-1 familly (aka CRISP family), signaling proteins, and transcription factors. For the following groups, we have annotated only a subset of genes in the group: amine oxidases, ribosomal proteins, amino acid catabolism, and lipid metabolism. Column P, "mountains", lists groups of putatively co-regulated genes from Kim, S. K., Lund, J., Kiraly, M., Duke, K., Jiang, M., Stuart, J. M., Eizinger, A., Wylie, B. N., & Davidson, G. S. (2001). Science 293 , 2087–2092. Column Q and R list the average ratio of signal from daf-c genotypes to wild-type, using all experiments. Column Q is the average expressed as a base 2 logarithm (>0 means the expression was higher in daf-c, <0 is higher in N2, and Column R is the average expressed as a fold change (for both columns, a positive number means the expression was higher in daf-c , a negative number means the expression was higher in N2). Column S is the p value for the values in column Q and R, using a t test, asking the question, given the standard deviation for the data, is the value significantly different from a ratio of 1 (or 0 for the base 2 log). For these calculations we used the local standard deviation and the global standard deviation as described in Jiang, M., Ryu, J., Kiraly, M., Duke, K., Reinke, V., & Kim, S. K. (2001). Proc. Natl. Acad. Sci. 98, 218–223. Column T is the number of successful experiments for each spot on the array. The numbers vary because the data for a particular spot may or may not be of acceptable quality for a given experiment. Some genes have a number greater than 10 (the total number of independent experiments) because the same PCR product was put on the array in two different locations. Columns U thru BH are the data for each spot for each experiment. The column labeled "CH1D_MEAN" is the data for wild-type sample, the column labeled CH2D_MEAN is the data for the daf-c mutant sample, the column labeled CORR is the correlation coefficient for the average ratio of channel 1 to channel 2 calculated pixel by pixel, and the column labeled Flag indicates the data quality. Any value other than 0 indicates that the data had problems and was not used for analysis. Columns BL to CG give the data broken down by genotype (for each of the three daf-c genotypes used in this study). For each genotype, the first three or four columns give the base 2 log of the ratio of channel 2 (from daf-c genotype) to channel 1 (from wild type) for each experiment. Blank cells indicate bad data, not used in the analysis. The five digit number in the headings of these colums refer to the experiment ID used to catalog data at the Stanford Microarray Database. The next column gives the average ratio, the next column the standard deviation and the next gives the number of successful experiments, and the next the p value, asking the question, given the standard deviation for the data, is the value significantly different from a ratio of 1 (or 0 for the base 2 log). Click here for file Additional File 2 Strongly regulated genes. Table S2 may be found online at: Table S2 has data identical to table 1, except that only genes for which the regulation was greater than 2.145 fold, with a p value less than 0.01 are shown. Click here for file Additional File 3 Regulated genes with putative DAF-16 binding sites. Table S3 can be found online at: All data in this table are taken from Table S1. The first 15 rows have data for downregulated genes that have a daf-16 binding site downstream of the stop codon. The next 134 rows have data for genes that are upregulated and have a daf-16 binding site upstream of the start codon. Column A, "SMD gene name" is the name given to the probe on the microarray in the Stanford Microarray Database. The probe is a PCR product from a pair of primers made to amplify the ORF named in this column (Reinke, V., Smith, H. E., Nance, J., Wang, J., Van Doren, C., Begley, R., Jones, S. J. M., Davis, E., Scherer, S., Ward, S., & Kim, S. K. (2000). Mol. Cell 6 , 605–616). Many of these ORF predictions have changed, and the current match is in column B. Column B, "wormbase gene name" gives the current gene prediction that is complementary to the probe on the array. Columns C-G are annotations downloaded from Wormbase in 2003, from April to June. The column headings are defined in Wormbase. Columns H-J are optimal DAF-16 binding sites, which are the sequence TTGTTTAC. Negative numbers are sites upstream of the start of translation, and positive numbers are downstream of the stop codon. These sites were predicted in Fall 2002 using data and software from (van Helden, J., André, B., & Collado-Vides, J. (2000). Yeast 16 ,177–187.). All sites within 2000 bp upstream and 300 bp downstream are shown, except for a few (<2%) that we were not able to match because of different annotation in our database and the rsat database. Column K is our annotation of function based on sequence similarity, annotation from wormbase, and published reports. Our annotation of the following gene types is complete, to the best of our knowledge: G protein coupled receptors, glutathione S transferases, cytochrome p450s, heat shock proteins, peroxidases, UGTs, epoxide hydrolases, collagens, cuticlins, NRF6 related, scl-1 familly (aka CRISP family), signaling proteins, and transcription factors. The following groups are incompletely annotated: amine oxidases, ribosomal proteins, amino acid catabolism, and lipid metabolism. Column L, "mountains", lists groups of putatively co-regulated genes from Kim, S. K., Lund, J., Kiraly, M., Duke, K., Jiang, M., Stuart, J. M., Eizinger, A., Wylie, B. N., & Davidson, G. S. (2001). Science 293 , 2087–2092. Column M lists the average ratio of signal from daf-c genotypes to wild-type, using all experiments. expressed as a fold change (a positive number means the expression was higher in daf-c , a negative number means the expression was higher in N2). Column N is the p value for the values in column M, using a t test, asking the question, given the standard deviation for the data, is the value significantly different from a ratio of 1 (or 0 for the base 2 log). For these calculations we used the local standard deviation and the global standard deviation as described in Jiang, M., Ryu, J., Kiraly, M., Duke, K., Reinke, V., & Kim, S. K. (2001). Proc. Natl. Acad. Sci. 98 , 218–223. Click here for file Additional File 4 Regulated regulatory genes. This table lists all transcription factors, signaling molecules, possible hormone biosynthetic enzymes, and other regulatory genes that are significantly regulated in our data. Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524168.xml

527878

Atherosclerotic renal artery stenosis: one year outcome of total and separate kidney function following stenting

Background Renal artery stenosis (RAS) is a known cause of hypertension and ischemic nephropathy. Stenting of the artery is a valid approach, in spite of cases of unexpected adverse evolution of renal function. Methods In this study, 27 patients with unilateral RAS were subjected to stenting and followed for a period of one year, while 19 patients were observed while on medical treatment only. The group of 27 patients, 67.33 ± 6.8 years of age, creatinine of 2.15 ± 0.9 mg/dl, following stenting, were followed at intervals with biochemical tests, renal scintigraphy and doppler ultrasonography. The control group (70.0 ± 6.1 years, creatinine 1.99 ± 0.7 mg/dl) was also followed for one year. Result One year after stenting mean creatinine clearance (Ccr) increased from 36.07 ± 17.2 to 40.4 ± 21.6 ml/min (NS). Arterial BP, decreased after 1,3,6, and 12 months (p < 0.05). The number of antihypertensive drugs also decreased (p < 0.05). A significant increase in proteinuria was also observed. In the control group both Ccr, BP and proteinuria did not show significant changes. Based on renal scintigraphy and Ccr at subsequent times, it was possibile to evaluate the timecourse of renal function in both kidneys of the stented patients. In the stented kidneys Ccr increased significantly. On the controlateral kidney a decrease of renal function (p < 0.05) was observed. Resistance index appeared to be a risk factor of the functional outcome. Conclusions Stenting of RAS due to atherosclerosis is followed by stabilization or improvement of Ccr, mainly at the stented kidney, while contralateral renal function showed a decrease.

Background Renal artery stenosis due to atherosclerotic changes of the renal arteries has become a serious concern as a cause of hypertension and renal ischemia, resulting frequently in end-stage renal failure [ 1 , 2 ]. Several epidemiologic studies have shown the elevated prevalence of ischemic nephropathy, with special regard to atherosclerotic renal artery stenosis, in elderly patients [ 3 , 4 ]. Instead of the classical surgical approach, percutaneous balloon angioplasty or endovascular stenting have recently become accepted procedures in the attempt to revascularize the stenotic kidney and prevent chronic renal insufficiency. However, in spite of the arterial dilatation obtained with these procedures, there is still some doubt that the long-term outcome is in general satisfactory [ 5 ]. There is currently no clear evidence that such interventions prevent further progressive decline of renal function. However the results have been somewhat different in different case series [ 6 , 7 ]. It is known that there are patients with satisfactory results in terms of improvement or stabilization of renal function, while some cases may deteriorate renal function in spite of the dilating procedure [ 8 , 9 ]. As for the results of stenotic artery dilatation procedure on blood pressure, most of reports have confirmed a significant fall in systolic and diastolic blood pressure [ 10 , 11 ], an important finding which however cannot justify the stenting procedure if not accompanied by a consensual improvement in kidney perfusion and stabilization or improvement of renal function. Therefore the purpose of many researchers has been to identify the risk factors which might exclude patients from the revascularizing procedure, due to predictable poor outcome. Radermacher et al. [ 12 ] have identified the resistance index (RI) as an important factor predicting the outcome of the stenting. In addition, a limited number of studies [ 9 , 13 , 14 ] have evaluated not only the overall renal glomerular filtration following the dilating procedure, but also the individual function of the stented and contralateral kidneys. The results are interesting since the behaviour of the two kidneys after the one-sided dilating procedure was found to be divergent. This study provides further data on the evaluation of the two kidneys with a follow-up of one year. Methods The study has been carried out prospectively on 46 patients affected by hemodynamically significant atherosclerotic renal artery stenosis, detected by Magnetic Resonance Angiography or Selective Digital Angiography. All the patients had a unilateral stenosis. 27 patients (diabetes mellitus in 8 cases) were subjected to stenting of the stenotic renal artery while 19 patients (diabetes mellitus in 9 cases) were kept on medical treatment only. Clinical data of the two groups are reported in table 1 . Patients were allotted to the control group in case of refusal of the invasive procedure. All patients had a stenosis judged by ultrasonography to be above 70%. Table 1 Clinical and biochemical data of the stented and control groups Stented Control Patient, n° 27 19 age, years 67,3 ± 6,8 70.0 ± 6,1 n.s M/F, n° 17/10 15/4 n.s. Systolic BP, mmHg 169,1 ± 23 165,8 ± 24,7 n.s. Diastolic BP, mmHg 89,1 ± 14,8 87,4 ± 13,4 n.s. Creatinine, mg/dl 2,15 ± 0,9 1,99 ± 0,7 n.s. Cr. Clearance, ml/min 36,1 ± 17,3 34,6 ± 15,6 n.s. Urea, mg/dl 73,2 ± 36,7 75,5 ± 29,2 n.s. Tot. Cholesterol, mg/dl 236,9 ± 33,8 241,1 ± 41,2 n.s. Tryglicerides, mg/dl 181,2 ± 87,6 166,1 ± 82,8 n.s. Sodium, mEq/L 139,7 ± 5,2 142,0 ± 3,3 n.s. Potassium, mEq/L 4,34 ± 0,5 4,7 ± 0,6 n.s. Proteinuria, mg/24 h 308 ± 323 545 ± 572 n.s. Uric acid, mg/dl 6,8 ± 1,4 6,6 ± 2,3 n.s. Resistance Index (DDS) 0,76± 0,11 (25) 0,79± 0,04 n.s. Severity of stenosis (%) 78,8 ± 8,66 (25) 79,06 ± 9,0 n.s. % % Smoker 14/27 51,9 12/19 63,2 n.s. Hypertension 25/27 92,6 19/19 100 n.s. Diabetes Mellitus 8/27 29,6 9/19 47,4 n.s. Dyslipidemia 17/27 63 15/19 78,9 n.s. Peripheral arterial insuff. 17/27 63 13/19 76,5 n.s. In all these patients the renal artery was approached through the femoral artery. French 6 guiding catheter (type "Cobra"or "Bates") was used for selective renal artery angiography and for positioning the stent. All stenotic lesions were repaired with stainless stent Express Vascular SD Monorail 5.5-6-15/20 mm.premounted on a balloon catheter on Choice extra support 014" guide. In these cases primary stenting was performed. The procedure requires, usually, an injection containing 30 ml of 50–50 mixture of isotonic contrast and normal saline. The patients were followed at the outpatient clinic of the Nephrology unit. Duplex-doppler sonography and renal scintigraphy were carried out basally and following 1,3,6 and 12 months after the stenting procedure. At the same times, biochemical parameters, like serum creatinine and proteinuria were measured. Creatinine clearance was evaluated with the Cockcroft and Gault formula [ 15 ]. Doppler ultrasonography was carried out after fasting, following a three days of a low fibre diet and without smoking for a minimum of six hours before the procedure. Patients were studied with an Acuson 120 XP/4 (Acuson Corp., Mountain View, CA), equipped with a 3.5 MHz transducer, with longitudinal anterior, lateral and oblique approach, with at least threefold sampling of parameters along the artery. The standard criteria for the diagnosis of significant renal artery stenosis have been previously reported [ 4 ]. Renal radionuclide scintigraphy was performed with a gamma camera (Starcam 4000, General Electric, USA) with 99mTc-DTPA or with 99mTc-MAG3 (mercaptoacetyltriglycine). MAG3 was chosen in patients with creatinine clearance <25 ml/min. Diuretics and/or ACE inhibitors were discontinued at least three days before, if treatment was underway. The criteria of positivity have previously been reported [ 4 ]. Evaluation of single kidney renal creatinine clearance was performed by measurement of creatinine clearance and simultaneous renal scintigraphy with MAG3, with percentage-wise function of each kidney, enabling to calculate the creatinine clearance pertaining to each kidney. The entire set of data for this evaluation was available in 21/27 patients. Statistical analysis. A descriptive univariate analysis, consisting in evaluation of percentages, means and standard deviations has been carried out as first step. To evaluate the dependence among the nominal variables, a Parson's Chi square test was also carried out. In case of not applicable Chi square test due to low theoretical frequencies (<5), Fischer exact test for tables 2 × 2 was employed. Comparison of means of the two groups was made. Since the data did not show a normal distribution, non parametric tests, as Wilcoxon rank-sum test for paired data and Mann-Whitney test, were employed. The significance level was <0.05, as usual. The more interesting significant and not significant data were represented graphically. The data were evaluated with the statistical package BMDP Release 7 (Cork, Ireland,1997). Results Clinical and biochemical data of the treated and control groups are reported in Table 1 . There was no significant difference between experimental and control groups. During 12 months observation period one patient of the stented group began dialysis treatment, while in the control group 4 patients died of cardiovascular events and one patient started the dialysis treatment. A significant drop in systolic and diastolic blood pressure at all control times compared to basal values, was found in the stented patients, while no significant blood pressure drop was found in the patients not undergoing the PTA stenting procedure (Table 2 ). A significant fall in number of antihypertensive drugs was also found at 3 and 6 months after the stenting. Table 2 Timecourse of blood pressure (mm Hg), number of antihypertensive drugs and proteinuria (mg/24 h) in the stented and control groups STENTED GROUP Systolic BP Diastolic BP n° antihypertensive drugs Proteinuria Basal 169 ± 23 89 ± 14,8 2,07 ± 1,1 309 ± 323 1 m 155 ± 12,9* 81,9 ± 6,77* 1,93 ± 1,07 764 ± 690* 3 m 157 ± 17,7* 82,0 ± 6,31* 1,6 ± 1,05* 1381 ± 2160 6 m 148 ± 12,2* 81,4 ± 5,39* 1,42 ± 0,93* 1743 ± 2884 12 m 152 ± 14,4* 81,2 ± 8,87* 1,7 ± 1,17 1377 ± 1643* CONTROL GROUP Systolic BP Diastolic BP n° antihypertensive drugs Proteinuria Basal 165,8 ± 24,6 87,4 ± 13,4 2,37 ± 0,83 545 ± 572 3 m 163,7 ± 16,5 83,0 ± 10,2 2,26 ± 0,93 534 ± 404 6 m 155,6 ± 20,4 88,1 ± 8,45 1,82 ± 0,73 812 ± 536 12 m 154,4 ± 20,8 87,1 ± 8,21 2,0 ± 0,94 225 ± 368 (*) = Significant difference compared to basal value (p < 0,05) A significant increase in proteinuria following the stenting was found at 1 and at 12 months, while the increment in proteinuria observed at the other control times was only borderline significant. There was no diference in the increase of proteinuria between patients with and without diabetes mellitus. No changes in proteinuria was observed in the control group. An increase in creatinine clearance and a slight fall in serum creatinine, however not reaching a significance level, was observed in the stented group, while no increment in creatinine clearance was found in the control group (Table 3 ). This observation is however limited by the fall in the number of the control group due to death and beginning of dialysis in a total of 5 patients. However the analysis of the separate renal function in the stented and non stented kidneys of the experimental group showed differences in behaviour at the two sides. An increment in the percentage of total glomerular filtration in the stented kidney as a group was found, while a significant fall in percentage of filtration was found on the controlateral side (Fig. 1 ). Table 3 Evolution of creatinine clearance (ml/min) and serum creatinine (mg/dl). STENTED GROUP CONTROL ROUP Global Ccr Serum Cr Global Ccr Serum Cr Basal 36,07 ± 17,2 (27) 2,15 ± 0,94 34,78 ± 15,5 (19) 1,99 ± 0,72 1 m 35,29 ± 16,39 2,32 ± 1,33 - - 3 m 34,78 ± 14,38 2,18 ± 0,76 29,99 ± 12,4 2,13 ± 0,73 6 m 37,03 ± 18,0 2,15 ± 0,94 33,84 ± 17,9 2,01± 0,66 12 m 40,42 ± 21,63 (26) 2,03 ± 0,73 34,78 ± 14,3 (14) 1,98 ± 0,56 Figure 1 Evolution of percent GFR in the stented and controlateral kidneys. In addition, patients with the stent were divided in those cases with a RI above 0.80 and cases with this parameter below 0.80. The patients with lower RI improved, on average, renal function while the patients with elevated RI had a worse outcome (Fig. 2 ). RI values correlated negatively with changes in creatinine clearance from baselines (r = -0.6712, p < 0.01)(Fig. 3 ). Figure 2 Timecourse of creatinine clearance in stented patients with Resistance index < and > of 0.80. Significance of the difference is indicated (*) Figure 3 Inverse correlation between Resistance index and changes in creatinine clearance after stenting Discussion The advantage deriving from positioning a stent in a significantly stenotic renal artery has been debated in recent years. Generally favorable results have been reported by Dorros et al [ 10 ] on a wide cohort of patients, with special regard to patients with preserved renal function. Lederman et al [ 8 ] have found either improvement or stabilization of renal function in 73 % of 300 patients with atherosclerotic renal artery stenosis, bilateral in 48% of cases. Beutler et al. [ 16 ] have found similar results on patients with atherosclerotic ostial renal artery stenosis. Perkovi et al [ 17 ] consider, as risk factors for an unfavorable outcome, diabetes mellitus, advanced age and renal failure, while the use of ACE inhibitors following the stenting procedure was protective toward death or deterioration of renal failure. Airoldi et al [ 9 ] have given a message of caution in extending the dilating procedure to all the patients with renal artery stenosis, due to the low rate of renal improvements and of fall in blood pressure, in their experience, with the finding of at least 20% of restenosis. On the contrary, renal function improvement or stabilization was found in 94% of cases by Rocha-Sing et al. [ 18 ], in patients who had a progressive decline of renal function prior to stent implantation. In our experience, the fall in blood pressure and of the number of antihypertensive drugs was confirmed. Our results on the overall outcome of renal function, over a one year observation period, in patients with one sided renal artery stenosis of atherosclerotic origin, have been satisfactory. As a risk factor of worse outcome, our data have confirmed that RI above 0.80, results in a less satisfactory outcome compared to patients with RI lower than 0.80, as already reported by Radermacher et al. [ 12 ]. The stenting procedure, in our experience, was not followed by restenosis or other ischemic complications. Stabilization of renal function observed in the control group should take into account the unfavorable outcome of 5 cases, four deaths and one starting dialysis during the observation period. Anyhow a rational selection of patients who might get benefit from the procedure is advocated. The differences in the percentage of complications following the stenting procedure, as reported in the literature [ 19 ], might suggest at least in part the possibility of differences in the individual surgeon's skill in positioning the stent. As for the increment in proteinuria observed following stenting of renal artery, this finding has not been reported previously, while in basal conditions significant proteinuria in patients with renal artery stenosis has been already reported in the literature [ 20 ]. Therefore proteinuria does not exclude the diagnosis of renal ischemia as a cause of renal failure. Proteinuria is probably connected with the type of renal lesions due to chronic ischemia, like focal and segmental glomerulosclerosis, or ischemic glomerular damage. Glomerular lesions resembling focal glomerulosclerosis have been reported in patients with renal artery stenosis [ 21 ]. The increase in proteinuria following stenting should probably be attributed to increased perfusion pressure in damaged sclerotic glomeruli. Less is known of the renal function in the kidney affected by the arterial stenosis, following the stenting procedure, compared to the contralateral kidney. In all our patients renal artery stenosis was of atherosclerotic origin, without cases of fibromuscular dysplasia. In general, no adverse events were found following the stenting in the patients closely followed for one year. In our experience, there were no apparent cases of cholesterol embolization, of thrombosis of the artery, of occlusion of the stent, or of dissecation of the renal artery. The function of the stented kidney improved in most of patients while a reduction of renal function was observed in the controlateral kidney. The overall renal function was stable. Similar findings have been published by Airoldi et al., Leertouwer et al., and La Batide-Alanore et al. [ 9 , 13 , 14 ]. However their patient cohorts were rather different. One third of the cases of Airoldi et al. [ 9 ] were affected by fibromuscular hyperplasia. In only seven of the 27 patients a Palmaz stent was inserted. The increment in glomerular filtration rate of the stenotic kidney was more evident in the cases with fibromuscular dysplasia. Also in the Leertouwer et al. experience [ 14 ], renal artery dilatation was carried out in atherosclerotic renal artery stenosis, while the average age of the patients was younger than in our experimental group. Dilatation of the artery was able to induce an improvement of glomerular filtration rate of the treated kidney, although the overall glomerular filtration rate did not change. In La Batide et al. experience [ 13 ], 14/32 patients had renal artery stenosis due to fibromuscular hyperplasia and also the average age was decisively younger than our experimental group. Therefore, in our cases, with selective atherosclerotic renal artery disease and an older age, an improvement in function of the stenotic kidney following the stenting procedure was also observed and deserves to be underlined. The reduction in contralateral kidney function has been attributed to ultrafiltration in the non stenotic kidney, declining after stenting of the stenotic contralateral renal artery. In addition, hemodynamic factors consequent to a decrease in renin-angiotensin activity could be considered as a factor. The involvement of the renin angiotensin system in renal artery stenosis should be suspected due to the significant fall in systolic and diastolic blood pressure following the procedure, an occurrence not found in the control group. Actually, a fall in plasma renin activity or concentration following dilatation of the stenotic artery has been reported by Airoldi et al [ 9 ] and by Leertouwer et al. [ 14 ]. Conclusions In conclusion, the stenting procedure of a stenotic renal artery does not seem to carry important risks, and is accompanied by a definite improvement of the stented kidney, with some reduction of the filtration rate of the controlateral kidney. This event cannot be considered unfavorable, since it denounces a condition of hyperfiltration of the kidney, probably, if left unchanged, able to induce a deterioration of renal function with time. Therefore, also in case of overall stabilization of renal function following the stenting procedure, improvement of the stented side and reduction of hyperfiltration on the contralateral side are both favorable evolutions for long-term success of the revascularization procedure. The results are less satisfactory in patients with RI >0.80. They should probably be excluded from the stenting procedure. Competing interests The author(s) declare that they have no competing interests. Authors' contributions GC, conceived the protocol and cohordinated the study EM, participated in the design of the study and in the cohordination CC, was encharged of the magnetic resonance image analysis RL, collaborated in the duplex doppler ultrasonography examination IN, was responsible of the statistical examination of data GR, responsible of renal scintigraphic investigation DS, performed all the biochemical tests AZ, performed arteriography and stenting of renal arteries RC, was active in the ultrasonography investigation and screening of patients with renal artery stenosis Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC527878.xml

555550

Chronic ethanol exposure increases microtubule content in PC12 cells

Background Chronic ethanol exposure has been shown to result in changes in neuronal cyto-architecture such as aberrant sprouting and alteration of neurite outgrowth. In PC12 cells, chronic ethanol treatment produces an increase in Nerve Growth Factor (NGF)-induced neurite outgrowth that appears to require the epsilon, but not delta, isoform of Protein Kinase C (PKC). Neurites contain a core of microtubules that are formed from polymerization of free-tubulin. Therefore, it would be expected that an increase in neurite outgrowth would correlate with an increase in microtubule content. We examined the effect of chronic ethanol exposure on microtubule content in PC12 cells and the role of PKC epsilon and delta in ethanol's effect on microtubule levels. Results Chronic ethanol exposure of wild-type and vector control PC12 cells resulted in a significant increase in microtubule content and a corresponding decrease in free tubulin. There was also a significant increase in microtubule content in PC12 cells expressing a dominate-negative inhibitor of epsilon PKC; cells which have previously been shown to have no ethanol-induced increase in neurite outgrowth. In contrast, ethanol had no effect on microtubule content in PC12 cells expressing a dominate-negative inhibitor of delta PKC. Conclusion These results suggest that chronic ethanol exposure alters the relative ratio of free tubulin to microtubule-associated tubulin, an important component of the cytoskeleton. Further, the data from the PKC dominant-negative cell lines suggest that the effects of ethanol on microtubule content do not correlate with the effects of ethanol on neurite outgrowth. The delta isoform of PKC appears to be necessary for the ethanol-induced increase in microtubule content. These studies demonstrate an effect of chronic ethanol exposure which may contribute to previously documented alterations of neuronal cyto-architecture.

Background Chronic ethanol exposure has been shown to cause damage to the adult and developing nervous system [ 1 , 2 ]. For example, in vivo chronic ethanol has been shown to cause aberrant sprouting of hippocampal neurites in developing rats [ 3 ], increase the length of dendrites in cerebellar Purkinje neurons [ 4 ], the size of synaptic terminals of cerebellar granule cells [ 5 ], and the number of dendritic spines on hippocampal dentate granule neurons in adult rats [ 6 ]. Furthermore, in vitro ethanol enhances neurite outgrowth in cultured rat cerebellar neurons [ 7 ]. Contrary to the enhancement of neurite outgrowth, other studies have shown that chronic ethanol exposure inhibits the growth of dendrites in CA1 hippocampal neurons and cerebellar Purkinje cells in vivo and inhibits chick spinal cord neurite formation in vitro [ 8 , 9 ]. However, the mechanisms underlying this alteration of dendrite formation induced by ethanol exposure remain unknown. PC12 cells have been used as a cell culture model system to study the underlying mechanisms of ethanol's alteration of neurite outgrowth [ 10 , 11 ]. PC12 cells are a rat chromaffin cell line that differentiate into neuronal-like cells in the presence of Nerve Growth Factor (NGF) [ 12 ]. Using these cells, chronic ethanol has been shown to enhance NGF-induced neurite outgrowth [ 10 , 11 ]. Thus, PC12 cells have proven to be a valuable system for studying the mechanisms underlying ethanol-induced enhancement of neurite outgrowth. Nerve growth factor-induced neurite outgrowth in PC12 cells involves an induction of microtubule assembly [ 13 , 14 ]. Microtubules are formed from α and β tubulin proteins, which form head to tail protofilaments [ 15 ]. Studies have shown that Protein Kinase C (PKC) activation enhances the polymerization of tubulin to form microtubules [ 16 - 19 ]. Schultz et al. [ 20 ] have also demonstrated that microtubules containing phosphorylated tubulin are more stable than those containing unphosphorylated tubulin, although it remains unclear whether tubulin phosphorylation is the cause or the result of microtubule stabilization. PKC also modulates the activity of several microtubule associated proteins, including those involved in microtubule polymerization and vesicle transport [ 21 - 24 ]. Specific isoforms of PKC have also been implicated in mediating NGF-induced neurite outgrowth. Using both antisense oligonucleotides and specific inhibitors of PKC delta, Corbit et al. [ 25 ] have demonstrated that this isoform of PKC is required for NGF-induced neurite outgrowth. Other studies have found that in PC12 cells which over-express PKC epsilon, there is an enhancement of NGF-induced neurite outgrowth, while PC12 cells which over-express a dominant negative inhibitor of PKC epsilon show an inhibition of neurite outgrowth [ 26 , 27 ]. Thus, both the epsilon and delta isoforms of PKC have been implicated in modulation of NGF-induced neurite outgrowth. It is possible that the enhancement of NGF-induced neurite outgrowth produced by chronic ethanol may be due to ethanol's known alterations of PKC signaling. Chronic ethanol has multiple effects on PKC, including altered PKC subcellular localization following chronic exposure [ 28 , 29 ]. Messing et al. [ 30 ] have shown that chronic ethanol exposure actually increases total cellular content of PKC delta and epsilon (membrane associated and cytosolic) in PC12 cells. However, other studies have shown that membrane-associated PKC activity is down-regulated following chronic ethanol exposure [ 31 ]. Thus, while total cellular content of PKC may increase with chronic ethanol exposure, membrane-associated PKC may be down-regulated. Interestingly, Hundle et al. [ 32 ] have shown, using PC12 cells which over-express an inhibitory fragment of either delta or epsilon PKC, that PKC epsilon is required for ethanol's enhancement of neurite outgrowth. In this study, we examined the effect of chronic ethanol exposure on the neuronal microtubule cytoskeleton using PC12 cells as a model system. Here we show that chronic ethanol exposure increases microtubule content, while decreasing free-tubulin content. Thus, it appears that ethanol enhances microtubule polymerization in PC12 cells. We also investigated the role of microtubule polymerization in mediating ethanol's effects on neurite outgrowth using PC12 cells which over-express an inhibitory fragment of either delta or epsilon PKC. Importantly, it is the PKC epsilon isoform which is required for ethanol's enhancement of neurite outgrowth. Here, we found that the PKC delta isoform, but not the PKC epsilon isoform, is required for the enhancement of microtubule polymerization following treatment with chronic ethanol. Thus, it appears that neurite outgrowth does not correlate with enhanced microtubule polymerization in PC12 cells. Results Chronic ethanol exposure increases microtubule content in PC12 cells For the following studies, we used 100 mM ethanol for four days as a chronic exposure; a dose and duration used by previous researchers to demonstrate ethanol's enhancement of neurite outgrowth [ 10 ]. This is a concentration of ethanol which can easily be achieved by chronic alcoholics (0.46 g/dl). Figure 1 demonstrates that following a 96 hour exposure to 100 mM ethanol, there was approximately a 13% increase in polymerized microtubules compared to control cells (t 9 = 5.2; p < 0.001; n = 10). Similarly, there was a significant decrease (about 15%) in free-tubulin concentration (t 9 = 5.7; p < 0.001; n = 10) following 96 hours of ethanol exposure. There was no effect on total tubulin expression following four days of chronic ethanol exposure (t 9 = 0.034; data not shown). Chronic ethanol exposure in PKC dominant-negative PC12 cells We next used PC12 cells which over-express the first variable domain of PKC epsilon or delta, which acts as an isozyme specific inhibitor of PKC epsilon or delta [ 32 ], respectively, to investigate the role of PKC in ethanol's enhancement of microtubule polymerization (Figure 2 ). Interestingly, in the cells which express the inhibitor of PKC epsilon (DNE cells), we found that chronic ethanol exposure significantly increased microtubule content (t 5 = 7.15; p < 0.001; n = 6) and decreased tubulin content (t 5 = 3.4; p < 0.01; n = 6) (similar to control PC12 cells). In the cells which express the inhibitor of PKC delta (DND cells), there was no significant effect on microtubule (t 5 = 0.02; n = 6) or tubulin (t 5 = 0.44; n = 6) content. Ethanol had no effect on total tubulin content in either the DNE (t 5 = 0.2) or DND cells (t 5 = 0.013). There was no difference between the control vector transfected PC12 cells and wild-type PC12 cells, thus these groups were combined and expressed in Figure 1 . Importantly, for both the DNE and DND experiments, these experiments were replicated in two different sub-clones of transfected PC12 cells. In other words, we had two strains of dominant-negative PKC epsilon cells (DNE1 and DNE4) and two strains of dominant-negative PKC cells (DND21 and DND24). Results were similar between the two epsilon lines and between the two delta lines. Discussion Cells maintain a balance between free tubulin in the cytoplasm and tubulin which is polymerized into microtubules of the cytoskeleton. In this study, we find that chronic ethanol exposure increases microtubule content while decreasing free tubulin content in PC12 cells. Ethanol appears to be enhancing polymerization of tubulin into microtubules. While there was no increase in total tubulin within the cells, there was a change in the proportion of tubulin in the polymerized (microtubule) versus non-polymerized state. We initially hypothesized that the increase in microtubule content was due to ethanol's enhancement of neurite outgrowth. Presumably, neurite outgrowth is not due to simple stability of microtubules, but to an increase in dynamic microtubule growth. Microtubules are particularly abundant along the axons of nerve cells and multiple labs have show that chronic ethanol exposure increases NGF-induced neurite outgrowth in PC12 cells [ 10 , 11 ]. Therefore, since the cells that were exposed to ethanol have increased neurite outgrowth, this could be reflected as an increase in microtubles. However, it is not clear that there would necessarily be a concomitant decrease in free tubulin content or what effect of ethanol may have on microtubule stability, per se. Our data from experiments utilizing PC12 cells which express dominant negative inhibitors of PKC do not seem to support the idea that the increased microtubule content reflects increased neurite outgrowth. Hundle et al. [ 32 ] have demonstrated the specificity of these inhibitory fragments by measuring PMA-induced translocation of PKC epsilon and delta to the particulate fraction of the cells using Western blotting with isoform specific antibodies. They show (see Figure 2 of Hundle et al. [ 32 ]) that PKC epsilon, but not PKC delta, translocation was specifically inhibited in the cells expressing the epsilon inhibitory fragment. Further, PKC delta, but not PKC epsilon, was specifically inhibited in the cells expressing the delta inhibitory fragment. Thus demonstrating that the delta and epsilon inhibitory fragments selectively inhibit PMA-induced translocation of their corresponding PKC isozymes. Hundle et al [ 32 ] has shown, using these same cells, that chronic ethanol exposure enhances NGF-induced neurite outgrowth in control cells and cells expressing a dominant-negative inhibitor of PKC delta (DND cells) but not in cells expressing the inhibitor of the epsilon isoform of PKC (DNE cells). While it was not a goal of the current study to measure neurite outgrowth, daily observation of the cells was that neurite outgrowth was most dramatic in wild-type cells, slightly less so in the DND cells, and very limited in DNE cells (C. Reiter-Funk, personal observation). The work of Hundle et al [ 32 ] suggests that PKC epsilon but not PKC delta is involved in the effect of ethanol on neurite outgrowth. Therefore, we would have predicted that ethanol would cause an increase in microtubule content in DND cells but not in the DNE cells which do not have increased neurite outgrowth. However, we found that chronic ethanol enhances polymerization of tubulin into microtubules in dominant-negative PKC epsilon cells but not dominant-negative delta cells. Thus it appears that the delta isoform is involved in ethanol's enhancement of microtubule polymerization. Further, our data suggest that neurite outgrowth does not correlate with enhanced microtubule polymerization in PC12 cells. Interestingly, PKC delta has previously been shown to be involved in NGF-induced neurite outgrowth in PC12 cells [ 25 ]. It should be noted however, that our experimental paradigm varied slightly from that of Hundle et al [ 32 ]. For example, we allowed our cells to differentiate for 4 days prior to beginning ethanol. Therefore, apparent discrepancies of our findings with those of Hundle et al [ 32 ] could be related to these differences. Based on our data, we speculate that ethanol's enhancement of microtubule polymerization may involve phosphorylation of tubulin by PKC. However, it should be noted that we have, thus far, not directly measured phosphorylation of tubulin. Studies have shown that chronic ethanol increases expression of delta and epsilon PKC in PC12 cells [ 30 ] and PKC activation enhances tubulin polymerization into microtubules [ 16 - 19 ]. Alternatively, chronic ethanol could be acting to alter important microtubule associated proteins. Many of these proteins, including Microtubule-associated Proteins (MAPS) are important for promoting microtubule assembly [ 34 , 35 ] and it has been shown that PKC phosphorylation mediates the assembly-promoting activity of these proteins [ 21 , 22 , 36 , 37 ]. Further studies are required to determine the mechanism of ethanol's enhancement of microtubule formation and the apparent role of delta PKC. Conclusion Our studies demonstrate that chronic ethanol alters the relative ratio of free versus microtubule-associated tubulin content in PC12 cells, resulting in an increase in microtubule content and a corresponding decrease in free tubulin. This alteration was found to occur in wild type cells, as well as those expressing a dominant-negative inhibitor of epsilon PKC but not in cells expressing a dominant-negative inhibitor of delta PKC. These ethanol-induced changes could be important during activity-dependent remodeling of synapses or developmental growth of axons and dendrites which may lead to cognitive dysfunction. Methods Materials Wild type PC12 cells were a gift from Nicholas Pantazis, Ph.D (University of Iowa). PC12 cell lines expressing dominant negative inhibitors of delta or epsilon PKC were a gift from Robert Messing, M.D (University of California, San Francisco). RPMI, Dulbecco's Modified Eagle Medium, and PenStrep were purchased from Gibco, NGF from R&D Systems, horse and fetal bovine serum was purchased from Hyclone, and laminin was purchased from Invitrogen. The microtubule assay kit was purchased from Cytoskeleton, Inc (#BK038). Cell Culture PC12 cells were cultured in RPMI medium supplemented with 10% horse serum, 5% fetal bovine serum, and 1% PenStrep. Medium for the dominant-negative PC12 cells also included G418 (250 μg/ml) for selection purposes. The dominant-negative lines are cells that have been transfected to express isozyme specific dominant-negative inhibitors of delta or epsilon PKC and have been used to show specific PKC isozyme involvement in multiple ethanol effects [ 32 , 33 ]. These cell lines stably express the fragments δV1 or εV1, which are derived from the first variable domains of delta PKC or epsilon PKC, respectively. There was also a cell line transfected with vector alone, which served as a control. Dr. Messing has shown that these fragments can function as isozyme specific inhibitors [ 32 , 33 ]. The cells were maintained in an incubator at 37°C in 5% CO 2 . For all experiments, PC12 cells were plated into six-well, laminin-coated (5 μg/ml) plates at a density of 270,000 cells/well. The cells were differentiated into neuronal-like cells with Nerve Growth Factor (25 ng/ml) for four days prior to addition of any chronic drug exposure. For chronic exposures, ethanol (100 mM) was added to the culture media (along with NGF) for 96 hours and plates (control and ethanol) were wrapped in parafilm to prevent ethanol evaporation. Media were changed every other day. Microtubule assay Measurement of microtubule and free-tubulin contents in cells are well-established assays. The materials are available as a commercially kit (Cytoskeleton, Inc). Microtubules are very sensitive to changes in temperature; therefore, all equipment and buffers were warmed to 37°C before use (unless otherwise indicated). To prevent free tubulin from polymerizing onto existing microtubules during the assay, lysis buffer was added at a ratio of 10 volumes of buffer to 1 volume of cell pellet. Cells were homogenized via syringe trituration and incubated for 10 minutes in lysis buffer (contents listed below). 10 μL of cell homogenates were saved for protein measurement using the Bradford Assay. Homogenized cells were then centrifuged at 100,000 × g for 30 minutes to separate microtubules from free-tubulin. The polymerized microtubules settle in the pellet, while the free-tubulin remains in the supernatant. Following centrifugation, the supernatant (free-tubulin) was removed and frozen until Western blot analysis. The pellet was resuspended in ice-cold water containing CaCl 2 (200 μM) and incubated for one hour. CaCl 2 acts to enhance microtubule depolymerization. Thus, the microtubules remaining in the pellet were depolymerized to free-tubulin. The samples were then centrifuged at 14,000 × g (4°C) for 10 minutes. The supernatant (containing free-tubulin representing the original microtubules) was collected and frozen. Tubulin concentrations in both fractions were measured using Western blotting as described below. Buffer contents Lysis buffer : LMS1 solution containing GTP (100 mM) + ATP (100 mM) + protease inhibitor cocktail (10 μM) + Okadaic Acid (100 nM). LMS1 : PIPES (100 mM) containing MgCl 2 (5 mM) + EGTA (1 mM) + glycerol (30%)+ Nonidet P40 (0.1%) + Triton X-100 (0.1%) + Tween 20 (0.1%) + beta-mercapto-ethanol (0.1%) + Antifoam (0.001%). Protease inhibitor cocktail : Pepstatin (1 μg/ml) + Leupeptin (1 μg/ml) + benzamidine (10 μg/ml) + tosyl arginine methyl ester (500 μg/ml). Measurement of tubulin concentrations Western blot analysis was used to determine tubulin content. Samples for Western blot were boiled with electrophoresis sample buffer (containing 25% glycerol, 5% beta-mercaptoethanol, 2% SDS, and 0.01% bromophenol blue in 62.5 mM Tris-HCl, pH 6.8). Equal amounts of protein (25 μg) were separated by SDS-PAGE using a 12% polyacrylamide gel. Molecular weight standards were also loaded. The separated proteins were transferred to nitrocellulose membranes using an electroblotting apparatus. The membrane was blocked [Tris-Buffered Saline (TBS) containing 4% nonfat dry milk and 0.05% Tween-20] for 20 minutes at room temperature. Membranes were then incubated in primary antibody (500 ng/ml; monoclonal anti-beta Tubulin; Cytoskeleton, Inc) overnight at 4°C. Following incubation in the primary antibody, membranes were washed with TBS/0.05% Tween-20 and incubated in secondary antibody (1:1000 dilution; HRP conjugated anti-mouse antibody; Santa Cruz). Membranes were again washed and tubulin bands were visualized by Enhanced Chemiluminescence (ECL) using standard luminol reagents captured using a BioRad Gel Doc 1000. The computer analysis software Molecular Analyst (version 2.1; BioRad) was used to quantify tubulin concentrations (relative density using the Volume Analysis setting with equal sized rectangular boxes set around individual bands). All blots were very clean and no filtering was needed. Data expression and statistical analysis Tubulin concentrations in both the free-tubulin and microtubule fractions were expressed as a percent control relative density and presented as mean (+/-SEM). The data represent multiple replicates of the same experiment (n values listed in figure legends). Student's t-tests were used to make comparisons between control and ethanol-treated PC12 cells. Each individual band analyzed was an independent extract from the same experiment, with the experiment being replicated 5 times. Each replication of the experiment was composed of 2 separate wells of cells for each treatment condition. For each of the 5 replications, the 2 control bands were averaged and each sample, including the controls, from that replication normalized to the average. The data was combined after all 5 replications. Authors' contributions CKR-F conceived and performed the experiments and participated in writing the manuscript. DPD participated in the design of the study and writing of the manuscript. Both authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555550.xml

519028

Synchronously diagnosed pre-sacral neurofibroma and cutaneous spitzoid melanoma: a fortuitous association?

Background At a U.S prevalence of 1 in 3000, Neurofibromatosis type-1 (NF-1) is a relatively common disorder. Amongst a variety of others, occurrence of 2 or more neurofibromas in the same patient represents one of the major diagnostic criteria for this disorder. Rarely, ocular, cutaneous or anorectal malignant melanomas may be identified in patients with NF-1, This rare association has caused controversy as to whether patients with NF-1 have an inherently higher risk for melanomas or whether the associations can be explained by chance alone. Case presentation The purpose of this report is to highlight the unusual confluence of rare clinicopathologic features in a patient without NF-1. The patient was diagnosed with an 8.5 cm pre-sacral neurofibroma and was shortly thereafter diagnosed with a cutaneous malignant melanoma showing spitzoid features. Pre-sacral neurofibromas are rare in patients without NF-1; likewise, malignant spitzoid melanoma, a controversial histopathological entity, is distinctly uncommon. Conclusions The synchronous diagnoses of these neural crest derived tumor entities in a patient without neurofibromatosis lends credence to the view that when these two lesions occur in patients with NF-1, the association is coincidental.

Background Neurofibromatosis type-1 (NF-1) is a common autosomal dominant disorder characterized by multiple neurofibromas, café-au-lait spots, freckling of the inguinal or axillary regions, gliomas, iris hamartomas, and malignant peripheral nerve sheath tumors [ 1 , 2 ]. Neurofibromas are typically well-delineated and are composed of an admixture of various cell types, such as Schwann cells, fibroblasts and perineural-like cells and cells showing intermediate features [ 1 , 2 ]. Although as outlined above, multiple neurofibromas are characteristic of patients with NF-1, however, most cases of neurofibroma which are diagnosed in general are sporadic in nature. The vast majority of neurofibromas are cutaneous and less commonly are intraneural, within the soft tissues or viscera. Presacral neurofibromas or neurofibromas with presacral involvement are uncommon in patients without NF-1, and have been the subject of sporadic case reports over the past half-century [ 3 - 16 ]. Likewise, spitzoid melanoma or melanomas showing spitzoid-like features form only a small percentage of all malignant melanomas. This diagnosis is based on the rare finding that some melanomas displays cytologic features that are similar to those identified in the benign Spitz nevus [ 17 ]. The controversy associated with this lesion stems from the fact that some dermatopathologists do not believe in its existence and prefer to designate melanocytic proliferations meeting traditional criteria for malignancy as malignant melanomas, irrespective of the Spitzoid features [ 18 ]. To our knowledge, a synchronous presacral neurofibroma and cutaneous spitzoid melanoma have never been reported in a patient without neurofibromatosis. More importantly, the absence of NF-1 in our patient may have implications for the potential association between malignant melanoma and NF-1. Case presentation In September 2000, a 35-year-old female without any history or clinical stigmata of NF-1 presented to her primary physician with complaints of a dull, localized left upper leg pain of several months' duration. An abdominal mass was palpated during a physical examination. Magnetic resonance imaging (MRI) as well as a computed tomographic (CT) scan of the abdomen and pelvis showed a large well-defined, near spherical mass in the left false pelvis which enhanced heterogeneously at a mean Hounsfield value of 44 units (Figure 1 ). The mass displaced the external iliac vein medially and psoas muscle laterally. It also abutted the upper surface of the left ovary without truly invading any of these or other surrounding structures. However, the mass was believed to be in the course of the left genitofemoral nerve and lumbar plexus. The decision was made to resect the mass. Intraoperatively, the tumor's capsule was found to be densely adhered medially to the external iliac vessels, with at least 10 external venous branches directly supplying the tumor. The tumor was carefully marsupialized out of the retroperitoneal area and the decision was made to leave the residual capsule, since an attempt at its removal would have entailed a highly morbid procedure that was not felt to be justified based on the histopathologic appearance of the tumor on frozen sections. Intraoperatively, a pigmented macular lesion with faintly irregular edges was noted in the left upper thigh, which was biopsied. Pathologic examination showed a malignant melanoma with spitzoid features. The precise circumstances regarding the duration of the lesion and whether there had been any increase in its size was unclear. She subsequently underwent a wide local excision (4 × 12 cm skin ellipse was removed) and sentinel lymph node biopsy, both of which showed no residual melanoma. The patient's postoperative course over the subsequent 2 years was remarkable for a relatively slow but progressive improvements in the neurologic symptoms related to her surgery. However, she showed no evidence of either tumor recurrence at last follow-up, 26 months postoperatively. Figure 1 Computed tomographic scan of the pelvis showing a large, well-circumscribed presacral mass Pathologic findings The resected mass was spherical, weighed 270 grams and measured 8.5 × 7 cm × 7 cm (figure 2 ). The specimen was sectioned to reveal a myxoid tan-yellow cut surface (figure 3 ). Microscopically, the specimen was uniformly hypocellular, and showed a haphazard admixture of wavy Schwann cells and collagen fibers dispersed in a mucopolysaccharide matrix (figure 4 ). No increased cellularity, nuclear pleomorphism, increased mitotic activity or tumor coagulative cell necrosis was identified. On immunohistochemistry the tumor was positive for S100, Neurofilament, vimentin, and negative for epithelial membrane antigen, in combination with the morphological features a diagnosis of neurofibroma was made. Pathologic examination of the skin lesion showed a malignant melanoma (6 mm in diameter) with spitzoid features: epithelioid and spindle atypical melanocytes growing in a solid, asymmetric, non-maturing pattern with deep mitotic figures (up to 3 mitotic figures/mm 2 ). The individual cells displayed nuclear pleomorphism with prominent nucleoli and the epithelioid forms showed abundant cytoplasm (Figures 5 , 6 , 7 , 8 ). Immunohistochemically, both the spindle and epithelioid cells showed strong and diffuse immunoreactivity for S100 and HMB-45. The proliferative index of the tumor was 30–40% as assessed with the immunohistochemical marker ki-67. This lesion was at a Clark's level IV and at a depth of 2.1 mm. Growth phase was vertical and ulceration was absent. A few tumor-infiltrating lymphocytes were present and there was no definitive evidence of regression. Figure 2 Macroscopic appearance of the external surface of the presacral mass Figure 3 The cut surface of the presacral mass showing glistening myxoid, tan-yellow appearance Figure 4 Microscopic appearance of the presacral mass showing a haphazard admixture of wavy Schwann cells and collagen fibers dispersed in a mucopolysaccharide matrix (Hematoxylin and Eosin, 20×) Figure 5 Photomicrographic panoramic view of the patient's cutaneous biopsy showing an asymmetric lesion with a basal confluent growth (Hematoxylin and Eosin 2×) Figure 6 Photomicrograph of the junctional component of the tumor showing Spitzoid features of the lesional cells. Note that the junctional nests do not display a uniform vertical orientation towards the epidermis, as is expected in most Spitz nevi. (Hematoxylin and Eosin 40×) Figure 7 Interemediate-power view of the cutaneous lesional cells, showing the admixture of spindle and epithelioid cells (Hematoxylin and Eosin 20×) Figure 8 Photomicrograph showing the cytologic features of the lesion. Note the nuclear pleomorphism and prominent nucleoli. This focus was near the deep edge of the lesion, reflecting a lack of histological maturation (Hematoxylin and Eosin 20×) Discussion The potential association between NF-1 and malignant melanoma has been the source of controversy in the medical literature. The common neural crest origin of these conditions has provided an attractive framework for this discussion. However, it is unclear whether patients with NF-1 have an inherently higher propensity to develop malignant melanomas than the general population. In a follow-up study of 70 NF-1 patients reported to the Swedish Cancer registry, 24% of the 70 patients developed 19 malignancies, only 1 of which was a melanoma [ 19 ]. However, the precise prevalence of melanomas in NF-1 patients is largely unknown. Most melanomas that arise in the setting of NF-1 are ocular. In a recent literature review, Honavar et al [ 20 ] identified only 19 reported cases overall. Cutaneous and anorectal melanomas have also been rarely reported in patients with NF-1 [ 21 - 27 ]. The rarity of this association given the frequency of NF-1 (1 in 3000) suggests that the probability of NF-1 patients developing malignant melanoma is no more than the general population. However, this needs to be tested in a rigorous population-based study. Ishii et al [ 21 ] recently reported a loss of heterozygosity (LOH) at the NF-1 gene in an anal melanoma occurring in an NF-1 patient. This suggests that the well-known Knudson's two-hit hypothesis may be operational, and that somatic loss of the second allele of the putative tumor suppressor function of the NF-1 gene causes development of this particular somatic malignancy. Again, more cases need to be studied to exclude the possibility that LOH for NF-1 occuring as a late event in the tumorigenesis of sporadic melanomas. Our case provides another framework for the discussion of the potential association between NF-1 and malignant melanoma. Our patient has no evidence of either of the neurofibromatosis syndromes. The rarity of the clinicopathologic features of both lesions identified in this patient (the unusual presacral location of the neurofibroma and the spitzoid melanoma) suggests that their association in this patient is coincidental, even though both are neural crest derived neoplasms. This presumption contradicts the notion that in patients with NF-1, malignant melanomas that rarely develop are part of their neurocristopathy. For residents of the United States, the lifetime probability of developing cutaneous melanoma is 1 in 55–82 (approximately 1.5%) [ 28 ]. As previously noted, NF-1 is a relatively common condition with a prevalence of 1 in 3000. Since no more than 100 cases of melanoma (all sites combined) developing in NF-1 patients have been reported, the incidence is significantly lesser than the 1.5% that can be attributed to chance alone. Although the patient described in this report did not have characteristic clinical features of NF-1, an important possibility that requires consideration is that she has segmental NF-1. Segmental NF-1 is thought to result from a post-zygotic mutation in the NF-1 gene resulting in a somatic mosaicism [ 29 , 30 ]. In these patients, characteristic NF-1-associated diseases are limited to a localized part of the body [ 29 , 30 ]. In our patient, a presacral neurofibroma was associated with an upper thigh cutaneous melanoma, putting both lesions in the general same region, albeit without true co-localization. Additionally, melanoma is not a diagnostic criteria for NF-1, as associated lesions are required to be for the definition of segmental NF-1. The location of the current patient's neurofibroma is also somewhat unusual for segmental NF-1. In two combined series that investigated 163 patients with segmental NF-1, there was not a single case of a retroperitoneal neurofibroma [ 29 , 30 ]. In one of these series [ 29 ], neurofibromas alone were the most common manifestation of segmental NF-1. However, in all such cases, the neurofibromas were either dermal, on major peripheral nerve trunks or both. Although these findings argue against segmental NF-1 in the current patient, the possibility certainly remains. Thus, the findings in this case should be viewed within the context of that possibility. In conclusion, we report here the previously unreported synchronous diagnosis of a presacral neurofibroma and a spitzoid malignant melanoma in a patient without NF-1. Furthermore, the finding of a sporadic neurofibroma and malignant melanoma occurring in a patient without NF-1 lends credence to the view that when these lesions occur in patients with NF-1, the association may be coincidental. Competing interests None declared. Authors' contributions OF and DH made substantial contributions to the intellectual content of the paper. Both co-wrote the manuscript. Both the authors have seen the final version of the manuscript and approved it for publication.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC519028.xml

555544

A case of invasive aspergillosis in CGD patient successfully treated with Amphotericin B and INF-γ

Background Chronic granulomatous disease (CGD) is a rare disorder of phagocytes in which absence of superoxide and hydrogen peroxide production in phagocytes predisposes patients to bacterial and fungal infections. The most common fungal infections in these patients are caused by Aspergillus species. Case presentation Here, we describe Aspergillus osteomyelitis of the ribs and hepatic abscess in a 5-year-old boy. The patient was successfully treated with Amphotericin B and INF-γ. Conclusion With respect to the high frequency of aspergillosis in the CGD patient, immune deficiency should be investigated in patients with invasive aspergillosis. Moreover, using antifungal drugs as prophylaxis can improve the quality of life in these patients.

Background Chronic granulomatous disease (CGD) is a rare inherited disorder of nicotinamide adenine dinucleotide phosphate (NADPH) oxidase complex of phagocytic cells resulting in failure to generate reactive oxidants and the absence of a respiratory burst [ 1 ]. The disease is characterized by recurrent or persistent intra-cellular bacterial and fungal infections. Approximately, the incidence of fungal infections in CGD patients has been reported up to 20% of infections [ 2 ]. Aspergillus spp are ubiquitous saprophytic fungi and are considered as the major causative fungal agent in these patients [ 2 , 3 ]. The spectrum of infection caused by Aspergillus species varies from flu-like pneumonia to life-threatening invasive aspergillosis [ 4 ]. The most common form of the aspergillosis in CGD patients is Aspergillus pneumonia which can be accompanied by dissemination to the ribs, chest wall and soft tissues [ 1 , 2 ]. Here, we describe a case of invasive aspergillosis in CGD patient with hepatic abscesses and osteomyelitis. Case Presentation A 5-year-old male patient was admitted to Children Medical Center (CMC) with inflammation and swelling in his left mandible and wrist without a history of trauma. In the past, he had suffered from several episodes of pneumonia which started at the age of seven months. On admission, laboratory findings included erythrocyte sedimentation rate (ESR) 84 mm/h, WBC count 12100/mm3 (61% neutrophils, 39% lymphocytes), hemoglobin 11.3 gr/dl and thrombocyte 386000/mm3. As the CRP analysis displayed 20 mg/dl, cephalexin (150 mg/kg/day) was initiated. In his roentgenogram, osteolytic lesions in the distal metaphase of hand and maxillary bone were observed. Considering history of several infections and multifocal osteomyelitis, bone biopsy was performed and his immune system function was evaluated. In the bone biopsy, non-necrotizing granulomatoid lesions were seen. The induration of purified protein derivative reaction was 10 mm diameter. Besides, HIV, hepatitis B surface antigen (HBs), rheumatoid factor and brucella agglutination tests were all negative. The serum IgG level was 1650 mg/dl (normal: 441–1135 mg/dl). IgM and IgA were in high normal range at 250 and 175 mg/dl, respectively. Because no defect was found in his humeral and cellular immunity, the phagocytic cells function was tested with a nitroblue-tetrazolium (NBT) slide test. Based on his hematological and immunological tests (NBT = 0), CGD was considered as underlying disease in this case. Regarding his NBT test, antibiotic therapy was changed from cephalexin to co-trimoxazole (20 mg/kg/day, iv) plus (along with) interferon-γ (50 microgram/m 2 every other day). After two weeks of treatment, the patient's condition improved and he was discharged with prescription of both cephalexin (100 mg/kg/day) and co-trimoxazole (10 mg/kg/day) to be taken orally as prophylaxis. The patient was readmitted to our center after eight months with a tender mass in his right upper quadrant (RUQ) (Fig 1 ). On admission, his major complaint was severe dyspnea, a persistent cough and also chest and abdominal pain in epigastric area which was started 10 days ago. He was placed on antibiotic therapy including cephalexin (100 mg/kg/day). A computerized tomography (CT) scan of the chest and abdomen was performed which revealed the hypodense area in liver (Fig 2 ). Adjacent to this opacity, involvement of lower right ribs and reaction to soft tissue were also observed, indicating ribs osteomyelitis. After sonography guided drainage of the above-mentioned hepatic abscess, a sample was sent to the Mycology Department in Tehran University of Medical Sciences. The microscopic examination of clarified specimen with KOH 10% indicated the branched, septated and dichotomous mycelia (Fig. 3 ). The remaining specimen was also cultured on Brain Heart Infusion agar (BHI), Sabouraud's dextrose agar (S) and Sabouraud's containing 0.005% chloramphenicol (Sc). The S and Sc culture media were incubated at 25°C and BHI at 37°C. The colonies grew rapidly, attaining the diameter of 5 cm within 5 days and their color was bluish green. Cellophane tape preparations and slide cultures demonstrated septated, branched and hyaline hyphae with rough-walled conidiophores and radiated conidial heads. Based on these microscopic and macroscopic findings, Aspergillus fumigatus was determined as causative agent in this case. Deoxycholate Amphotericin B (1 mg/kg/day, iv), interferon-γ (50 microgeram/m 2 every other day, sc) and rifampicin (10 mg/kg/day) were administered with diagnosis of invasive aspergillosis. The only adverse event observed during treatment was hypokalemia, which was adjusted by administration of potassium chloride 15%. One month after initiation of antifungal therapy, his follow-up CT scan of the abdominal and thoracal region demonstrated relative resolution of hepatic abscess. After four weeks of intravenous treatment, the patient's clinical condition improved. He was discharged upon his parents' responsibility while continuing taking rifampicin (10 mg/kg/day) for two more weeks as a treatment in addition to co-trimoxazole (5 mg/kg/day) and itraconazole (4 mg/kg/day) as long term prophylaxis. Figure 1 Subcutaneous swelling and granuloma formation in right upper quadrant. Figure 2 Computerized tomography showed a hypodense area in right lobe of liver with peripheral enhancement Figure 3 KOH 10% preparation of hepatic abscess showing dichotomous septated hyaline hyphae. Discussion CGD is a rare inherited immune disorder whose prevalence is estimated to be about 1/1,100,000 – 1/1,300,000 individuals worldwide [ 1 ]. Similar to the presented case, the most common form of CGD is X-linked recessive that consists of about two thirds of cases and the rest are autosomal recessive [ 5 ]. In the absence of minimal oxidative metabolism in CGD which can be ascertained easily using nitro blue tetrazolium (NBT) slide test, other immune mechanisms are triggered [ 6 ]. The relative evaluated immunoglobulin levels in the above-mentioned case might be due to persistent antigenic stimulation and it is a common phenomenon in all chronic infections. This defect is characterized by recurrent or persistent infections due to catalase-positive fungal and bacterial agents despite aggressive antibiotic therapy [ 1 , 6 ]. The incidence of aspergillosis in these patients has been reported to be 78% of all fungal infections [ 2 ]. Among Aspergillus spp , Aspergillus fumigatus is considered to be the predominant cause of invasive aspergillosis in CGD patients [ 1 , 7 ]. Pulmonary aspergillosis has been reported in CGD patients infected with Aspergillus fumigatus . As shown in this case, Aspergillus might spread from lungs to the bones of thoracic wall and cause osteomyelitis [ 7 - 9 ]. Although Aspergillus fumigatus is considered to be the most isolated species, Aspergillus nidulans osteomyelitis is reported to have a higher incidence and more mortality rate in these patients [ 7 , 9 ]. The treatment of infections in CGD patients is not easy. Since the underlying immunodeficiency is the most important factor with respect to the outcome of treatment, these patients should be treated either with immunomudulative agents such as recombinant INF-γ or with stimulating factors [ 10 ]. Recently, on the basis of cytochrome b (558) expression and NADPH oxidase activity, three different sub-type of X-linked chronic granulomatous disease were described [ 11 ]. Therefore, therapeutic response to INF-γ in this case and other X-linked CGD patients might be elucidated. Besides, similar to other systemic fungal infections, antifungal drugs such as amphotericin B should be added to therapeutic regimen of CGD patients with established invasive aspergillosis. Our patient responded to the above-mentioned therapeutic protocol and was discharged with long term anti-microbial and immunomudulatory prophylactic treatment as well as anti fungal drug [ 12 ] to enhance the quality of life and lessen the risk of re-infection.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555544.xml

548515

Supporter or obstructer; experiences from contact person activities among Swedish women with breast cancer

Background Swedish patient associations for breast cancer patients (PABCPs) offer patients with breast cancer unlimited meetings with a breast cancer survivor, a contact person (CP). We applied the voluntary action perspective in this interview study with members of Swedish PABCPs in order to explore how women with breast cancer experienced their contact with a CP from a PABCP. Methods Audio-taped narratives from 8 women were analysed using Reissman's monitoring and Gee's analysis structure. Results Three themes appeared: 1. Shared experiences give new perspectives on having cancer, 2. Feelings of isolation are a part of the identity of the illness and 3. Relations with others enable self-help. However, the relationship with the CP is sensitive to timing, correct information and understanding. Conclusions CPs act as sounding boards and should optimally have capacity for listening, gives support and act as partner in this conversation. On the other hand, CPs should be aware that their presence and limited general medical knowledge could at times disturb the patient's psychological recovery and strengthen feelings of isolation. Thus, PABCPs must be careful in selecting CPs and offer relevant educational activities related to the themes identified herein.

Background Proximity to individuals and society in order to reach out and provide optimal support is basic for patient associations for cancer patients (PACPs) [ 1 , 2 ]. Swedish patient associations for breast cancer patients (PABCPs) offer breast cancer patients' unlimited meetings with a breast cancer survivor, a contact person (CP). The present study focuses on women with breast cancer and their experiences from having a CP. In Sweden, an association consists of a number of individuals who work together in an organised form towards a common vision [ 3 ]. The Scandinavian concept of an 'association' is different from that in Western Europe and the U.S.[ 4 , 5 ]; in Scandinavia the collective rather than the individual aspect is emphasized with focus on social activities, educational workshops, and support groups [ 6 , 7 ]. However, also individual activities exist within the associations and the CP activities in Swedish PABCPs constitute an example hereof. This activity is inspired by the Reach to Recovery Program (RtoRP), a world-wide rehabilitation program initiated in the U.S. in 1952 and was later established also in Europe. RtoRP is based on the opportunity to meet with individuals with similar experiences and thereby learn different ways of dealing with disease-related feelings and problems [ 8 , 9 ]. Rehabilitation from breast cancer is a requirement for becoming a CP in a PABCP according to the national organisation for Swedish breast cancer associations, (BRO) [ 2 , 10 ]. Listening, supporting, and acting as a partner in conversations and discussions with a breast cancer patient is common to all CPs, although the forms may differ, such as one-to-one meetings or group meetings, between different PABCPs. The 28 of the total 32 PABCPs in Sweden (according to BRO) have appointed specially selected persons to coordinate the CP activities. In addition to self-rehabilitation for breast cancer, the PABCPs also provide education that includes basic psychological knowledge, medical information, organization of and contacts with the health-care system for the CPs [ 10 ]. This training program is aimed at preparing the CPs for their task to support individuals in a similar situation. Hence, the CP activity provides an opportunity for cancer patients to meet individuals with similar experiences and to learn different ways of dealing with feelings and problems [ 11 - 13 ]. Treatment and voluntary action provide different perspectives in studies of self-help groups or mutual support groups in the voluntary sector. We have, in accordance with other researchers' views, chosen to apply the voluntary action perspective in this study of CPs activities within the voluntary sector [ 14 - 16 ]. Support in self-help groups, in which the members share and articulate common experiences, is often viewed as a variant of professionally led group therapy [ 16 - 19 ]. However, the treatment perspective in the latter type of groups is based on the elements of the intervention that lead to cure and on the result obtained. In contrast, the voluntary action perspective of the self-help group is based on a mutual relationship with focus on benefits and aspects related to the individual or to the group [ 16 , 20 , 21 ]. What experiences do the individuals have in their contact with a CP and what do the individuals (in terms of their breast cancer illness) gain from their contact with the CPs from a PABCP? We applied the voluntary action perspective in this interview study of with the aim to explore how women with breast cancer experienced their contact with a CP from a PABCP. Methods Informants and data collection Women with breast cancer who had personal experience from CP activities were informants in this study. To qualify for participation, the informants were required not to be involved in treatment (surgery, chemotherapy, or radiotherapy) of the primary tumour. The cancer diagnosis should be within 3 years of the data collection in order to recall the illness and treatments obtained. We approached volunteers responsible for contact and visiting activities in 3 PABCPs who in their turn contacted 8 women who were judged eligible for the study. These women were given information on the purpose of the study and the interview procedure, and confirmation of confidential treatment of the women's information was sent by the PACP's. The recipients of the letter were informed that it was sent by the PABCP, but they were offered to contact the researcher with questions. The informants were asked to give their decision to the author (CC) by telephone within two weeks and all 8 individuals invited chose to participate in the study. The 8 informants had a mean age of 56 (range 39–69) years and demographic characteristics are presented in Table 1 . Of these individuals, 3 were diagnosed between 3 and 4 years prior to the study, but since these individuals provided extensive narratives they were allowed to stay in the study. Of the 8 individuals, 5 had experience of contact with one single CP, two of several CPs (group sessions), and one had experience of open CP meetings. Table 1 Demographic characteristic of informants ( n = 8) Characteristics Number Age: 39 – 69 (years) 56 (mean) Married and children 8 Time since diagnosis : 18 – 49 (month) 34 (mean) Treatment: Operation 8 Cytotoxic therapy 4 Radial therapy 7 Hormonal therapy 2 Experience of CP: Open meetings 1 One single CP 5 Several CP 2 All the interviews were held in the home districts of the participants. The data were based on audio-taped narratives and takes into account people's natural way of constructing and interpreting experience and provides the opportunity to take relevant contextual factors into account [ 22 ]. The personal narrative refers to brief and topically specific discrete stories recapitulating specific events that the narrator had experienced [ 22 , 23 ] (cf Mishler [ 24 ] narratives about extra exceptional life events). The informants were asked to tell about their experiences from having a CP. The initial exhortation, "can you tell me about..." was intended to encourage the informants to tell their story in their own way, which would allow analysis from the voluntary action perspective [ 16 ]. Each narrative lasted 30 – 50 minutes, and was transcribed verbatim. Analysis Transcription is a part of the analysis, regardless of whether the emphasis is on the content or the form [ 25 ]. This study focused on the contents of the stories. We used Gee's analysis structure [ 26 ], which is adapted to long oral stories, and gives attention to how a story is told and organizes the narrative into sequences, "stanzas" (poetic units) that often consist of four lines and represent a condensation of the narration. The narration also contains small interruptions that embody the basic themes. The basic theme often consists of a single line or sentence, the "coda" [ 26 ] which gives instant information about the meaning of the narration [ 22 ]. To discover the rhythm and to gain an understanding of each narration, the first author (CC) listened to the taped narratives. The two authors CC and KN thereafter read and re-read the transcribed text several times to identify the meanings (the stanzas) and the basic themes (the codas) of the stories (for examples, see Table 2 ). This monitoring (Riessman's term for reading and interpreting a text) emphasized how the narrative consisted of stanzas and codas [ 22 , 26 ]. Table 2 Illustrations of structured transcribed text into stanzas and codas Transcribed text Stanza and coda /.../ It can have do with prostheses, cytotoxins and radiation therapy, then that I haven't had to do but then there's so much we have in common, so the pain over getting the diagnosis and experience of the operation and – worry about the future Stanza There's so much we have in common pain over the diagnosis experience of the operation and worry about the future And it feels incredibly good to be able to talk to someone who has the exact same experience , even though the health care personnel have a lot of experience – but that is – that is a different kind of experience, since it's more observational or from the outside /.../ (Informant 8) Coda And it feels incredibly good to be able to talk to someone who has the exact same experience The authors examined and discussed the stanzas and codas on the basis of their contents and relevance in relation to the aim. The final monitoring was done to find the informants' meaning according to their breast cancer when they met another individual with experience of breast cancer, i.e. a CP. The coda was used as the base in the final thematic monitoring [ 22 ] and we found 3 themes, including sub-themes, that illustrated the individuals' experience of CPs. Stanzas were used later in this text to exemplify what emerged in the themes. Ethical permission for the study was obtained from the Ethics committee at Lund University (LU 605-01). Results The thematic contents of the narratives' stanzas fell into the following themes, which were related to the purpose of the study: 1. Shared experiences give new perspectives on having cancer 2. Feelings of isolation are a part of the identity of the illness 3. Relations with others enable self-help Themes and sub-themes are presented in Table 3 . Table 3 Themes and sub-themes Theme Sub-theme 1. Shared experiences give new perspectives on having cancer 1:1 Feelings of not being alone 1:2 Cancer can be survived 2. Feelings of isolation are a part of the identity of the illness 2.1 Being different, unique and odd 2:2 One among others like oneself 3. Relations with others enable self-help 3:1 Talking with others, asking questions and learning 3:2 Right time for making contact 3:3 The contact person provides support 3:4 Indifference in the contact with the CP 1. Shared experiences give new perspectives on having cancer 1:1 Feelings of not being alone The meeting with the contact person can have emotional significance in terms of understanding that one is not alone in one's illness. Emotions came out in one case by a postcard sent by the CP. The following stanza from the informant exemplifies the CPs' activity. I got a card from my CP a (certain motif) where she wrote (name) I know what you're going through I'm thinking about you (informant 2) The receiver kept the card and it gave her a warm feeling long after the most difficult time of the cancer illness was over. 1:2 Cancer can be survived The contribution of the cancer association and the CPs to giving the women insight about their illness can take place through invitations to rehabilitation activities, such as physical therapy groups. One woman said that, in spite of her having many questions, she did not feel ready to have contact with the association other than coming to the therapy sessions, which gave that woman sufficient and obvious evidence that she was not alone. At the water therapy sessions, she was able to see that it was possible to survive cancer and said "it's nice to be able to, like feel that fellowship". The following stanza illustrates the activity. when we sit in the sauna – so everybody is looking - some have no breasts and some they've taken parts of everybody looks a little strange (informant 7) 2. Feelings of isolation are a part of the identity of the illness 2:1 Being different, unique, and odd One woman said that her contact with her CP gave a long-lasting and negative experience of the meeting. After having received advice from a nurse, she had herself contacted the CP. Instead of getting access to other women's experiences through the CP she was left with an even stronger feeling of not only being ill but also of being odd. This feeling was caused by the CP not knowing that it was possible to have bilateral breast cancer at the same time (which this woman had). The CP had reacted with alarm – a reaction that frightened her and made her even more scared. These unpleasant feelings were still present after a year and because of the negative reactions she did not seek further contact with the CP, which led to a sense of embarrassment since she had initiated contact. The narrative illustrates the feelings isolation was strengthened by the CP and how the process of relating to the new disease identity was damaged. Another narrative came from a woman who felt odd because she did not identify herself as being ill. She felt that her cancer was less serious since she did not need further treatment after the operation. This feeling of being isolated gave her a great deal to think about, which she could talk about with her CP. The reaction of being ill came late, according to the informant, and was triggered when she wanted to donate blood. When she filled in the health certificate she realized she was not permitted to do so. The CP was available at that time to talk about the incident and the following two stanzas illustrate the importance that the CP was ascribed: it feels like she [the CP] understands what I mean, like she understands that I can float above it and thought like it wasn't so serious (informant 8) yes, I think that she took me seriously when I felt like it was hard not to be able to identify myself in the group [the PACP group] (informant 8) 2:2 One among others like oneself One woman described being one among others like oneself by talking about the special feeling that came over her when she was welcomed into the association. It was not necessary to say that she had had a breast cancer operation since everyone there knew; this was obvious in the association. Another woman described the importance of meeting people in the same age and situation in life, which took place at regular theme meetings for younger women. A younger woman exemplified the importance of being the same age by saying that women of the same age had common questions to discuss – questions that could not be talked about with health professionals. The following two stanzas illustrate how the women used the CP meetings to talk with women with similar problems: it could for example be how I could make up my eyes now when I don't have any eyelashes (informant 6) if there was anyone who had children because the ones I'd met were only older people who didn't have children living at home (informant 6) 3. Relations with others enable self-help 3:1 Talking with others, asking questions, and learning At each contact meeting, the women felt that they could gain knowledge through the different experiences that were described about different forms of treatment and their effects on well-being. One woman describes how she could balance her worry in this environment by talking about what she thought about, which is illustrated in the stanzas below: how will I go through this psychologically in the future will I trust this answer or how much will I worry (informant 2) alot of questions like that that I've thought about and that I could talk about with other people (informant 2) Another woman's reactions to some of the meetings were feelings of sadness, while other meetings could give her hope for the future. However, while the meetings made her react differently, she felt it was important to come to the CP meetings regularly to see with her own eyes that the other women were alive and led good lives. It came out in another narrative how it is gradually possible to talk with any of the members of the association that have experience even though it is the CP that is the person one initially talks with. Thus in the long run the experiences are felt as being most important, where the CP acts to open the door to them. 3:2 Right time for making contact Time, in the sense of a particular phase in the process, is important when the women want to make contact with others with experience of having breast cancer. One woman felt that she was not ready to contact the association and meet others with similar experience despite having many questions. At first, she said, she wanted most just to crawl into herself and manage on her own without involving other people. Nevertheless, this woman was content having received a card from the CP early in the process, telling her about the possibility of establishing a contact. The possibility of getting a contact is thus given higher priority than hearing others' experiences at an early stage. 3:3 The contact person provides support One woman evaluated the meeting with the CP positively in part because the CP was a trained CP and in part because the CP knew what she needed to hear. This is illustrated in the following stanza: it was most of all good to hear [name of CP] maybe because she had training knew what you needed to hear (informant 8) Another woman described how she had used her CP as a "sounding board" in her choice of surgical method. It comes out in this narrative that the sounding board function and not the counselling function made the woman finally decide in favour of a surgical method that the CP had not recommended on the basis of her own experience. The fact that the CP had given an opinion helped the woman to form an opinion herself. Thus it is not necessary to share the exact same experience; at times it is enough to share the experience of cancer. One of the women said that she would have appreciated there being someone with experience to contact when she was forced to wait for a long period before having her operation. Access to a CP for one's own personal problems is illustrated in the following stanza: I know she's the contact person which makes me think that I can turn to her and talk about my special problems (informant 8) 3:4 Indifference in the contact with the CP Women describe not only the benefit of the CP meetings but also say that there can be a feeling of more or less indifference in the contact with the CP. In one example the CP acted as a guide to the association's premises but had otherwise not spoken of anything in particular. Another woman had come into contact with the association's CP at an educational class but had never contacted a CP herself. For these women it was sufficient to contact with other women in educational classes organised by professionals. Discussion Since the CP activity is central within the PABCPs, we aimed to investigate how women with breast cancer described their experiences from having contact with a CP. The narratives allowed the women to reflect and formulate themselves through this opportunity to tell their own stories with the aim to analyse their experiences [ 24 ]. The strength of the narrative method lies in studying a person's identity in times of changing circumstances as in this case, where the individual with a diagnosis of breast cancer meets another individual with the same experience [ 22 ]. The experiences described from CP meetings reflect how shared experiences give new perspectives on having cancer and how it is possible to help oneself through the relation to the CP. PABCPs offer meetings with survivors, but it may be difficult to predict the optimal time for each individual to establish such contact since individuals react differently to the diagnosis and to the cancer treatment it requires. The narratives in our study illustrate the importance of meeting women of the same age and in a similar life situation and of being in an environment (CP meeting) that is free of need to explain. Meeting others and experiencing their' reactions help these women feel normal [ 16 ]. "Catching one's breath" from the isolation that the situation creates seems, according to the respondents, to be needed during certain periods. To become aware of not being alone with the disease, to receive visible evidence there of, and to realize that it is possible to survive appeared most important to the women in our study. The fellowship of existential uncertainty could be discerned in several of the narratives. However, although some women expressed a need to meet with others in order to reduce feelings of isolation, they did not always initiate a contact with the CP. Our results demonstrate that CPs could act as sounding boards in treatment issues and that this could be important for the woman's possibility to make her own decision. When these data have been presented during further contacts with PACPB members the findings were perceived as recognizable and the members reported similar experiences (data not shown). However, other aspects and reactions may be identified in other types of patient's associations or in other cultures. Indeed, considering the large number of PABCP in Europe multicentre studies with such a focus would be of interest [ 27 ]. Even though all patients do not wish to be confronted with the experiences of others via personal contact with a CP, they may be ready to be confronted with the disease in other ways. A confrontation with women who have undergone partial or complete mastectomies can take place in water therapy sessions and in the sauna. The results show that this meeting also gave a feeling of fellowship. This may mean that the person who chooses only this kind of activity is ready only for visual impressions. Breasts mean different things to different women, and only the individual woman can define how great a handicap breast surgery is. Langellier and Sullivan [ 28 ] studied illness narratives among breast cancer patients that showed clusters of meanings: the medicalized breast, the functional breast, the gendered breast and the sexualized breast. Compared to other research these results suggest both greater and fewer problems with femininity, sexuality and body image than have earlier been presumed. It is also important that a CP is able to manage timing, in the sense of the phase in the process at which contact is made with others who have experience or when the person is ready to confront others' experiences. Great sensitivity towards individual needs is required and there are individual variations as to the optimal time for a meeting, which are related to when the individual is ready to encounter other people's experiences. In terms of being ready, thoughts arise as to the many visual and auditory impressions that cancer patients receive during a treatment session, e.g. in a waiting room and when given treatment, regardless of whether they are ready for these or not. It is conceivable that some individuals do not have any need to meet with other people and share their experiences since recovery can also occur without involving others. However, it is not only the initial need for contact that varies. Women also have different needs in terms of the duration of the contact. For one person, contact with a CP may simply mean guidance in the association's activities while for another a CP can act as a support person for a period of several years and the information-seeking behaviour can also change over time [ 29 ]. For the woman who felt great isolation in the illness the meeting with the CP strengthened the feelings of isolation, which demonstrates how the meeting with the CP led to a non-intended consequence. This demonstrates the need for the PABCPs to be meticulous in the choice of CPs; the requirements must not only be that they have been rehabilitated but also that they have substantial knowledge about cancer, treatment possibilities, and psychological reactions to disease. Although self-help may come through a relation with other individuals with a shared experience [ 30 ] the associations' CPs should be aware that patients' meeting with them does not necessarily provide support, but may strengthen feelings of isolation. The results also demonstrate that some questions could not be answered by health-care professionals, but rather by individuals with personal experience from cancer. Hence, new knowledge developed in the meeting with the CP (cf Borkmans [ 14 ] thinking concerning the meaning perspective) and it is therefore important that health-care professionals allow and perhaps also encourage cancer patients to participate in patient's associations and thereby share their experiences with others [ 31 ]. Worry was balanced by their own experience being reflected in the experience of others, which indicates that CPs have a function in areas that professionals by tradition control, such as in information about treatments and outcome. Our findings confirm results that indicate that support in the form of social relationships with other breast cancer women empowers these women by giving them abilities to cope and adopt supportive roles towards each other during treatment [ 32 , 33 ]. Furthermore the emotional support that is given in connection with other survivors is important for navigating the short and long term impacts of cancer as well as the benefits from rehabilitation [ 34 , 35 ]. Similar observations have been made among individuals with prostate cancer, where shared experiences give reassurance, helps alleviate anxiety, and provides the participants with a more positive outlook [ 36 ] and self-help has a potential that could be strengthened in cancer care [ 37 ]. Conclusions Our study on patient's experience from the CP activity within PABCPs – based on narratives from 8 patients with breast cancer – show that shared experiences give new perspectives on having cancer, that feelings of isolation are part of the identity of the illness, and that relations with other individuals may enable self-help. The CP is thus important for the breast cancer patient since it helps these individuals to gain a perspective on their disease, to realize that they are not alone, to provide hope for survival, and support in the feelings of isolation that are part of the identity of cancer. From the patient's perspective, it is important that the health care system provides information on the CPs, whose responsibility it is to listen, support and act as conversational partners, and the importance of having access to other persons' experiences. The CP serves as a counsellor and needs to have an understanding for and knowledge about the patient's needs and expectations since self-help may come about in this relation. The CPs should also be aware that their presence and a possible lack of knowledge can sometimes disturb the psychological management of the disease, and contrary to the intention, strengthen feelings of isolation. The CP must also be ready to confront others' experiences, but also needs to understand that not all individuals have such a need. The CP must be able to offer many different kinds of help related to feelings of isolation, survival, and rehabilitation, and may thereby act as a sounding board for women's experiences and a shelter for emotional expressions. Hence, to optimise rehabilitation for individuals with breast cancer, PABCP should carefully select and educate CP's, and an exchange of experiences between the PABCP's and the health care system may contribute to this process. List of abbreviations PABCP Patient associations for breast cancer patients CP Contact person Competing interests The author(s) declare that they have no competing interests. Authors' contributions CC: study design, data collection, data analysis, and writing the manuscript. MN: supervising the study and participation in writing the manuscript. KN: study design, data analysis, supervising the study, and participation in writing the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548515.xml

514530

Sad, blue, or depressed days, health behaviors and health-related quality of life, Behavioral Risk Factor Surveillance System, 1995–2000

Background Mood disorders are a major public health problem in the United States as well as globally. Less information exists however, about the health burden resulting from subsyndromal levels of depressive symptomatology, such as feeling sad, blue or depressed, among the general U.S. population. Methods As part of an optional Quality of Life survey module added to the U.S. Behavioral Risk Factor Surveillance System, between 1995–2000 a total of 166,564 BRFSS respondents answered the question, "During the past 30 days, for about how many days have you felt sad, blue, or depressed?" Means and 95% confidence intervals for sad, blue, depressed days (SBDD) and other health-related quality of life (HRQOL) measures were calculated using SUDAAN to account for the BRFSS's complex sample survey design. Results Respondents reported a mean of 3.0 (95% CI = 2.9–3.1) SBDD in the previous 30 days. Women (M = 3.5, 95% CI = 3.4–3.6) reported a higher number of SBDD than did men (M = 2.4, 95% CI = 2.2–2.5). Young adults aged 18–24 years reported the highest number of SBDD, whereas older adults aged 60–84 reported the fewest number. The gap in mean SBDD between men and women decreased with increasing age. SBDD was associated with an increased prevalence of behaviors risky to health, extremes of body mass index, less access to health care, and worse self-rated health status. Mean SBDD increased with progressively higher levels of physically unhealthy days, mentally unhealthy days, unhealthy days, activity limitation days, anxiety days, pain days, and sleepless days. Conclusion Use of this measure of sad, blue or depressed days along with other valid mental health measures and community indicators can help to assess the burden of mental distress among the U.S. population, identify subgroups with unmet mental health needs, inform the development of targeted interventions, and monitor changes in population levels of mental distress over time.

Background Mood disorders are a major public health problem in the United States as well as globally, imposing a substantial burden of disability, impaired quality of life, and death if they remain untreated [ 1 - 3 ]. National estimates for 12-month prevalence of depressive disorders for adults aged 18 and over range between 6.3%-11.3% depending on the assessment tools, criteria used, and populations studied [ 1 , 4 - 6 ]. The lifetime prevalence of six selected mood disorders, including major depressive episode, dysthymia, and bipolar disorder as assessed by the Diagnostic Interview Schedule [ 7 ] among 7,667 respondents aged 17–39 years to the third National Health and Nutrition Examination Survey, was 8.6% for major depressive episode; 7.7% for severe major depressive episode; 6.2% for dysthymia; 3.4% for combined major depressive episode and dysthymia, 1.6% for any bipolar disorder, and 11.5% for any mood disorder [ 2 ]. In the Alameda County Study, 6.6% of men and 10.1% of women aged 50 years or older met DSM-III-R and DSM IV [ 8 , 9 ] symptom criteria for major depression within the past two weeks [ 10 ]. By 2020, depression will become the second leading cause for disease burden [ 11 ] Mental health disorders due to depression, anxiety and substance use are not only burdensome by themselves, but they can complicate existing physical disorders and also increase risk for other physical comorbidity [ 3 , 12 , 13 ]. For example, psychological distress might interfere with medication adherence for an existing disorder such as hypertension, but also increase the likelihood of adopting unhealthy behaviors such as smoking, excessive alcohol use, or overeating that can further impair physical health. Both major depressive disorder and subsyndromal levels of depression are associated with similar demographic, social, psychiatric and physical health predictors [ 14 , 15 ]. Results from the 1980–1985 Epidemiologic Catchment Area Study indicated that almost 30% of the population reported having experienced a period lasting at least two weeks in their lifetime when they felt sad, blue or depressed or lost interest in previously pleasurable things or activities [ 16 ]. The inclusion of lesser levels of depressive symptomatology when calculating estimates of the prevalence of diagnosable depression could inflate such estimates [ 17 ]. However, an examination of subsyndromal levels of depression to determine what proportion of the population is at risk for major depressive disorder can be useful for communities interested in preventing depression [ 12 ]. Examining subsyndromal depression and its associated risk with unhealthful behaviors, furthermore, can highlight associations between feeling sad, blue or depressed and behaviors risky to health [ 3 ], and is of public health interest [ 12 , 18 - 20 ]. For at least one year since 1995, more than one-third of state health departments have assessed the number of recent days that adults "felt sad, blue or depressed" using the Behavioral Risk Factor Surveillance System (BRFSS). This measure has good construct validity when compared with other BRFSS health-related quality-of-life (HRQOL) domains related to mental health [ 21 ], and has acceptable reliability and criterion validity when compared with the mental health scales of the Medical Outcomes Study Short-Form 36 (SF-36) [ 22 ] and with the Center for Epidemiologic Studies Depression Scale (CES-D) [ 23 ] among older, low-income African-American men [ 24 ]. Using a large multi-state sample, this study is the first to focus on the prevalence of self-reported "sad, blue or depressed" days (SBDD) overall and in sociodemographic subgroups in the United States. It also examines the construct validity of the measure. Methods The BRFSS, which is designed to monitor behavioral health risks in the United States, is an annual random-digit-dialed telephone survey of the non-institutionalized civilian population aged 18 years or older conducted in all states and the District of Columbia [ 25 ]. As part of an optional Quality of Life survey module that was added to the BRFSS and used in 38 states and the District of Columbia in one or more years from 1995 through 2000, a total of 166,564 BRFSS respondents answered the question, "During the past 30 days, for about how many days have you felt sad, blue, or depressed?" The module also contains questions on the number of recent days of pain, anxiety, sleeplessness and on other HRQOL domains. Respondents answered standard BRFSS questions about age, race/ethnicity, education, employment, income, marital status, health status, physical health, mental health, activity limitation, access to care, and the presence of certain health conditions such as hypertension and diabetes. Respondents also answered questions about how often they engaged in behaviors risky to health, such as smoking and binge drinking, and in health promoting behaviors, such as using a seat belt and exercising. Each respondent's body mass index (BMI), (weight in kilograms divided by the square of height in meters), was classified according to the National Institutes of Health criteria as either underweight (<18.5 kg/m 2 ), normal weight (18.5< 25.0 kg/m 2 ), overweight (25.0 to < 30.0 kg/m 2 ), or obese (≥ 30.0 kg/m 2 ) [ 26 ]. Individual responses were weighted to reflect the age and sex distribution of each state's population during each survey year. To account for the BRFSS's complex sample survey design, means (M) and 95% confidence intervals (CI) for SBDD and other HRQOL measures were calculated using SUDAAN (Research Triangle, release 8.0.0, Research Triangle Park, NC: 2001). Because mean SBDD varied by five-year age groups, the analyses were directly standardized to the age distribution of adults aged 18 years or older from the 2000 U.S. Census to control for confounding by age. Results Respondents reported a mean of 3.0 (95% CI = 2.9–3.1) SBDD in the previous 30 days. About 43.4% of respondents reported one or more SBDD including 7.9% who reported 14 or more SBDD. Women (M = 3.5, 95% CI = 3.4–3.6) reported a higher number of SBDD than did men (M = 2.4, 95% CI = 2.2–2.5) (Table 1 ). Young adults aged 18–24 years reported the highest number of SBDD, whereas older adults aged 60–84 reported the fewest, with the gap in mean SBDD between men and women decreasing with increasing age (Table 1 ). Table 1 Mean number of days U.S. adults felt sad, blue or depressed (Behavioral Risk Factor Surveillance System)* Males Females Males & Females N Mean 95% CI N Mean 95% CI N Mean 95% CI Demographic, behavioral risk group variable 5 year age group 18–19 yrs. 1,704 2.8 2.4–3.2 1,914 4.5 4.0–5.0 3,618 3.6 3.2–3.9 20–24 yrs. 4,899 2.8 2.5–3.0 6,697 4.0 3.7–4.2 11,596 3.4 3.2–3.6 25–29 yrs 6,439 2.4 2.2–2.6 8,881 3.4 3.2–3.6 15,320 2.9 2.7–3.1 30–34 yrs. 7,163 2.2 2.0–2.4 10,016 3.6 3.4–3.8 17,179 2.9 2.7–3.0 35–39 yrs. 7,763 2.5 2.3–2.6 11,061 3.9 3.7–4.1 18,824 3.2 3.0–3.3 40–44 yrs. 7,589 2.6 2.4–2.8 10,430 3.7 3.5–3.9 18,019 3.2 3.0–3.3 45–49 yrs. 6,729 2.7 2.5–2.9 9,297 3.8 3.5–4.0 16,026 3.2 3.0–3.4 50–54 yrs. 5,722 2.6 2.4–2.9 7,986 3.8 3.5–4.1 13,708 3.2 3.0–3.4 55–59 yrs. 4,525 2.2 1.9–2.4 6,424 3.6 3.3–3.9 10,949 3.0 2.7–3.2 60–64 yrs. 3,806 2.2 1.9–2.6 5,489 3.2 2.9–3.4 9,295 2.7 2.5–2.9 65–69 yrs. 3,733 1.8 1.6–2.1 5,905 2.6 2.4–2.9 9,638 2.3 2.1–2.5 70–74 yrs. 3,232 1.9 1.6–2.2 5,629 2.8 2.5–3.1 8,861 2.4 2.2–2.6 75–79 yrs. 2,287 2.2 1.8–2.5 4,627 2.8 2.5–3.1 6,914 2.5 2.3–2.8 80–84 yrs. 1,270 2.3 1.8–2.8 2,949 2.6 2.3–3.0 4,219 2.5 2.2–2.8 85+ yrs. 616 2.4 1.7–3.1 1,782 3.1 2.6–3.7 2,398 2.9 2.4–3.4 All categories 67,477 2.4 2.2–2.5 99,087 3.5 3.4–3.6 166,564 3.0 2.9–3.1 Race/ethnicity Asian/Pacific Islander 1,098 1.6 1.2–2.0 1,216 2.5 2.0–2.9 2,314 2.0 1.7–2.3 White 54,502 2.3 2.2–2.4 77,761 3.3 3.3–3.4 132,263 2.8 2.7–2.9 Hispanic 4,395 2.9 2.5–3.2 6,455 4.3 3.9–4.6 10,850 3.6 3.3–3.8 Black 5,852 2.9 2.7–3.2 11,463 4.5 4.2–4.7 17,315 3.8 3.6–4.0 Native American Indian/Alaska Native 679 3.0 2.3–3.7 956 5.0 4.1–5.9 1,635 4.0 3.4–4.6 Other 431 3.0 2.0–4.0 538 5.2 4.0–6.4 969 4.2 3.2–5.1 All categories 66,526 2.4 2.3–2.5 97,851 3.6 3.4–3.7 165,346 3.0 2.9–3.1 Education < High school 7,772 3.7 3.4–4.0 12,468 5.6 5.3–5.9 20,240 4.7 4.5–4.9 HS graduate 20,727 2.6 2.4–2.7 32,902 3.8 3.7–3.9 53,629 3.2 3.1–3.3 Some college 17,770 2.4 2.2–2.5 27,700 3.4 3.3–3.5 45,470 2.9 2.8–3.0 College graduate 21,068 1.7 1.6–1.8 25,826 2.4 2.3–2.5 46,894 2.0 1.9–2.1 All categories 67,337 2.4 2.3–2.5 98,896 3.6 3.5–3.7 166,233 3.0 2.9–3.1 Employment Employed (wages) 41,465 1.9 1.8–2.0 51,567 2.9 2.8–3.1 93,032 2.4 2.3–2.5 Self-employed 7,497 2.3 2.0–2.5 5,354 3.0 2.7–3.3 12,851 2.5 2.3–2.7 Retired 12,054 3.6 2.4–4.9 19,880 2.4 1.5–3.2 31,934 3.0 2.0–4.0 Student 1,997 2.0 1.6–2.4 3,019 3.8 3.1–4.5 5,016 3.3 2.7–3.9 Homemaker 168 2.7 1.6–3.9 11,549 3.5 3.3–3.8 11,717 3.5 3.3–3.7 Unemp. < 1 yr. 1,238 4.5 3.9–5.1 2,094 6.4 5.5–7.3 3,332 5.7 4.9–6.5 Unemp. ≥1 yr. 766 5.3 4.4–6.2 1,657 6.3 5.6–7.0 2,423 6.1 5.5–6.7 Unable to work 2,225 9.6 8.5–10.6 3,878 10.7 10.0–11.5 6,103 10.2 9.5–10.8 All categories 67,410 2.4 2.3–2.5 98,998 3.6 3.5–3.7 166,408 3.0 2.9–3.1 Income <$15,000 5,194 5.4 5.0–5.8 13,012 6.5 6.2–6.8 18,206 6.1 5.8–6.3 $15,000–$24,999 10,765 3.3 3.1–3.4 18,662 4.5 4.3–4.7 29,427 3.9 3.8–4.0 $25,000–$49,999 23,046 2.2 2.1–2.4 29,419 3.2 3.0–3.3 52,465 2.7 2.6–2.8 ≥$50,000 20,046 1.7 1.6–1.8 21,382 2.4 2.3–2.6 41,428 2.0 1.9–2.1 All categories 59,051 2.5 2.4–2.6 82,475 3.6 3.5–3.7 141,526 3.0 2.9–3.1 Marital status Currently married 38,756 2.1 1.8–2.3 49,671 3.0 2.8–3.1 88,427 2.5 2.4–2.6 Never married 14,429 3.0 2.8–3.2 15,847 3.6 3.3–3.8 30,276 3.3 3.1–3.4 Divorced 8,060 3.6 3.1–4.2 13,600 5.0 4.7–5.2 21,660 4.4 4.1–4.8 Unmarried couple 1,536 3.8 2.6–5.0 2,008 4.7 3.7–5.6 3,544 4.5 3.6–5.5 Widowed 2,980 4.2 3.4–5.0 14,562 6.6 5.5–7.7 17,542 6.0 5.1–6.9 Separated 1,609 5.4 4.6–6.1 3,164 6.3 5.7–6.9 4,773 6.0 5.5–6.5 All categories 67,370 2.4 2.3–2.5 98,852 3.6 3.5–3.6 166,222 3.0 2.9–3.1 Participate in any physical activity Yes 36,039 2.1 2.0–2.2 49,472 3.1 3.0–3.2 85,511 2.6 2.5–2.7 No 13,388 3.1 3.0–3.3 22,607 4.6 4.4–4.8 35,995 3.9 3.8–4.1 All categories 49,427 2.4 2.3–2.5 72,079 3.6 3.5–3.7 121,506 3.0 2.9–3.1 Body mass index Underweight 608 4.6 3.6–5.6 3,192 4.5 4.0–5.0 3,800 4.5 4.1–5.0 Normal 23,561 2.5 2.4–2.6 47,187 3.0 2.9–3.1 70,748 2.8 2.7–2.9 Overweight 30,305 2.1 2.0–2.2 25,987 3.8 3.6–3.9 56,292 2.7 2.6–2.8 Obese 12,295 2.8 2.6–2.9 17,031 5.0 4.7–5.2 29,326 3.8 3.7–4.0 All categories 66,769 2.4 2.3–2.5 93,397 3.6 3.5–3.7 160,166 3.0 3.0–3.1 Ever drank ≥ 5 drinks in past 30 days at once Yes 6,403 2.6 2.4–2.9 2,968 5.0 4.4–5.5 9,371 3.3 2.9–3.6 No 18,206 2.2 2.0–2.3 29,554 3.3 3.2–3.4 47,760 2.8 2.7–2.9 All categories 24,609 2.3 2.2–2.4 32,522 3.4 3.3–3.5 57,131 2.8 2.7.2.9 Smoking status Never smoked 31,088 1.9 1.8–2.0 57,611 3.0 2.9–3.1 88,699 2.6 2.4–2.7 Former smoker 18,973 2.2 2.1–2.4 19,639 3.5 3.4–3.7 38,612 2.9 2.8–3.0 Smokes <1 pack/day 7,112 3.1 2.8–3.3 11,357 5.0 4.7–5.2 18,469 4.1 3.9–4.3 Smokes ≥1 pack/day 8,751 4.2 3.8–4.6 8,688 6.1 5.7–6.4 17,439 5.0 4.7–5.2 All categories 65,924 2.4 2.3–2.5 97,295 3.6 3.4–3.7 163,219 3.0 2.9–3.1 Seatbelt use Always 7,960 2.2 2.0–2.4 13,769 3.1 2.9–3.2 21,729 2.7 2.6–2.8 Nearly always 2,142 2.0 1.7–2.3 2,469 3.5 3.1–3.9 4,611 2.7 2.4–2.9 Sometimes 1,287 2.8 2.4–3.3 1,291 4.4 3.8–4.9 2,578 3.5 3.1–3.8 Seldom 698 3.1 2.4–3.8 578 5.5 4.5–6.4 1,276 3.9 3.4–4.5 Never 901 4.0 3.3–4.7 639 6.1 5.0–7.1 1,540 4.7 4.1–5.3 All categories 12,988 2.4 2.3–2.5 18,746 3.4 3.2–3.5 31,734 2.9 2.8–3.0 Time when could not afford to get medical care in past year Yes 5,629 5.4 5.0–5.7 11,867 6.8 6.5–7.1 17,496 6.2 6.0–6.5 No 61,740 2.1 2.0–2.2 87,069 3.1 3.0–3.2 148,809 2.6 2.5–2.7 All categories 67,369 2.4 2.3–2.5 98,936 3.6 3.4–3.7 166,305 3.0 2.9–3.1 Have any kind of health plan? Yes 58,645 2.2 2.1–2.3 87,693 3.3 3.2–3.4 146,338 2.8 2.7–2.9 No 8,657 3.7 3.4–4.1 11,236 5.1 4.7–5.4 19,893 4.4 4.1–4.6 All categories 67,302 2.4 2.3–2.6 98,929 3.6 3.4–3.7 166,231 3.0 2.9–3.1 Self-rated health Excellent 16,485 1.4 1.3–1.5 22,518 1.9 1.8–2.0 39,003 1.6 1.5–1.7 Very good 22,980 1.7 1.6–1.8 33,099 2.7 2.6–2.8 56,079 2.2 2.1–2.3 Good 18,980 2.5 2.3–2.6 28,081 3.9 3.7–4.0 47,061 3.2 3.1–3.3 Fair 6,441 4.8 4.5–5.1 11,041 7.2 6.8–7.5 17,482 6.0 5.8–6.3 Poor 2,464 11.3 10.2–12.4 4,154 11.2 10.3–12.1 6,618 11.2 10.5–12.0 All categories 67,350 2.4 2.3–2.5 98,893 3.6 3.4–3.7 166,243 3.0 2.9–3.1 Note. *Selected U.S. states, 1995–2000; all variables except age-groups are age-adjusted. Asians/Pacific Islanders reported the fewest SBDD (M = 2.0, 95% CI = 1.7–2.3), whereas Hispanics, Blacks, American Indians and Alaska Natives, and non-whites of another race/ethnicity reported about 4–5 SBDD; in the last three of these groups, the gaps between men and women were larger. Adults with more education reported fewer SBDD, with the gap between men and women diminishing with more education. Respondents who were unemployed or unable to work reported more SBDD than the employed. The number of SBDD decreased with increasing levels of annual household income. Widowed or separated adults reported about 6 SBDD, whereas respondents who were currently married reported the fewest number of SBDD (M = 2.5); the gap between men and women was least among those who had never married. SBDD was associated with an increased prevalence of behaviors risky to health, extremes of BMI, less access to health care, and worse self-rated health status. Respondents who reported physical inactivity, binge drinking, seldom-use or nonuse of seatbelt, or any or more cigarette smoking reported substantially higher numbers of SBDD than those who did not report engaging in these risky behaviors (Table 1 ). Subjects who were either underweight or obese reported a higher number of SBDD than those of normal weight or overweight. Obese women reported more SBDD (M = 5.0, 95% CI = 4.7–5.2) than obese men (M = 2.8, 95% CI = 2.6–2.9). However, underweight men and women reported the same number (about 4.5) of SBDD. Respondents who could not afford to see a physician at least once during the past year or who had no health care insurance coverage reported more SBDD than those who could afford to see a physician or had such coverage. Although subjects with excellent self-rated health status reported 1.6 SBDD, those with poor health reported 11.2 SBDD, with the largest gap (2.4 days) occurring between men and women with fair health status. Additional Construct Validity SBDD were associated with other physical and mental HRQOL domains in expected ways. Mean SBDD increased with progressively higher levels (i.e., 0, 1–2, 3–13, 14–29, and 30 days) of physically unhealthy days, mentally unhealthy days, unhealthy days, activity limitation days, anxiety days, pain days, and sleepless days (Table 2 ). Similarly, subjects who reported more days when they felt "very healthy and full of energy" reported fewer SBDD. The gap between men and women was smaller at lower levels of these other HRQOL domains. Table 2 Mean number of days adults felt sad, blue or depressed by HRQOL* and sex** HRQOL variable Males Females Male & Females N Mean 95% CI N Mean 95% CI N Mean 95% CI Physically unhealthy days 0 days 48,188 1.7 1.6–1.8 63,192 2.6 2.5–2.7 111,380 2.2 2.0–2.3 1–2 days 6,792 2.3 2.1–2.5 10,855 3.0 2.8–3.2 17,647 2.6 2.5–2.8 3–13 days 6,371 3.6 3.4–3.9 12,631 4.6 4.4–4.8 19,002 4.2 4.1–4.4 14–29 days 1,959 5.9 5.3–6.5 4,614 7.5 7.0–8.0 6,573 6.9 6.5–7.3 30 days 3,539 8.1 7.3–8.8 6,207 9.3 8.6–9.9 9,746 8.7 8.3–9.2 all categories 66,849 2.4 2.3–2.5 97,499 3.5 3.4–3.6 164,348 3.0 2.9–3.1 Mentally unhealthy days 0 days 50,042 1.1 1.0–1.2 63,446 1.5 1.4–1.6 113,488 1.3 1.2–1.4 1–2 days 5,446 2.1 1.9–2.2 9,644 2.2 2.0–2.3 15,090 2.1 2.0–2.2 3–13 days 6,717 4.9 4.6–5.2 14,219 5.2 5.0–5.3 20,936 5.0 4.8–5.2 14–29 days 2,194 11.1 10.5–11.8 5,318 12.6 12.2–13.1 7,512 12.1 11.7–12.5 30 days 2,366 16.1 15.3–16.9 5,173 17.2 16.6–17.7 7,539 16.8 16.3–17.2 all categories 66,765 2.4 2.3–2.5 97,800 3.5 3.4–3.6 164,565 3.0 2.9–3.1 Unhealthy days + 0 days 38,831 0.9 0.8–1.0 45,983 1.3 1.2–1.4 84,814 1.1 1.0–1.2 1–2 days 7,675 1.4 1.3–1.5 11,093 1.5 1.4–1.6 18,768 1.5 1.4–1.6 3–13 days 10,963 2.9 2.8–3.0 20,785 3.4 3.3–3.5 31,748 3.2 3.1–3.5 14–29 days 3,344 6.7 6.3–7.2 7,935 7.7 7.4–8.0 11,279 7.4 7.1–7.6 30 days 5,446 10.7 10.2–11.3 10,678 12.4 12.1–12.8 16,124 11.7 11.4–12.1 all categories 66,259 2.4 2.3–2.5 96,474 3.5 3.4–3.6 162,733 3.0 2.9–3.1 Activity limitation days 0 days 56,202 1.7 1.6–1.8 77,711 2.6 2.5–2.7 133,913 2.1 2.0–2.2 1–2 days 3,556 2.5 2.2–2.7 6,446 3.3 3.1–3.5 10,002 3.0 2.8–3.1 3–13 days 3,462 5.3 4.9–5.7 7,169 6.6 6.3–6.9 10,631 6.1 5.8–6.4 14–29 days 1,305 10.2 9.4–11.0 2,887 11.8 11.2–12.5 4,192 11.2 10.7–11.7 30 days 1,962 12.3 11.3–13.4 3,204 13.6 12.7–14.5 5,166 13.0 12.2–13.7 all categories 66,487 2.4 2.3–2.5 97,417 3.5 3.4–3.6 163,904 3.0 2.9–3.1 Anxiety days 0 days 32,664 0.6 0.5–0.7 38,600 0.8 0.7–0.9 71,264 0.7 0.6–0.8 1–2 days 11,739 1.1 1.0–1.2 17,592 1.3 1.2–1.4 29,331 1.2 1.1–1.3 3–13 days 14,076 3.1 3.0–3.3 25,130 3.7 3.6–3.8 39,206 3.4 3.3–3.5 14–29 days 3,946 9.4 8.9–9.9 7,834 10.2 9.9–10.6 11,780 9.9 9.6–10.1 30 days 4,018 13.9 13.3–14.5 8,017 15.0 14.6–15.5 12,035 14.5 14.2–14.9 all categories 66,443 2.4 2.3–2.5 97,173 3.6 3.4–3.7 163,616 3.0 2.9–3.1 Pain days 0 days 52,475 1.7 1.6–1.8 72,697 2.6 2.5–2.7 125,172 2.2 2.0–2.2 1–2 days 4,421 2.6 2.3–2.8 7,143 3.5 3.3–3.8 11,564 3.1 2.9–3.3 3–13 days 4,815 3.8 3.6–4.1 8,598 5.5 5.2–5.8 13,413 4.8 4.5–5.0 14–29 days 1,762 6.8 6.1–7.4 3,642 8.6 8.1–9.2 5,405 7.9 7.4–8.3 30 days 3,363 9.1 8.3–9.9 5,612 10.3 9.6–11.0 8,975 9.7 9.2–10.3 all categories 66,836 2.4 2.3–2.5 97,692 3.6 3.4–3.7 164,528 3.0 2.9–3.1 Sleeplessness days 0 days 24,107 1.6 1.5–1.7 31,809 2.2 2.1–2.3 55,916 1.9 1.8–2.0 1–2 days 7,750 1.1 1.0–1.2 10,410 1.6 1.5–1.8 18,160 1.4 1.3–1.5 3–13 days 20,007 2.1 2.0–2.2 29,594 2.8 2.7–2.9 49,601 2.5 2.4–2.6 14–29 days 8,407 5.0 4.6–5.3 13,633 6.0 5.8–6.3 22,040 5.5 5.3–5.7 30 days 6,347 6.4 6.0–6.8 12,087 7.9 7.6–8.3 18,434 7.3 7.0–7.6 all categories 66,618 2.4 2.3–2.5 97,533 3.6 3.4–3.7 164,151 3.0 2.9–3.1 Vitality days 0 days 6,247 7.1 6.6–7.5 11,536 9.0 8.6–9.4 17,783 8.2 7.9–8.5 1–2 days 1,450 5.9 5.2–6.6 3,118 7.4 6.8–8.0 4,568 6.8 6.4–7.3 3–13 days 8,912 4.2 4.0–4.4 14,809 5.4 5.2–5.6 23,721 4.9 4.7–5.0 14–29 days 28,280 1.7 1.6–1.8 41,496 2.5 2.3–2.6 69,776 2.1 2.0–2.2 30 days 20,990 1.1 0.9–1.2 24,687 1.3 1.2–1.4 45,677 1.2 1.1–1.3 all categories 65,879 2.4 2.3–2.5 95,646 3.6 3.4–3.7 161,525 3.0 2.9–3.1 Note. *Health-Related Quality of Life (HRQOL) questions are: Now thinking about your physical health which includes physical illness and injury, for how many days in the past 30 days was your physical health not good?; Now thinking about your mental health which includes stress, depression, and problems with emotions, for how many days in the past 30 days was your mental health not good?; During the past 30 days, for about how many days did poor physical or mental health keep you from doing your usual activities, such as self-care, work, or recreation?; During the past 30 days, for about how many days did pain make it hard for you to do your usual activities, such as self-care, work, or recreation?; During the past 30 days, for about how many days have you felt worried, tense, or anxious?; During the past 30 days, for about how many days have you felt you did not get enough rest or sleep?; During the past 30 days, for about how many days have you felt very healthy and full of energy? **Behavioral Risk Factor Surveillance System; Selected U.S. states 1995–2000; all variables are age-adjusted. + A calculated measure which results from the sum of physically unhealthy days and mentally unhealthy days with a maximum of 30 unhealthy days for an individual. Discussion In our study, U.S. adults reported an average of about 3 days during the past 30 days when they felt "sad, blue or depressed." Our results are consistent with previous studies documenting the increased prevalence of depressive symptoms among the following groups: women [ 27 - 31 ], in certain minority racial and ethnic groups [ 32 ], people with lower levels of education and income [ 32 , 33 ], people of lower employment status [ 27 , 34 , 35 ], people formerly married or living together but not married [ 27 , 36 ], and in those with limited or no access to health care [ 32 , 37 , 38 ]. The gap in the number of SBDD between men and women was less pronounced as socioeconomic status improved. Respondents who reported a higher number of SBDD also reported engaging in unhealthy behaviors such as cigarette smoking, binge drinking, and physical inactivity. Underweight and obese adults also reported higher numbers of SBDD than did normal or overweight adults. These findings extend previous public health studies that have documented an association between self-reported mental distress and behaviors risky to health [ 39 - 42 ]. Given the cross-sectional design of the BRFSS, we were unable to determine whether risky behaviors preceded or followed SBDD. Nonetheless, our findings provide additional evidence for the association of considerable public health importance between negative mood and unhealthful behaviors [ 43 ]. For example, in the prospective Stirling County Study (1952–1992), subjects who became depressed were more likely to initiate smoking, continue smoking, and refrain from quitting smoking than those who had never become depressed [ 44 ]. Negative mood adversely influences self-efficacy to adopt and maintain healthful behaviors and may thwart other self-motivating processes (e.g., attitudes, outcome expectations, and goals) associated with engaging in healthful behaviors [ 45 ]. Perceived inefficacy can foster additional despondency. This finding has implications for public health interventions. For example, psychosocial interventions that elicit positive emotions, instill confidence in adopting health-promoting behavior, and improve people's coping skills might be more effective for individuals with despondent mood than interventions designed to arouse fear regarding the consequences of engaging in risky behaviors–which can foster inefficacy and increased despondency [ 45 ]. Our findings support the construct validity of the SBDD measure in this study because SBDD were associated with other physical and mental HRQOL domains in expected ways. Groups with progressively higher numbers of physically unhealthy days, activity limitation days, and pain days reported a higher number of SBDD. Moreover, these associations were more pronounced with mentally unhealthy days and anxiety days, than with physically unhealthy days, activity limitation days, and pain days. We found an exception to the linear relationship between SBDD and HRQOL measures with our measure for sleeplessness. Adults who reported 1–2 days of sleeplessness reported fewer SBDD than those who reported no days of sleeplessness. Sleep disturbance, both insomnia and hypersomnia are symptoms of depression. Those reporting no days of sleeplessness, but more SBDD, might be those with hypersomnia. Additional studies are warranted to examine this hypothesis. Besides the cross-sectional design, this study has other limitations. Only 38 states and the District of Columbia included the HRQOL supplemental module that assessed SBDD. All states and the District of Columbia, however examined mentally unhealthy days–the number of days respondents experienced poor mental health due to stress, depression or problems with emotions. Mean mentally unhealthy days in the states that assessed SBDD with the HRQOL supplemental module did not differ significantly from that in states that did not. Given the positive correlation between mentally unhealthy days and SBDD (r = 0.6), states that did not assess SBDD would most likely report similar SBDD as states that did include this measure, suggesting similar study results had all states assessed SBDD. Second, BRFSS excludes people who do not have telephones, live in institutions, and persons younger than 18 years. Third, BRFSS may under represent the severely impaired because functional capacity is required to participate in BRFSS. Including this group however, would probably only strengthen the associations we found because the variability of SBDD would increase because the severely impaired would be more likely to report more SBDD. Finally, because our findings on SBDD are based on respondents' self-reports rather than on professionally administered psychiatric evaluations, people who experience SBDD may differ from people with clinical depression. Conclusion The 1999 Surgeon General's report states that mental health and mental illness "are not polar opposites but may be thought of as points on a continuum" [ 1 ]. Although most people who report feeling sad, blue or depressed several days each month probably do not have a diagnosable mental disorder, those above a certain threshold of SBDD might be at increased risk for mental illness and physical illness. Additional studies that examine this hypothesis are warranted. Findings from this study, moreover, highlight the relationship between feeling sad, blue or depressed and engaging in risky behaviors, thereby suggesting the need for appropriately designed interventions specifically targeted to a person's individual and social context [ 18 ]. Use of this measure of "sad, blue or depressed days" along with other valid mental health measures can help to assess the burden of population mental distress, identify subgroups with unmet mental health needs, inform the development of targeted interventions, and monitor changes in population levels of mental distress over time [ 12 ]. Future research might examine in more detail the associations among SBDD, anxiety, vitality, and sleeplessness and their ability to assess mood, anxiety, and sleep disorders. It would also be useful to examine the prevalence and demographic characteristics of those who report 14 or more SBDD, and the criterion validity of this measure with other screening instruments and clinical assessments. While SBDD does not provide a strict measure of diagnosable depression as would validated screening and diagnostic assessments, SBDD and other measures such as activity limitations, alcohol or substance abuse, physical inactivity, and employment status can be useful community indicators for addressing the prevention and treatment of depressive symptoms and associated comorbidity [ 12 ]. Abbreviations SBDD sad, blue or depressed days HRQOL health-related quality of life BRFSS Behavioral Risk Factor Surveillance System CDC Centers for Disease Control and Prevention

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514530.xml

543454

Expressional patterns of chaperones in ten human tumor cell lines

Background Chaperones (CH) play an important role in tumor biology but no systematic work on expressional patterns has been reported so far. The aim of the study was therefore to present an analytical method for the concomitant determination of several CH in human tumor cell lines, to generate expressional patterns in the individual cell lines and to search for tumor and non-tumor cell line specific CH expression. Human tumor cell lines of neuroblastoma, colorectal and adenocarcinoma of the ovary, osteosarcoma, rhabdomyosarcoma, malignant melanoma, lung, cervical and breast cancer, promyelocytic leukaemia were homogenised, proteins were separated on two-dimensional gel electrophoresis with in-gel digestion of proteins and MALDI-TOF/TOF analysis was carried out for the identification of CH. Results A series of CH was identified including the main CH groups as HSP90/HATPas_C, HSP70, Cpn60_TCP1, DnaJ, Thioredoxin, TPR, Pro_isomerase, HSP20, ERP29_C, KE2, Prefoldin, DUF704, BAG, GrpE and DcpS. Conclusions The ten individual tumor cell lines showed different expression patterns, which are important for the design of CH studies in tumor cell lines. The results can serve as a reference map and form the basis of a concomitant determination of CH by a protein chemical rather than an immunochemical method, independent of antibody availability or specificity.

Background Chaperones (CH) and heat shock proteins (HSPs) play important roles in tumor biology and still are holding centre stage. The heat shock response was discovered in 1962 by Ritossa [ 1 ], who reported that elevated temperature led to the appearance of a new 'puffing' pattern in the salivary gland polytene chromosomes of Drosophila busckii . Since then, efforts from a large number of investigators have shown that the heat shock response is ubiquitous and highly conserved. It is observed in all organisms from bacteria to plants and animals. CH form an essential defense mechanism for protection of cells from a variety of harmful conditions, including temperature elevation or heat shock, decrease in pH, hypersalinity, alcohols, heavy metals, oxidative stress, inhibitors of energy metabolism, fever or inflammation [ 2 , 3 ]. This broad spectrum of functions gave rise to the term 'molecular chaperone' an entity that acts to assist other proteins' folding and maturation in the cell. However, not all HSPs are CH and not all CH are HSPs [ 4 ]. Genetic studies showed that most HSPs are essential to life. They are believed to play an indispensable role in the conformational maturation of a nascent polypeptide chain in prokaryotic and eukaryotic cells. Traditionally, HSPs are grouped into five major families according to molecular weights. They were designated HSP90 (heat shock protein of apparent molecular weight 90 kDa), HSP70 (70-kDa HSPs), HSP60 (60-kDa HSPs), HSP40 or DnaJ (40-kDa HSPs), and the small heat shock proteins (sHSPs) [ 5 - 7 ]. The connection of HSPs with tumor immunity was discovered in the 1980s [ 8 - 10 ]. It was found that structurally unaltered HSPs which are purified from tumor cells could immunize animals to generate tumor-specific immunity whereas corresponding preparations from normal tissues did not. Many recent interesting observations have been made with regards to CH's ability to regulate tumor biology. HSP70 and other CH are known to be determinants of cell death and cell transformation processes. Elevated expression of HSP70 and HSP90 in tumor cells was detected in several cases [ 11 , 12 ]. Recently, it has been recognised that HSPs regulate apoptosis. HSP27 and HSP70 are antiapoptotic, while HSP60 and HSP10 are proapoptotic. The ability of HSPs to protect cells from stressful stimuli suggests that these proteins play a role in tumorigenicity, with the fact that cells or tissues from various tumors have been shown to express unusually high levels of one or more HSPs. Experimental models support the role of HSPs in tumorigenesis since HSP27 and HSP70 have been shown to increase the tumorigenic potential of rodent cells in syngeneic hosts [ 13 , 14 ]. The contribution of HSPs to tumorigenesis may be attributed to their pleiotropic activities as molecular chaperones that provide the cancer cell with an opportunity to alter protein activities, in particular components of the cell cycle machinery, kinases and other proteins implicated in tumor progression. HSP70 chaperone activity may also influence tumorigenesis by regulating the activity of proteins that are involved in the cell cycle machinery [ 15 ]. Clinically, in a number of cancers such as leukaemia, breast cancer and endometrial cancer, an increased level of HSP27, relative to its level in non-transformed cells has been detected [ 16 ]. In addition, increased expression of HSP70 has also been reported in high-grade malignant tumors such as breast and endometrial cancer, osteosarcoma and renal cell tumors [ 17 - 19 ]. HSP70 levels correlate with malignancy in osteosarcoma and renal cell tumors; its expression is paradoxically associated with improved prognosis [ 18 , 20 ]. HSP90 and HSP60 are also over-expressed in breast tumors, lung cancer, leukaemias and Hodgkin's disease [ 19 , 21 - 23 ]. The molecular basis for over-expression of HSPs in tumors is not completely understood and may have different etiologies. For example, overexpression may be due to the suboptimal cellular environment in poorly vascularised hypoxic tumors or to growth conditions within the solid tumor [ 24 ]. Many studies have focused on the critical role of chaperones in protein folding, their relevance in protein conformational diseases and tumorigenesis. There is, however, no systematic information available on their expressional pattern in individual tumor cell lines in a wide range of tumors. This study addresses the question of differential chaperone expression studied in ten different tumor and three normal cell lines using 2-DE and matrix-assisted laser desorption/ionization-mass spectrometry (MALDI-TOF/TOF) allowing concomitant determination of many CH at the protein chemical level rather than by immunochemical methods, independent of antibody availability and specificity. Results Chaperone proteins were taken from the list of all identified proteins from ten tumor and three normal cell lines using a predetermined list of expected CH based upon our own experiments, databases and literature. Identified CH proteins made up approx. 12% of all identified proteins in all cell lines studied. All major housekeeping proteins (cytoskeleton and metabolic) expected to occur in any cell lines were present in all cell lineages studied (data not shown). A series of chaperone proteins with different expression patterns in ten different tumor and three normal cell lines using 2-DE and MALDI-MS were identified and listed (see Additional file 1 and 2 ) in Table 1 and 1-1. Chaperone proteins were classified according to their domains. Most proteins have similar p I values and molecular weights with theoretical value. The observed p I was represented in Table 2 (see Additional file 3 ) with the total score and the number of peptides matched. The expressional patterns of each tumor and normal cell line are shown in Figure 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 . Figure 1 Saos-2 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of Saos-2 cell line: Swiss prot accession numbers are used to identify proteins. Figure 2 SK-N-SH cell Colloidal Coomassie Blue stained 2D gels representing protein maps of SK-N-SH cell line: Swiss prot accession numbers are used to identify proteins. Figure 3 HCT116 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of HCT 116 cell line: Swiss prot accession numbers are used to identify proteins Figure 4 CaOv-3 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of CaOv-3 cell line: Swiss prot accession numbers are used to identify proteins Figure 5 A549 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of A549 cell line: Swiss prot accession numbers are used to identify proteins Figure 6 HL-6 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of HL-6 cell line: Swiss prot accession numbers are used to identify proteins Figure 7 A-375 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of A-375 cell line: Swiss prot accession numbers are used to identify proteins Figure 8 A-673 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of A-673 cell line: Swiss prot accession numbers are used to identify proteins Figure 9 MCF-7 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of MCF-7 cell line: Swiss prot accession numbers are used to identify proteins Figure 10 Hela cell Colloidal Coomassie Blue stained 2D gels representing protein maps of Hela cell line: Swiss prot accession numbers are used to identify proteins Figure 11 HK-2 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of Kidney HK-2 cell line: Swiss prot accession numbers are used to identify proteins Figure 12 Lymphocyte 3610 cell Colloidal Coomassie Blue stained 2D gels representing protein maps of Lymphocyte 3610 cell line: Swiss prot accession numbers are used to identify proteins Figure 13 Hs 545 SK cell Colloidal Coomassie Blue stained 2D gels representing protein maps of Fibroblast Hs 545 SK cell line: Swiss prot accession numbers are used to identify proteins Proteins with HATPase_C and HSP90 domains Prominent members of the HSP90 family of proteins are heat shock protein 90-alpha (HSP90α), heat shock protein-beta (HSP90β) and endoplasmin (GRP94) [ 25 ]. The two HSP90 isoforms are essential for the viability of eukaryotic cells. They are rather abundant constitutively, making up 1–2% of cytosolic proteins, and can be further stimulated in their expression level by stress. HSP90α and HSP90β were expressed in HCT116 and Hela cells. In addition, HSP90α was also detected in A-673, MCF-7 and A549 cell lines and HSP90β was in Saos-2, SK-N-SH, HL-60 and A375 cell lines, respectively. The isoform of HSP90β (Acc.No.Q9NTK6) was also detected in A-673 cells and is represented in Figure 8 . All cell lines except adenocarcinoma cells (CaOv-3) expressed endoplasmin (GRP94) protein that functions in the endoplasmic reticulum [ 26 ] and the protein similar to tumor rejection antigen is observed in the osteosarcoma (Saos-2) cell line represented in Figure 1 . Heat shock protein 75 kDa is named tumor necrosis factor type 1 receptor associated protein or TRAP1. The 2.4-kilobase TRAP-1 mRNA was variably expressed in skeletal muscle, liver, heart, brain, kidney, pancreas, lung, and placenta. TRAP-1 mRNA was also detected in each of eight different transformed cell lines [ 27 ]. At the protein level, it was expressed in ovary (CaOv-3), lung (A549), skin (A-375), rhabdomyosarcoma (A-673), breast cancer (MCF-7) and cervix carcinoma (Hela) cells. Proteins with HSP 70 domain The HSP70 family constitutes the most conserved and best studied class of HSPs. Human cells contain several HSP70 family members including stress inducible HSP70, constitutively expressed HSC70, mitochondrial HSP75, and GRP78 localised in the endoplasmic reticulum [ 28 ]. HSP70 has been shown to increase the tumorigenicity of cancer cells in rodent models [ 13 ]. All ten tumor cells expressed HSC70 proteins and the isoform of HSC70 was also observed in promyeloblasts (HL-60). The 105-kDa protein shows high similarity with HSP90 on peptide mapping with trypsin digestion. Except for the molecular mass, the physicochemical properties of the 105-kDa protein are similar to those of HSP90, although it has a HSP70 domain. This is detected in the brain, but not in the liver, lung, spleen, kidney, ovary, or uterus, in contrast to the wide distribution of HSP90 [ 29 ]. In case of tumor cells, it was observed in adenocarcinoma (CaOv-3), rhabdomyosarcoma (A-673), breast cancer (MCF-7) and cervix carcinoma (Hela) cells and is shown in Figure 4 , 8 , 9 and 10 . HSP70 kDa protein 1(HSP70-1) stabilises preexistent proteins against aggregation and mediate the folding of newly translated polypeptides in the cytosol as well as within organelles. HSP70-1 in mitochondria and endoplasmic reticulum plays an additional role by providing a driving force for protein translocation. They are involved in signal transduction pathways in cooperation with HSP90 and participate in all these processes through their ability to recognise non-native conformations of other proteins. Eight cell lines except SK-N-SH and CaOv-3 cells expressed HSP70-1. Heat shock 70 kDa protein 4 was observed in SK-N-SH, HL-60 and A-673 cells and is represented in Figure 2 , 6 and 8 . Stress-70 protein (GRP75) was proposed to be implicated in the control of cell proliferation and cellular aging [ 30 ] and was observed in eight cell lines except MCF-7 and Hela cells. Heat shock-related 70 kDa protein 2 and 150 kDa oxygen-regulated protein were detected in SK-N-SH and MCF-7 cells. The best characterised GRP is a 78 kDa protein known as GRP78, which is identical to BiP, the immunoglobulin heavy chain binding protein [ 31 ]. Since GRP78 shares similar function and 60% amino acid identity with HSP70 [ 32 ], it is also categorised within the HSP70 multi-gene family [ 33 ]. GRP78 / BiP protein were expressed in all ten tumor cells. Proteins with Cpn60_TCP1 domain Proteins with a Cpn60_TCP1 domain are involved in chaperonins, which belong to the 55–64 kDa family of HSP or stress proteins [ 34 ]. Mammalian HSP60, also called chaperonin, is mostly contained within the mitochondrial matrix, although it has also been detected in extramitochondrial sites. HSP60 participates in the folding of mitochondrial proteins and facilitates proteolytic degradation of misfolded or denatured proteins in an ATP-dependent manner. The chaperone function of HSP60 is regulated by HSP10, which binds to HSP60 and regulates substrate binding and ATPase activity [ 7 ]. In this study, all cells except MCF-7 expressed 60-kDa heat shock protein. The chaperonin-containing T-complex polypeptide has many subunits. Among these, alpha, beta, gamma, epsilon and zeta units as well as the isoform of epsilon subunit were shown in 2-DE gels of tumor cell lines. T-complex protein 1, zeta subunit was detected in all ten tumor cell lines. Proteins with DnaJ and DnaJ_C domains Among numerous co-chaperones for HSC70 [ 35 ], the DnaJ family is an essential group.. The human DnaJ (HSP40) family [ 36 ] is a noncanonical member of DnaJ, which lacks the zinc-finger domain. DnaJ homolog subfamily A member 1 and 2 were detected in A549 and SK-N-SH cell lines and DnaJ homolog subfamily B member 11 was found in SK-N-SH, A549 and Hela cells (Figure 2 , 5 and 10 ). Proteins with a thioredoxin domain Protein disulfide isomerase, protein disulfide A3 and A6 containing 2 thioredoxin domains are members of the protein disulfide isomerase family and rearrange both intra-chain and inter-chain disulfide bonds in proteins to form the native structures. Protein disulfide isomerase was expressed in CaOv-3, A549 and HL-60 cells. Protein disulfide isomerase A3 was detected in nine cells except Hela cell and protein disulfide isomerase A6 was observed in HL-60, A-375, A-673 and Hela cell lines. Proteins with a TPR domain Stress-induced-phosphoprotein 1 showing 9 TPR domains was expressed in all cell lines except Saos-2 cell. Hsc70-interacting protein with 3 TPR domains was detected in HL60 and A-375 cell. FK506-binding protein 4 has 3 TPR and 2FKBP_C domains. This protein was observed in SK-N-SH, A549, HL-60, A-375 and MCF-7 cells. Proteins with other domains Heat shock 27 kDa protein (HSP27) is involved in the stabilisation of cytoskeletal proteins and in the protective mechanisms against oxidative stress by abolishing the burst of intracellular reactive oxygen species (ROS) [ 37 ]. HSP27 is the only protein detected among small heat-shock proteins in SK-N-SH, HCT, A549, HL-60, MCF-7 and Hela cell. Peptidyl-prolyl-cis-trans isomerase A contains a Pro_isomerase domain and accelerates folding of proteins. It was observed in SK-N-SH, A549, HL-60, A-375 and A-673 cell lines. ERP29_C has been recently characterised as a novel 29 kDa endoplasmic reticulum protein that is widely expressed in rat tissues. It plays an important role in the processing of secretory proteins within the ER [ 38 ]. SK-N-SH, A549, A-375, MCF-7 and Hela cells expressed ERP29 protein. Prefoldin subunit 2 has a KE2 domain but profoldin subunit 3 has a prefoldin domain. Two of them were expressed in Hela cells and profoldin subunit 2 was also detected in SK-H-SH cell. The protein DUF704 is an activator of 90 kDa heat shock protein ATPase homolog 1. This protein was observed in bone marrow neuroblastoma (SK-N-SH) and Hela cells. BAG-1 binds the ATPase domains of HSP70 and Hsc70, modulating their chaperone activity and functioning as a competitive antagonist of the co-chaperone Hip. The human BAG-1, BAG-2, and BAG-3 proteins bind with high affinity to the ATPase domain of Hsc70 and inhibit its chaperone activity. All these proteins contain a conserved 45-amino acid region near their C termini (the BAG domain) that binds Hsc70/HSP70, but they differ widely in their N-terminus [ 39 ]. This protein was observed in A549 and Hela cells. GrpE protein homolog1 with GrpE domain and HSPC015 with DcpS domain were observed in A-673 and HL-60 cell exclusively. Major differences between the ten tumor and the three normal cell lines were observed: 21 CH were observed in tumor cell lines only and vice versa, two CH were expressed in three normal cell lines exclusively (see Additional files 1 and 2 ). Discussion The main outcome of the study is the generation of tumor cell specific patterns of chaperone and heat shock protein expression. The results form the basis for designing chaperone protein expression studies needed to evaluate the role of these structures in tumor biology by providing an analytical tool for the concomitant determination of CH, unambiguously identifying CH by a protein chemical rather than an immunochemical technique, independent of antibody availability and specificity. It must be mentioned that only high abundance CH have been detected by this method and that the generated individual maps show relative abundance proteins only. A series of relatively abundant CH were presented by more than one spot indicating the presence of splicing variants or posttranslational modifications including glycosylation, phosphorylation, methylation, oxidation, truncation, to name a few and the molecular diversity is currently subject of detailed studies using advanced proteomic tools as MS-MS sequencing and Q-TOF instrumentation. Carbamylation during sample preparation may have been contributing to electrophoretic shifts as well Methodologically, identification of proteins by MS-MS is a sound approach and the major (inherent) problem with the interpretation may be at the cell culture level: the standard protocols for cultivating the individual cell lines according to the supplier were followed, but the conditions for the individual cell cultures vary. Different antibiotics or fungizides used may well lead to differences in CH expression and indeed, the use of geldanamycin has been already reported to induce heat shock expression in brain tissue [ 40 ]. Other tentative confounding factors as e.g. cell cycle differences, differences in growth and proliferation have to be taken into account, but it was the aim of the study to examine CH expression in cell lines in the absence of stressors or metabolic derangement, warranted by the use of standard protocols, wherefrom toxic effects are expected. Moreover, a list of non-tumor cell lines has been studied for CH expression already using this principle as well as comparable analytical technique and this may allow some comparison between tumor and non-tumor cell line CH profiles. Herein, we tested three more normal cell lines and observed a large number of differentially expressed CH between tumor and normal cell lineages. We are aware of the fact that the already reported CH expression patterns of five cell lines [ 41 ] and the three normal cell lines are not sufficient to find out differences between normal and tumor cell CH expression in general. We used herein cell types widely used in life sciences as e.g. fibroblasts, lymphocytes and kidney cells, that are representing well-characterised normal cell lines. It may be impossible to show specific differences between normal and tumor cell lines as the stem cells from which tumor cell lines originate are not generally known. Conclusions Basically, differences between individual CH expression patterns may be due to different functional roles in individual cells and the presence of specific proteins in the individual cell lineages: there are CH to specifically protect individual proteins as e.g. specific chaperoning of tubulin beta 1 by the TCP complex [ 42 ] and of procollagen by HSP47 [ 43 ]. Differential CH expression may reflect or lead to tumor biological characteristics including dignity and carcinogenesis, and the method given herein provides the possibility and option to test these characteristics in a high-throughput performance. The specific nature of cell line patterns is given by the observation that only heat shock cognate 71 kDa protein and TCP zeta subunit protein were expressed non-specifically in all ten cell lines studied and previous protein expression profiles of CH in tumor cells or tissues are hereby extended and confirmed. Absence of a CH in an individual cell line could be explained by poor resolution in an individual gel but gel quality was checked by comparing general patterns and therefore this explanation is rather unlikely The main focuses in CH protein research maybe now to investigate detection of more CHs, probably by prefractionation into different compartments [ 44 ], to characterise the splice variants expressed at the protein level and to evaluate post-translational modifications that may be responsible for a significant part of multiple expression forms. Methods Cell culture Ten different tumor cell lines were purchased from American Type Culture Collection (ATCC). The cell lines and their ATCC numbers are given in Table 3 (see Additional file 4 ). SK-N-SH (bone marrow neuroblastoma) and Hela (cervix carcinoma) cell lines were grown in Minimum Essential Medium (Eagle) with 2 mM L-glutamine and Earle's Basic Salt Solution (BSS) adjusted to contain 1.5 g/L sodium bicarbonate, 0.1 mM non-essential amino acids, and 1 mM sodium pyruvate, with 10% fetal bovine serum (FBS). The same conditions were used to culture the MCF-7 cell line except for supplementing 10% FBS with 0.01 mg/ml bovine insulin. HCT 116 (colorectal carcinoma) and Saos-2 (osteosarcoma) cell lines were cultured in McCoy's 5a medium with 90% 1.5 mM L-glutamine and 10% FBS. Ham's F12K medium with 2 mM L-glutamine adjusted to contain 1.5 g/L sodium bicarbonate and 10% FBS were used for the A549 (lung carcinoma) cell culture. HL-60 (acute promyelocytic leukaemia) cells were cultured with Iscove's modified Dulbecco's medium with 4 mM L-glutamine adjusted to contain 80% 1.5 g/L sodium bicarbonate and 20% FBS. A-673 (rhabdomyosarcoma), Caov-3 (ovarial adenocarcinoma) and A-375 (malignant melanoma) cell lines were cultured in DMEM with 4 mM L-glutamine adjusted to contain 1.5 g/L sodium bicarbonate and 4.5 g/L glucose with 10% FBS. Three normal cell lines, fibroblasts, lymphocytes and kidney cells were used: Fibroblasts, Hs 545 SK ATCC CRL-7318 (Manassas, VA), were obtained from human skin and grown in monolayers in Dulbecco's modified Eagle's medium (DMEM) containing 4 mM L-glutamine adjusted to contain 1.5 g/L sodium bicarbonate, 4.5 g/L glucose, with 10% fetal bovine serum (FBS), Penicillin and Streptomycin (GIBCO BRL) according to standard techniques Lymphocyte cell line 3610 is a spontaneously EBV transformed cell line from a patient with osteosarcoma and was obtained from the St. Anna Kinderspital-Forschungsinstitut (Vienna, Austria). The cell line was established from peripheral heparinized blood by a density gradient centrifugation using Ficoll-Paque (AMERSHAM BIOSCIENCE, Uppsala, Sweden) and grown in RPMI 1640 with 10% FBS, 70 μM gentamicin sulfate and 2 mM glutamine at a density of 2 × 10 6 cells per ml in 96 well plates. HK-2 cells (Human Kidney 2) are proximal tubular epithelial cells derived from normal kidney and were purchased from ATCC CRL-2190. Cells were cultured in Keratinocyte-Serum Free Medium (GIBCO-BRL 17005–042) with 5 ng/ml recombinant epidermal growth factor (positive for alkaline phosphatase, gamma glutamyltranspeptidase, leucine aminopeptidase, acid phosphatase, cytokeratin, alpha 3 beta 1 integrin, fibronectin; negative for factor VIII-related antigen, 6.19 antigen and CALLA endopeptidase) and 0.05 mg/ml bovine pituitary extract. All cell cultures were maintained in a humified atmosphere of 5% (v/v) CO 2 in air at 37°C and logarithmically growing cells were harvested by trypsinisation. Sample preparation Harvested cells were washed three times with 10 mL PBS (phosphate buffered saline) (Gibco BRL, Gaithersburg, MD, USA) and centrifuged for 10 min at 800 g at room temperature. The supernatant was discarded and the pellet was suspended in 1 ml of sample buffer consisting of 40 mM Tris, 7 M urea (Merck, Darmstadt, Germany), 2 M thiourea (Sigma, St. Louis, MO, USA), 4% CHAPS (3-[(3-cholamidopropyl) dimethylammonio]-1-propane-sulfonate) (Sigma, St. Louis, MO, USA), 65 mM 1,4-dithioerythritol (Merck, Germany), 1 mM EDTA (ethylenediaminetetraacetic acid) (Merck, Germany), protease inhibitors complete (Roche, Basel, Switzerland) and 1 mM phenylmethylsulfonyl fluoride (PMSF). The suspension was sonicated for approximately 30 sec in an ice bath. After homogenisation samples were left at room temperature for 1 h and centrifuged at 14,000 rpm for 1 h. The supernatant was transferred into Ultrafree-4 centrifugal filter units (Millipore, Bedford, MA), for desalting and concentrating proteins. Protein content of the supernatant was quantified by the Bradford protein assay system [ 45 ]. The standard curve was generated using bovine serum albumin and absorbance was measured at 595 nm. Two-dimensional gel electrophoresis (2-DE) Samples prepared from each cell line were subjected to 2-DE as described elsewhere [ 46 ]. 1 mg protein was applied on immobilised pH 3–10 nonlinear gradient strips in sample cups at their basic and acidic ends. Focusing was started at 200 V and the voltage was gradually increased to 5000 V at a rate of 3 V/min and then kept constant for a further 24 h (approximately 180,000 Vhs totally). After the first dimension strips (18 cm) were equilibrated for 15 min in a buffer containing 6 M urea, 20% glycerol, 2% SDS, 2% DTT and then for 15 min in the same buffer containing 2.5% iodoacetamide instead of DTT. After equilibration, strips were loaded on 9–16% gradient sodium dodecylsulfate polyacrylamide gels for second-dimensional separation. Gels (180 × 200 × 1.5 mm) were run at 40 mA per gel. Immediately after the second dimension run, gels were fixed for 18 h in 50% methanol, containing 10% acetic acid, the gels were stained with Colloidal Coomassie Blue (Novex, San Diego, CA) for 12 h on a rocking shaker. Molecular masses were determined by running standard protein markers (Biorad Laboratories, Hercules, CA) covering the range 10–250 kDa. p I values were used as given by the supplier of the immobilised pH gradient strips (Amersham Bioscience, Uppsala, Sweden). Excess of dye was washed out from the gels with distilled water and gels were scanned with an ImageScanner (Amersham Bioscience). Electronic images of the gels were recorded using Adobe Photoshop and Microsoft Power Point softwares. Matrix-assisted laser desorption ionisation mass spectrometry (MALDI-MS) Spots visualised by Colloidal Coomassie Blue staining were excised with a spot picker (PROTEINEER sp™, Bruker Daltonics, Germany), placed into 96-well microtiter plates and in-gel digestion and sample preparation for MALDI analysis were performed by an automated procedure (PROTEINEER dp™, Bruker Daltonics) [ 44 , 47 ]. Briefly, all visible spots were excised and washed with 10 mM ammonium bicarbonate and 50% acetonitrile in 10 mM ammonium bicarbonate. After washing, gel plugs were shrunk by addition of acetonitrile and dried by blowing out the liquid through the pierced well bottom. The dried gel pieces were reswollen with 40 ng/μl trypsin (Roche Diagnostics, Penzberg, Germany) in enzyme buffer (consisting of 5 mM Octyl β-D-glucopyranoside (OGP) and 10 mM ammonium bicarbonate) and incubated for 4 h at 30°C. Peptide extraction was performed with 10 μl of 1% TFA in 5 mM OGP. Extracted peptides were directly applied onto a target (AnchorChip™, Bruker Daltonics) that was load with α-cyano-4-hydroxy-cinnamic acid (CHCA) (Bruker Daltonics) matrix thinlayer. The mass spectrometer used in this work was an Ultraflex™ TOF/TOF (Bruker Daltonics) operated in the reflector for MALDI-TOF peptide mass fingerprint (PMF) or LIFT mode for MALDI-TOF/TOF with a fully automated mode using the FlexControl™ software. An accelerating voltage of 25 kV was used for PMF. Calibration of the instrument was performed externally with [M+H] + ions of angiotensin I, angiotensin II, substance P, bombesin, and adrenocorticotropic hormones (clip 1–17 and clip 18–39). Each spectrum was produced by accumulating data from 200 consecutive laser shots. Those samples which were analysed by PMF from MALDI-TOF were additionally analysed using LIFT-TOF/TOF MS/MS from the same target. A maximum of three precursor ions per sample were chosen for MS/MS analysis. In the TOF1 stage, all ions were accelerated to 8 kV under conditions promoting metastable fragmentation. After selection of jointly migrating parent and fragment ions in a timed ion gate, ions were lifted by 19 kV to high potential energy in the LIFT cell. After further acceleration of the fragment ions in the second ion source, their masses could be simultaneously analysed in the reflector with high sensitivity. PMF and LIFT spectra were interpreted with the Mascot software (Matrix Science Ltd, London, UK). Database searches, through Mascot, using combined PMF and MS/MS datasets were performed via BioTools 2.2 software (Bruker). A mass tolerance of 100 ppm and 1 missing cleavage site for PMF and MS/MS tolerance of 0.5 Da and 1 missing cleavage site for MS/MS search were allowed and oxidation of methionine residues was considered. The probability score calculated by the software was used as criterion for correct identification. The algorithm used for determining the probability of a false positive match with a given mass spectrum is described elsewhere [ 48 ]. List of abbreviations CH chaperone HSP heat shock protein 2-DE two dimensional gel electrophoresis MALDI-MS matrix-assisted laser desorption ionisation mass spectrometry OGP octyl-β-D-glucopyranoside CHCA α-cyano-4-hydroxy-cinnamic acid PMF peptide mass fingerprint GRP glucose regulated protein TCP t-complex protein Competing interests The author(s) declare that they have no competing interests. Authors contributions JKM did data mining and contributed to the preparation of the manuscript. LAS performed protein extraction, two-dimensional electrophoresis and data handling. MFC carried out MALDI-TOF-TOF analyses. IS developed methodology for studying proteins from cell lines. GL initiated and planned the study developing the concept, supervised 2-DE, mass spectrometry, creating data and the manuscript. All authors have read and approved the final manuscript. Supplementary Material Additional File 1 Table 1. Identified proteins in different human tumor cell lines: Saos-2, SK-N-SH, HCT 116, CaOv-3, A549, HL-60, A-375, A-673, MCF-7 and Hela. Click here for file Additional File 2 Table 1-1. Identified proteins in different normal cell lines: Kidney, Lymphocyte, Fibroblast Click here for file Additional File 3 Table 2. Theoretical molecular weight, theoretical p I , observed p I , total score and peptide matched of molecular chaperones in tumor cell lines Click here for file Additional File 4 Table 3. The list of tumor cell lines Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC543454.xml

534804

Early and long-term outcome of elective stenting of the infarct-related artery in patients with viability in the infarct-area: Rationale and design of the Viability-guided Angioplasty after acute Myocardial Infarction-trial (The VIAMI-trial)

Background Although percutaneous coronary intervention (PCI) is becoming the standard therapy in ST-segment elevation myocardial infarction (STEMI), to date most patients, even in developed countries, are reperfused with intravenous thrombolysis or do not receive a reperfusion therapy at all. In the post-lysis period these patients are at high risk for recurrent ischemic events. Early identification of these patients is mandatory as this subgroup could possibly benefit from an angioplasty of the infarct-related artery. Since viability seems to be related to ischemic adverse events, we initiated a clinical trial to investigate the benefits of PCI with stenting of the infarct-related artery in patients with viability detected early after acute myocardial infarction. Methods The VIAMI-study is designed as a prospective, multicenter, randomized, controlled clinical trial. Patients who are hospitalized with an acute myocardial infarction and who did not have primary or rescue PCI, undergo viability testing by low-dose dobutamine echocardiography (LDDE) within 3 days of admission. Consequently, patients with demonstrated viability are randomized to an invasive or conservative strategy. In the invasive strategy patients undergo coronary angiography with the intention to perform PCI with stenting of the infarct-related coronary artery and concomitant use of abciximab. In the conservative group an ischemia-guided approach is adopted (standard optimal care). The primary end point is the composite of death from any cause, reinfarction and unstable angina during a follow-up period of three years. Conclusion The primary objective of the VIAMI-trial is to demonstrate that angioplasty of the infarct-related coronary artery with stenting and concomitant use of abciximab results in a clinically important risk reduction of future cardiac events in patients with viability in the infarct-area, detected early after myocardial infarction.

Background Management of acute myocardial infarction (AMI) has underwent considerable changes in the last two decades and the management of patients with AMI has become more established [ 1 , 2 ]. In patients with ST-segment elevation myocardial infarction (STEMI), primary angioplasty is becoming first-choice therapy [ 2 - 4 ]. However, because of the low availability of such treatment, even in developed countries, most patients with STEMI are reperfused with intravenous thrombolysis. Many patients do not receive a reperfusion therapy at all [ 5 , 6 ]. Data from large clinical trials indicate that after successful thrombolysis more than 50% of patients have a significant residual stenosis and about 30% of patients have spontaneous or inducible ischemia [ 7 - 9 ]. In this group reocclusion of the infarct-related artery is a potential threat, as it is associated with recurrent ischemia or recurrent infarction [ 10 , 11 ]. Therefore, early risk assessment is of great importance, especially in patients treated with thrombolysis and patients who did not receive reperfusion therapy. This risk assessment should be followed by an effective treatment strategy in order to prevent recurrent cardiac events and deterioration of left ventricular function [ 12 , 13 ]. As these recurrent events are mainly due to the presence of an unstable residual stenosis of the infarct-related coronary artery, an effective therapy should include an invasive procedure like angioplasty to optimize coronary flow to the infarct area in high risk patients [ 11 ]. Trials have been performed to study the effect of routine angioplasty on clinical outcome early after AMI treated with thrombolysis [ 14 - 20 ]. Most of these studies failed to show improvement of clinical outcome with a standard invasive approach. Possible explanations for these results are the unselected approach, and the high risk profile of balloon angioplasty in the early days, when abciximab and stents were not available. Because angioplasty always carries an inherent risk, it remains important to select those patients after a recent myocardial infarction, who will actually benefit from angioplasty of the infarct-related artery. To date, non-invasive risk stratification after AMI has mainly focused on exercise stress testing. The inability to exercise, the low diagnostic accuracy and resting ECG abnormalities, however, remain important limitations in the detection of ischemia [ 21 ]. Several post-infarction observational studies investigated viability as prognosticator and showed that the presence of viability in the infarct area was highly predictive for future coronary events like recurrent ischemia, recurrent infarction, left ventricular failure, and death [ 22 - 31 ]. Viable tissue is potentially jeopardized by an unstable residual stenosis in the infarct-related coronary artery. In a meta-analysis of non-randomized data by Iskandrian, the impact of revascularization on clinical outcome in patients with viability after AMI was studied [ 32 ]. A significant decrease in future cardiac events was observed in the patients with viability who were revascularized. In contrast, the outcome in patients without viability in the infarct area did not change by an invasive strategy. In this context it should be noted that other post-infarction studies have shown viability to be associated with an improved prognosis [ 33 - 36 ]. Two studies demonstrated that especially patients with severe LV dysfunction and viability show a better survival than patients with LV dysfunction but without viability [ 34 , 36 ]. The reason for this paradox is not quite understood. However, it can be argued that patients with viability have a potential of recovery of LV function (spontaneous or by revascularization), thereby improving their survival [ 37 ]. Thus, on one hand viability may improve survival by recovery of function especially in patients with moderate to severe LV dysfunction, on the other hand viability is related to a worse prognosis by increased risk of recurrent ischemic events. Based on the aforementioned assumptions, we initiated a clinical trial to investigate the benefits of percutaneous coronary intervention (PCI) of the infarct-related artery in patients with viability detected in the early, subacute phase of myocardial infarction. To demonstrate viability, we use low-dose dobutamine echocardiography (LDDE). This test can safely be performed 48 hours after acute MI [ 23 , 27 , 30 , 31 ]. It is a well validated bedside test with a diagnostic accuracy of about 80%, which is comparable to scintigraphical techniques (SPECT/PET) [ 38 ]. Coronary stenting will be performed with concomitant use of abciximab. After stenting, oral clopidogrel is given in a standard way. With this approach the lowest possible periprocedural event rate will be attained, with a low rate of target vessel revascularization (TVR) [ 39 - 43 ]. We hypothesize that, in order to prevent future cardiac events, PCI is only useful in patients with viability in the infarct zone early after myocardial infarction. Methods Patient Selection Patients admitted to any of the participating centers with an acute or recent myocardial infarction, who are not treated by direct or rescue angioplasty, and who are stable during the first 48 hours after the acute event, are screened for the study. Patients < 80 years of age are considered suitable for the study when they have definite myocardial infarction, as demonstrated by an significant rise in creatine kinase-MB levels (twice the upper limit of normal: ULN), 1 mm ST segment elevation in two or more standard leads or 2 mm ST segment elevation in two contiguous chest leads, and/or the development of Q waves. The criteria for exclusion are: viability testing technically not possible (poor echo-window), contraindications for dobutamine echocardiography (arrhythmia), and coronary angiography (severe diabetic nephropathy or known contrast-allergy), serious life-threatening non-cardiac illness, ECG abnormalities making the evaluation of the ST segment impossible (left bundle branch block, pacemaker), and an unreliable follow-up. The study complies with the Declaration of Helsinki and all ethics committees of the participating centers have approved the protocol. All eligible patients provide written informed consent. Study design The study is a prospective, multicenter, randomized, controlled clinical trial. In the VIAMI-trial, patients who are admitted to the hospital with an acute myocardial infarction and who did not undergo primary or rescue PCI, are evaluated by LDDE within 3 days of admission. Patients with unequivocal signs of viability in the infarct-area are randomized to an invasive or a conservative treatment strategy. In the invasive strategy patients undergo coronary angiography with the intention to perform PCI with stenting of the infarct-related coronary artery. In the conservative group an ischemia-guided approach is adopted with stress testing before discharge from the hospital. When the test is highly suggestive for ischemia, coronary angiography will be performed. If revascularization is performed, this will be scored as a secondary endpoint. Patients without viability will serve as a registry group with long-term follow-up (Fig 1 ). These patients are assigned to the conservative group in order to prevent physicians' bias during the trial. Figure 1 Flow chart The primary endpoint is the composite of death from any cause, recurrent infarction, and unstable angina. The secondary endpoints are need for revascularization, the occurrence of angina pectoris (CCS classification), and the incidence of heart failure (NYHA classification). Left ventricular function is also evaluated as determined by echocardiography at 3 months, 6 months, and 1 year follow-up. A reinfarction is diagnosed if there is an increase in the total creatine kinase and MB isoenzyme activity (2 times ULN) and either a history of ischemic chest discomfort, or electrocardiographic changes indicative for transmural ischemia or necrosis. For the diagnosis of unstable angina, the patient must be hospitalized with ischemic chest pain or discomfort occurring at rest or with minimal exertion. In addition, the need for intravenous medical intervention and/or objective evidence of myocardial ischemia is required. For extensive description of the end points definitions, see Table I . Table I Primary end point definitions Definition of reinfarction 1. Reinfarction during hospitalization for index infarct and not related to revascularization procedures - Either ischemic type of chest discomfort or new electrocardiographic changes indicative for transmural ischemia or necrosis with an increase in the total creatine kinase and MB isoenzyme activity. The activity of CK-MB has to be at least 2 times the upper limit of normal and more than 50% above the previous baseline value. 2. Reinfarction discharge for index infarct and not related to revascularization procedures - Either a history of ischemic chest discomfort, usually lasting > 20 minutes, or classic electrocardiographic changes indicative for transmural ischemia with an increase in the total creatine kinase and MB isoenzyme activity of at least 2 times the upper limit of normal. - New abnormal Q-waves (amplitude ≥ 1/3 of total QRS amplitude and ≥ 0.04 seconds) in ≥ 2 contiguous leads. 3. Periprocedural reinfarction during revascularization after index infarct PCI - Classic electrocardiographic changes indicative for transmural ischemia with an increase in the total creatine kinase and MB isoenzyme activity of at least 2 times the upper limit of normal. - New abnormal Q-waves (amplitude ≥ 1/3 of total QRS amplitude and ≥ 0.04 seconds) in ≥ 2 contiguous leads. CABG - Classic electrocardiographic changes indicative for transmural ischemia with an increase in the total creatine kinase and MB isoenzyme activity of at least 5 times the upper limit of normal. - New abnormal Q-waves (amplitude ≥ 1/3 of total QRS amplitude and ≥ 0.04 seconds) in ≥ 2 contiguous leads. Definition of unstable angina - Ischemic type of chest discomfort at rest or with minimal exertion, with a duration of at least 15 minutes. The presenting symptoms must represent a change from the patients' usual anginal pattern. - Either a need for intravenous medical intervention and/or objective evidence of myocardial ischemia (dynamic ST changes in ≥ 2 contiguous electrocardiographic leads, or an abnormal elevation of Troponin-T or Troponin-I without a significant rise of CK MB isoenzyme activity) PCI, percutaneous coronary intervention; CABG, coronary artery bypass grafting; CK-MB, creatine kinase myocardial enzyme A clinical endpoints and safety monitoring committee will continuously check and monitor all events in a blinded manner. Reinfarction after revascularization will be scored as described in Table I . Major and minor bleeding complication in the invasive treated group are defined according to the criteria of the Thrombolysis in Myocardial Infarction trials [ 44 ]. All patients will be treated with aspirine, beta blockers, angiotensin-converting-enzyme inhibitors and statins as accepted in international guidelines [ 2 , 4 ]. Low-dose dobutamine echocardiography If the inclusion criteria are met and none of the criteria for exclusion, viability testing by LDDE is performed within 3 days of admission. Preferably, beta-blockers are withdrawn 24 hours before the test [ 23 , 27 , 30 , 31 ]. Discontinuation of beta-blockers seems to to enhance LDDE accuracy [ 45 , 46 ]. Before the administration of dobutamine, a baseline echocardiogram is obtained according to the guidelines of the American Society of Echocardiography [ 47 ]. Five standard views are obtained: the parasternal long-axis and short-axis view and the apical two, three- and four-chamber view. A 16-segment model is used in which the apex is divided in 4 segments. Segmental wall motion and thickening is scored according to a 4-point scale: 1 = normal, 2 = hypokinetic, 3 = akinetic, and 4 = dyskinetic. Left ventricular volumes and ejection fraction are measured by use of the modified Simpson's rule algorithm from orthogonal apical long-axis projections as recommended by the American Society of Echocardiography. Dobutamine is administrated intravenously at doses of 5, 10, and 15 μg/kg/min, for 5 minutes at each dose. When a 10% increase in heart rate is not achieved with 15 μg/kg/min, a 5-minute infusion with 20 μg/kg/min can be used as the final stage of the procedure. Viability is defined as the improvement of wall motion abnormalities in two or more segments of the infarct zone. Changes from hypokinesia to normokinesia and from dyskinesia or akinesia to hypo- or normokinesia are considered an improvement in wall motion abnormality. Dyskinesia changing to akinesia is not considered as an improvement. Patients are continuously monitored by a 12 lead ECG and blood pressure is recorded at the end of each stage. All views are recorded at rest and during dobutamine on VHS videotape. All videotapes are sent to the core-lab (VU University medical center) and will be analyzed by 2 experienced observers. A third observer is used in case of disagreement to reach consensus. For subsequent off-line analysis, the echocardiographic images will be digitized from VHS videotape and transferred to a working station (Enconcert ® by Philips). The images will be displayed side-by-side in a quadscreen format to facilitate the comparison of images. Coronary angiography and angioplasty Angiography and angioplasty will be performed as soon as possible after randomization. Coronary angiography will be performed according to standard protocols. To determine the severity of culprit lesions, quantitative coronary arteriography (QCA) will be performed to measure percent diameter stenosis, reference diameter and cross-sectional area stenosis. The degree of stenosis will be determined in the view in which the stenosis is most severe. When feasible, PCI will be performed (with or without etc) when there is a significant (≥ 50%) stenosis or occlusion of the infarct-related coronary artery, with the intention to perform primary stenting of the infarct-related artery, with concomitant use of abciximab, according to the EPILOG protocol [ 39 ] After stenting, all patients recieve oral clopidogrel in a standard way. In case of severe 3-vessel disease or significant left main stem stenosis, where PCI is judged to be a high risk, coronary artery bypass grafting will be considered. Follow-up Patients will be followed for a period of three years. Follow-up data will be obtained of all patients during visits at the outpatient clinic in the first year, and by telephone interview in the second and third year of follow-up. Left ventricular function (volumes and ejection fraction) is determined by echocardiography at 3 months, 6 months, and 1 year of follow-up. Statistical design The VIAMI-trial is conducted to investigate the differences in clinical outcome between an invasive and a conservative strategy in patients with demonstrated viability in the infarct-area. The expected event rate in the viability positive group is estimated to be 35 percent. To demonstrate with a power of 80% (α = 0.05, two-sided) that PCI leads to a 50% reduction in event rate in the invasive group compared to the conservative group, 200 patients will be needed in each group. Therefore, we intend to randomize a total of 400 patients in this trial, with interim analysis after 200 included patients. The formal stopping rules for the study are the following: If one of the treatment strategies appears significant superior at interim analysis (P ≤ 0.01), the study will be stopped. Statistical analysis Baseline descriptive data will be presented as mean ± standard deviations (SD). Differences in clinical and echocardiographic variables will be assessed by unpaired Student's t test. Differences between proportions will be assessed by chi-square analysis; a Fisher's exact test will be used when appropriate. Event-free survival curves are computed with the Kaplan-Meier method, and the differences between these curves are tested with a Mantel-Cox log-rank test. A primary endpoint analysis is planned at 30 days, 6 months, and 1 year of follow-up. Subgroup analyses are planned to determine whether the treatment effect is more or less pronounced in certain subgroups. The data will be subgrouped by sex, age, diabetes, anterior infarction, time from onset of symptoms to treatment, and the use of thrombolytics. All analyses will be performed on an intention-to-treat basis. Also, the outcome per-protocol will be evaluated. Such an analysis seems worthwhile, as it will reflect the true influence of PCI on post-thrombolytic ischemic events. Current status Enrollment of patients started April 1, 2001. Recently, an interim analysis was performed after the inclusion of 200 patients, having a 30 day follow-up. The Clinical Event Committee recommended continuation of the trial. Conclusion The VIAMI-trial is the first multicenter, randomized, controlled clinical trial, investigating the clinical benefits of percutaneous catheter intervention of the infarct-related artery in patients with demonstrated viability in the early, subacute phase of myocardial infarction. Competing interests The authors declare that they have no competing interests. Authors' contribution RBvL recruited and analysed all data and drafted the manuscript. GV have been involved in drafting the article and revised it critically. OK and JGFB revised the manuscript critically. CAV and FCV have given final approval of the version to be publiced.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534804.xml

519000

Developmental Context Determines Latency of MYC-Induced Tumorigenesis

One of the enigmas in tumor biology is that different types of cancers are prevalent in different age groups. One possible explanation is that the ability of a specific oncogene to cause tumorigenesis in a particular cell type depends on epigenetic parameters such as the developmental context. To address this hypothesis, we have used the tetracycline regulatory system to generate transgenic mice in which the expression of a c-MYC human transgene can be conditionally regulated in murine hepatocytes. MYC 's ability to induce tumorigenesis was dependent upon developmental context. In embryonic and neonatal mice, MYC overexpression in the liver induced marked cell proliferation and immediate onset of neoplasia. In contrast, in adult mice MYC overexpression induced cell growth and DNA replication without mitotic cell division, and mice succumbed to neoplasia only after a prolonged latency. In adult hepatocytes, MYC activation failed to induce cell division, which was at least in part mediated through the activation of p53. Surprisingly, apoptosis is not a barrier to MYC inducing tumorigenesis. The ability of oncogenes to induce tumorigenesis may be generally restrained by developmentally specific mechanisms. Adult somatic cells have evolved mechanisms to prevent individual oncogenes from initiating cellular growth, DNA replication, and mitotic cellular division alone, thereby preventing any single genetic event from inducing tumorigenesis.

Introduction The frequency of cancer development varies depending on the age of the host. In humans, the most common childhood cancers include tumors of the hematopoietic system, nervous system, and skeletal muscle system. In contrast, in the adult population, solid tumors of the lung, colon, breast, and prostate are more common. Differences in types of cancers in hosts of different ages may reflect the abundance of cells in a differentiative state susceptible to tumorigenesis ( Greaves 1986 ; Klein and Klein 1986 ). Indeed, many reports document that oncogene activation generally induces tumorigenesis in immature cellular lineages ( Adams et al. 1985 ; Spanopoulou et al. 1989 ; Pelengaris et al. 1999 ; Blyth et al. 2000 ). We recently reported that upon oncogene inactivation tumor cells can differentiate into mature cells, and in this new differentiative context the reactivation of an oncogene fails to restore tumorigenesis ( Jain et al. 2002 ). Based on these results, we speculate that only specific differentiative windows provide the correct epigenetic program to permit oncogene activation to initiate and sustain tumorigenesis. Here we directly evaluate whether an oncogene's ability to induce tumorigenesis depends on the differentiative context when this oncogene first becomes activated. We have examined the ability of the C-MYC oncogene to induce tumorigenesis in mice of different ages using a novel conditional transgenic model system for C-MYC –induced hepatocellular carcinoma (HCC). C-MYC (now referred to as MYC ) is a member of a family of proto-oncogenes comprising C - MYC, N-MYC, and L-MYC. MYC encodes a transcription factor that, as part of a heterodimeric complex with MAX, regulates the expression of a multitude of genes involved in regulating cellular proliferation and growth ( Johnston et al. 1999 ; Grandori et al. 2000 ; Oster et al. 2002 ; Pelengaris et al. 2002a ). Overexpression of MYC is commonly associated with tumorigenesis. MYC exerts its neoplastic function by inducing autonomous cellular proliferation and cellular growth, blocking differentiation, and inducing genomic destabilization ( Dang 1999 ; Felsher and Bishop 1999a ; Grandori et al. 2000 ; Oster et al. 2002 ; Pelengaris et al. 2002a ; Karlsson et al. 2003 ). It is generally assumed that MYC is restrained from causing tumorigenesis because it concomitantly induces cellular proliferation and apoptosis ( Pelengaris et al. 2002a ). HCC is a common and generally incurable human malignancy of epithelial cells ( Thorgeirsson and Grisham 2002 ). HCC has been strongly associated with viral infections such as hepatitis B and C; exposure to toxins, such as alcohol, aflatoxin, and phenobarbital; and exposure to various carcinogens such as polyvinyl chloride ( Thorgeirsson and Grisham 2002 ). Interestingly, the ability of these carcinogens, such as the hepatitis viruses, to induce HCC depends on the host age during infection. Hepatitis B infection acquired during neonatal development versus adulthood results in a several-magnitude increased risk of HCC ( Chang et al. 1991 ). These results suggest that there are differentiative windows during liver development that may be more susceptible to neoplastic transformation. Human tumors have been analyzed extensively for genetic events associated with HCC ( Buetow et al. 1989 ; Tsuda et al. 1990 ; Boige et al. 1997 ; Piao et al. 1997 ). MYC oncogene activation is one of the more common events in the pathogenesis of HCC. MYC overexpression in human HCC is most commonly associated with genomic amplification ( Abou-Elella et al. 1996 ; Kawate et al. 1999 ). Human HCCs exhibit amplification of MYC in up to 50% of tumors ( Abou-Elella et al. 1996 ; Kawate et al. 1999 ). The presence of MYC amplification in HCC portends a more advanced and aggressive clinical phenotype ( Abou-Elella et al. 1996 ). Thus, the MYC oncogene appears to play a critical role in the pathogenesis of HCC. The most compelling evidence that MYC is causally associated with the etiology of HCC comes from animal models ( Sandgren et al. 1989 ; Fourel et al. 1990 ; Murakami et al. 1993 ; Morgenbesser and DePinho 1994 ; Sargent et al. 1996 , 1999 ; De Miglio et al. 1999 ; Santoni-Rugiu et al. 1999 ; Renard et al. 2000 ). MYC is frequently activated through insertional mutagenesis mediated by the hepadnavirus in woodchuck liver tumors ( Fourel et al. 1990 ; Renard et al. 2000 ). Carcinogen-induced HCC in Wistar rats is associated with MYC amplification and overexpression ( De Miglio et al. 1999 ). The overexpression of MYC (as a transgene) and other oncogenes (e.g., RAS, T antigen) in murine hepatocytes results in HCC ( Sandgren et al. 1989 ). The latency of HCC in these transgenic mice is long, but is greatly accelerated by the transgenic overexpression of transforming growth factor-alpha ( Murakami et al. 1993 ; Sargent et al. 1996 , 1999 ; Santoni-Rugiu et al. 1999 ). These results highlight that the activation of MYC alone is not sufficient to induce HCC. Traditional transgenic systems that have been previously used to study the role of oncogenes in tumorigenesis continuously overexpress transgenes and hence preclude the investigation of the initial and developmentally specific consequences of oncogene activation. To investigate the developmentally specific consequences of MYC overexpression in the pathogenesis of HCC in vivo, we used transgenic mice in which the MYC proto-oncogene is conditionally regulated via the tetracycline regulatory system (Tet system) ( Felsher and Bishop 1999b ). We found that the ability of MYC to induce cellular proliferation versus cellular growth, and consequently its ability to induce tumorigenesis in murine hepatocytes, was dependent on the age of the host. Our results have implications for the mechanisms by which MYC and other oncogenes initiate and are restrained from causing tumorigenesis. Results A Conditional Model System for MYC -Induced HCC We used the Tet system to conditionally express MYC in murine hepatocytes ( Kistner et al. 1996 ; Felsher and Bishop 1999b ). We mated the transgenic line TRE- MYC ( Felsher and Bishop 1999b ), which contains the tetracycline response element adjacent to the human MYC cDNA, with the transgenic line LAP- tTA ( Kistner et al. 1996 ), which contains a liver-specific enhancer that drives the expression of the tetracycline-transactivating protein. Mice possessing both transgenes exhibited increased expression of the MYC transgene in their hepatocytes ( Figure 1 A). Mice possessing either transgene alone did not overexpress MYC and lacked evidence of morbidity or mortality. Similarly, mice possessing both transgenes that were treated with doxycycline to suppress MYC transgene expression did not exhibit a phenotype. Thus, we have developed a transgenic model that enables us to conditionally regulate MYC expression in murine hepatocytes. Figure 1 MYC Overexpression in Adult Hepatocytes Results in HCC (A) Western blot analysis demonstrating that mice transgenic for both LAP- tTA and TRE-MYC conditionally express MYC protein in their hepatocytes in the absence (−) but not in the presence (+) of doxycycline. (B) Adult mouse with MYC -induced liver tumor. (C) Histology of an adult MYC -induced liver tumor. (D) Gross pathology of an adult liver tumor transplanted subcutaneously into a scid mouse. (E) Histology of an adult tumor transplanted subcutaneously into a scid mouse. To investigate if MYC overexpression is sufficient to induce HCC in our model system, we removed doxycycline treatment in adult mice (6–12 weeks of age) transgenic for both TRE- MYC and LAP- tTA . Ninety percent of adult mice overexpressing MYC succumbed to liver tumors with a mean latency of 35 weeks. At necropsy, mice exhibited marked gross enlargement of the liver with multiple tumor masses ( Figure 1 B). The normal liver architecture was disrupted by nodular tumors with histological features typical of HCC. Tumors were composed of dysplastic nests of cells with large pleomorphic nuclei, delicate vesicular chromatin, and very prominent nucleoli ( Figure 1 C). Tumors could invade into the abdomen and the lung ( Figure 2 A– 2 C). These features demonstrated that MYC overexpression in adult mice resulted in HCC. Figure 2 MYC-Induced Hepatic Tumors Are Invasive and Metastatic (A) Adult mouse with MYC -induced liver tumor that has metastasized to the abdomen and the lungs. (B) Histology of an adult MYC -induced lung metastasis. (C) Histology of an adult MYC -induced liver tumor. (D) Gross pathology of a liver tumor from a neonatal host transplanted subcutaneously into a scid mouse. To confirm that these tumors were malignant, we transplanted them subcutaneously into scid mice. Tumors formed in the inoculated mice after an 8–10-week latency (see Figure 1 D). The transplanted tumors displayed identical histology to the primary transgenic tumor (see Figure 1 E versus 1 C). Normal adult hepatocytes failed to induce tumors when inoculated into scid mice. We conclude that MYC overexpression in adult hepatocytes results in the formation of highly malignant liver cancers with features consistent with human HCC. Developmental State of the Host Influences on Frequency and Latency of Tumor Onset To determine if the developmental state of host hepatocytes influenced the ability of MYC activation to induce tumorigenesis, we induced MYC in cohorts of different ages ( Figure 3 A). Mice that overexpressed MYC during embryonic development of the liver succumbed to neoplasia within 10 d of birth. Mice in which MYC was activated at birth (neonates) succumbed to neoplasia within 8 weeks. Mice in which MYC was induced at 4 weeks or 6–12 weeks of age developed tumors after a mean latency of 15 and 35 weeks, respectively ( Figure 3 A). We conclude that the ability of MYC activation to induce tumorigenesis in hepatocytes is inversely correlated with the developmental age of the host. Figure 3 MYC's Ability to Induce HCC Is Inversely Correlated with the Age of the Host at the Time of MYC Activation (A) Survival of transgenic mice demonstrates that tumorigenesis in the liver is inversely correlated with the age of the host at the time of MYC induction. Shown are cases where MYC was constitutively expressed (▪), newborn mice in which MYC was activated at birth (▴), young mice in which MYC was activated at 4 weeks of age (•), adult mice in which MYC was activated at 6–12 weeks of age (♦), and transgenic mice treated with doxycycline (□). Cohorts consisted of 15–30 mice. Mice were scored when moribund. MYC transgene expression was induced to similar levels in the differently aged cohorts of mice. Survival time is measured as the time after MYC induction. (B) Western blot examining total MYC protein levels (human MYC and endogenous murine c-MYC) in mice when MYC is induced during embryonic development, activated at birth, and activated during adulthood (after 10 weeks of age). Adult mice exhibited a progressive increase in MYC protein levels during the course of MYC induction, with a significant increase in MYC protein in tumors. MYC protein levels in neonatal mice in which MYC was activated at birth were slightly increased at 10 d of age, and significantly increased at 18 d of age when these mice developed liver tumors. Liver tumors in 2- and 6-d-old neonatal mice that overexpressed MYC during embryonic development exhibited MYC protein levels similar to those observed in neonatal and adult tumors. (C) Real-time PCR analysis showing human MYC RNA levels in mice after different durations of MYC transgene induction. Adult livers exhibited a small increase in MYC RNA levels upon MYC activation, and a much greater increase in MYC RNA in MYC-induced tumors (black bars). In neonatal mice in which the MYC transgene was induced at birth, MYC RNA levels rose after 10 d of MYC activation. When these neonatal mice developed liver tumors, they exhibited MYC RNA levels similar to those seen in adult tumors (gray bars). Mice in which MYC was overexpressed during embryonic development developed liver tumors by 2 d of age and exhibited MYC RNA levels similar to those observed in neonatal and adult tumors (white bars). One possible explanation for our results was that the levels of MYC induction were different in embryonic and neonatal versus adult hosts. To address this possibility, we examined total MYC protein levels by Western analysis using a polyclonal antibody that recognizes both the human c-MYC protein and the endogenous murine c-MYC ( Figure 3 B). In neonatal and adult mice, MYC protein levels were induced at similar levels ( Figure 3 B). In tumors from embryos, neonatal, and adult mice, MYC protein levels increased an additional 5- to 10-fold over the levels observed in nontransgenic and in MYC -induced nonneoplastic livers. Tumorigenic conversion of hepatocytes was associated, in all age groups of mice, with further increases in the levels of MYC protein ( Figure 3 B). We obtained similar results by quantitative PCR analysis of mRNA expression of the human MYC transgene ( Figure 3 C). In neonatal and adult mice, MYC transgene expression was induced at similar levels. In tumors from embryos, neonatal, and adult mice, the levels of MYC transgene RNA increased an additional 10-fold over the levels observed in nonmalignant hepatocytes ( Figure 3 C). Hence, tumorigenic conversion of hepatocytes was associated, in all age groups of mice, with further increases in the levels of MYC transgene expression. However, differences in the ability of MYC activation to initiate tumorigenesis in mice of different ages did not appear to be related to differences in the levels of induction of MYC transgene expression. The increased levels of MYC expression we observed in tumors likely reflect that the proliferating tumor cells express more abundant levels of the MYC transgene than normal hepatocytes. This observation is consistent with observations described in other transgenic models in which expression of transgenes is generally higher in tumors than it is in the normal cellular counterparts ( Weiss et al. 2003 ). MYC Activation in Embryonic and Neonatal Hepatocytes Induces Cellular Proliferation and Tumorigenesis To evaluate how MYC 's ability to induce tumorigenesis is influenced by the age of mice, we investigated the initial consequences of MYC activation in hepatocytes during different developmental periods. Mice that overexpressed MYC during embryonic development were born with livers similar in weight and gross architecture to normal age-matched livers, yet exhibited increasing numbers of neoplastic cells from birth through the first week of life associated with progressive abdominal enlargement. At necropsy, abdominal enlargement was associated with marked hepatomegaly with a 5-fold increase in total liver weight ( Figure 4 A, and see Figure 8 B below). Although these livers were larger, the gross architecture was preserved ( Figure 4 A, MYC ON versus MYC OFF). When we examined the histology of the livers in which MYC was overexpressed during embryogenesis, we found that they resembled liver cancers ( Figure 4 B versus 4 C) similar to the MYC -induced HCCs we observed in adult mice (see Figure 1 C versus 4 C). Hence, MYC overexpression appears to induce cellular proliferation in neonatal hepatocytes that progresses rapidly to neoplasia. Figure 4 MYC Activation during Embryonic Development Induces a Rapid Onset of Neoplasia (A) A normal neonatal liver and a neonatal liver in which MYC was activated embryonically. (B) Histology of a normal neonatal liver in which MYC was not activated. (C) Histology of a neonatal liver in which MYC was activated embryonically. (D) DNA content of normal neonatal hepatocytes. (E) DNA content of neonatal hepatocytes in which MYC was activated embryonically. (F and H) Ki67 immunofluorescence and DAPI staining corresponding to a normal neonatal liver. (G and I) Ki67 immunofluorescence and DAPI staining corresponding to a neonatal liver in which MYC was activated embryonically. Figure 8 MYC Activation in Adult Hepatocytes Causes Cellular Hypertrophy (A) Relative volumes of neonatal hepatocytes and nuclei after MYC activation. Data are expressed as normalized volume plus or minus the standard error of the mean. The volume was normalized by dividing each measurement by the mean volume of normal 1-d-old neonatal mice. Three livers were measured per time point. T, tumor. (B) Neonatal liver weights of normal and MY C-activated livers. Three to five livers were weighed per time point. Data are expressed as the mean weight (grams) plus or minus the standard error of the mean. (C) Relative volumes of adult hepatocytes and nuclei after MYC activation. Volumes of cells are expressed as the mean volume divided by the mean volume of hepatocytes from normal mice plus or minus the standard error of the mean. Cells from two to three livers were measured per time point. (D) Adult liver weights after MYC activation. A total of nine livers were measured per time point after MYC activation. Data are expressed as the mean weight (grams) plus or minus the standard error of the mean. To determine if MYC was inducing changes in cell cycle transit, we measured the DNA content of isolated nuclei from normal hepatocytes and hepatocytes in which MYC was activated during embryonic development ( Figure 4 D and 4 E). Normal neonatal hepatocytes mostly contained 2N DNA content, consistent with most of the cells residing in G1 ( Figure 4 D). A minority of hepatocytes exhibited 4N and 2N–4N DNA content, demonstrating that few cells were in G2/M and S phase, respectively. In contrast, upon MYC activation the proportion of neonatal hepatocytes with 2N–4N DNA content substantially increased, suggesting that an increased number of cells were in S phase ( Figure 4 E). In order to confirm that MYC activation caused tumorigenesis by inducing cell proliferation, we performed Ki67 immunofluorescence and DAPI staining in tumors induced by MYC activation during embryonic development and in age-matched nontransgenic livers. Indeed, there was evidence for increased hepatocyte proliferation in the MYC-induced neonatal tumor, as demonstrated by an increase in Ki67-positive cells ( Figure 4 G and 4 I versus 4 F and 4 H). We conclude that MYC activation during embryonic development causes neonatal hepatocytes to undergo DNA replication, cell cycle transit, proliferation, and almost immediate neoplastic conversion. To confirm that MYC activation during embryonic development induced tumorigenesis in neonatal livers, we transplanted neoplastic hepatocytes into scid mice. We found that neoplastic neonatal hepatocytes readily formed tumors, whereas the transplantation of normal neonatal hepatocytes did not form tumors (see Figure 2 D and unpublished data). Therefore, MYC overexpression during embryonic development of the murine liver causes hepatocellular tumorigenesis within the first 10 d of birth ( Figure 3 ). We conclude that MYC overexpression results in rapid neoplastic conversion of neonatal hepatocytes. Mice in which MYC was activated at birth exhibited progressive abdominal enlargement during their second and third weeks of life, and they showed signs of tumorigenesis by 18 to 40 d of age. When these mice developed tumors their livers were ten times the normal size, were paler, exhibited a multitude of coalescing tumor nodules, and preserved a normal gross architecture (unpublished data). We did not observe any histological changes in the liver after 10 d of MYC activation ( Figure 5 A versus 5 B); however, by 18 d of MYC activation the histology resembled liver cancers ( Figure 5 C versus 5 D), similar to the MYC-induced HCCs we observed in adults ( Figure 5 D versus 1 C). Figure 5 MYC Activation at Birth Induces Proliferation of Neonatal Hepatocytes (A) Histology of a normal 10-d-old neonatal liver. (B) Histology of a 10-d-old liver in which MYC was activated at birth. (C) Histology of a normal 18-d-old neonatal liver. (D) Histology of a MYC-induced neonatal liver tumor that developed after 18 d of MYC overexpression; MYC was activated at birth. (E–L) Ki67 immunofluorescence (E–H) and DAPI staining (I–L) of normal neonatal hepatocytes (E, G, I, and K), MYC -activated hepatocytes (F and J), and MYC-induced neonatal tumors (H and L). Upon initial MYC activation in neonatal mice, there was a small increase in Ki67-positive cells. MYC-induced neonatal tumors exhibited much higher levels of Ki67-positive cells. We performed Ki67 immunofluorescence and DAPI staining in order to determine if MYC overexpression in neonatal livers was inducing hepatocyte proliferation. At 10 d of age the liver is in an active state of proliferation; thus, there was little detectable difference in the number of Ki67-positive cells between the MYC -activated and nontransgenic livers ( Figure 5 E and 5 I versus 5 F and 5 J). However, once these livers became neoplastic, there was a great increase in the number of Ki67-positive cells ( Figure 5 G and 5 K versus 5 H and 5 L). MYC Activation in Adult Hepatocytes Induces Cellular Growth, but Not Proliferation We examined the initial consequences of MYC overexpression in adult hepatocytes. In contrast to the rapid neoplastic conversion we observed in embryonic or neonatal hepatocytes ( Figures 4 and 5 ), MYC overexpression in adult hepatocytes caused a marked cellular growth, accompanied by an even greater relative nuclear growth, as observed by histological analysis ( Figures 6 – 8 ). The effects of MYC overexpression on the size of adult hepatocytes depended on the duration of MYC activation. After 2 weeks of MYC overexpression, no changes were observed in cell size compared to normal hepatocytes ( Figure 8 C and unpublished data). However, after 4–8 weeks of MYC activation, adult hepatocytes exhibited increased cell and nuclear size ( Figure 6 B versus 6 A and Figure 8 C). Similar results were observed in over 20 different mice. Similarly, we observed that MYC induces hypertrophy of hepatocytes by flow cytometry analysis ( Figure 7 ). Further duration of MYC activation did not induce further cell growth, as measured up to 50 weeks of MYC activation (unpublished data). Thus, there may be an absolute limit to the ability of MYC to induce liver growth. MYC activation in adult hepatocytes was not associated with a change in overall liver weight ( Figure 8 D). Since the cells were bigger, but the overall weight of the liver did not increase, we infer that the total number of hepatocytes was unchanged or slightly decreased. One possible explanation for these results is that MYC induced apoptosis, as described below. Figure 6 MYC Activation in Adult Hepatocytes Induces Increased Cell Size and Endoreduplication, and Only Results in Cell Proliferation upon Neoplastic Conversion of Hepatocytes (A) Histology of a normal liver. (B) Histology of a liver 2 months after MYC activation. (C) Histology of a MYC-induced liver tumor. (D and G) Ki67 immunofluorescence and DAPI staining of a normal adult liver. (E and H) Ki67 and DAPI staining of an adult liver after 8 weeks of MYC activation. (F and I) Ki67 and DAPI staining of a MYC-induced adult tumor. (J) DNA content measured in normal hepatocytes. (K) DNA content measured after MYC induction for 2 months. (L) DNA content of a representative MYC-induced liver tumor. Figure 7 MYC Activation in Adult Hepatocytes Induces Increased Cell Size A histogram obtained by FACS forward versus light scatter analysis of adult hepatocytes from normal FVB/N livers (green), doxycycline-treated transgenic livers (red), and livers in which the MYC transgene was overexpressed for 3 months (blue). The x -axis represents cell size and the y -axis represents cell count. Adult mice were matched for age. To examine if MYC activation induced proliferation of adult hepatocytes, we measured Ki67 expression by immunofluorescence. We did not observe increased Ki67 expression when MYC was overexpressed in the adult liver ( Figure 6 E and 6 H versus 6 D and 6 G). Only upon neoplastic conversion of hepatocytes was there evidence for increased hepatocyte proliferation ( Figure 6 F and 6 I). We conclude that MYC overexpression in adult hepatocytes induces increased nuclear and cell growth, but not cell proliferation. Our observations are consistent with previous reports that MYC activation induces cell growth ( Mateyak et al. 1997 ; Iritani and Eisenman 1999 ; Johnston et al. 1999 ; Grandori et al. 2000 ; Kim et al. 2000 ). MYC activation in adult hepatocytes eventually culminated in tumorigenesis, demonstrating that some adult hepatocytes acquire the ability to undergo cell division. To confirm this, we measured the nuclear and cellular sizes in liver tumors. When we examined the cell size in ten different tumors from adult hosts, we found that in all tumors, the cell size was reduced to below normal and the nuclear size was similar to that of normal hepatocytes ( Figures 6 C and 8 C). We were also able to confirm that the cell size of tumor cells was reduced to below normal by FACS forward versus side scatter (unpublished data). We conclude that MYC -induced malignant conversion of adult hepatocytes is associated with the acquired ability to undergo mitotic division. MYC Overexpression in Adult Hepatocytes Results in Endoreduplication To further define the consequences of MYC activation on the cell cycle, we examined the DNA content of isolated nuclei from normal and MYC -activated adult hepatocytes. We expected that if adult hepatocytes were restrained from undergoing mitotic division, MYC activation might result in endoreduplication. Age-matched normal hepatocytes exhibited a 2N DNA content consistent with most of the cells residing in G1, and there was no evidence for cells in S or G2/M ( Figure 6 J). After MYC activation for 2 months, we found that almost all nuclei had a 4N, 8N, or 12N DNA content, suggesting that the cells replicated their DNA repeatedly without dividing ( Figure 6 K). Almost no cells contained the intermediate DNA content (2N–4N), demonstrating that very few cells were in S phase at any given time. We conclude that MYC activation induces endoreduplication of the genomes of normal adult hepatocytes. Our results are consistent with reports that MYC overexpression can enforce DNA replication, resulting in endoreduplication in normal cells ( Cerni et al. 1986 ; Mai et al. 1996 ; Chernova et al. 1998 ; Felsher et al. 2000 ). We reasoned that if MYC was causing endoreduplication in adult hepatocytes by arresting cell division and enforcing DNA replication, then upon neoplastic conversion these hepatocytes must acquire the ability to undergo mitotic division and would no longer endoreduplicate. As predicted, tumors did not exhibit evidence for endoreduplication ( Figure 6 L). Greater than 70% of the tumor cells contained a 2N–4N DNA content and none of the cells contained greater than 4N DNA content. The majority of tumor cells were in S phase ( Figure 6 L). Hence, MYC -induced tumorigenesis in adult hepatocytes is associated with the acquired ability to divide mitotically. MYC Overexpression in Adult Hepatocytes Does Not Induce Apoptosis MYC -induced apoptosis is an important mechanism that restrains MYC from causing tumorigenesis ( Evan et al. 1992 ; Pelengaris et al. 2000 , 2002b ). We reasoned that MYC may induce cellular hypertrophy, but not an increase in liver mass, because MYC induces apoptosis. Normal neonatal and adult hepatocytes did not undergo apoptosis, as demonstrated by TUNEL assay or DAPI staining ( Figure 9 A, 9 B, 9 E, and 9 F). Surprisingly, we could not find evidence that MYC induced apoptosis in adult hepatocytes by TUNEL assay or DAPI staining after 2, 4, or 8 weeks of MYC induction prior to tumor formation ( Figure 9 G and 9 H and unpublished data). Contrary to what we expected, MYC activation was only associated with increased apoptosis in the hepatocytes of liver cancers ( Figure 9 C, 9 D, 9 I, and 9 J). Hence, apoptosis is not necessarily the mechanism restraining MYC from causing tumorigenesis, at least in hepatocytes ( Pelengaris et al. 2002a ). However, we recognize that MYC could be inducing low levels of apoptosis in hepatocytes, perhaps not easily detected by TUNEL, since apoptotic cells may be rapidly eliminated from the liver through the host reticulo-endothelial system. Such a low level of apoptosis still could explain why in the adult liver hepatocytes become hypertrophic but the liver mass does not increase. Figure 9 MYC Activation Does Not Induce Apoptosis in Murine Hepatocytes TUNEL assay (A, C, E, G, and I) and DAPI staining (B, D, F, H, and J) of normal (A, B, E, and F) and MYC -activated (C, D, and G–J) hepatocytes of neonatal (A–D) and adult (E–J) mice. After 4 weeks of MYC activation in adult hepatocytes, there was no evidence of apoptosis either by TUNEL assay (G) or by DAPI staining of nuclei (H). MYC activation is associated with increased apoptosis with the neoplastic conversion of neonatal (C) and adult (I) hepatocytes. Representative data from one of three experiments are shown. Identical results were seen when MYC was activated for 2 or 8 weeks. Loss of p53 Function Cooperates with MYC to Induce Tumorigenesis in Adult Mice Previously, we have shown that the loss of p53 function is required to permit the cell division of normal mouse and human fibroblasts overexpressing MYC ( Felsher et al. 2000 ). We speculated that loss of p53 function might be similarly required for MYC activation to induce cell proliferation and tumorigenesis in hepatocytes. First, we examined if MYC activation affected p53 protein expression. We found that MYC activation was associated with an increase in p53 protein levels in adult hepatocytes, as measured by Western analysis ( Figure 10 A). Conversely, tumors in adult mice frequently exhibited reduced levels of p53 protein expression ( Figure 10 A and unpublished data). Not all tumors exhibited reduced p53 expression. Since p53 mediates its function largely through inducing the transcription of many different genes, we evaluated if tumors exhibited a loss of p53 function by measuring the expression of these target genes. We found by Northern analysis that the p53 target genes, p21 and MDM2, were induced upon MYC activation in adult livers ( Figure 10 C). Conversely, MYC -induced adult HCCs frequently exhibited reduced or no expression of p53 downstream targets. Notably, a tumor that exhibited high p53 protein levels, tumor 2264, exhibited a loss in expression of p53 target genes. In contrast, tumors arising in neonatal mice expressed p53 protein and exhibited the induction of p53 target genes ( Figure 10 B and 10 C). We conclude that the adult, but not neonatal MYC -induced liver tumors require the loss of p53 function for tumorigenesis. Hence, HCCs that arise in adult versus neonatal hosts appear to occur through genetically distinct mechanisms. Figure 10 MYC Activation Induces p53 Function, and Loss of p53 Function Is Necessary for MYC to Induce Tumorigenesis in Adult Hepatocytes (A) Western blot analysis for p53 protein expression after MYC activation for 1 month, 2 months, and in MYC-induced tumors. As a positive control, we used a lymphoma cell line that overexpresses a mutant p53, kindly provided by Dr. Kevin Smith. (B) Western blot analysis for p53 protein expression in a normal neonatal liver, in a neonatal liver in which MYC was activated during embryonic development that was obtained from a 2-d-old mouse, and in MYC-induced neonatal tumors. As a positive control, we used a lymphoma cell line generated in our lab that overexpresses a mutant p53. (C) Northern blot analysis for p21 and MDM2 in neonatal and adult liver tumors. (D) Survival of adult mice after activation of MYC in the presence of the wild type or the loss of one p53 allele. (E) Loss of heterozygosity analysis of MYC/p53 +/− tumors by PCR analysis. To directly address if loss of any of p53's functions accelerates the ability of MYC to induce HCC in adult mice, we generated transgenic mice that overexpressed MYC in their hepatocytes in the absence of one p53 allele. We mated LAP- tTA /TRE- MYC mice with p53 +/− mice that were in the FVB/N background. We activated MYC in mice when they were 6 weeks old and monitored them for morbidity. We found that even the loss of a single p53 allele was sufficient to reduce the mean latency of tumor onset in 6-week-old adult mice from 20 weeks to 15 weeks ( Figure 10 D). We found by PCR that tumors did not generally inactivate the second allele through deletion ( Figure 10 E). Our results extend previous findings that suggest that the lack of p53 function cooperates with MYC to induce HCC ( Klocke et al. 2001 ). We conclude that even a slight reduction of p53 function greatly facilitates the ability of MYC to induce tumorigenesis in adult hepatocytes. Partial Hepatectomy Accelerates MYC-Induced Tumorigenesis Our results suggest that the ability of MYC to induce tumorigenesis in hepatocytes depends on the developmental context. We recognized that an alternative explanation is that MYC induces tumorigenesis more readily in hepatocytes that are already proliferating. Adult hepatocytes are known to undergo rapid proliferation in response to partial hepatectomy resulting in the complete regeneration of the liver within 2 weeks of surgical removal ( Michalopoulos and DeFrances 1997 ; Kountouras et al. 2001 ). We found that MYC activation in adult mice that have undergone partial hepatectomy exhibited a reduced latency of tumor induction in comparison with adult mice that had not undergone surgery (mean latency of 14 weeks versus 35 weeks). However, this latency of tumorigenesis in adult mice after partial hepatectomy was still up to two magnitudes longer than what was observed when MYC was activated in embryonic and neonatal mice (<10 d and 4 weeks, respectively) ( Figures 11 and 3 A). In addition, tumors in mice that had undergone partial hepatectomy, unlike tumors arising in neonatal mice, were multifocal, suggesting that tumorigenesis was occurring infrequently ( Figure 11 A versus 4 A). Similarly, upon histological analysis, after MYC was activated for 7 weeks in mice that had undergone partial hepatectomy, mice exhibited many individual foci of HCC ( Figure 11 B and 11 C). Finally, in adult mice after partial hepatectomy, but not in neonatal mice, areas of the liver that had not undergone neoplastic conversion clearly exhibited increased cellular hypertrophy, and hence were unable to undergo mitotic division (unpublished data). We conclude that the ability of MYC to induce tumorigenesis in adult hepatocytes is accelerated after partial hepatectomy when adult hepatocytes are proliferating, but other developmentally specific parameters are more important in determining when oncogene activation will induce tumorigenesis than the ability of hepatocytes simply to proliferate. Figure 11 Partial Hepatectomy Accelerates the Ability of MYC to Induce HCC in Adult Mice (A) Liver from an adult mouse 7 weeks after MYC activation exhibited no gross phenotypic changes (left), whereas the liver from an adult mouse 7 weeks after MYC activation that had undergone partial hepatectomy exhibited a multifocal liver tumor (right). (B) The histology of the liver from an adult mouse 7 weeks after MYC activation exhibited no evidence of a tumor. (C) The histology of a liver from an adult mouse after MYC activation that had undergone partial hepatectomy exhibited a multifocal HCC. (D) Survival after MYC activation in adult mice that have either not undergone surgery (▪) or undergone a partial hepatectomy (□). Results are pooled from two independent experiments with a total of ten mice per group. Discussion Developmental Context Influences MYC 's Ability to Induce Cellular Proliferation Here we demonstrate that MYC 's ability to induce cellular growth versus proliferation and tumorigenesis depends on the differentiative context of a cellular lineage. Our results have general implications for the mechanisms by which MYC and other oncogenes induce and are restrained from causing tumorigenesis. We found that the initial consequences of MYC overexpression, as well as its ability to induce tumorigenesis in murine hepatocytes, depend on the host age. The activation of MYC during embryonic development and at birth caused gross liver enlargement and rapid emergence of neoplasia. This increased liver size resulted from increased cell number or cellular hyperplasia. In contrast, in adult murine hepatocytes, MYC activation induced cell growth, resulting in cellular hypertrophy and endoreduplication. MYC overexpression caused HCC in adult mice less frequently than in neonates, and only after a prolonged latency. We conclude that the ability of MYC to induce cellular proliferation and tumorigenesis appears to be determined by the relative developmental maturation of a cellular lineage. We recognized that an alternative explanation for our results is that only hepatocytes that are actively proliferating are susceptible to neoplastic transformation upon MYC induction. We were able to address this possibility directly by examining the consequences of MYC overexpression in adult hepatocytes after partial hepatectomy, which resulted in the immediate induction of adult hepatocyte proliferation associated with complete liver regeneration within 2 weeks. Adult mice only exhibited a modest increased susceptibility to MYC -induced tumorigenesis observed when MYC was induced in embryonic mice. This is despite the fact that after partial hepatectomy the total number of proliferating hepatocytes would be at least two magnitudes greater than in a neonatal or embryonic liver, whose total mass is much smaller. We conclude that developmentally specific parameters other than proliferation are more likely to play a causative role in the differential susceptibility of embryonic and neonatal hepatocytes to MYC -induced tumorigenesis. Several different development-related factors could account for our observations. The relative ability of MYC to induce tumorigenesis in developmentally immature hosts may reflect the relative increased abundance of immature hepatocytes. Homeostatic mechanisms that regulate liver size and hepatocyte proliferation may permit MYC-induced proliferative expansion in the developing livers of young hosts and restrain proliferation in the mature liver of adult hosts. Differences in rates of tumorigenesis may be related to telomerase function that monitors the distance from senescence, as recently described ( Artandi et al. 2000 ). An unlikely possibility is that higher doses of tetracyclines may impede liver carcinogenesis ( NTP 1989 ). The conditional model system that we have described should be useful for addressing these different possible mechanisms. We conclude that mechanisms that regulate mitotic division play a critical role in preventing potent oncogenes, such as MYC, from inducing cancer in adult somatic cells. MYC activation alone is capable of enforcing DNA replication, but not cell division ( Johnston et al. 1999 ; Felsher et al. 2000 ). Mitotic arrest may represent a critical fail-safe mechanism. If MYC were capable of enforcing mitotic division as well as cell growth and DNA replication, then aberrant activation of MYC alone would be sufficient to induce tumorigenesis. Activation of MYC in immature hepatocytes, cells already committed to a program of cellular proliferation, is sufficient to induce tumorigenesis. Activation of MYC in adult cells must require additional genetic events that permit mitotic cell division. Notably, our interpretation of our findings is supported by similar observations previously described in keratinocytes ( Gandarillas et al. 2000 ). We now can offer an explanation for what previously have been described as discordant results between reports that the MYC oncogene regulates cellular proliferation and other reports that MYC regulates cell growth ( Mateyak et al. 1997 ; Iritani and Eisenman 1999 ; Johnston et al. 1999 ; Grandori et al. 2000 ; Kim et al. 2000 ; Trumpp et al. 2001 ). MYC generally appears to coordinate cell growth with DNA replication ( Mateyak et al. 1997 ; Iritani and Eisenman 1999 ; Johnston et al. 1999 ; Grandori et al. 2000 ; Kim et al. 2000 ; Trumpp et al. 2001 ). However, contrary to what has been described, we show that the ability of MYC to induce cell division may depend on the developmental context of the cell. In adult hepatocytes, which normally do not proliferate, MYC activation can induce cell growth and DNA replication, but not cell division. In embryonic or neonatal hepatocytes, which are intrinsically committed to cellular proliferation, MYC activation induces cell growth, DNA replication, and mitotic division. Hence, the consequences of MYC activation appear to depend on the previous commitment of cells in a specific developmental state to a cellular program capable only of growth versus growth and cellular proliferation. It is not clear whether the hepatocytes that ultimately give rise to the tumors we observed derive from mitotically arrested diploid hepatocytes or from polyploid hepatocytes that acquire the capacity to undergo mitotic division. MYC Is Restrained from Inducing Proliferation by an Arrest in Cell Division Many reports document that p53 functions as a general surveillance checkpoint that prevents oncogenes from inducing tumorigenesis ( Sherr 1998 ; Vogelstein et al. 2000 ; Wahl and Carr 2001 ). MYC activation is known to induce p53 function ( Chernova et al. 1998 ; Zindy et al. 1998 ; Schmitt et al. 1999 ; Felsher et al. 2000 ; Grandori et al. 2000 ; Oster et al. 2002 ; Pelengaris et al. 2002a , 2002b ), which in turn has been shown to cause apoptosis ( Evan et al. 1992 ; Zindy et al. 1998 ; Schmitt et al. 1999 ; Grandori et al. 2000 ; Oster et al. 2002 ; Pelengaris et al. 2002a , 2002b ). In contrast to these reports, we found that MYC is restrained from causing cell division in adult hepatocytes, at least in part, through a p53-dependent mechanism. Our results are consistent with many reports that demonstrate that p53 regulates checkpoints during DNA replication and during mitosis ( Wahl and Carr 2001 ). We conclude that apoptosis is not the mechanism that precludes MYC from inducing neoplasia in hepatocytes. Rather, apoptosis is associated with MYC -induced neoplastic progression. Notably, similar results have been described in MYC-induced breast cancer and lymphoma ( McCormack et al. 1998 ; Blyth et al. 2000 ). There are several possible explanations for this discordance between our results and many previous reports. MYC may only induce apoptosis in some cellular lineages or only in particular differentiative contexts. In this regard, we have shown that upon inactivating MYC , tumor cells differentiated, and upon MYC reactivation in this new differentiative state, cells underwent apoptosis ( Jain et al. 2002 ). Thus, whether oncogene activation induces proliferation, growth, arrest, or apoptosis may depend on the gene expression program of a cell associated with a particular differentiative state ( Felsher 2003 ). We found that the ability of MYC overexpression to induce tumorigenesis in adult hepatocytes often requires the loss of p53 function. MYC -induced HCC exhibited reduced p53 protein expression and transcriptional activity. The introduction of a single mutant p53 allele greatly accelerated MYC 's ability to induce tumorigenesis. Hence, in adult hepatocytes, loss of p53 function appears to be necessary to permit MYC activation to induce cellular proliferation and tumorigenesis. Our results here are similar to our previous observations that normal mouse and human fibroblasts overexpressing MYC replicate and even endoreduplicate their DNA, but are unable to undergo mitotic division unless p53 function has been lost ( Felsher et al. 2000 ). Specific Developmental Contexts Permit Oncogene-Induced Tumorigenesis We may be able to explain why the frequency and spectrum of neoplasia vary with the host's age. In general, children are susceptible to tumors of the hematopoietic system, musculo-skeletal system, and central nervous system, whereas adults are susceptible more frequently to tumors derived from epithelial lineages, such as colon, lung, breast, prostate, and liver. The ability of oncogenes to induce cancer may be dependent upon the differentiative context. Frequently, oncogene activation has been associated with the malignant conversion of immature cellular compartments ( Adams et al. 1985 ; Spanopoulou et al. 1989 ; Pelengaris et al. 1999 ; Blyth et al. 2000 ). We may be able to explain why neonatal versus adult humans that have become infected with hepatitis B succumb to HCC with a 100-fold-increased frequency and reduced latency of tumor onset ( Chang et al. 1991 , 1997 ). We infer that oncogenes in general are more potent in inducing cancer in hepatocytes of younger hosts because these cells are committed to a developmental program permissive to tumorigenesis. Cancers frequently correspond to the malignant expansion of specific immature differentiative states within a given cellular lineage ( Greaves 1986 ; Klein and Klein 1986 ). It would be advantageous for adult somatic cells, which are required to be long lived, to acquire mechanisms that prevent single oncogenes from inducing inappropriate cellular proliferation and thereby tumorigenesis. Two such mechanisms have been proposed: Some oncogenes induce premature senescence ( Serrano et al. 1997 ; Lin et al. 1998 ; Zhu et al. 1998 ; Dimri et al. 2000 ), and other oncogenes induce apoptosis ( Pelengaris et al. 2000 ). Here we have provided evidence for a different mechanism: The aberrant activation of proto-oncogenes, such as MYC, in the developmental context of adult somatic cells appears to be inherently prohibited from inducing mitotic division. The intrinsic inhibition of a proliferative program in mature somatic cells may be a more parsimonious mechanism than apoptosis as a means of restraining tumorigenesis because it would permit otherwise normal fully differentiated somatic cells to continue to operate despite their acquisition of an activating mutation in an oncogene. In some cases, it would be advantageous to arrest cells rather than induce their apoptosis. Immature cells, which are committed to a program of proliferative expansion, are inherently more susceptible to oncogene-induced autonomous proliferation and tumorigenesis. In immature cells, apoptosis may be the only mechanism that could prevent tumorigenesis. Thus, the differentiative state and epigenetic program of a cell influences whether an oncogene induces cellular senescence, mitotic arrest, or apoptosis. Generally, oncogenes may cause tumorigenesis most readily in developmental contexts that provide a gene expression program permissive to tumorigenesis ( Jain et al. 2002 ; Felsher 2003 ). Pathologic conditions that trigger the expansion of immature cells—tissue injury and regeneration, infection, and autoimmune processes—may be associated with cancer because they change the cellular state, now permitting a single oncogene, such as MYC, to initiate tumorigenesis. In this regard, we found that MYC accelerated tumorigenesis in adult hosts that had undergone partial hepatectomy and were undergoing liver regeneration—a state that induces robust cellular proliferation. However, tumorigenesis was not accelerated to the same degree as observed when MYC was activated in embryonic or neonatal hosts. Therefore, inducing the capacity of a cell to proliferate alone is unlikely to be sufficient to confer susceptibility to tumorigenesis. It has long been appreciated that conferring the ability of a cell to proliferate is not sufficient to induce tumorigenesis. Other developmentally specific parameters may play a more critical role in defining when oncogene activation results in tumorigenesis. The model system we have developed should prove useful in defining how particular developmental contexts and pathologic states contribute to tumorigenesis. Materials and Methods Transgenic mice The TRE- MYC transgenic line generated for these experiments was described previously ( Felsher and Bishop 1999b ). The LT-tTA transgenic line was kindly provided by H. Bujard ( Kistner et al. 1996 ). The p53 +/− mice were generously provided by A. Bradley. Mice were mated and screened by PCR. MYC expression was activated by removing doxycycline treatment (100 μg/ml) from the drinking water of mice transgenic for both TRE- MYC and LAP-tTA. Tumorigenicity assays MYC was activated in the liver by removing doxycycline treatment from the water. MYC was activated during embryonic development by removing doxycycline treatment from the mating cage before conception. MYC was activated at birth by removing doxycycline from the mating cage immediately after the birth of the litter. Mice were monitored daily and were sacrificed when moribund. During necropsy, liver tissues were saved by fixation in 10% buffered formalin or by being snap frozen in liquid nitrogen. For transplantation, liver tumor specimens were sliced into small pieces, incubated first in calcium-free Hank's Buffered Saline Solution (HBSS) on a stirring plate at 37 °C for 20 min, then washed and resuspended in 1× digestion buffer with 1.5 mg/ml collagenase, and incubated on a stirring plate at 37 °C for an additional 40 min. The solution was then filtered through a 100-μm filter, washed, and resuspended in PBS twice. After the first wash, the solution was resuspended in 10 ml of PBS. After the second filtering and wash, the solution was resuspended in 500 μl of PBS. A quantity of 10 6 cells was injected subcutaneously into scid mice (250 μl per injection) using a 1-ml syringe and 27-gauge needle. Mice showed signs of tumorigenesis within 2–3 weeks of inoculation. Mice were sacrificed when tumors reached 2 cm in size. To prepare 5× digestion buffer, 1.12 g of KCl, 37.94 g of NaCl, 0.69 g of NaH2PO 4 *H 2 O, and 9.9 g of dextrose monohydrate were dissolved in 1 liter of H 2 O. Partial hepatectomy The median and left lateral lobes comprise about 70% of the liver and their removal is recognized classically as a partial hepatectomy. We performed a one-third partial hepatectomy, removing only the median lobe of the liver. Mice were anesthetized with 16 mg/kg ketamine/xylazine. An incision was made through the midline ventral abdominal skin and abdominal muscles, extending from just above the xiphoid cartilage to about halfway towards the base of the tail. A small bolster was placed under the thorax, causing the liver to fall forwards away from the diaphragm. The liver was pushed out and the suspensory ligaments were cut with blunt-end scissors. The median lobe was raised and a ligature was tied around it with the blood vessels at the base. The liver lobe was removed by cutting close to the ligature with sharp scissors. The bolster was removed and the muscle and skin incisions were closed. The mice were monitored hourly for pain and dehydration after the procedure. Histology Liver tissues were fixed in 10% buffered formalin for 24 h and then transferred to 70% ethanol until embedding in paraffin. Tissue sections 4 μm thick were cut from paraffin-embedded blocks and placed on glass slides. Hematoxylin and eosin (H&E) staining was performed using standard procedures. The Stanford Histology Core laboratory prepared paraffin sections and performed H&E staining. We measured Ki67 expression by immunofluorescence using a mouse anti-human Ki67 monoclonal antibody (BD Biosciences, Palo Alto, California, United States). We used the Vector M.O.M. Basic Kit (Vector Laboratories, Burlingame, California, United States). Sections were deparaffinized with xylene and rehydrated through graded alcohol washes, followed by antigen retrieval in a microwave for 15 min in Vector Antigen Unmasking solution (H-3300). The slides were then incubated in 100 mM glycine for 2 × 8 min to reduce fluorescent background. Slides were blocked by incubation in avidin for 10 min followed by biotin for 10 min using the Dako biotin blocking system (DAKO Corporation, Carpinteria, California, United States) and subsequently incubated for 1 h in M.O.M. IgG-blocking reagent diluted 1:4 in PBS. Slides were then incubated for 1 h in mouse anti-human Ki67 monoclonal antibody diluted 1:100 in M.O.M. diluent. Slides were washed in TBST for 3 × 5 min to reduce background and were then treated with M.O.M. biotin-labeled anti-mouse IgG, diluted 1:250 in M.O.M. diluent. Following another 3 × 5 min of TBST washes, slides were incubated for exactly 45 min in Cy3-conjugated streptavidin diluted 1:800 in PBS (Amersham Biosciences, Piscataway, New Jersey, United States) in the dark. To visualize nuclei, slides were counterstained with 0.2 μg/ml DAPI. Ki67-positive cells were visualized by fluorescence microscopy. Western blot analysis Western analysis was performed using conventional techniques. Liver tissues were disrupted and protein was isolated using a pestle and tube homogenizer in NP-40 lysis buffer. Equal protein was loaded in each lane, as quantitated by the Bicinchoninic Acid (BCA) Protein Assay (Pierce, Rockford, Illinois, United States). Proteins were electrophoresed on 10% Tris-HCl polyacrylamide gels at 100 V for 60 min and transferred on PVDF membranes at 100 V for 60 min. The membrane was blocked in 5% nonfat dry milk solution in TBS at 4 °C overnight. MYC protein expression was detected using the C-19 rabbit polyclonal antibody that recognizes mouse and human MYC (Santa Cruz Biotechnology, Santa Cruz, California, United States). p53 protein expression was detected using the NCL-p53-CM5p rabbit polyclonal antibody (Vector Laboratories). As a positive control, we used a hematopoietic tumor previously shown to overexpress p53 that was generously provided by Dr. Kevin Smith. Cell and nucleus size measurements Images of H&E-stained liver sections were made with a Nikon Eclipse E800 microscope utilizing a Spot RT Slider digital camera (Diagnostic Instruments, Sterling Heights, Michigan, United States) and Spot Advanced Software (version 3.2.4, Diagnostic Instruments). To estimate cellular and nuclear volume, the radii of at least five hepatocytes per field were measured in at least three fields. Hepatocyte isolation Hepatocytes were collected by a two-step in situ perfusion technique ( Seglen 1976 ; Bumgardner et al. 1990 ). First, the inferior vena cava was cannulated and the liver perfused with calcium-free EGTA buffer followed by calcium-containing collagenase buffer (Invitrogen, Carlsbad, California, United States). After the perfusion, the liver was excised and mechanically disrupted in Williams' E medium (Sigma-Aldrich, St. Louis, Missouri, United States). The resulting slurry was filtered through a 40-μm filter, and viable hepatocytes were isolated by Percoll gradient centrifugation. The pelleted hepatocytes were washed serially with Williams' E medium and counted prior to further analyses. Cell size measurements by FACS analysis Cell size was determined by FACS forward versus light scatter of isolated hepatocytes utilizing a Becton Dickinson FACSCaliber (BD Biosciences). Data were analyzed using Cellquest v3.3 Software (BD Biosciences). DNA content Nuclei were prepared for staining by touching a cut piece of liver to superfrost slides. The smears were then air dried, fixed in formalin for 5 min, and stored in 70% ethanol. The nuclei were stained for DNA content analysis according to the Feulgen technique ( Oppedal et al. 1988 ). Apoptosis assay Apoptosis was detected using terminal deoxyribonucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) staining, using the In Situ Death Detection Kit (Boehringer Mannheim, Indianapolis, Indiana, United States). In order to visualize the nuclei, cells were counterstained with DAPI (0.2 μg/ml). TUNEL-positive cells were visualized by fluorescence microscopy. Northern blot analysis Northern blotting and probing were performed using standard methods. RNA samples were isolated according to the Trizol product manual specifications using a Kontes 1-ml glass tissue homogenizer. A formaldehyde, 1% agarose gel was used to run the Northerns, and transferring was done overnight in 20× SSC. Blots were washed in 2× SSC, cross-linked twice in a Stratalinker UV source, and pre-hybed and hybridized using the UltraHyb (Ambion, Austin, Texas, United States) product specifications. cDNAs corresponding to p21, MDM2, and glyceraldehye-3-phosphate dehydrogenase (GAPDH) were used as probes ( Macleod et al. 1995 ; el-Deiry 1998 ). The probes were generated through random priming reactions. Kodak BioMax MS film was used to expose the blots. Loss of p53 heterozygosity analysis by PCR To evaluate loss of heterozygosity of liver tumors derived from mice heterozygous for p53 deletion, PCR analysis was performed as previously described ( Timme and Thompson 1994 ). RNA isolation and quantification Total cellular RNA was isolated from snap-frozen liver tissue using the Invitrogen Micro-to-Midi Total RNA Purification System according to the manufacturer's instructions. The amount of total RNA isolated from tissues was quantified using spectrophotometric OD 260 measurements. The quality of RNA was measured on a formaldehyde, 1% agarose gel. Real-time PCR The RNA of human c- MYC and rodent GAPDH were measured by real-time quantitative RT-PCR using the 5′ nuclease technology on an ABI PRISM 7900HT Sequence Detection System (Applied Biosystems, Foster City, California, United States). The probe sequences for human c-MYC were forward primer 5′-CCCCTGGTGCTCCATGAG-3′ and reverse primer 5′-GCCTGCCTCTTTCCACAGA-3′. The human c- MYC probe, 5′-TCCTCCTCAGAGTCGC-3′, was labeled with FAM dye-MGB. A VIC TaqMan rodent GAPDH control reagents kit (Applied Biosystems) was used to measure mouse GAPDH. The RNA was reverse transcribed using the High-Capacity cDNA Archive Kit (Applied Biosystems) according to the manufacturer's protocol with a minor modification, the addition of RNase inhibitor (Applied Biosystems) at a final concentration of 1 U/μl. Samples were incubated at 25 °C for 10 min and 37 °C for 180 min. PCR reactions were prepared in a final volume of 20 μl, with final concentrations of 1× TaqMan Universal PCR Master Mix (Applied Biosystems) and cDNA derived from 20 ng of input RNA as determined by spectrophotometric OD 260 measurements. Thermal cycling conditions comprised an initial UNG incubation at 50 °C for 2 min, AmpliTaqGold DNA polymerase activation at 95 °C for 10 min, 40 cycles of denaturation at 95 °C for 15 s, and annealing and extension at 60 °C for 1 min. Each measurement was performed in triplicate and the threshold cycle (C t ), the fractional cycle number at which the amount of amplified target reached a fixed threshold, was determined. For calibration and generation of standard curves, we used cDNA prepared from Universal Human Reference RNA (Stratagene, La Jolla, California, United States) and cDNA prepared from Universal Mouse Reference RNA (Stratagene). Universal Human Reference RNA was used for human c-MYC and Universal Mouse Reference RNA was used for rodent GAPDH. Supporting Information Accession Numbers Swiss-Prot accession numbers ( http://us.expasy.org/sprot/ ) for the loci discussed in this paper are the following: Cdkn1a (mouse), P39689; Mdm2 (mouse), P23804; Myc (human), P01106; Myc (mouse), P01108; and Tp53 (mouse), P02340.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC519000.xml

544850

The quantal theory of how the immune system discriminates between "self and non-self"

In the past 50 years, immunologists have accumulated an amazing amount of information as to how the immune system functions. However, one of the most fundamental aspects of immunity, how the immune system discriminates between self vs. non-self, still remains an enigma. Any attempt to explain this most intriguing and fundamental characteristic must account for this decision at the level of the whole immune system, but as well, at the level of the individual cells making up the immune system. Moreover, it must provide for a molecular explanation as to how and why the cells behave as they do. The "Quantal Theory", proposed herein, is based upon the "Clonal Selection Theory", first proposed by Sir McFarland Burnet in 1955, in which he explained the remarkable specificity as well as diversity of recognition of everything foreign in the environment. The "Quantal Theory" is built upon Burnet's premise that after antigen selection of cell clones, a proliferative expansion of the selected cells ensues. Furthermore, it is derived from experiments which indicate that the proliferation of antigen-selected cell clones is determined by a quantal, " all-or-none ", decision promulgated by a critical number of cellular receptors triggered by the T Cell Growth Factor (TCGF), interleukin 2 (IL2). An extraordinary number of experiments reported especially in the past 20 years, and detailed herein, indicate that the T cell Antigen Receptor (TCR) behaves similarly, and also that there are several critical numbers of triggered TCRs that determine different fates of the T cells. Moreover, the fates of the cells appear ultimately to be determined by the TCR triggering of the IL2 and IL2 receptor (IL2R) genes, which are also expressed in a very quantal fashion. The "Quantal Theory" states that the fundamental decisions of the T cell immune system are dependent upon the cells receiving a critical number of triggered TCRs and IL2Rs and that the cells respond in an all-or-none fashion . The "Quantal Theory" accounts fully for the development of T cells in the thymus, and such fundamental cellular fates as both "positive" and "negative" selection, as well as the decision to differentiate into a "Regulatory T cell" (T-Reg). In the periphery, the "Quantal Theory" accounts for the decision to proliferate or not in response to the presence of an antigen, either non-self or self, or to differentiate into a T-Reg. Since the immune system discriminates between self and non-self antigens by the accumulated number of triggered TCRs and IL2Rs, therapeutic manipulation of the determinants of these quantal decisions should permit new approaches to either enhance or dampen antigen-specific immune responses.

Introduction Perhaps one of the most unique and fundamental aspects of the immune system is the clonal nature of the response to the introduction of antigen. The "Clonal Selection Theory" as originally formulated by Burnet stated that the immune system is made up of cells, each of which are only capable of reacting with a single antigenic molecule[ 1 ]. Thus, Burnet improved upon the "Natural-selection theory of antibody formation" proposed by Neils Jerne[ 2 ], by giving the immune response a cellular basis. Also, Burnet introduced the notion that after antigen selection, the reactive cell clones proliferate, and the resulting expanded population of cells would only then be capable of removing the offending foreign antigen. Consequently, one of the most crucial decisions required of an antigen-selected cell is whether to undergo cell cycle progression. Ultimately this decision determines one of the other most fundamental characteristics of the immune system, the ability to discriminate between self vs. non-self. At the time that Burnet formulated the concept of clonal selection in the mid 1950s, the identity of the cells comprising the immune system was unknown. Plasma cells had been found to be the source of antibodies[ 3 ], but lymphocytes had not yet been identified as the precursors of plasma cells. Moreover, thymic-derived lymphocytes (T cells) and bone marrow-derived lymphocytes (B cells) were not to be discovered for almost two more decades [ 4 - 6 ]. In the intervening 50 years since Burnet formulated his theory, much has been discovered and even more has been proposed to explain how the immune system functions, especially how the discrimination between self vs. non-self is made. Since 1959, several modifications of Burnet's original model have been offered, each of which introduced additional cells in an attempt to explain how the entire system could react with absolutely everything in the environment, but not react with any molecules comprising self [ 7 - 15 ]. However, none of the models proposed thus far have focused on one of the most fundamental aspects of the Clonal Selection Theory as originally formulated by Burnet, i.e. the molecular forces driving the proliferative expansion of the antigen-selected cell clones. We now know that lymphocytes make up the immune system, and we know the molecular structures of the antigen receptors expressed by both T cells and B cells. Consequently we know that Burnet was largely correct. Each cell expresses a unique antigen receptor that has the capacity to bind only a single antigenic epitope. The exception to this rule is the "allelic inclusion" of two T Cell Receptor (TCR) α-chains, resulting in approximately 30% of α/β-chain bearing T cells as having dual antigenic specificity. As well, it is known that T cells recognize epitopes comprised of short peptides of only a few amino acid residues bound to molecules encoded by the Major Histocompatibility gene Complex (MHC), while B cells recognize both the tertiary surface structure of larger molecules, as well as linear determinants of molecules. Despite this information, it still remains unknown how the immune system discriminates between non-self vs. self-molecules, in that the molecular nature of the T cell Antigen Receptors (TCR) and the B cell Antigen Receptors (BCR) that recognize both self- and non-self- epitopes are identical. Moreover, the molecular natures of self-epitopes and non-self -epitopes are identical. Accordingly, it is even more perplexing as to how the immune system manages this discrimination. One key to understanding the way in which the immune system operates is the observation that the capacity to recognize and respond to antigen is dependent on the dose of antigen introduced[ 16 ]. Thus, there appears to be several outcomes possible, such that at low antigen doses there is no detectable response, and with increasing doses of antigen there may be the induction of tolerance, while even higher doses are necessary to trigger an immune response, which involves more and more antigen-reactive cells as the antigen dose increases. At very high antigen doses there may even be a "paralysis" induced. Another key to understanding the immune response resides in the realization that the absolute frequency of potential antigen-reactive lymphocytes is very low before the introduction of antigen, on the order of 1 in a million cells up to 1 in 10,000 cells (i.e. 10 -6 to 10 -4 ). However, after antigen selection, proliferative clonal expansion increases the frequency of antigen-reactive cells to as high as 1 in 10 cells, an astonishing 1–100,000-fold increase [ 17 - 20 ]. Thus, the decision by individual cells to proliferate in response to recognition of an antigen is the critical decision at the cellular level that controls the ability of the whole immune system to discriminate between self vs. non-self. Until the discovery of mitogenic cytokines, it was assumed that antigens are solely responsible for stimulating proliferation. It is now known that there are antigen-activated cytokines with T cell Growth Factor (TCGF) activity that provide molecular signals that markedly stimulate cell cycle progression. Since the principle cytokine with TCGF activity driving T cell proliferation is the interleukin-2 (IL2) molecule[ 21 , 22 ], the molecular mechanism whereby IL2 promotes cell cycle progression is of utmost importance. Accordingly, to understand how the immune system discriminates between self vs. non-self antigens, the IL2 molecule is first examined, particularly how IL2 promotes the proliferation of antigen-selected cells. Then, the cellular and molecular determinants of IL2 production and IL2R expression are traced. Ultimately, the maturation of T cells in the thymus must be considered, to understand how cells determine whether to produce IL2 or not, and whether to respond to it or not. Most of the discussion that follows is focused on T cells, but the general principles developed apply to B cells as well. The Quantal Nature of IL2-Promoted T Cell Proliferation At the level of the individual cell, the proliferative response to IL2 is quantal , i.e. it is "all-or-none"[ 23 , 24 ]. However, to proceed beyond simply a description of this phenomenon, one must understand the molecular basis for the response of individual cells to IL2, so as to predict the behavior of the immune system as a whole. Soon after the development of the radiolabeled-IL2 binding assay[ 25 ], which permitted IL2 receptors to be quantified and defined for the first time, experiments could be performed to ascertain how the concentrations of IL2 that bind to IL2Rs compare with the IL2 concentrations that promote T cell proliferation. It was revealed that the binding and biological response curves are coincident, as shown in Figure 1 [ 25 ]. The IL2 concentration-dependent response ranges from 1–100 pM, and the 50% effective concentration (EC 50 ) equals the equilibrium dissociation constant (Kd) of the IL2-IL2R interaction, both of which are ~ 10 pM. Although this is a very high affinity for a ligand-receptor interaction (e.g. most TCR-peptide-MHC affinities are ~ a million-fold lower), it is noteworthy that in molecular terms, the EC 50 /Kd = 6 billion molecules/mL. Figure 1 The IL2 binding and biological response curves are coincident. Radiolabeled IL2 and purified homogeneous IL2 were used in parallel experiments with the same IL2R+ T cell population to determine the relationship between IL2 binding and IL2-promoted T cell proliferation as monitored by 3 H-TdR incorporation. From reference 25. Of course, these experiments were performed using cell populations , and symmetrically sigmoid log-dose response curves are quite familiar in these circumstances. However, from these experiments it was not possible to discern whether the low thymidine incorporation found at low IL2 concentrations was due to all of the cells in population incorporating only a little thymidine, or whether only a few of the cells could respond at low IL2 concentrations. Therefore, it was not until the IL2 dose-response relationship could be examined at the single cell level did it become apparent that the cell populations are comprised of individual cells that differ markedly as to their responsiveness to the mitogenic ligand. Thus, as shown in Figure 2 using propidium iodide staining of DNA and the flow cytometer to compare with thymidine incorporation, it was found that some cells of an asynchronously proliferating population respond by proliferating to very low IL2 concentrations, e.g. only 1 pM, while others need 100-fold higher IL2 concentrations. Moreover, the marked heterogeneity in IL2 responsiveness could not be explained on a genetic basis, since even cloned cell populations behaved in an identical fashion. Figure 2 The IL2 biological dose-response relationship is determined at the single cell level by a marked heterogeneity of responsiveness. Asynchronously proliferating IL2R+ cells were exposed to varying concentrations of IL2 for 18 hours. Subsequently, cell aliquots were either pulsed for 4 hours with 3 H-TdR, or stained with propidium iodide prior to single cell analysis by flow cytometry. The cell population incorporates 3 H-TdR in an IL2 concentration dependent manner, and the amount of 3 H-TdR incorporation at each IL2 concentration is determined by the absolute number of cells that have entered S-phase, as indicated by the single cell analysis by propidium iodide staining. From reference 26. Once it is realized that the effective IL2 concentrations span 2 orders of magnitude, the question becomes why some cells are capable of responding at only 1 pM, while others require IL2 concentrations that are 100-fold higher. The only logical answer to this question is that there must be intrinsic cellular differences, and that these differences are manifest in molecules that are critical for signaling cell cycle progression. Therefore, in experiments focused on understanding how IL2 promotes T cell proliferation after antigen activation, we found that in addition to the affinity of the IL2/IL2R interaction, and the concentration of IL2, the other variables involved are the IL2 receptor (IL2R) density, and the duration that the IL2 and the IL2R molecules interact[ 26 ]. In order to design experiments to approach these variables, it was necessary to use cell populations that were synchronized in early G 1 , so as to follow the rate at which individual cells progressed through G 1 into S-phase. Thus, if G 0/1 synchronized IL2R+ cells are exposed to two different IL2 concentrations, one receptor saturating and another only half-saturating; the receptor-saturating concentration promotes cell cycle progression twice as rapidly as the half-saturating IL2 concentration. However, as the ligand concentration dependency of cell cycle progression of cell populations was well known, these results were not surprising. However, it was surprising to find that if one exposes two cell populations, one with a higher IL2R density than another, to the same receptor saturating IL2 concentration, the cell population with the higher IL2R density traverses G 1 and enters S-phase more rapidly than the population with the lower IL2R density (Figure 3 ). Figure 3 The effect of IL2R density on the rate of T cell cycle progression. Two G 0/1 synchronized T cell populations that differed 3-fold in mean IL2R density were exposed to an IL2R saturating concentration (250 pM) for 48 hours and 3 H-TdR incorporation was monitored as indicated at 1 hour intervals. The cells with the higher IL2R density (solid circles) entered S-phase before the cell population with the lower IL2R density (solid triangles). From reference 26. These observations predict that the basis for the characteristic sigmoid IL2 log-dose response curve (Figure 1 ) is only explicable because of heterogeneity of IL2R density within a given population of T cells. The heterogeneity of IL2R density/cell is readily appreciated from examination of the flow cytometry plot of the log-normal distribution of IL2Rs as shown in Figure 4 . Thus, at low IL2 concentrations, i.e. ~ 1 pM, only cells with the highest IL2R density are capable of responding. As the cell population is exposed to increasingly higher IL2 concentrations, cells with lower IL2R densities will meet the quantal requirement to enter the cell cycle. Finally, as the IL2 concentration reaches levels that saturate all IL2Rs, even cells with the lowest IL2R density can reach the critical number. Figure 4 IL2R density determined at the single cell level by flow cytometry. IL2R+ T cells were labeled with anti-Tac (CD25) monoclonal antibody and analyzed by single cell flow cytometry. The IL2R density varies among cells within he population over 3 orders of magnitude. From reference 26. These findings indicate that cells reach a decision to undergo cell cycle progression based on some critical number of IL2-IL2R intermolecular reactions at the cell surface. Also, they indicate that the duration of the IL2-IL2R interaction plays a role, such that it appears that if a cell has a low density of IL2Rs, the critical number of IL2-IL2R interactions can still be reached, but a longer time interval is necessary. Also, the data indicate that if one interrupts the IL2-IL2R interaction before the critical number of interactions is attained, then cell cycle progression will not occur, as shown for a 3-hour exposure in Figure 5 . In other words, the cell " counts " the number of triggered receptors and waits until the requisite number has accumulated before proceeding beyond G 1 to S-phase[ 24 , 27 ]. Figure 5 The effect of varying the IL2 exposure period on the proliferative response of G 0/1 synchronized IL2R+ T cells. Aliquots of synchronized cells were exposed to IL2 for varying intervals (3, 6, 11, 26 hours) then washed and placed into culture without IL2 and pulsed with 3 H-TdR. Symbols: IL2 exposure, 3 hr (solid circles), 6 hr (open circles), 11 hr (solid triangles), and 26 hr (open triangles). Inset, shows the 3 H-TdR incorporation of each cell population in response to an IL2R saturating IL2 concentration (250 pM) monitored for 1 hr at the times indicated. From reference 26. IL2/IL2R binding is rapid and comes to steady state within 10 minutes[ 25 , 28 ], yet several hours of IL2 exposure are necessary to reach the critical number of IL2/IL2R interactions required to trigger cell cycle progression. Therefore, it must be concluded that the requisite number of IL2Rs is not present on any cells before the addition of IL2. Instead, new receptors must be continuously synthesized, expressed on the cell surface and serially engaged over several hours to finally reach the quantal number. Accordingly, the decision to divide is regulated with exquisite high fidelity, and the decision must be quantal. Otherwise a cell might only partially replicate its DNA before undergoing cytokinesis, a situation clearly incompatible with life. An estimate of this critical number of IL2/IL2R interactions required can be calculated[ 24 ], knowing the initial mean number of receptors (~ 750 Rs/cell), and the rate of internalization and degradation of IL2 bound IL2Rs from the cell surface at steady state (t 1/2 = 15 minutes). Thus, the rate constant, κ = ln2/15 min = 4.67 × 10 -2 min -1 , governs the rate of new receptor synthesis necessary to maintain the surface expression at steady state. Moreover, 11 hours (660 minutes) of IL2 exposure is necessary to trigger 50% of the cells within the population to undergo cell cycle progression (Figure 5 ). Thus, the mean number of triggered IL2Rs necessary is: R# (R/cell) = κ × R# @ steady state = 4.67 × 10 -2 min -1 × 750 R/cell = 35 R/cell/min × 660 min = 23,100 triggered R/cell It follows that if the mean number of IL2Rs at steady state is lower, e.g. only 375 Rs/cell, 22 hours would be required to reach the quantal number and if the rate of new receptor synthesis is doubled, maintaining twice as many, 1,500 sites/cell, then only 5.5 hours would be required to trigger the cells. Intracellular Sensors of the Extracellular Signals There are 3 chains that make up the trimeric IL2R that has an extremely high affinity for IL2, termed α (CD25)[ 29 ], β (CD122)[ 30 ] and γ (CD132)[ 31 ]. Careful kinetic and equilibrium binding experiments using radiolabeled IL2 revealed that the α chain contributes a very rapid association rate (κ = 10 7 M -1 sec -1 ), while the β chain contributes a slow dissociation rate (κ' = 10 -4 sec -1 ), which together yield the high affinity (Kd = κ'/κ = 10 -11 M) for the ligand observed at steady state[ 28 ]. Recent experiments focused on the energetics of assembly of IL2/IL2R signaling complexes have revealed that in solution, the IL2R α and β chains bind to one another, whereas the α chain does not bind to the γ chain[ 32 ]. Moreover, the γ chain can only bind to α,β dimers or isolated β chains if IL2 has already bound to these receptor chains. These data support the interpretation that the α,β dimer probably is formed on the surface of antigen-activated T cells, and serves as the initial receptor complex capable of binding IL2. Moreover, the 10–20-fold excess expression of the α chain vs. the β chain favors the formation of the α,β heterodimer by the law of mass action. Subsequently, the IL2, α,β trimeric complex then can attract and bind the γ chain, forming a quaternary complex that is capable of signaling the cell interior. These energetic experiments provide an explanation for the IL2-dependency of signaling, since signaling only occurs when the γ-chain and the β chain are brought into close proximity, thereby activating the tyrosine-specific kinases, JAK 1 and 3, which are already bound to the β and γ chains respectively[ 33 ]. Accordingly, the assembly and maintenance of this energetically stable multicomponent macromolecular signaling complex is a fundamental requirement for the cell ultimately to make the quantal decision to divide. The important downstream events in IL2-promoted T cell cycle progression are the JAK-dependent activation of at least two distinct proliferative signaling pathways. One is mediated by the transcription factor STAT5, while the other is mediated by the adapter molecule Shc, which activates phosphatidylinositol 3-kinase (PI3K). Both cyclin D2 and D3 are expressed in response to IL2R triggering[ 34 ], and recent studies have shown that the STAT5 and PI3K pathways play distinct, but coordinated roles in the quantal IL2R induction of progression through the Restriction Point (R-point) in the G 1 phase of the cell cycle [ 35 - 37 ]. In addition to Shc, STAT5 also facilitates the activation of the PI3K pathway by a delayed mechanism that requires protein synthesis, and PI3K activity is essential for the induction of cyclin D2 expression by STAT5. PI3K activity is required for the optimal binding of RNA polymerase II to the promoters of cyclin D2 as well as other IL2/STAT5-induced genes. Because of these findings, it has been proposed that the D cyclins serve as intracellular sensors of the extracellular signals [ 38 ] generated at the cell surface by the formation of the stable quaternary IL2/IL2R signaling complex[ 39 ]. Thus, cyclin D2 and D3 complex with the cyclin-dependent kinases (cdk) 4 & 6 and the p27 protein, thereby forming active kinases that initiate phosphorylation of the Rb proteins that repress the E2F transcription factors. The E2Fs are already bound to response elements that regulate the expression of multiple genes, the expression of each of which is critical for both nucleotide synthesis and DNA replication. Also, the E2F transcription factors regulate the expression of multiple genes required for formation of the Pre-replication Complexes (Pre-RCs), which must first assemble and then disassemble at sites on DNA termed Origins of Replication (Ori) before the DNA strands can separate, thereby allowing DNA duplication. Accordingly, the cyclin D/cdk/p27-dependent phosphorylation of Rb has been proposed to initiate the passage through the R-point, and has been described as a quantal molecular "binary switch"[ 39 , 40 ]. However, recent experiments with mice that have had all 3 of the cyclin D genes deleted have revealed that although hematopoiesis is dependent on the coordinated expression of the cyclin D genes, nonhematopoietic cells can proliferate in the absence of the D-type cyclins and their cyclin-dependent kinases 4 and 6 (cdk4/6)[ 41 , 42 ]. Even so, mouse embryo fibroblasts lacking type D cyclins proliferate more slowly to stimulation by serum by comparison to their wild type counterparts. Therefore, it appears that there are at least 2 distinct pathways whereby extracellular signals can trigger G 1 progression, one involving cyclin D and another that is cyclin D independent. At this time, it remains unknown as to whether T cells can be stimulated to proliferate by both cyclin D-dependent and independent pathways, or only by cyclin D-dependent (and therefore IL2R/STAT5-dependant) pathways. However, it is clear that T cells are in G 0 until activated by initial signals received via the TCR, so that it still remains possible that T cells may be capable of proliferating in response to both TCR- and IL2R-derived signals[ 33 , 43 , 44 ]. Alternatively, the TCR may be responsible for the G 0 to G 1 transition, while the IL2R is responsible for G 1 progression to the R-point and S-phase transition. However, another possibility remains that it still may well be that TCR-derived signals can initiate early rounds of cell division, but that for a fully developed clonal expansion, IL2/STAT5-dependent signals are necessary. If so, one would predict that TCR-mediated cell cycle progression is STAT5- and cyclin D-independent, so that the role of IL2 is to markedly accelerate and extend cell cycle progression initiated by the TCR. The Quantal Regulation of IL2 Gene Expression Once one realizes that the quantal decision of T cells to proliferate is based upon a critical number of IL2-IL2R interactions, it becomes immediately obvious that the availability of IL2, together with the extent of IL2R expression, ultimately determines whether a immune response occurs that will be detectable at the systemic level. Since the discovery [ 45 - 51 ] and elucidation of the nature of the T Cell antigen Receptor (TCR) over the past 20 years [ 52 - 56 ], data have accumulated indicating that the regulation of IL2 gene expression and IL2R gene expression is under a tight and complex cell surface signaling mechanism that involves not only the TCR, but other surface molecules as well. Before the elucidation of the structure and function of the receptors involved in antigen recognition, it was difficult to envision how the IL2/IL2R system is regulated. However, the principles from the quantal IL2-IL2R signaling of cell cycle progression can now be extrapolated to this more complex signaling system, and herein resides the ultimate control of "self-non-self" recognition and response. Like the IL2R[ 24 , 27 ], the TCR also is capable of counting the number of antigen interactions so as to acquire the critical number of triggered receptors necessary for IL2 gene expression. Since the cloning and sequencing of the IL2 cDNA[ 57 ] and gene[ 58 ], detailed studies have revealed the nature of the molecules controlling IL2 expression. There are 3 distinct response elements in the promoter region of the IL2 gene that bind members of distinct families of transcriptional activating factors [ 59 - 61 ]. These factors include Activating Protein-1 (AP-1), Nuclear Factor of Activated T cells (NF-AT), and the Nuclear Factor kappa B/Rel (NF-κB). Individual IL2 transcription factors from these three families cannot bind stably to their target DNA response elements in vivo without coengagement of each of the distinct factors that bind at neighboring sites[ 62 ]. Also, if the members of any one of these factors is prevented from binding to the IL2 promoter region, there is a marked attenuation of IL2 gene transcription[ 63 ]. Moreover, even after the factors have bound, inactivation of any of the three transcription factors pharmacologically extinguishes the binding of all three factors, thereby aborting transcription[ 64 ]. Therefore it has been proposed that there is a nonhierarchical, cooperative enhancement of binding at the IL2 gene locus, and that this binding and transcriptional activation of IL2 gene expression is consequently, quantal. The quantal binding of IL2 transcription factors to the IL2 enhancer and promotion of IL2 gene expression has been extended by studies on the allelic expression of the IL2 genes. Under optimal TCR stimulating conditions, where antigen is in excess, and Antigen Presenting Cells (APCs) are not limiting, there is biallelic expression of the IL2 genes[ 65 ]. By comparison, when TCR stimulatory conditions are suboptimal, then expression can be monoallelic, and as well, fewer cells will be activated to express IL2. However, once engaged, the rate of IL2 production per cell remains constant. In this respect, IL2 gene expression per cell is always quantal. The activation of NF-AT, AP-1 and NF-κB/Rel family members is controlled by signaling pathways triggered by the TCR, as well as costimulatory molecules and coinhibitory molecules. Thus, TCR triggering promotes the rapid influx of calcium into the cell, which activates the phosphatase calcineurin, thereby dephosphorylating and activating NF-AT[ 66 ], while other TCR-triggered pathways activate the kinases p56 lck , ZAP-70, PLCγ, and Protein Kinase C-θ (PKC-θ), which simultaneously promote the activation and translocation of AP-1 to the nucleus, and the activation of NF-κB/Rel family members that are already present in the cytoplasm bound to the Inhibitors of κB (IκB)[ 67 ]. In addition, stimulation of co-stimulatory receptors, particularly CD28, by the B7 ligands expressed by APCs activates PI3K, thereby further activating both AP-1 and NF-κB/Rel members [ 68 - 70 ]. The Phenomenon of Anergy ("Abnormal Inactivity") Soon after the derivation of the first T cell clones[ 71 ], experiments with cloned helper T cells revealed that high concentrations of specific antigenic peptide (i.e. 1–100 μM) would lead to unresponsiveness, i.e the incapacity to produce IL2 or to proliferate when subsequently exposed to a stimulatory concentration of antigenic peptide (i.e.1 nM-1 μM)[ 72 ]. The suppressive effect was peptide and clone specific, took several hours to develop, and was long lasting, up to 7 days in vitro . It could not be ascribed to nonspecific toxicity and cell death, in that the cells were still capable of proliferating in response to IL2 added exogenously. Subsequently, anergy (defined as a state of proliferative unresponsiveness to normal mitogenic activation) was shown to be produced by delivery of "Signal-1" (i.e. TCR), without "Signal-2" (i.e. CD28)[ 11 ]. More recently, it has been shown that it is possible to induce anergy pharmacologically by stimulating calcium flux using calcium ionophores without activating the TCR or CD28[ 66 , 73 , 74 ]. In this instance, NF-AT is activated and translocates to the nucleus in the absence of activation of AP-1 and NF-κB/Rel. Activation of NF-AT without the participation of AP-1 and NF-κB/Rel results in the proteolytic degradation of PLCγ and PKC-θ [ 74 ], as well as the transcriptional activation of a unique set of genes that collectively suppress the capacity of the cell to respond to subsequent full TCR/costimulatory activation[ 73 ]. This induced "anergic state" is stable and is manifest by the inability to maintain a stable immunologic synapse, which precludes expression of the IL2 gene, so that anergic cells will not proliferate in response to a subsequent full TCR/CD28 stimulation. However, like the high antigen dose anergy, if IL2 is supplied exogenously, the proliferative block is bypassed, so that the anergized cells are still capable of proliferating in response to the IL2 signals. These findings are reinforced by other experiments, which reveal that inhibitors of calcineurin, such as cyclosporine-A and tacrolimus (FK506), which prevent the activation of NF-AT, also prevent the induction of anergy[ 75 ]. As well, T cells from NFAT-1 (-/-) mice are resistant to anergy induction by calcium ionophores[ 76 , 77 ]. Moreover, NF-AT-1 (-/-) mice exhibit a syndrome characterized by the accumulation of hyperactivated T cells[ 76 , 77 ]. Thus, in situations where calcium-mediated activation of calcineurin and NF-AT predominates, and pairing with AP-1 and NF-κB/Rel transcription factors does not occur, such as in situations with little of no costimulatory activation, a biochemical milieu exists that favors the creation of an anergic state. By comparison, NF-κB/Rel appears to be the most critical of the three families of transcription factors involved in the activation of IL2 gene expression[ 78 ]. Moreover, of the 5 members of the NF-κB/Rel family, c-Rel is the most important, and also the critical transcription factor activated by costimulatory signals[ 79 ]. C-Rel expression is restricted to cells of the lymphoid and myeloid lineages, whereas the other NF-κB family members are expressed ubiquitously in almost all tissues. T cells from c-Rel (-/-) mice cannot express the IL2 gene or proliferate in response to activation via full TCR/co-stimulation. However, they can proliferate normally if IL2 is supplied exogenously[ 79 ]. As well, c-Rel binds to the costimulatory response element IL2 CD28RE with a high affinity (Kd = 25 nM), while NF-κB/Rel p50/p65 heterodimers bind to this response element with a 10-fold lower affinity[ 80 ]. Accordingly, for productive CD4+ T cell activation manifest by IL2 gene expression and proliferative clonal expansion, the minimum requirements are optimal activation of the TCR via peptide-MHC complexes, and costimulation via activation of CD28 by the APC B7 ligands. In the case of a low affinity TCR-MHC-peptide interaction, or in the absence of costimulation, NF-AT activation may predominate and anergy can result[ 81 ]. The actual critical number of TCR/CD28 activating signals that result in the quantal expression of the IL2 gene have not been determined, but by extrapolation from the parameters regulating IL2-IL2R activation, it is logical that peptide-MHC ligand concentration, TCR receptor density and affinity, as well as duration of the ligand-receptor interaction, will dictate the number of triggered TCR/CD28 interactions, which determine whether the cell becomes anergized vs. activated to produce IL2, to express IL2Rs and to proliferate. The Regulation of IL2R Gene Expression Resting T cells that have not been recently activated via TCR/CD28 do not express detectable high affinity IL2Rs[ 25 ]. A recent study carefully examined resting T cells isolated from human peripheral blood for expression of the 3 chains of the IL2R by flow cytometry[ 82 ]. To ensure that the T cells represented only unactivated, truly "resting" T cells, any in vivo activated cells were removed using monoclonal antibodies reactive with the transferrin receptor (CD71), known to be an early TCR/CD28 activation molecule. Both CD4+ and CD8+ resting T cells have undetectable surface or cytoplasmic IL2R α and β chains, as monitored using very sensitive flow cytometry methods. By comparison, IL2R γ chains are detectable in the cytoplasm, but undetectable on the cell surface[ 83 ]. Upon activation via the TCR/CD28, the expression of the genes encoding both the IL2R α and β chains occurs, and γ chains are rapidly mobilized to the cell surface, so that high affinity trimeric IL2Rs are expressed, and the cells are competent to respond to IL2 by proliferating. The regulation of α chain (CD25) gene expression is under the control of the TCR/CD28 via activation of NF-κB/Rel, AP-1 and NFAT, which interact with 2 distinct REs [ 84 ]. Therefore, it appears that the IL2Rα chain gene is coordinately regulated along with the IL2 gene by the same signaling pathways emanating from the TCR/CD28 receptors that activate the same families of transcription factors regulating the IL2 gene. However, it has not been determined whether the critical number of TCR/CD28 receptors necessary to trigger the IL2 gene and the IL2Rα chain gene are similar. Most experience suggests that there is a much lower number of triggered TCR/CD28 receptors regulating IL2Rα gene expression as compared with IL2 gene expression, but this question needs to be examined directly. In addition to the TCR, IL2 enhances IL2Rα chain gene expression as much as 10–20-fold [ 85 - 87 ]. This IL2 effect on α-chain expression is readily appreciated by flow cytometry, in that IL2 shifts the mean fluorescence intensity more than an order of magnitude. As well, it is noteworthy that the IL2Rα chain expression is ~ 10-fold higher than expression of either the IL2Rβ or the IL2Rγ chains, so that when using flow cytometry to detect each of the chains, the IL2Rα chain (CD25) is always predominant. As well, the number of functional high affinity trimeric IL2Rs is determined by the number of β and γ c chains, which are limiting. The energetics of IL2 chain assembly has now provided an explanation for the excess α chains, in that the law of mass action favors the formation of α,β heterodimers, to which IL2 binds and then recruits the γ chain, thereby forming the stable quaternary signaling complex. The In Vivo Functions of IL2 Before the identification of the IL2 molecule in 1983[ 57 , 88 ], it was assumed that antigens stimulate T cell proliferation, and that mitogenic cytokines, which were first described almost 20 years earlier[ 89 , 90 ], functioned simply to amplify the signals already initiated by antigen[ 22 ]. Thus, when purified IL2 became available[ 88 ] and the IL2 receptor (IL2R) had been discovered[ 25 ], experiments became possible for the first time to determine the molecular mechanism whereby antigen activated T cells are stimulated to proliferate. Initial studies revealed that although antigen is necessary to activate T cells to leave G 0 and to enter early G 1 , cell cycle progression through G 1 to S-phase and mitosis appeared to be mediated by IL2 upon binding to the IL2R[ 26 , 91 , 92 ]. Resting T cells were found to be IL2R negative and IL2 unresponsive[ 25 ], while purified, homogeneous IL2 was capable of promoting long-term T cell proliferation of mitogen- or antigen-activated T cells. By comparison, mitogen or antigen alone could not sustain long-term T cell growth[ 26 ]. Moreover, immunosuppressive pharmacological agents such as glucocorticoids[ 93 ] were found to inhibit T cell proliferation by preventing IL2 production but not IL2 responsiveness. As well, experiments with monoclonal antibodies that block either IL2[ 88 ] or the IL2R[ 29 ] were found to inhibit T cell cycle progression after mitogen or antigen activation. These experiments all suggested that antigen per se could not promote T cell proliferation, and suggested that IL2 drives T cell proliferation after the initial antigen activation. Even so, the monoclonal IL2- or IL2R-reactive antibodies suppressed proliferation by > 90–95%, but never completely abrogated proliferation, leaving open the possibility that either the TCR itself or other mitogenic cytokines might also be operative. In 1991, with the advent of IL2 gene deletion through genetic recombination, it became possible to test definitively the functional importance of IL2, both in vitro and in vivo , at least in mice. Initial in vitro experiments testing cells from IL2 (-/-) mice for proliferation in response to activation by the mitogenic lectin Concanavalin-A (Con-A) revealed a 70–75% diminution of tritiated thymidine incorporation, but not a complete abrogation[ 94 ]. Therefore, these experiments suggested that perhaps the TCR really was capable of promoting proliferation, and that IL2 functioned to merely amplify the response, as assumed originally. Alternatively, it was also considered possible that perhaps IL2 was not the only cytokine with TCGF activity operative to promote T cell cycle progression, and additional cytokines such as IL4[ 95 ] or IL7[ 96 ], which were thought originally to be primarily B cell stimulators, might also be playing roles. Even so, it was still a great surprise when initial in vivo experiments with IL2 (-/-) mice infected with Vaccinia Virus (VV) and Lymphocytic Choriomeningitis Virus (LCMV), revealed that the generation of antigen-specific effector cytolytic T cell activity was reduced by only ~ 30% as monitored by Cytolytic T Lymphocyte (CTL) 51 Cr-release assays[ 97 ]. Moreover, neutralizing IgG antibody responses to Vesicular Stomatitis Virus (VSV) infection, a T-helper-dependent function, were delayed but not reduced. Other in vivo experiments with staphylococcal super antigens indicated that CD4+ T cells doubled normally, while CD8+ T cells from IL2 (-/-) mice were only ~ 50% of wild type[ 98 ]. These findings led to the interpretation that in vivo IL2 is redundant for the generation of immune responses, and that the TCR or other cytokines with TCGF activity could substitute for IL2. However, upon subsequent and more extensive testing of IL2 (-/-) mice, it was found that in the absence of IL2, the marked proliferative expansion of LCMV-induced CD8+ T cells was virtually eliminated, the total cytolytic effector capacity was reduced by > 90%, and IFN-γ production resulting from T cell activation was dramatically inhibited[ 99 , 100 ]. Moreover, IL2 (-/-) mice permitted prolonged viral replication compared with (+/+) and (+/-) controls, which could clear the virus within a few days. Therefore, all of these data indicated that IL2 may not be the sole cytokine with TCGF activity, but it is one of the principle TCGFs responsible for the maximal proliferation of antigen selected mature peripheral T cells, as well as their differentiation to effector cells, both in vitro and in vivo . Moreover, IL2 (-/-) mice are immunocompromised without IL2; thereby indicating that the TCR or other yet undiscovered cytokines cannot fully substitute for IL2 in vivo . The IL2 Deficiency Autoimmune Syndrome Since IL2 (-/-) mice are immunocompromised, it was entirely unexpected to find that a syndrome of lymphocyte hyperactivity and apparent autoimmunity appears as the mice mature beyond puberty [ 101 ]. Thus, although lymphocyte development during embryogenesis is grossly unperturbed by the absence of IL2, generalized lymphoid hyperplasia ensues after the first several weeks and months of life, and T cells that express activation markers accumulate in the secondary lymphoid tissues. As well, autoimmune antibody-mediated hemolytic anemia appears, in addition to antibodies reactive to self-molecules, such as DNA and other nuclear antigens. Similar findings with mice deleted of the IL2R α chain (CD25)[ 102 ], β chain (CD122)[ 103 ], the JAK3 protein kinase[ 104 , 105 ], and the transcription factors STAT5a/b[ 106 ] all supported the idea that IL2 or one of the other interleukins that signal via the IL2R γ c chain [ 107 ] somehow determines the selection of mature cells in the thymus, and in the absence of postnatal IL2 expression, the immune system begins to react to self as if it is nonself. The accumulation of T cells with an activated phenotype in the setting of an initial lymphopenia and immunodeficiency, has recently been attributed to compensatory over stimulation by the cytokines responsible for homeostatic proliferation, e.g. IL7, IL15 and IL21, all of which signal via the γc chain[ 108 ]. Of these cytokines, IL7 and IL15 signal via STAT5, whereas IL21 signals via STAT1 and STAT3. Accordingly, IL21 stimulation via STAT1&3 may well be responsible for the hypersensitivity lymphoproliferative syndrome that is common to the IL2, IL2R and signaling (-/-) phenotype. These observations in mice were reinforced by a report of a human homozygous mutation of CD25[ 109 ]. A male child of first cousin parentage presented at age 6-months with increased susceptibility to viral, bacterial, and fungal infections, suffering from cytomegalovirus pneumonitis, persistent oral thrush, candida esophagitis, and adenovirus gastroenteritis, chronic diarrhea, and failure to thrive. From the age of 8-months, lymphadenopathy and hepatosplenomegally became apparent. In vitro assays demonstrated a reduced responsiveness to stimulation by anti-CD3 (11% of control) and phytohemagglutinin (20% of control). Severe immunodeficiency was proven by the patient's inability to reject an allogeneic skin graft. Paradoxically, despite the obvious immunodeficiency, there was a normal sized thymus and lymphocytic infiltration of multiple tissues, including lung, liver, gut, soft tissue and bone. These findings were interpreted as possibly a result of a failure of negative selection of potential autoreactive cells in the thymus, as well as an inability to control autoreactive cells in the periphery, perhaps due to the absence of CD4+CD25+ Regulatory T cells. Regulatory T Cells (T-Regs) Soon after it was demonstrated that IL2 (-/-) and IL2Rα or β chain (-/-) mice develop an autoimmune syndrome, it was reported that immunocompromised ( nu/nu ) mice would also develop a wide spectrum of both organ-specific and systemic autoimmune diseases if they received normal cell populations from which CD4+CD25+ T cells were eliminated[ 110 ]. Furthermore, reconstitution of CD4+CD25+ T cells in the transferred cell populations prevented the development of autoimmunity. Subsequently, we found that IL2 treatment of IL2 (-/-) mice before day 10 after birth prevents the onset of the syndrome of lymphocyte activation and autoimmunity[ 111 ]. As well, thymocytes and spleen cells from IL2-treated IL2 (-/-) mice transferred to IL2 (-/-) recipients delayed the development of the autoimmune syndrome. These data suggested that IL2 treatment induced some normal cellular maturation/differentiation step in the transferred IL2 (-/-) cells that subsequently prevented the cells in the IL2 (-/-) mice from responding to self-antigens[ 111 ]. In other reports from as early as the 1960s[ 112 ], it was established that neonatal thymectomy in the first few days after birth can lead to subsequent widespread autoimmune phenomena, such as hemolytic anemia, thyroiditis, gastritis, oophoritis, or orchitis [ 113 - 115 ]. These two observations, i.e. autoimmune phenomena arising in both IL2 (-/-) mice and neonatal thymectomized mice, were connected when it was demonstrated that T cells expressing the IL2Rα chain (CD25) ontogenically begin to appear in the normal periphery immediately after day 3 of life, rapidly increasing within 2 weeks to adult levels, which comprise ~ 10% of CD3+ cells [ 116 ]. As well, neonatal thymectomy on day 3 eliminates CD25+ T cells from the periphery, and injection of CD25+ T cells from normal adult donors into day-3 neonatally thymectomized mice prevents the development of autoimmunity, while injection of CD25- T cells does not. These observations were interpreted as consistent with the notion that neonatal thymectomy on day 3 can eliminate or reduce the autoimmune preventative CD25+ T cells, thereby leading to unchecked activation of the self-reactive T cells produced before neonatal thymectomy. Together with the observations on the IL2 treatment of IL2 (-/-) mice[ 111 ], these experiments fix the source of these CD25+ cells within the thymus, and also imply that IL2 is a necessary component in their development. CD4+CD25+ "suppressor" regulatory T cells (T-Regs) could also be demonstrated in the secondary lymphoid organs of normal adult mice, and an in vitro assay was devised to test their suppressive activity[ 117 ]. CD4+CD25+ cells typically represent ~ 10%–15% of CD4+ T cells in lymph nodes from 8–10 week old mice. Paradoxically in view of their expression of CD25, these cells appear to be resting as well as anergic, in that they cannot proliferate in response to soluble and solid-phase anti-CD3 or Con-A. As well, even though the cells express the IL2R α chain, they cannot proliferate in response to exogenous IL2. However, these cells can be activated and proliferate in response to a combination of anti-CD3 + IL2. These data suggest that resting, anergic CD4+CD25+ T cells do not express the IL2R β and γ chains, but they can be induced to express them after activation of the TCR. When co-cultured with CD4+CD25- cells, these CD4+CD25+ cells were found to markedly suppress the proliferative response of the CD25- cells, provided they are stimulated with low concentrations of soluble, but not solid-phase anti-CD3. This inhibition was dependent on cell-cell contact, not soluble factors, and dependent on the suppression of IL2 production by the CD25- responding cells. The inhibition could be bypassed by the addition of IL2, or costimulation with anti-CD28. These data were interpreted as consistent with the idea that CD4+CD25+ cells in normal unimmunized animals represent a distinct lineage of "professional" suppressor cells that have matured in the thymus. These T-Reg cells have been termed "naturally occurring" (nT-Reg). It has also been reported that CD4+ T cells with regulatory function can be generated in vitro by the activation of mature peripheral CD4+CD25- T cells[ 118 ]. Thus, both regulatory and effector cells can, in principle, be generated from the same mature CD4+ T cell precursors. It has been postulated that these "inducible" cells (iT-Regs) might be triggered by "low-affinity or altered TCR signal transduction"[ 118 ], in that the conditions that favor the generation of T-Regs ex vivo from mature CD4+CD25- T cells, include antigen in the presence of immunosuppressive cytokines such as IL10 and TGFβ, immunosuppressive agents such as vitamin D3 and dexamethasone, CD40-CD40L blockade or immature DC populations[ 119 , 120 ]. Furthermore, it was reported recently that subcutaneous infusion of low doses of antigenic peptide by means of osmotic pumps over 14 days transforms mature peripheral T cells into CD4+CD25+ "suppressor cells" that can persist for long periods of time (i.e. several months) in the absence of antigen and confer specific immunological tolerance upon challenge with immunogenic doses of antigen[ 121 ]. Therefore, it appears that both in vitro and in vivo , low antigen concentrations can promote the differentiation of mature peripheral CD4+ T cells to express CD25, and to become "suppressor" cells rather than effector cells. The dependency on the thymus for maturation of CD4+CD25+ T-Regs, as well as the dependency upon IL2, has been underscored by experiments with IL2β chain (-/-) mice made transgenic for the IL2Rβ chain under the influence of the proximal lck promoter, so that mature trimeric IL2Rs capable of signaling are only expressed in the thymus[ 122 , 123 ]. Transgenic expression of the IL2Rβ chain in IL2Rβ chain (-/-) thymocytes corrects the lack of CD4+CD25+ peripheral T cells in IL2Rβ chain (-/-) mice and prevents lethal autoimmunity. These experiments further emphasize the unique contribution of IL2 to the development of CD4+CD25+ T-Regs, in that the other γ c chain cytokines, e.g. IL4 and IL7, IL9, and IL21 do not compensate for the lack of T-Regs in IL2Rβ chain (-/-) mice. Suboptimal antigen concentrations and IL2 are both required for the generation and activity of T-Regs, and both in the thymus and in the periphery. However, the forces governing the simple generation of anergic and nonfunctional CD4+CD25+ T cells, as compared with the determinants of the generation of activated proliferating, suppressive CD4+CD25+ "Regulatory T cells", vs. CD4+CD25+ activated proliferating functional effector T cells remain to be defined. Because each of these cell fates is functionally distinct, it is reasonable to hypothesize that these distinct cell fates are determined by different numbers of triggered TCRs and IL2Rs. The Transient Nature of the In Vitro T Cell Proliferative Response: Feedback Inhibition of IL2 Gene Expression Upon productive activation of T cells in vitro via TCR/CD28, there is a characteristic transient expression of the IL2 gene, such that IL2 mRNA first becomes detectable within 6 hours, and then peak levels occur after 12 hours, with a subsequent decline to undetectable levels by 24 hours[ 21 ]. Detectable IL2 protein in the culture media follows a similar, but delayed course with peak concentrations found after 24 hours, and by 48 hours barely detectable levels remain. Expression of high affinity trimeric IL2Rs follow a similar, but delayed transient expression course, with peak levels expressed at 24–48 hours, followed by a slow decline over several days[ 91 ]. By comparison, expression of the IFN-γ gene follows a much more protracted course, with detectable expression still evident after several days[ 124 ]. The mechanisms accounting for the transient expression of the IL2 gene and the genes encoding the IL2R chains have not been apparent, until recently. It is now realized that surface molecules of the coinhibitory CTLA-4 family[ 125 ] appear later after TCR/CD28 activation, first evident after ~ 24 hours with peak levels at 48–96 hours. In addition to CTLA-4, which has a 10-fold higher affinity for the B7 molecules expressed by APCs, the coinhibitory receptor PD-1[ 126 ], as well as the ligands reactive with this receptor, PDL-1[ 126 ] and PDL-2[ 127 ], appear on activated T cells. Still a third, later appearing coinhibitory receptor, BTLA, has also recently been described as expressed on both antigen-activated T cells and B cells[ 128 ]. This coinhibitory ligand-receptor family of molecules belongs to the Ig superfamily and the B7/CD28 costimulatory ligand/receptor family[ 129 , 130 ]. However, the CTLA-4 family of receptors does not function to deliver a costimulatory signal, as does CD28. Instead, members of the CTLA-4 family have inhibitory signaling motifs in their cytoplasmic domains and they have been shown to localize in close proximity to CD28, where they compete with activating signals from CD28. For example, by binding the phosphatase SHP-2, the CTLA-4 family of molecules inhibits the positive signals emanating from the TCR/CD28 at a very proximal position within the signaling cascade, thereby extinguishing the expression of the IL2 gene. Once IL2 production is shut down, because IL2 is internalized and degraded with a half-time (t 1/2 ) of ~ 2 hours, IL2 is consumed very rapidly, resulting in cessation of IL2-promoted cell cycle progression and eventually apoptosis due to the lack of IL2-promoted anti-apoptosis gene expression, such as Bcl-X L . The coinhibitory ligand/receptor pairs identified thus far can be shown to exert their negative effects by attenuating IL2 gene expression, but the administration of IL2 can bypass this block, in that the cells are still IL2-responsive. A similar phenomenon has been found with regard to the effects of T-Regs. T-Regs shut down IL2 gene expression, via a cell-cell contact mechanism that has not yet been delineated. However, IL2 supplementation will bypass the suppressive effects of T-Regs, thereby allowing for T cell proliferation. Also, in both instances, activation of CD28 via monoclonal antibodies serves to counteract the blockades, and permits the suppressed cells to both express the IL2 gene and to proliferate. Accordingly, it would seem a plausible hypothesis that the cell-cell contact mechanisms employed by T-Regs to inhibit IL2 gene expression are mediated by members of the CTLA-4 family of coinhibitory ligands/receptors, either those already identified, or others yet to be identified. This is a particularly attractive hypothesis, in that the ligands that trigger the coinhibitory PD-1 receptor are also expressed by TCR/CD28-activated T cells[ 131 ]. The Transient Nature of the In Vivo T Cell Proliferative Response And "Adaptive Tolerance" With the advent of the ability to label cells with 5- and 6-carboxyfluorescein diacetate succinimidyl ester (CFSE), combined with the use of gene deleted mice, it has been possible to follow the proliferation of T cells in vivo , and to test which proliferative signals are operative. In this regard, T cells from IL2Rβ (-/-) mice that have had the IL2Rβ chain expressed only during thymopoiesis do not express detectable IL2Rβ chains in the periphery, and therefore are incapable of delivering a proliferative signal. However, these cells are capable of undergoing 1–2 divisions upon stimulation with anti-CD3 + anti-CD28[ 132 ]. Similar findings have been reported in systems using TCR transgenic mice and adoptive transfer experiments when either IL2 or CD25 have been deleted[ 133 , 134 ]. One interpretation is that TCR/CD28 activation may be capable of initiating 1–2 rounds of cell division, but IL2 appears necessary for maximal and sustained proliferation. Alternatively, other, yet undiscovered cytokines with TCGF activity may be responsible for activating the initial proliferative responses to antigen activation. After the initial proliferative clonal expansion of antigen-selected cells, the fate of the expanded effector cells has been found to depend greatly upon whether the antigen is cleared or whether it persists. In experimental viral infections where the virus is cleared rapidly within the first week after infection, expanded CD4+ and CD8+ effector cell populations undergo a contraction, with the loss of as much as 90% of the expanded effector cells[ 18 ]. We have found that this contractive phase is attributable to cytokine withdrawal apoptosis, in that the administration of IL2 during this phase prevents the contraction[ 135 ]. Others have shown a similar protective effect of IL2 after both CD4+ and CD8+ cells are expanded in response to activation by staphylococcal superantigen[ 136 ]. Subsequent studies have revealed that the residual populations of expanded effector cells eventually differentiate to "central" memory cells, which have the capacity for maintenance of the population size via slow proliferative renewal and as well, the capacity to respond to the reintroduction of antigen by rapidly producing IL2, proliferating and differentiating to effector cells[ 137 ]. With persistence of antigen, the fate of the expanded effector T cell populations changes dramatically. Instead of differentiating into responsive memory cells, the cells revert to a state of unresponsiveness, which has been termed "exhaustion" by those studying experimental persistent viral infections [ 138 , 139 ]. This exhausted state is manifested by an early loss of the capacity to produce IL2 and to proliferate. As antigen persists, the cells gradually lose their capacity to lyse target cells and to secrete antiviral effector cytokines such as TNF-α and IFN-γ. Eventually, clones of virus-specific cells can undergo apoptosis, which may be attributed to Activation-Induced Cell Death (AICD), leading to clonal disappearance[ 140 ]. A similar phenomenon has been described in experiments employing a paired transgenic model (TCR and Ag)[ 141 ]. In this model, CD4+ TCR-Tg T cells from antigen-naïve animals are labeled in vitro with CFSE, then transferred into recipient mice expressing low levels (~ 100 pM) of antigenic pigeon cytochrome c peptide. The cells become activated, express the IL2R α-chain (CD25) and CD69, and proliferate for the first 4 days, eventually expanding ~ 100-fold. However, over the subsequent 10–14 days, the cells lose expression of CD25 and cell recovery declines by ~ 50%. Thereafter, the cells are said to be in the "adaptive phase", which is characterized by a hyporesponsiveness to antigen in vitro , and is manifest by a decreased capacity to produce IL2 and other cytokines, and to proliferate. This adaptive state persists in the host with chronic expression of the antigen, and in contrast to a similar paired transgenic model in a B cell system[ 142 ], is not associated with a decrease in the level of expression of the TCR[ 143 ]. However, the adaptive state is dependent upon continuous exposure to antigen; upon transfer to an antigen-negative host, the hyporesponsive state reverts. Moreover, the adaptive state is similar to a desensitization phenomenon akin to tachyphylaxis, and studies have revealed a down regulation of the TCR signaling molecules, involving an early block in tyrosine kinase activation, which primarily inhibits calcium mobilization, thereby suggesting that the desensitization involves the adaptation of the TCR signaling apparatus to the chronic persistence of low levels of antigen[ 143 , 144 ]. It is important to emphasize that the adaptive state is only a relative hyporesponsiveness, as compared to either the naïve situation or the host with central memory cells[ 141 , 143 ]. If cells are exposed to higher concentrations of peptide in vitro , i.e. between 1 nM and 1 μM, a response can be detected, but the antigen dose-response curve is shifted 100–300-fold to the right. In addition, the adaptive phenomenon cannot be ascribed to active suppression by a T-Reg differentiative process, in that the adapted cells do not express CD25, and in vivo experiments have excluded a suppressive mechanism. What Determines the "Strength" of the Signal? From the discussion thus far, it must be apparent that the strength of the signals delivered to T cells ultimately determines the outcome, i.e. either anergy, or activation of the IL2 gene, and if activation occurs, the duration that it persists. Thus, considerations of the ligand concentrations available, the receptor affinities and numbers expressed, and the duration of the ligand-receptor interactions again become important. The duration of signaling via TCR/CD28 is known to be a major determinant of the magnitude of IL2 production and thus the extent of T cell proliferation. Even in the 1970s the magnitude of the proliferative response after mitogenic lectin administration was found to be directly related to the duration of lectin stimulation[ 145 ]. Thus, removal of the mitogenic lectin within the first 24 hours of stimulation attenuates the proliferative response markedly. With the discovery that T cell proliferation after mitogen or antigen activation is principally mediated by IL2[ 22 ], it seemed obvious that a certain time interval was necessary after TCR/CD28 triggering to provide for maximal expression of the IL2 and IL2R genes. Indeed, experiments proved this to be the case[ 146 ]. Even so, several hours seemed somewhat prolonged, given that early biochemical events such as calcium flux and kinase activation were detectable within minutes after TCR engagement. Moreover, IL2 gene transcription could be detected within 45 minutes, using sensitive techniques[ 147 ]. A series of experiments reported in the mid 1990s began to provide an explanation for this perplexing problem. By equating "triggered" TCRs with their internalization and disappearance from the cell surface after ligand activation, it was shown that a single peptide-MHC complex is capable of serially triggering up to ~ 200 TCRs[ 148 , 149 ]. Furthermore, like the IL2Rs, T cells appeared to be able to "count" the number of triggered TCRs, and responded by proliferating when ~ 8,000 TCRs were triggered[ 149 , 150 ]. Other experiments showed that the duration of antigenic stimulation was one of the most critical parameters determining the fate of naïve and effector T cells, i.e. whether they would be activated or deleted[ 151 ]. However, these observations were not linked to the IL2/IL2R system. With the discovery of Supramolecular Activation Clusters (SMACs), an additional level of complexity was added[ 152 ]. Also termed the "Immunological Synapse", the specialized junction between a T cell and an APC consists of a central cluster of T cell receptors together with costimulatory and coinhibitory receptors, surrounded by a ring of adhesion molecules[ 153 ]. Recent experiments, which employed peptide antigen-specific TCR αβ transgenic T cell blasts labeled with CFSE, the relationship between the duration of the TCR signal and the extent of the proliferative response could be examined further at the single cell level. Between 10–24 hours of continuous TCR stimulation by MHC-peptide in a stable SMAC is necessary to promote maximal IL2 production and proliferation[ 154 ]. Regarding the density of the TCR on responding T cells, it is important to restate and emphasize that the TCR density follows the same log-normal distribution as detailed for the log-normal distribution of IL2Rs, even on cloned T cells. Thus, there is at least a 2-log 10 difference in TCR density among potential responding T cells, and consequently, those cells with the highest density of TCRs, will be capable of responding to lower concentrations of MHC-peptide epitopes, and also capable of responding more rapidly than cells with lower TCR densities to an optimal pMHC concentration. Detailed studies examining the cytokine response of murine T cell clones to graded peptide antigen concentrations have revealed a hierarchical organization of TCR signal-dependent response thresholds for elicitation of different cytokines in individual cells[ 155 ]. IL2 production was found to remain constant per cell as the ligand concentration was increased, with the primary change being in the number of cells making IL2 at a fixed level. Therefore, the decision to produce IL2 is a quantal decision on the part of each individual cell within the cloned population. Exactly what leads to the heterogeneity of the quantal response of IL2 gene expression on the part of the cells within the cloned cell population has not been determined, but very similar findings have been reported for IFN-γ production by a human T cell clone in response to graded doses of antigenic peptide[ 156 ]. The peptide dose that stimulated 5% vs. 95% of the T cells was found to span over a log, and the response on the part of the cells comprising the population was quantal; i.e. at low antigen concentrations fewer cells expressed IFN-γ and as the antigen concentration was increased, an increasing number of cells expressed IFN-γ. To explain these results it was postulated that the intraclonal heterogeneity in antigen responsiveness could result from the different numbers of TCRs expressed by individual cells. However, this conjecture has not been examined directly. Exactly the same considerations hold for the distribution of costimulatory and coinhibitory molecules. Thus, there is interplay between all of these receptors, which ultimately impacts the "strength" of the signal[ 157 ] and the duration that the signal must be applied to productively signal the IL2 gene response elements, and eventually lead to an "activated" T cell. Obviously, if any of these parameters are limiting, the activation events may favor the delivery of abortive transcriptional activating signals, which may favor anergy or differentiation to T-Regs, rather than activation[ 141 ]. As already detailed, if CD4+ T cells are activated via the TCR without adequate stimulation via CD28, such as may occur if self peptide is presented on immature dendritic cells as APCs, the conditions would favor the predominant activation of NF-AT yet inadequate AP-1 and NF-κB/Rel activation, which would promote anergy and perhaps even the irreversible differentiation of most of the cells to T-Regs. Thus, the concentration of antigen, the availability of adequate costimulatory molecule function, the affinity and density of the TCR/cell, as well as the duration that the TCR is triggered all influence signal "strength". With regard to the affinity of the TCRs for the peptide-MHC complex, it is important to note that the TCR does not undergo somatic hypermutation and "affinity maturation" as does the BCR. Accordingly, the Kd of the TCR is fixed after recombination and rearrangement in the thymus, and is relatively low by comparison with the Kd of antibody molecules, which have ~ 1000-fold higher affinity for binding antigen. Measurements of the equilibrium dissociation constants of isolated agonist pMHC binding to TCR molecules by surface plasmon resonance have revealed Kds in the range of 1–100 μM, with k off rate constants of ~ 0.01–0.1 s -1 , yielding t 1/2 ~ 7–70 secs[ 158 ]. In contrast, antagonistic MHC-peptide-TCR interactions off-rates are ~ 10-fold faster than agonistic interactions. Thus, off-rate constants of ~ 5 sec -1 have been found, which yield t 1/2 of only ~ 0.15 seconds. Of course, the TCR does not bind MHC-peptide complexes in isolation or in solution. The formation of the immunological synapse greatly alters the way in which TCRs engage antigens and the way in which they are triggered[ 55 ]. Thus, T cell clones that have TCRs with a Kd = 1 μM binding to MHC-peptide in isolation can be triggered at peptide concentrations ranging as much as 100-fold lower in vivo . Furthermore, new studies have made it possible to "count" the exact number of ligands that a T cell encounters on another cell, and then monitor the consequences of that interaction with respect to the increase of intracellular calcium concentration[ 55 ]. It has been found that only 10 MHC-peptide ligands are sufficient to provide for the formation of a stable immunological synapse and sustained calcium flux for several hours. However, below this critical number of MHC-peptide ligands, only transient calcium increases occur, and a stable synapse does not form. Thus, an abortive signaling process appears to ensue, which could very well lead to the activation of NF-AT without adequate levels of AP-1 or NF-κB/Rel, which would be insufficient for IL2 and IL2Rα chain gene expression, thereby promoting anergy/T-Reg differentiation. Recent experiments focused on how the TCR can respond to such low concentrations of agonist peptides indicate that the slower off rate of the agonist pMHC/TCR interaction allows the juxtaposition of CD4 with bound lck to the agonist pMHC/TCR, so that endogenous pMHC/TCR, which are in a large excess, can form a dimeric signaling complex comprised of agonist pMHC and endogenous pMHC[ 159 ]. Then, the endogenous pMHC with its fast off rate can trigger many TCRs serially and greatly amplify the TCR signals. Given the long duration necessary to trigger a response, a kinetic model has been proposed to account for how serially triggered TCRs that interact very briefly with peptide-MHC complexes, then are rapidly internalized and degraded can be counted by the T cell, and how transient signaling events can be accumulated over time and integrated into a quantal response[ 160 ]. The model is based on a process first described in neuronal cell activation termed 'temporal summation'. The signaling events originating from successively triggered TCRs build up, with each adding to the falling phase of the one before. In this way, small and short signals that alone are unable to trigger a response can be summed up over time eventually to reach the level sufficient to trigger the quantal response, in this instance IL2/IL2R gene expression. All of these considerations lead one to the conclusion that like the IL2/IL2R-determined quantal decision to undergo cell cycle progression, there appear to be quantal decisions operative at the level of the immunological synapse that lead to distinct cell fates, which in turn are ultimately determined by IL2 and IL2R gene expression. Thus, if the agonist pMHC ligand concentration is low, only T cells with a high density of TCRs will form stable synapses that will result in sustained activation of the IL2 gene and thereby cell cycle progression. It follows that at the same limiting agonist pMHC concentrations, cells with lower TCR densities may have abortive expression of the IL2 gene, which would favor differentiation to T-Regs, while cells with still lower TCR densities would not successfully trigger expression of the IL2 gene, thereby favoring the triggering of differentiation to an anergic state and unresponsiveness. Thus, there are at least 3 distinct cell fates that are determined by the accumulated number of triggered TCRs, which is determined by the agonist pMHC concentration, TCR density and the duration of the pMHC/TCR interaction. The Quantal Numbers of Triggered Receptors are Specified in the Thymus The structure and function of the immunological synapse essentially determines the fate of T cells as they mature in the thymus. Again, this cell fate determination is linked to IL2 and IL2R gene expression. From the above discussion, it is now clear that like the quaternary IL2/IL2R complex, the immunological synapse is a dynamic multicomponent molecular complex, the stability of which requires ongoing signaling through the TCR for stable calcium mobilization and kinase activation occurring over several hours. As well, the immunological synapse modulates the overall level of mature T cell activation by integrating positive (costimulatory) signals and negative (coinhibitory) signals from a variety of surface receptors. In this regard, it is noteworthy that the synapse that forms between thymocytes and thymic stromal cells differs qualitatively from that observed between mature peripheral T cells and peripheral APCs[ 161 ]. One reason that this may occur relates to the lack of expression of the costimulatory B7 molecules on thymic stromal cells[ 162 ]. In this regard, on would predict that thymocytes would not be activated to produce IL2 very readily, due to the lack of CD28/B7 costimulation. From detailed experiments performed primarily with mice made transgenic for the TCR, it is now clear that the interaction of the TCR expressed on developing immature thymocytes with self peptide-MHC molecules expressed on thymic stromal cells is essential for the selection of those cells that ultimately are destined to leave the thymus and populate the periphery [ 163 - 166 ]. Thus, after productive rearrangement and expression of the αβ chains of the TCR, four fates are possible. If there is little or no affinity of the TCR for self peptide-MHC molecules, the T cells undergo apoptosis as a result of a lack of signal generation. However, if there is a "weak interaction" between the TCR and nonagonist or antagonist self peptide-MHC molecules (i.e. those molecules that have a rapid off-rate from binding to the TCR), the maturing T cells are "positively selected" to survive. Although it still remains controversial, most data are consistent with the notion that an intermediate strength of signal leads to the differentiation of T-Regs. By comparison, if the αβ TCR encounters an agonistic self peptide-MHC interaction, i.e. one that has a slower off-rate and a higher affinity, "negative selection" occurs and these T cells are induced to undergo apoptosis. Consequently, only those T cells that have "nonagonistic" reactivity with self peptide-MHC molecules make up the T cell repertoire. With regard to the generation of quantal cellular responses, it is noteworthy that IL2 has now been implicated to be involved in both positive and negative selection, as well as T-Reg differentiation. Recent experiments focused on the signals generated in thymocytes leading to positive selection have revealed that signaling via calcium and calcineurin is necessary for positive selection but dispensable for negative selection[ 167 ]. For example, deletion of the regulatory subunit-B1 of calineurin in thymocytes leads to loss of activation of NF-ATc proteins and also inefficient ERK activation, but normal activation of NF-κB/Rel. Of interest, positive selection was found to be markedly deficient in these animals, but negative selection remained intact. As already discussed, experiments performed with IL2, IL2R, JAK3, and STAT5 (-/-) mice have all now demonstrated that the IL2/IL2R interaction is unnecessary for positive selection. In the absence of signals generated via the IL2/IL2R interaction, positive selection proceeds unimpeded, so that during development and after birth, a normal number and composition of cells mature in the thymus and populate the peripheral lymphoid tissues. Thus, if the "weak" signals between the TCR and self peptide-MHC are not strong enough to trigger expression of the IL2 and IL2R genes, the cells are permitted to survive, and leave the thymus to populate the secondary lymphoid tissues. Should agonistic nonself-peptide-MHC complexes be introduced that are capable of binding to the TCR with high affinity, these T cells, which are non-reactive with "nonagonistic" self peptide-MHC complexes, are not anergic. Rather, if stimulated by non-self agonistic peptide, they are fully capable of expressing IL2 and IL2R genes, and of undergoing IL2-dependent proliferative expansion and differentiation to effector cells in the periphery. However, the development of T-Regs (i.e. CD4+CD25+ cells) is dependent on the IL2/IL2R interaction. In the absence of IL2 or functional IL2Rs, T-Regs are low or absent, both in the thymus and in the periphery. Moreover, if the IL2/IL2R interaction is restored, either genetically or pharmacologically, then T-Regs are reconstituted and the autoimmune phenomena are delayed or prevented altogether[ 106 , 111 , 168 - 170 ]. As well, the function of T-Regs in the periphery is also totally dependent on IL2/IL2R signaling, so that if potential positively selected autoreactive T cells are not continuously suppressed by T-Regs, the IL2 (-/-) syndrome of lymphoid hyperplasia and autoimmunity will occur. At this juncture, it is logical to propose that the number of triggered TCRs and IL2Rs receptors necessary to generate T-Regs must be higher than the number required to generate simple "positive selection, in that IL2/IL2R gene expression must be triggered[ 73 , 171 ]. In this regard, it has been reported that α chain allelic "inclusion" results in a lower density of TCRs/cell, in that the β chains are paired with 2 distinct α chains in ~ 30% of αβ TCRs[ 172 ]. Upon introduction of antigenic peptide, these TCR bi-allelic cells can escape negative selection, presumably because their epitope-specific TCR density is only half normal, and accordingly they would not accumulate the same number of triggered TCRs as a mono-allelic cell. This difference could be responsible for the generation of T-Regs. It also appears that differentiative signals triggered by the IL2/IL2R interaction are necessary to promote the differentiation to an anergic and suppressive T-Reg cell[ 81 ]. Furthermore, the number of triggered TCR receptors must be lower than those necessary to trigger "negative selection". These findings have led some investigators to propose that the main "nonredundant" function of IL2 is to promote the development and function of T-Regs[ 173 , 174 ]. "Negative selection" of potential self-reactive T cells has been proposed to occur via a TCR-triggered apoptosis. The exact molecular mechanisms responsible for this effect still remain obscure, but apoptosis appears to occur when there is a strong agonistic TCR-self-peptide-MHC interaction, which should trigger maximal IL2 and IL2R gene expression. However, data have been reported indicating that negative selection of CD8+ T cells proceeds normally without IL2[ 175 ]. By comparison, it remains controversial whether the IL2/IL2R interaction is necessary for negative selection of CD4+ T cells[ 176 ]. Quite convincing data have revealed a role for IL2 in CD4+ T cell negative selection [ 177 ] using nontransgenic and transgenic IL2-sufficient and deficient animal model systems. It could be shown that during TCR-mediated thymocyte apoptosis, IL2 protein is expressed and detectable in situ in the thymus, and apoptotic thymocytes up-regulate the expression of IL2Rs. Furthermore, IL2R+ CD4CD8 double-positive and CD4 single-positive thymocytes undergoing apoptosis bind and internalize IL2. As well, IL2-deficient thymocytes are resistant to TCR/CD3-mediated apoptotic death, which is overcome by providing exogenous IL2 to IL2 (-/-) mice. Finally, disruption or blockade of IL2/IL2R interactions in vivo during antigen-mediated negative selection rescues MHC class II restricted thymocytes from apoptosis. Thus, all of these findings provide evidence for the direct involvement of the IL2/IL2R signaling pathway in the deletion of self-reactive double-positive and CD4 single-positive T cells[ 177 ]. Accordingly, these data are all entirely consistent with the notion that the CD4+ T cell hyperplasia and autoimmunity observed in IL2 (-/-), IL2R (-/-), and IL2 signaling (-/-) mice are attributable, at least in part, to inefficient deletion of strongly agonistic self-reactive CD4+ T cells, as well as deficient maturation of T-Regs. How Can the Immune System Ever Discriminate Between "Self & Non-Self" Peptides? The foregoing considerations lead one to the realization that there are no known molecular mechanisms that can explain how the TCR can discriminate qualitatively between peptides of self-origin vs. peptides of nonself-origin. Both of these ligands are identical in structure, i.e. they are both peptides. Moreover, the αβ TCRs are also identical structures, whether they recognize self or non-self peptides bound to MHC. It follows that all of the data and logic support a quantitative mechanism of discrimination based upon the accumulated number of triggered TCRs and IL2Rs, as shown in Figure 6 . Moreover, each triggered cellular differentiative fate of survival, death, anergy, or proliferative expansion, is quantal. Figure 6 The number of triggered TCRs and IL2Rs determine quantal T cell fates in both the thymus and the periphery. On each plot, the number of triggered TCRs and IL2Rs increase from bottom to top. The different quantal cell fates are dictated by a definite number of triggered Rs as depicted. Both in the thymus and in the periphery, there are 3 cellular fates specified by an increasing number of triggered TCRs, which dictates whether IL2 is produced and how much IL2 is produced. Thus, ultimately, the number of IL2-triggered IL2Rs determines the critical quantal fate decisions. A similar conclusion was introduced recently, with the difference that TCR avidity (i.e TCR affinity × density) was postulated to dictate the cell fates, but no role was postulated for IL2[ 178 ]. Since ultimately, self-nonself discrimination of the immune system depends on proliferative expansion of antigen-selected clones, the connection between the number of triggered TCRs and IL2Rs offers a molecular explanation for the quantal cellular response. If T cells that have potential reactivity with self peptide-MHC ligands exit the thymus having escaped negative selection, these T cells will populate the secondary lymphoid tissues in the periphery at very low frequencies, ~ 1 in a million lymphocytes. Thus, these cells make up the TCR repertoire in the periphery, and as long as the distinct self-pMHC complexes remain below the critical number necessary for the formation of a stable synapse, and the TCR-pMHC off-rate is rapid, there is no necessity to introduce any additional mechanism to allow the immune system to ignore self. Instead, in the periphery the immune system only recognizes peptides, whether self or non-self, which are present at a high enough concentration to attain the critical density of 10 peptide-MHC molecules/synapse and as well, that generate a slow enough off-rate from the TCR to form a stable synapse, so that the critical number of triggered TCR/CD28 receptors for activation is reached. The only caveat beyond these considerations is that if an abortive pMHC/TCR synapse forms, the cell receives signals that it interprets as instructions to become anergic, or possibly to differentiate to become suppressive cells (T-Regs), thereby solidifying the non-reactivity on the part of the host to these peptides. Conclusions The "Quantal Theory" states that the fundamental decisions of the T cell immune system are dependent upon the cells receiving a critical number of triggered TCRs and IL2Rs and that the cells respond in an all-or-none fashion. Any reductionist approach to understand how the immune system discriminates self from non-self must begin with a systemic immunological response that most closely correlates with immunity. Thus, the "Quantal Theory" is based on Burnet's axiom that the proliferative expansion of antigen-selected clones is central to the generation of a protective immune response[ 1 ]. Secondly, a successful reductionist theory must explain how individual cells of the immune system make the decision to proliferate or not. As each decision of the individual cell is quantal, one must explain the molecular basis of the quantal cellular decision. At each decision point, the assembly of essentially irreversible multicomponent macromolecular complexes underlies the quantal cellular responses. Given the data that the simple IL2/IL2R interaction promotes the quantal decision to undergo cell cycle progression by reaching a critical number of triggered IL2Rs[ 26 ], it follows that the quantal decision to express the IL2 gene and the IL2R genes is similarly regulated by a critical number of triggered TCR/CD28 molecules. Thus, the TCR/CD28-triggered expression of the IL2 and IL2R genes is pivotal for the quantal cellular decisions in the thymus that determine distinct fates such as positive selection (survival), the differentiation to T-Regs (anergy), and negative selection (apoptosis), while it is also pivotal for the quantal cellular decisions in the periphery that determine whether to remain unresponsive (survival), to differentiate to T-Regs (anergy), or to begin proliferating (immunity). Accordingly, the "Quantal Theory" offers a unifying explanation at the molecular level that provides the cellular mechanisms for the immune system as a whole to make the quantal discrimination between self antigens and nonself antigens. These considerations lead to the speculation that non-self peptides that are introduced in low enough concentrations may well be perceived by the immune system as "self" and will generate tolerance. Thus, we now have a molecular, cellular and immunological explanation for the phenomenon of "Low Zone Tolerance", first demonstrated by Mitchison 40 years ago[ 16 ]. Accordingly, one might have a means to tolerize individuals to specific defined peptides that may be useful in the treatment of allergy, autoimmunity, and allograft rejection. In contrast, a similar situation could be operative in the tumor-bearing host, and in the host infected chronically with viruses such as the Human Immunodeficiency Virus, and Hepatitis C Virus. In these instances, low persistent antigen levels may well serve to maintain a state of "low zone" tolerance. Accordingly, the question arises as how to break this tolerant state? Competing interests The author declares that he has no competing interests.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544850.xml

548298

First documented cure of a suggestive exogenous reinfection in polymyositis with same but multidrug resistant M. tuberculosis

Background MDR Mycobacterium tuberculosis is the major cause of treatment failure in tuberculosis patients, especially in immunosuppressed. We described a young polymyositis patient on immunosuppressive therapy who was started with antituberculosis therapy as a susceptible strain of M. tuberculosis was isolated from a single cutaneous abscess in his neck and from regional lymph nodes. Case presentation He had non-reactive miliary tuberculosis and multiple cutaneous abscesses 6 months later with the same strain, which was resistant this time to 9 antituberculosis drugs. We described clinical presentation, radiological and laboratory work-up, treatment and follow-up as the patient was cured after 1.5 years with 6 antituberculosis drugs. Conclusion To our knowledge, this is the first reported case where an immunosuppressed patient with suggestive exogenous reinfection within 6 months with the same but MDR strain of M. tuberculosis was cured. Intense management and regular follow up were important since the patient was a potent source of MDR M. tuberculosis infection and there was limited choice for therapy.

Background World Health Organization published the Global Tuberculosis Control Report (2003) on 'World Tuberculosis Day' where India is ranked number one in the world for high incidence of smear positive cases of pulmonary (about 0.9 million) and extrapulmonary TB (about 0.2 million) every year [ 1 ]. Undoubtedly, TB itself has reemerged in India as a serious problem since 1985 with the advent of HIV/AIDS [ 2 ]. However, cutaneous TB is still rare in countries like India (0.10%), Hong Kong (0.07%) and Madrid (0.14%) [ 2 ] and it manifests either as a true bacterial invasion or as a tuberculid (hypersensitivity reaction) with primary focus elsewhere. Evidences of bacterial invasion are found in lupus vulgaris, the most common manifestation (55%) in patients with cutaneous TB (1975–95) in northern India [ 2 ], primary chancre, tuberculous verrucosa cutis, scrofuloderma, tuberculous cutis orificialis and tuberculous cutis miliaris disseminates [ 3 ] whereas erythema induratum (Bazin disease) and lichen scrofulosorum are tuberculid lesions [ 2 ]. Moreover, atypical mycobacteria such as M. kansasii , M. scrofulaceum , rather than M. tuberculosis are the most common etiological agents for cutaneous TB in HIV/AIDS and other immunocompromised patients. WHO defines acquired drug resistance as the isolation of drug-resistant M. tuberculosis from a patient who has been treated for TB for 1 month or longer [ 4 ] and primary drug resistance as the isolation of drug-resistant strain from a patient without a history of previous treatment. The selection of M. tuberculosis with mutations conferring resistance to antitubercular drugs may result from poor management including the prescription of incorrect regimens and non-compliance with treatment. We report a rare case of reinfection with same but MDR M. tuberculosis in a polymyositis patient on immunosuppressive therapy resulting in multiple cutaneous abscesses and miliary TB. Case Presentation A 23-year-old polymyositis patient with suggestive clinical symptoms, muscle biopsy, enzyme assay, needle EMG and NCV test was on immunosuppressive drugs such as corticosteroids at 1 mg/kg/day for 3 to 4 weeks, tapered gradually over a period of 10 weeks to 1 mg/kg every alternate days and methotrexate at 7.5 mg weekly with gradual increase to 25 mg weekly for a period of 1 year when he experienced slightly improved muscle power. He had single cutaneous abscess 6 months later on the right side of the neck with no sinus formation and cervical and axillary lymphadenopathy. He was HIV-negative with normal CXR and there was no AFB in three consecutive (spot-early morning-spot) sputum samples (induced in two occasions). However, the aspirated pus from cutaneous abscess and biopsy of the axillary lymph node showed plenty of M. tuberculosis as culture was confirmed by niacin and nitrate reduction tests, rapid bacteriophage assay (FAST Plaque Tuberculosis kit, BIOTEC Laboratories Ltd, UK) and RFLP analysis (figure 1 ) using species-specific probes for devR (Differentially expressed virulent gene) [ 5 ]. In RFLP analysis, the devR gene (Gene bank nucleotide sequence accession no. U22037) encodes a response regulator that is part of a two-component signal transduction system of M. tuberculosis . The authenticity of the amplified product was established by hybridization of immobilized PCR products to an internal oligonucleotide, devR1 , mapping within the devR gene. The chromosomal DNA was prepared by chloroform-isoamyl alcohol DNA extraction and 4.5 μg of DNA from the isolate was restricted with Pvu II. Separation of Pvu II-restricted DNA by electrophoresis, Southern blot hybridization with a 513-bp PCR probes for devR ( devRf , 5' GGTGAGGCGGGTTCGGTCGC 3'; devRr , 5' CGCGGCTTGCGTCCGACGTTC 3') and chemiluminescence detection were done according to the standard method recommended for the DNA fingerprinting of M. tuberculosis [ 6 ]. Histopathologically there were nonspecific necrosis with disintegrating polymorphs, very few granulomatous cells and no epithelioid cells in the biopsy specimen. USG of kidney, liver or spleen revealed no abscess. The strain was susceptible to the 4 first line drugs [isoniazide (H), rifampicin (R), ethambutol (E) and pyrazinamide (Z)] by BACTEC 460TB system. Treatment with 4-drug regimen (H = 300 mg/d, R = 450 mg/d, E = 800 mg/d and Z = 1.5 g/d) for 8 months (2EHRZ/6HR) [ 7 ] with first 2 months supervised showed clinical improvement. However, the patient returned 6 months later with multiple cutaneous abscesses, mainly on his back, left thigh and left arm and miliary mottling in CXR. He denied any irregularity in antituberculosis therapy. All immunosuppressive drugs (corticosteroids and methotrexate) were discontinued as he was readmitted. Isolates from 2 of 3 consecutive sputum samples and aspirated pus samples from all 5 completely drained abscesses were reconfirmed as the identical M. tuberculosis strain by phage assay and RFLP analysis (figure 1 ). The isolate was resistant to 4 first-line drugs by BACTEC 460TB system and to 9 drugs, tested into Loewenstein Jensen media (Hi-media, Mumbai, India): H (1 μg/ml), H (10 μg/ml), R (20 μg/ml), R (50 μg/ml), E (2 μg/ml), E (10 μg/ml), Z (50 μg/ml, at pH 5.5), streptomycin (5 μg/ml), streptomycin (50 μg/ml), paraaminosalicylate (2.5 μg/ml), cycloserine (30 μg/ml), amikacin (700 μg/ml) and ciprofloxacin (12.5 μg/ml). Treatment was continued with H (5 mg/kg/d), R (10 mg/kg/d), Z (30 mg/kg/d), E (15 mg/kg/d), kanamycin (15 mg/kg IM 5 times weekly) and sparfloxacin (500 mg/d) for 18 months with first 2 months under supervision. A slow but complete clinical, microbiological and radiological cure after 1.5 year was followed up for another 16 months. Discussion India is an endemic country for TB with an estimated 20,000 infectious cases and 1–3.3% of new cases of MDR TB every year [ 8 ]. Infections are common with more than one strain of M. tuberculosis during the same episode (multiple infection), in different lesions (multiple infection) or during successive episodes (reinfection) in HIV-positive as well as -negative individuals [ 9 ]. In this study identical strains of M. tuberculosis were isolated at 6 months interval, but contrary to the first, the strain in the second episode was resistant to all the first line and most of the second line antitubercular drugs. As far as could be ascertained, this is the first case of isolation of MDR strain of M. tuberculosis from a HIV-negative but immunosuppressed patient who had an infection with the same, but susceptible strain 6 months before. As evident from the study, the chance of exogenous reinfection with the same but drug-resistant strain from some undetected source within the endemic community is the most likely explanation. This explains the acquisition of resistance against all those drugs, to which the patient was not exposed earlier. Reinfection though occurs usually after first two to five years in immunocompetent hosts, may progress to active disease at any time after treatment has been discontinued and even during treatment for active tuberculosis [ 10 ]. Moreover, an ongoing tuberculous infection and simultaneous immunosuppressive therapy might significantly divert the immune response, thereby increasing the overall susceptibility to 'superinfection' [ 10 ] with the same but MDR strain. The chances of 'simultaneous infection' [ 11 ] with both the strains (susceptible and MDR) could be another possibility. Drug-susceptibility testing could discriminate simultaneous infection with different susceptibility profiles if individual colonies from the isolate were tested whereas in our case the sensitivity pattern of the susceptible strain might have been obtained at first attempt. Even RFLP and phage typing were not enough to determine simultaneous infections or reinfection (superinfection). Endogenous reactivation which is higher than the rate of exogenous reinfection in endemic countries [ 12 , 13 ] with multi- "drug resistance in previously treated case"[ 14 ] might be a remote possibility. The history of regular medication itself might be notoriously misleading in patients with multidrug resistance [ 14 ]. However, it is difficult still to explain the acquisition of multidrug resistance of the strain in endogenous reactivation in such a short period against those drugs, to which the patient was not exposed. Any switch in specimens or cross-contamination of cultures in laboratories were unlikely since no other sample to switch or contaminate was identified. Niacin test (95% positive [ 15 ]) and nitrate reduction test (97% positive [ 15 ]) differentiated M. tuberculosis from other mycobacteria in M. tuberculosis complex whereas RFLP and phage typing excluded atypical mycobacteria, like M. kansasii and M. scrofulaceum . The overall acuracy of the drug susceptibility test ranges between 84–100%, since mycobacteria often clump [ 16 ]. We tried thorough vortexing with glass beads to obtain homogenous inoculum suspensions. It was a rare case of miliary TB of non-reactive type [ 17 ] with MDR M. tuberculosis invasion of the skin without any muscle involvement and sinus formation as the biopsy of the axillary lymph node showed non-specific necrosis containing disintegrating polymorphonuclear leukocytes and enormous number of AFB. This type is more often seen with severe HIV infection than in patients with immunosuppressive therapy [ 18 , 19 ], where the liver and the spleen are most commonly involved followed by the lung, the bone marrow and the kidney [ 17 ], granulomas and epithelioid cells are lacking and CXR shows inconspicuous diffuse mottling. In our case, there was no other organ involvement. Moreover, the rapid skin and lung involvement despite adequate antituberculosis therapy suggests that the tubercle bacilli were somehow protected from the drugs in the cutaneous and subcutaneous tissue ('paradoxical expansion of disease during therapy' [ 20 ]) either due to polymyositis-associated subcutaneous calcification or due to obstruction in the small arteries with the large mass of bacilli leading to necrosis and abscess formation in the surrounding areas. This is probably the first reported case of cure where all first and most of the second line drugs (a total of 9 antitubercular drugs) were found to be resistant. However, the 6-drug antituberculosis therapy might not be considered as the only reason for cure, since 4 first line drugs showed high-level resistance and chances of cross-resistance between closely-related aminoglycosides, amikacin and kanamycin and quinolones, sparfloxacin and ciprofloxacin could not be ignored [ 21 , 22 ]. Discontinuation of immunosuppressive therapy with a reconstitution of the immune system and complete surgical drainage of the abscesses might have proved to be an important adjunct to the treatment. Conclusions Severe immunosuppression may lead to disseminated TB such as miliary TB or other rare types of extra-pulmonary TB such as cutaneous abscesses. Follow-up of patients is important, and if response to treatment is poor, adherence to treatment, drug resistance and other possible reasons such as continuation of immunosuppressive therapy should be considered. In this case intervention by drainage of abscesses, discontinuation of immunosuppressive treatment and possibly long-term treatment with additional second line antituberculosis drugs eventually lead to cure. Abbreviations AFB – Acid fast bacilli AIDS – Acquired immunodeficiency disease syndrome CXR – Chest radiograph DNA – Deoxyribonucleic acid EMG – Electromyography HIV – Human immunodeficiency virus MDR – Multidrug resistant M. tuberculosis - Mycobacterium tuberculosis NCV – Nerve conduction velocity PCR – Polymerase Chain Reaction RFLP – Restriction-fragment length polymorphism TB – Tuberculosis UK – United Kingdom USG – Ultrasonography WHO – World Health Organisation Competing interests The author(s) declare that they have no competing interests. Authors' contributions CM conceived the study, carried out the case study and follow-up, in the clinical as well as microbiological aspects, formatted the study design, performed sensitivity of the organism, participated in the FAST plaque assay and molecular identification method and drafted the manuscript. AG carried out FAST plaque assay and molecular identification method. AA participated in the design of the study and acted as an overall supervisor. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548298.xml

517944

Competitive release of drug resistance following drug treatment of mixed Plasmodium chabaudi infections

Background Malaria infections are often genetically diverse, potentially leading to competition between co-infecting strains. Such competition is of key importance in the spread of drug resistance. Methods The effects of drug treatment on within-host competition were studied using the rodent malaria Plasmodium chabaudi . Mice were infected simultaneously with a drug-resistant and a drug-sensitive clone and were then either drug-treated or left untreated. Transmission was assessed by feeding mice to Anopheles stephensi mosquitoes. Results In the absence of drugs, the sensitive clone competitively suppressed the resistant clone; this resulted in lower asexual parasite densities and also reduced transmission to the mosquito vector. Drug treatment, however, allowed the resistant clone to fill the ecological space emptied by the removal of the sensitive clone, allowing it to transmit as well as it would have done in the absence of competition. Conclusion These results show that under drug pressure, resistant strains can have two advantages: (1) they survive better than sensitive strains and (2) they can exploit the opportunities presented by the removal of their competitors. When mixed infections are common, such effects could increase the spread of drug resistance.

Background Malaria infections often consist of more than one parasite genotype [ 1 - 3 ]. Humans represent ecological niches for co-infecting malaria parasites, with shared predators (immune responses) and limited resources, so that competition between co-infecting malaria strains is likely to be intense [ 4 ]. Such competition could strongly affect the relative transmission of newly arisen drug-resistant strains, and thus the spread of drug resistance [ 5 ]. Resistant and sensitive strains will co-occur in the same host both when de novo mutations arise, and when hosts acquire resistant and sensitive strains from one mosquito bite simultaneously or from different mosquito bites. In the absence of drug treatment, the transmission success of the resistant strain will depend on its intrinsic fitness and competitive ability. However, if drug treatment does occur, the resistant strain has two potential fitness advantages. First, it will better survive the drug than the sensitive strain. Second, treatment can remove drug-sensitive competitors, thus freeing up ecological space for the resistant strains to occupy; this would increase the relative transmission of the drug-resistant strain. This second effect, well recognized in theory, has the potential to greatly enhance the rate of spread of drug resistance in a population [ 5 ]. However, there is no direct experimental evidence that removal of competitors by drug treatment does enhance the transmission of drug-resistant parasites. This paper reports the first direct experimental demonstration that competitive release of drug-resistant strains can occur following drug treatment. Competition between drug-sensitive and -resistant malaria clones was studied using the rodent malaria Plasmodium chabaudi . This parasite is commonly used as a model for human malaria [ 6 ], and has been extensively used to study drug resistance [ 7 ]. In the absence of drugs, the drug-resistant clone is competitively suppressed by a drug-sensitive clone [ 8 ]. Here, competition between the two strains in drug-treated and untreated mice is compared. Methods Two genetically distinct Plasmodium chabaudi chabaudi clones were used: an AS clone resistant to the antifolate drug pyrimethamine [ 9 ], and AJ, a sensitive clone. These clones will be referred to as R (for resistant) and S (for sensitive) from hereon. Hosts were eight weeks old CBA/Ca inbred female mice (Ann Walker, University of Edinburgh; Harlan, England). Two experiments were performed. In the first, two groups of five mice were infected with 10 6 R parasites, and two groups with 10 6 R + 10 6 S parasites, as described elsewhere [ 8 ]. One group from each of these two infection types was drug-treated within three hours of inoculation and again on days 1, 2 and 3 PI (post-infection), using an oral administration of 8 mg pyrimethamine per kg mouse body weight. Asexual parasite densities and gametocyte densities – the latter being the transmission stages to the mosquito – were monitored using microscopic examination of thin blood smears and determination of red blood cell densities using flow cytometry (Beckman Coulter), as described elsewhere [ 8 ]. Real-time quantitative PCR was used to distinguish and quantify R and S parasites in mixed infections [ 8 , 10 ]. This protocol cannot distinguish between asexual parasites and gametocytes, but real-time PCR data were used as estimates of asexual densities, because gametocyte densities were 2–3 orders of magnitude lower than asexual densities and thus a negligible component of overall parasite numbers. For each infection, two phases were distinguished: the acute phase, involving the first wave of parasites, and the chronic phase, beginning when parasite numbers began to recover after the collapse of that first wave around day 12. All parasites had disappeared below detectable levels after 50 days. In the second experiment, two groups of nine mice were infected as above with either R parasites or R+S parasites. The subsequent transmission success of clone R was assayed by allowing batches of 30 starved Anopheles stephensi mosquitoes to feed on 3 mice from each group on each of days 7, 14, and 21 PI, as described elsewhere [e.g. [ 11 ]]. Eight days after the feeds, mosquitoes were dissected, and DNA extracted from midguts carrying oocysts. Real-time quantitative PCR was subsequently used to determine the prevalence of clone R in these mosquitoes. All procedures were regulated and carried out under the British Home Office Animals (Scientific Procedures) Act 1986. Results Two untreated mice infected with R+S parasites died on days 10 and 11 PI respectively, and were excluded from the analysis. In untreated mice, there were far fewer R parasites during the acute phase in mixed infections with clone S than in R-only infections (Figures 1a,1c ). However, in drug-treated mice, where S parasites were entirely removed by pyrimethamine (none of the PCR reactions performed were positive for clone S), there were as many R parasites in mixed infections as there were in R-only infections (Figures 1b,1c ; Drug treatment × Alone/Mixed interaction: F 1,14 = 14.4, p = 0.002). Thus, R parasites were competitively suppressed in mixed infections in untreated mice, but this suppression was negated when mice were treated with pyrimethamine, which effectively removed S parasites. Figure 1 Log asexual parasite densities of the resistant clone R over time in untreated (a) and drug-treated (b) mice infected with R alone or a mixture of R+S clones, and total numbers of R parasites produced over the acute (c) and chronic phases (d). All data points (mean ± 1 s.e.m.) are based on 5 replicate mice, except for mixed infections in untreated mice in (a) (4 mice on day 11 and 3 mice from day 12 onwards) and (c) and (d) (3 mice). As the limit of detection was 100 parasites per μl blood, y-axes in (a) and (b) start at 2. During the chronic phase, clone R was more numerous in untreated mice in mixed infections than in single-clone infections (due to the parasite peak around day 21; Figures 1a,1d ). Thus, in untreated mice in the chronic phase, clone R did not suffer from competition, and actually benefited from the presence of clone S (facilitation). In drug-treated mice, however, R parasites were similarly numerous in mixed- and single- clone infections (Figures 1b,1d ; Drug treatment × Alone/Mixed interaction: F 1,14 = 13.8, p = 0.002). The large peak of R parasites in the chronic phase in the untreated mixed infections around day 21 (Figure 1a ) coincided with a large peak of gametocytes, the transmissible stages of the parasite (Figure 2a ). This was in contrast with single-clone infections of R in untreated mice, and infections in drug-treated mice, where gametocytes were mainly produced around day 14 (Figures 2a,2b ). Overall, gametocyte numbers were the same for all four infection types (p > 0.05 for both Drug treatment and Alone/Mixed). Whether clone R really suffered from competitive suppression by clone S in untreated mice thus depends on how many of the gametocytes around day 21 were of the R genotype, and on how transmissible they were. Figure 2 Log gametocyte densities over time (mean ± 1 s.e.m.) for untreated (a) and drug-treated (b) mice. In (a) gametocyte densities for mixed R+S infections reflect overall R+S gametocytes, as the PCR assay could not distinguish between these (see text); in (b) all gametocytes are produced by clone R, as clone S was cleared from mixed infections. All data points are based on 5 replicate mice, except for mixed infections in untreated mice in (a): 4 mice on day 11 and 3 mice from day 12 onwards. As the limit of detection was 100 gametocytes per μl blood, y-axes start at 2. The second experiment assessed transmission to mosquitoes on days 7, 14 and 21 PI. It was found that the resistant clone R infected far fewer mosquitoes from mixed infections than from single infections (figure 3 ; Alone/Mixed: p = 0.002), indicating that transmissibility of gametocytes produced around day 21 was low, probably as a result of transmission-blocking immunity [ 12 ]. Thus, the competitive suppression of the resistant clone in untreated infections translated into reduced transmission success. Figure 3 Proportions of mosquitoes infected with the resistant clone R (mean and 95% confidence interval); mosquitoes fed either on mice infected with clone R alone or mice infected with a mixture of clones R and S. Means are based on 9 mice (3 on day 7, 3 on day 14 and 3 on day 21 PI) from which totals of 205 (R alone) and 216 (mixed R+S) mosquitoes took a blood meal. Infection with clone R was assessed by real-time PCR. Discussion These results show that drug treatment of malaria infections can severely affect ecological interactions between co-infecting strains. The drug-resistant clone was competitively suppressed by the drug-sensitive clone in untreated mice, in terms of both within-host growth and transmission to the mosquito vector. However, drug treatment removed that competitive suppression, and allowed the resistant clone to fill the ecological space emptied, giving it a substantial and additional fitness benefit in addition to the simple survival advantage conferred by resistance. Thus, under drug pressure, resistant strains can have two advantages: they survive better than sensitive strains and they can exploit the opportunities presented by the removal of their competitors, thereby increasing their relative transmission. Competition was studied between two unrelated clones, and thus did not reflect the situation in which a resistant clone arose de novo [ 13 ], but it seems likely that the competitive release following drug therapy would also apply there. Competitive release following drug treatment will greatly enhance the spread of drug resistance [ 5 ]. Also, with only the resistant strain left in the host, the probability of outbreeding is reduced, thus reducing the chances of meiotic recombination destroying multi-locus resistance [ 14 ]. In combination, these two processes could enhance the spread of drug resistance, especially in areas with high numbers of strains per infection [ 5 ]. Of course, this is an argument for judicious use of drugs, not their non-use. Clearance of drug-sensitive strains from mixed infections might enhance the spread of drug resistance, but this has to be offset against the short-term public health benefits, such as reducing overall malaria prevalence. In these experiments, the drug-sensitive clone was also the more virulent clone [ 8 ], and when it was cleared from mixed infections by drug treatment, mice were less sick, in that they lost less weight and became less anaemic (results not shown). In this experiment, mice were drug-treated before symptoms occurred, resulting in competitive release. This situation perhaps best mimics the case of prophylactic drug use, or what might occur to new co-infections in high transmission areas where drug use is common. A battery of more complex experiments will be necessary to determine if competitive release occurs when treatment follows symptoms, and when drugs are used to treat semi-immune individuals. The facilitation observed in chronic infections (Figures 1a,1d ) suggests the situation might be very complex. Within-host competition in P. chabaudi is now firmly established [ 8 , 15 , 16 ]. Evidence for competition between co-infecting genotypes in human malaria infections is necessarily indirect, but consistent with this [ 4 ]. In older children and adults, for example, parasite densities do not increase with increasing numbers of clones, thus indicating that parasite clones are not regulated independently [ 17 ]. Given this, and the high frequency of mixed infections in human malaria [ 1 - 3 , 18 ] often consisting of both resistant and sensitive genotypes [ 19 ], and the fact that genetic diversity can be altered by antimalaria prophylaxis [ 20 ], it is highly likely that competitive release of drug resistance also occurs in human malaria. Indeed, a recent study has already implicated release of within-host competition as a key-factor in the spread of drug resistance in Uganda [ 21 ]. Authors' contributions JCdR and RC designed and performed the first experiment, while JCdR and ASB performed the second experiment. JCdR analysed the results and drafted the manuscript. ASB developed the real-time PCR assays for analysis of parasite populations inside mosquitoes. AFR assisted in designing both experiments and writing the manuscript. All authors read and approved of the final version of the manuscript. Competing interests None declared.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517944.xml

479045

Evolution of a Primate Defense against Intragenomic Infiltrators

Anyone who uses a word processor is likely thankful for the spell checker program. But that autocorrect function can introduce errors, “correcting” the spelling of words to fit its stored repertoire, which is decidedly limited. Take that one step further and imagine a rogue program that destroys the coherence and meaning of your prose by swapping out one letter for another throughout the document. That's the situation retroviruses like the human immunodeficiency virus (HIV) face during the course of their infectious cycle, when a protein encoded by the host genome slips into the virus, mutates the virus's genetic material, and alters the viral genome. The gene, APOBEC3G , belongs to a family of primate genes that produce enzymes (in this case, APOBEC3G) that “edit” DNA and RNA, by slipping into viral particles and inducing mutations that replace one base (cytosine) with another (uracil) as the virus undergoes reverse transcription in the host cell's cytoplasm. The edited virus fails to replicate. HIV, in turn, generates a protein called Vif that binds to the APOBEC3G enzyme and targets it for degradation, thereby eliminating its antiviral activity. Since the protein-binding regions that govern these interactions have a direct effect on the fitness of both virus and host, one would expect to see the proteins angling for advantage, with Vif maximizing its ability to recognize APOBEC3G and APOBEC3G doing its best to evade Vif. Such battles are thought to result in frequent mutations that alter the amino acids involved in the interaction; the perpetuation of such advantageous mutations is called positive selection. In this issue of PLoS Biology , Sara Sawyer, Michael Emerman, and Harmit Malik investigate the genetic roots of this battle for evolutionary advantage and find something surprising. As predicted, the APOBEC3G gene is under strong positive selection. But that selection appears to predate the existence of HIV-type viruses. To characterize the selective pressures on APOBEC3G evolution, Sawyer et al. analyzed the gene from twelve primates—New World monkeys, Old World monkeys, and great apes, including humans—spanning 33 million years of evolution. Most of the primate lineages showed evidence of positive selection, indicating that the gene has been under pressure to adapt throughout the history of primate evolution. But viruses like HIV have been found in only five of the primates studied—three African monkeys, chimpanzees, and humans—and appear to be at most one million years old. And HIV infection in human populations is too recent to account for the positive selection of APOBEC3G in humans—so what has been fueling APOBEC3G' s rapid evolution? Genetic conflict between the host antiviral editing enzyme APOBEC3G, and the viral Vif protein leads to rapid fixation of amino acid replacements in both proteins APOBEC3G and Vif interact in T-cells, but the fact that selective pressure on APOBEC3G has been constant over the course of primate evolution suggests that another force is also acting on the gene. Sawyer et al. propose that this force is most likely occurring in germline cells (sperm and egg precursors), which also produce high levels of APOBEC3G and can pass mobile genetic elements on to the next generation. Despite being non-infectious, these elements increase their own copy number in the host genome, moving from one part of the genome to another. The human genome is littered with such “retrotransposons,” and it is these mobile genetic elements, the authors conclude, that likely antagonize APOBEC3G . One class of retrotransposons, called human endogenous retroviruses, acts in many ways like foreign retroviruses. A retrovirus emanating from one's own genome poses less of an immediate threat than a retrovirus like HIV. But the constant efforts of the endogenous retrovirus to “jockey for evolutionary dominance,” the authors conclude, could eventually take a toll and would be expected to provoke efforts to contain it. And it may be that this ancient intragenomic conflict endowed APOBEC3G with the means to do battle with foreign retroviruses like HIV. Sawyer et al. also found evidence that five other APOBEC human genes appear to be engaged in similar conflicts. Combined with the finding that rodents have only one APOBEC3G gene and that five out of the six human APOBEC3 genes have been under positive selection, these results suggest that this gene family expanded in mammalian evolution as a means of defending the germline from the promiscuous intrusions of mobile genetic elements.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC479045.xml

545964

Personal health promotion at US medical schools: a quantitative study and qualitative description of deans' and students' perceptions

Background Prior literature has shown that physicians with healthy personal habits are more likely to encourage patients to adopt similar habits. However, despite the possibility that promoting medical student health might therefore efficiently improve patient outcomes, no one has studied whether such promotion happens in medical school. We therefore wished to describe both typical and outstanding personal health promotion environments experienced by students in U.S. medical schools. Methods We collected information through four different modalities: a literature review, written surveys of medical school deans and students, student and dean focus groups, and site visits at and interviews with medical schools with reportedly outstanding student health promotion programs. Results We found strong correlations between deans' and students' perceptions of their schools' health promotion environments, including consistent support of the idea of schools' encouraging healthy student behaviors, with less consistent follow-through by schools on this concept. Though students seemed to have thought little about the relationships between their own personal and clinical health promotion practices, deans felt strongly that faculty members should model healthy behaviors. Conclusions Deans' support of the relationship between physicians' personal and clinical health practices, and concern about their institutions' acting on this relationship augurs well for the role of student health promotion in the future of medical education. Deans seem to understand their students' health environment, and believe it could and should be improved; if this is acted on, it could create important positive changes in medical education and in disease prevention.

Background Our purpose was to describe both typical and outstanding personal health promotion environments experienced by medical students in U.S. medical schools. Our interest in health promotion among medical students was based on compelling data showing that physicians who have healthy personal habits are more likely to encourage patients to adopt related habits [ 1 ]. However, despite the clear possibility that promoting medical student health should therefore be an innovative, efficient, and effective way to improve patient outcomes, no one has examined the extent to which Deans or students believe that this concept is enacted in medical school. Methods We collected information through four different modalities: a literature review, a written survey of medical school Deans and students, focus groups of preclinical and clinical medical students and dean, and site visits at and interviews with medical schools with reportedly outstanding student health promotion programs. Medical student and dean surveys There were 17 respondents to the Dean Survey (DS), representing 12 of the 16 schools in the nationally representative Healthy Doc (HD) project [ 2 ]. Two deans responded from Mercer, RWJ/UMDNJ, Tulane, UCLA, and University of Pennsylvania, while Colorado, Creighton, Emory, Georgetown, Loma Linda, Medical College of Georgia, and University of Rochester each had one dean respond. It was not always clear whether the Dean of Curriculum or the Dean of Student Affairs was the respondent, therefore we did not differentiate in the analyses by dean type. We also compared Deans' responses with responses (83% response rate) from the 1336 medical students in the Class of 2003 in these Deans' schools, as they were about to begin on wards. All medical students in that class were eligible to complete a self-administered questionnaire covering personal and professional health promotion topics. Our sample of schools was designed to be representative of all U.S. medical schools in our geographic distribution, age (our freshman average was 24 vs. 24 nationally), school size (our schools averaged 563 medical students/school vs. 527 nationally), NIH research ranking (our average was 64 vs. 62 nationally), private/public school balance (51% in private schools vs. 41% nationally), under-represented minorities (13% Blacks, Hispanics, and Native Americans, vs. 11% nationally), and gender (45% women vs. 43% nationally) 5–7 Methodology for gathering medical student data in HD has been more fully described elsewhere [ 2 ]. DS data were collected between February 2002 and April 2003. In analyses comparing DS and HD data, DS schools with two respondents were first averaged so that each school is represented by one value (since repeated measures analysis was not available for the desired analyses). Variation between deans representing a school was quite low for all but one pair. By averaging for the five dean pairs (and consequently having a sample size of 12 rather than 17), the tests are conservative. Student opinion scores were also averaged for each of the twelve schools from which we received Dean responses; these averages were then correlated with the Dean's scores using Spearman's correlation method. For questions with fairly uniform responses by either Deans or students, Wilcoxon's Signed Rank Test was used to test if there were consistent differences between student and Dean opinion. The two variables to be correlated were ordinal variables, each with 5 levels. The type of correlation method was therefore limited to a non-parametric method. Additionally, the raw student data was clustered by school, requiring methods suitable for correlated data. Since the non-parametric method needed is not available for correlated data, we determined that the best method was to take the student mean values at each school to correlate with the dean values. While this ignored the student variability within school, this deficit was balanced by the fact that the much smaller n would require much stronger evidence of a relationship to evince a significant result. Deans were also asked to rate their school relative to other schools. To compare these ratings to students' opinions, schools were ranked using their mean student scores on each question related to prevention and healthy activities encouraged by the school. All 16 schools in the HD cohort were used in the ranking process (1 = highest, 16 = lowest), not just the 12 schools represented by the responding deans, as the 16 were the intended sample, and are representative of US medical schools [ 2 ]. Therefore, the twelve schools for which we have Dean data could have rank values between 1 and 16. For Deans' survey questions without comparative HD data, only simple descriptive statistics are presented. Medical student and dean focus groups For our focus groups (conducted in 2002), we identified opportunities where there would be a wide and nationally representative range of medical schools. The first focus group was convened at the AMSA Chapter Officers' Training Conference (COC) attended by student leaders (primarily rising second years) from every U.S. osteopathic and allopathic medical school. AMSA invited a random sampling of those attending the COC to participate in the focus group. Since the first focus group of students attracted 10 first and second year students, the second focus group was a random sample of 12 clinical students; both student focus groups had an even gender mix. Because Philadelphia has so many medical schools (five), we sampled for the second focus group from those Philadelphia students who were listed in AMSA's membership database. Deans of Primary Care were invited to the third focus group convened at the annual conference of the Association of American Medical Colleges. AMSA used the list of Primary Care Deans and invited a random sample of them to attend the focus group; four attended. An outside contractor (Bennett, Petts & Blumenthal) assisted AMSA in developing the focus group guide, conducted all three focus groups, transcribed the conversations and analyzed the notes for trends in responses. Site visits and interviews In 2002–2003, we identified medical school campuses with intensive programs in medical student well-being through literature and web searches, recommendations from project advisory panel members, results from the Association of Academic Health Centers' American Network of Health Promoting Universities assessment, and participants in the HRSA-funded UME-21 project. Site visits and in-depth interviews were conducted using a protocol which sought information and recommendations on the following topics: • Student well-being programming, including the policies, activities, and evaluation for such efforts as stress reduction, exercise, diet, and mentoring. • Prevention in the curriculum using the Healthy People 2010 objectives and how the various topics are integrated, taught, and evaluated. • Deans' office support (including financial) for prevention in the curriculum and student wellness activities. • Student assessments and recommendations regarding their schools' efforts. Results Survey of Deans and medical students Most surveyed Deans reported that their schools generally support students' health, though fewer Deans believe that their school encourages healthy eating (Table 1 ). Both Deans and students rate their programs rather positively, and their responses are very highly correlated, though Deans consistently rate their programs even more positively than do students (Table 2 ). Table 1 Deans' beliefs regarding their schools' student health promotion efforts % (n) Strongly agree Agree Neither agree nor disagree Disagree Strongly disagree 1.1 Overall, our medical school encourages students to lead healthy lives. 35 (6) 53 (9) 6 (1) 6 (1) 0 1.2 Our medical school curriculum emphasizes preventive medicine in medical practice. 29 (5) 53 (9) 12 (2) 6 (1) 0 1.3 Our medical school encourages extracurricular activities that promote medical students' health. 35 (6) 35 (6) 18 (3) 6 (1) 6 (1) 1.4 Our medical school tries to minimize student stress. 41 (7) 41 (7) 12 (2) 0 6 (1) 1.5 Our medical school has a good system to help students cope with stress. 29 (5) 47 (8) 18 (3) 6 (1) 0 1.6 Our medical school encourages students' healthy eating. 12 (2) 47 (8) 35 (6) 6(1) 0 1.7 Our medical school encourages students to exercise. 24 (4) 53 (9) 12 (2) 6 (1) 6 (1) 1.8 Our medical school discourages students from smoking 41 (7) 47 (8) 12 (2) 0 0 Table 2 Mean scores* for and correlation coefficients ¶ between Deans' and students' responses to statements concerning health promotion at medical school. Deans' mean score Students' mean score r p-value Overall, our medical school encourages students to lead healthy lives. 1.8 2.5 .87 .0002 Our medical school curriculum emphasizes preventive medicine in medical practice. 1.9 2.3 .51 .0912 Our medical school encourages extracurricular activities that promote medical students' health. 2.1 2.7 .54 .0681 Our medical school tries to minimize student stress. 1.8 3.0 .91 <.0001 Our medical school has a good system to help students cope with stress. 2.0 2.9 .70 .0110 Our medical school encourages students' healthy eating. 2.3 3.1 .74 .0064 Our medical school encourages students to exercise. 2.1 2.9 .48 .1139 *Responses were scored 1 for "strongly agree", continuing to 5 for "strongly disagree". Therefore higher scores indicate less agreement with the statement. ¶ Spearman's correlation coefficients. Deans were essentially unanimous in agreeing that faculty members should model healthy behaviors, and that schools should promote health with their students (Table 3 ). However, Deans felt less strongly regarding the need for more training in prevention for primary care physicians, or that a physician must have a healthy lifestyle to effectively counsel patients on healthy lifestyles (Table 3 ). Students also agreed with these statements, but generally to a lesser extent than Deans (Table 4 ). Table 3 Deans' opinions on the role medical schools and physicians should play in promoting healthy behaviors/prevention % (n) Strongly agree Agree Neither agree nor disagree Disagree Strongly disagree 1.9 Medical school faculty members should set a good example for medical students by practicing a healthy lifestyle. 59   (10) 35 (6) 6 (1) 0 0 1.10 Medical schools should encourage students and residents to practice healthy lifestyles. 65   (11) 35 (6) 0 0 0 1.11 Primary Care physicians need more training in prevention. 29 (5) 59 (10) 12 (2) 0 0 1.12 In order to effectively encourage patient adherence to a healthy lifestyle, a physician must adhere to one him/herself. 18 (3) 65 (11) 12 (2) 6 (1) 0 Table 4 Dean and student opinions on the need for schools and faculty to promote healthy lifestyles, the need for more prevention training, and the connection between a physician's healthy lifestyle and his/her counseling efficacy. Deans' mean score Students' Mean score Wilcoxon signed rank test p-value Medical school faculty members should set a good example for medical students by practicing a healthy lifestyle. 1.4 2.1 .0015 Medical schools should encourage their students and residents to practice healthy lifestyles. 1.3 1.9 .0015 Doctors need more training in prevention. 1.8 2.1 .0342 In order to effectively encourage patient adherence to a healthy lifestyle, a physician must adhere to one him/herself. 2.1 2.2 .3804 Three-quarters of Deans believed that their medical schools' attitude toward alcohol was that drinking in moderation was acceptable, though students had more mixed impressions about schools' alcohol attitudes (Table 5 ). Deans believed that their schools did average or better on nearly all health promotion activities (Table 6 ), and students' and Deans' assessments of their schools are highly correlated (Table 7 ). Table 5 Deans' and students' impressions of their medical schools' attitudes about alcohol use Deans Students No obvious attitude 18% 25% Students shouldn't drink at all 6% 13% Drinking in moderation is acceptable 76% 50% Drinking is a good release 0% 11% Table 6 Deans' comparisons of their medical school vs. other medical schools "My school does this (circle choice) compared to other schools." Much more Some-what more An average amount Somewhat less Much less %(n) 3.1 Encourages students to lead healthy lives. 24 (4) 29 (5) 41 (7) 6 (1) 0 3.2 Emphasizes preventive medicine in medical practice. 6 (1) 29 (5) 53 (9) 12 (2) 0 3.3 Encourages extracurricular activities that promote medical students' health. 24 (4) 18 (3) 41 (7) 18 (3) 0 3.4 Encourages students to exercise. 6 (1) 38 (6) 38 (6) 19 (3) 0 3.5 Helps students minimize/cope with stress. 24 (4) 47 (8) 18 (3) 6 (1) 6 (1) 3.6 Discourages students from smoking. 12 (2) 41 (7) 41 (7) 6 (1) 0 3.7 Discourages drinking as a release for students. 6 (1) 12 (2) 65 (1) 18 (4) 0 3.8 Encourages students' healthy eating. 6 (1) 24 (4) 65 (11) 6(1) 0 Table 7 Comparing Dean's perceptions of their school's health promotion in relation to that of other medical schools* with school rankings based on students' opinions. "My school r ¶ p-value ...encourages students to lead healthy lives." .78 .0026 ...emphasizes preventive medicine in medical practice." .45 .1433 ...encourages extracurricular activities that promote medical students' health." .70 .0118 ...encourages students to exercise." .77 .0051 ...helps students minimize/cope with stress." .77/.75 φ .0033/.0047 ...encourages students' healthy eating." .68 .0151 *In which response possibilities were: much less, less, average, more, much more. ¶ Spearman's correlation coefficient. φ The Deans' survey asked one question that queried both minimizing and coping with stress, while students were asked one about each aspect of stress. Correlations are presented for minimizing stress and coping with stress, respectively. We also asked a few narrative questions of the Deans only. Deans indicated that Student Affairs and Student Health offices most often had responsibility for handling medical student wellness (responses of 10 and 4 Deans, respectively). Funds for student wellness activities primarily came from student fees and University budgets (9 and 13 Deans, respectively). Activities' effectiveness was usually unassessed, though some Deans used occasional surveys, data from health programs, student evaluations, and student feedback at meetings/events to help evaluate their programs. Focus groups Our three hours of focus groups yielded little information about students' perceptions of the relationships between their personal and clinical health promotion practices; most students either had not considered this link, or had little to say about it. A few preclinical students reported that their personal wellness is generally linked to their competence as physicians, asserting that "if we sacrifice our own health from studying too long, staying up too late, stressing out too much about exams, we can't take care of other people if we don't watch our own health first." Several clinical students stated that wellness was difficult to achieve ("We're really stressed, basically"), and that having access to help/mentorship might help promote wellness for them: "I [would like] having a designated person to whom students can turn at any time. That would be a hotline . . . A counselor." Deans generally agreed with the concept of putting a mentoring support system in place. However, both students and deans see few resources in the medical schools directed toward student wellness and what programming that is offered is reactive and small in nature. Both the students and the deans discussed wellness in terms of stress and mental well-being, rather than including physical health factors such as nutrition and exercise. Students felt that the best way to teach prevention would be through skill development and role modeling from faculty who incorporate prevention into their practice. The deans proposed that prevention be integrated throughout the curriculum and not be offered as a separate course; students concurred that more prevention instruction would be optimal and acknowledged that a separate course gives the impression that the content is less important and optional. Site visits We visited three medical schools with especially good and abundant practices around medical student health (Emory, Mercer, and Loma Linda Universities), and several other schools with some activities that seemed also to merit mention. These schools were selected for in-depth interviewing, with the best practices outlined in Table Five being used on medical school campuses. Conclusions Prior literature [ref ] has typically examined limited populations of medical students regarding personal health promotion, with few assessments of student well-being or of the success of various interventions, so only limited conclusions can be drawn (a situation that will be improved with this and other publications from HD). However, some trends may be emerging, such as students' health practices being good in some spheres [ 2 ], but not being maintained in medical school [ 3 ] and residency [ 4 , 5 ], with an increase in alcohol consumption, and a decrease in socialization and exercise[ 6 ]. Poor medical student health habits also include maladaptive behaviors such as students going to school when sick, self-prescribing, and under-using medical care [ 7 ]. While medical students' positive health behaviors may be encouraged by their expanding knowledge and peer and role model support [ 2 ], some students may avoid treatment because of concerns that others' knowledge of their illness may place them in academic jeopardy [ 8 ]. Medical student and physician health is of inherent interest, but it is especially of concern because of the well-documented link between physicians' personal health practices and their patient counseling practices [ 1 ]. Despite the clear need in medical school for an emphasis on student wellness, the number of health promotion programs is declining[ 9 , 10 ]: competing demands for faculty time and financial resources are barriers to program implementation, and there is virtually no systematic study of the effects of such programs beyond our HD work with surveying students' counseling practices and validating these surveys with simulated patients (in review). We found consistent support from both Deans and students for medical schools' encouraging healthy student behaviors, though modest follow-through on this support. Though students seemed to have thought little about the relationships between their own personal and clinical health promotion practices, we were especially impressed with the Deans' unanimity that faculty members should model healthy behaviors. The deans' support of the relationship between physicians' personal and clinical health practices, and concern about their institutions' acting on this relationship bodes well for the role of HD principles in the future of medical education. The correlation between students' and deans' responses suggests that deans understand well their students' health environments. If acted on, this finding (coupled with deans' beliefs that the environment can and should be improved) could create important positive changes in medical education and in disease prevention. Competing interest The author(s) declare that they have no competing interests. Authors' contributions EF co-developed the protocol, helped guide analyses, and drafted and revised the manuscript. JH co-developed the protocol, obtained funding, and helped edit the manuscript. LE co-developed the protocol, performed analyses, and helped edit the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545964.xml

548501

Gene expression profiling revealed novel mechanism of action of Taxotere and Furtulon in prostate cancer cells

Background Both Taxotere and Capecitabine have shown anti-cancer activity against various cancers including prostate cancer. In combination, Taxotere plus Capecitabine has demonstrated higher anti-cancer activity in advanced breast cancers. However, the molecular mechanisms of action of Taxotere and Capecitabine have not been fully elucidated in prostate cancer. Methods The total RNA from PC3 and LNCaP prostate cells untreated and treated with 2 nM Taxotere, 110 μM Furtulon (active metabolite of Capecitabine), or 1 nM Taxotere plus 50 μM Furtulon for 6, 36, and 72 hours, was subjected to Affymetrix Human Genome U133A Array analysis. Real-time PCR and Western Blot analysis were conducted to confirm microarray data. Results Taxotere and Furtulon down-regulated some genes critical for cell proliferation, cell cycle progression, transcription factor, cell signaling, and oncogenesis, and up-regulated some genes related to the induction of apoptosis, cell cycle arrest, and differentiation in both cell lines. Taxotere and Furtulon also up-regulated some genes responsible for chemotherapeutic resistance, suggesting the induction of cancer cell resistance to these agents. Conclusions Taxotere and Furtulon caused the alternation of a large number of genes, many of which may contribute to the molecular mechanisms by which Taxotere and Furtulon inhibit the growth of prostate cancer cells. This information could be utilized for further mechanistic research and for devising optimized therapeutic strategies against prostate cancer.

Background Prostate cancer is the most common cancer and the second leading cause of cancer related deaths in men in the United States with an estimated 230,110 new cases and 29,500 deaths in 2004 [ 1 ]. Initial treatment for prostate cancer is usually androgen-ablative therapy, radiotherapy or radical prostatectomy and the patients respond to androgen-ablative therapy in the beginning of treatment. However, many patients eventually fail this therapy and die of recurrent androgen-independent prostate cancer and metastasis [ 2 ]. Up to 30% of men undergoing radical prostatectomy will also relapse, often as a result of micrometastatic cancer present at the time of surgery [ 3 ]. The efficacy of cytotoxic chemotherapy for treatment of hormone-refractory prostate cancer has been tested in clinical trials. In general, response rates of <10% were observed in single-agent studies [ 2 ]. Thus, there is a tremendous need for the development of mechanism-based targeted strategies for treatment of prostate cancer. Taxotere, a member of taxane family, is semi-synthesized from an inactive taxoid precursor extracted from the needles of the European yew, Taxus baccata . Its known basic mechanism of action is that it binds to tubulin and deranges the equilibrium between microtubule assembly and disassembly during mitosis [ 4 ]. Stabilization of microtubules by Taxotere impairs mitosis and exerts an anticancer effect in tumors [ 4 ]. Taxotere has shown clinical activity in wide spectrum of solid tumors including breast, lung, ovarian, prostate cancers, etc [ 5 , 6 ]. In metastatic breast, lung, and ovarian cancer, randomized trials have shown that Taxotere-containing therapies are superior to or as effective as established standard chemotherapeutic regimens and are often associated with an improved safety profile [ 6 ]. Clinical trials have also found that weekly Taxotere in patients with metastatic hormone-refractory prostate cancer is associated with improvements in clinical benefit response and quality of life [ 7 , 8 ]. Thus, Taxotere is currently considered to be among the most important anticancer drugs in cancer chemotherapy. In addition to its effects on microtubules, Taxotere also induces apoptosis with down-regulation of bcl XL and bcl-2, and upregulation of p21 WAF1 and p53 [ 9 , 10 ]. We have previously reported that Taxotere down-regulates some genes for cell proliferation, mitotic spindle formation, transcription factors, and oncogenesis, and up-regulates some genes related to induction of apoptosis and cell cycle arrest in prostate cancer cells, suggesting the pleiotropic effects of Taxotere on prostate cancer cells [ 11 ]. Capecitabine is an orally administered systemic prodrug of 5'-deoxy-5-fluorouridine (5-DFUR or Furtulon) which is converted to 5-fluorourasil (5-FU) [ 12 ]. Capecitabine is readily absorbed from the gastrointestinal tract. In human and animals, carboxylesterase hydrolyzes much of Capecitabine to 5'-deoxy-5-flurocytidine (5-DFCR). Cytidine deaminase, an enzyme found in most tissues including tumors, subsequently converts 5-DFCR to 5-DFUR. The enzyme, thymidine phosphorylase (dThdPase), then hydrolyzes 5-DFUR to the active drug 5-FU both in vivo and in vitro . After being converted to 5-FU, Capecitabine inhibits essential cellular biosynthetic processes and is incorporated into DNA to inhibit normal bioprocess function of the cell [ 13 ]. Capecitabine has shown anti-tumor activity in various cancers including prostate cancer [ 14 - 16 ]. 5-FU-based chemotherapy improves overall and disease-free survival of patients with cancer. However, response rates for 5-FU-based chemotherapy are low and a large number of tumors eventually becomes resistant to 5-FU [ 13 , 17 ]. Clinical trials showed that chemotherapeutic combination with Taxotere and Capecitabine resulted in improved objective response rate and overall survival without a significant increase in the treatment related adverse effects in metastatic breast cancer and advanced non-small cell lung carcinoma [ 18 , 19 ]. However, the molecular mechanism(s) of action of Taxotere and Capecitabine have not been fully elucidated. In this study, we utilized high-throughput gene chip, which contains 22,215 known genes, to determine the alternation of gene expression profiles of hormone insensitive (PC3) and sensitive (LNCaP) prostate cancer cells exposed to Taxotere and Furtulon. The purpose of this study was: 1) to identify novel genes that have key roles in cancer cell killing and resistance induced by Taxotere and/or Furtulon; 2) to test whether similar genes are altered by Taxotere and Furtulon; 3) to test whether combination therapy alters genes that may reflect better treatment outcome or may provide information whether combination therapy could be antagonistic; 4) finally to provide molecular information for further mechanistic investigation and future clinical application. Methods Cell culture and growth inhibition PC3 (ATCC, Manassas, VA) and LNCaP (ATCC) human prostate cancer cells were cultured in RPMI-1640 media (Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum and 1% penicillin and streptomycin in a 5% CO 2 atmosphere at 37°C. Taxotere (Aventis Pharmaceuticals, Bridgewater, NJ) was dissolved in DMSO to make 4 μM stock solution. Furtulon (Roche, Palo Alto, CA) was dissolved in PBS to make 100 mM stock solution. For growth inhibition, PC3 and LNCaP cells were treated with Taxotere (1, 2, and 4 nM), Furtulon (50, 100, and 200 μM), or 1 nM Taxotere plus 50 μM Furtulon for one to three days. Control PC3 and LNCaP cells received 0.01% DMSO or 0.1% PBS for same time points. After treatment, PC3 and LNCaP cells were incubated with MTT (0.5 mg/ml, Sigma, St. Louis, MO) at 37°C for two hours and then with isopropyl alcohol at room temperature for one hour. The spectrophotometric absorbance of the samples was determined by using ULTRA Multifunctional Microplate Reader (TECAN, Durham, NC) at 595 nm. The concentrations of Taxotere and Furtulon used for our in vitro studies are easily achievable in humans, suggesting that our experimental results are relevant for human applications. The experiment was repeated three times and a t test was performed to verify the significance of cell growth inhibition after treatment. Microarray analysis for gene expression profiles PC3 and LNCaP cells were treated with 2 nM Taxotere, 110 μM Furtulon, or 1 nM Taxotere plus 50 μM Furtulon for 6, 36, and 72 h. Total RNA from each sample was isolated by Trizol (Invitrogen, Carlsbad, CA) and purified by RNeasy Mini Kit and RNase-free DNase Set (QIAGEN, Valencia, CA) according to the manufacturer's protocols. cDNA for each sample was synthesized by Superscript cDNA Synthesis Kit (Invitrogen, Carlsbad, CA) using the T7-(dT) 24 primer instead of the oligo(dT) provided in the kit. Then, the biotin-labeled cRNA was transcripted in vitro from cDNA by using BioArray HighYield RNA Transcript Labeling Kit (ENZO Biochem, New York, NY), and purified by RNeasy Mini Kit. The purified cRNA was fragmented by incubation in fragmentation buffer (40 mM Tris-acetate pH 8.1, 100 mM KOAc, 30 mM MgOAc) at 95°C for 35 min and chilled on ice. The fragmented labeled cRNA was applied to Human Genome U133A Array (Affymetrix, Santa Clara, CA), which contains 22,215 human gene probes, and hybridized to the probes in the array. After washing and staining, the arrays were scanned. Two independent experiments were performed to verify the reproducibility of results. Microarray data normalization and analysis The gene expression levels of samples were normalized and analyzed by using Microarray Suite, MicroDB™, and Data Mining Tool software (Affymetrix, Santa Clara, CA). The absolute call (present, marginal, absent) and average difference of 22,215 gene expressions in a sample, and the absolute call difference, fold change, average difference of gene expressions between two or several samples were normalized and identified using these software. Statistical analysis of the mean expression average difference of genes, which show >2 fold change, was performed using a t test between treated and untreated samples. Clustering and annotation of the gene expression were analyzed by using Cluster and TreeView [ 20 ], Onto-Express [ 21 ], and GenMAPP [ 22 ]. Genes that were not annotated or not easily classified were excluded from the functional clustering analysis. Real-time RT-PCR analysis for gene expression To verify the alterations of gene expression at the mRNA level, which appeared on the microarray, we chose representative genes (Table 1 ) with varying expression profiles for real-time RT-PCR analysis. Two micrograms of total RNA from each sample were subjected to reverse transcription using the Superscript first strand cDNA synthesis kit (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. Real-time PCR reactions were then carried out in a total of 25 μL reaction mixture (2 μl of cDNA, 12.5 μl of 2X SYBR Green PCR Master Mix, 1.5 μl of each 5 μM forward and reverse primers, and 7.5 μl of H 2 O) in SmartCycler II (Cepheid, Sunnyvale, CA). The PCR program was initiated by 10 min at 95°C before 40 thermal cycles, each of 15 s at 95°C and 1 min at 60°C. Data were analyzed according to the comparative Ct method and were normalized by actin expression in each sample. Melting curves for each PCR reaction were generated to ensure the purity of the amplification product. Western blot analysis We also conducted Western Blot analysis to verify the alterations of genes at the level of translation for selected genes with varying expression profiles. The PC3 and LNCaP cells were treated with 1 and 2 nM Taxotere or 50 and 110 μM Furtulon for 24, 48, and 72 hours. After treatment, the cells were lysed in 62.5 mM Tris-HCl and 2% SDS, and protein concentration was measured using BCA protein assay (PIERCE, Rockford, IL). The proteins were subjected to 10% or 14% SDS-PAGE, and electrophoretically transferred to nitrocellulose membrane. The membranes were incubated with anti-cathepsin C (1:200, Santa Cruz, Santa Cruz, CA), anti-p16 (1:200, Santa Cruz, Santa Cruz, CA), anti-IKKα (1:100, Santa Cruz, Santa Cruz, CA), anti-p21 WAF1 (1:500, Upstate, Lake Placid, NY), anti-Bax (1:10000, Trevigen, Gaithersburg, MD), anti-survivin (1:200, Alpha Diagnostic, San Antonio, TX), anti-CDC2 (1:200, Santa Cruz, Santa Cruz, CA), anti-cyclin A (1:250, NeoMarkers, Union City, CA), anti-cyclin B (1:200, Santa Cruz, Santa Cruz, CA), anti-cyclin E (1:250, NeoMarkers), and anti-β-actin (1:10000, Sigma, MO) primary antibodies, and subsequently incubated with secondary antibody conjugated with fluorescence dye. The signal was then detected and quantified by using Odyssey infrared imaging system (LI-COR, Lincoln, NE). Results Cell growth inhibition MTT assay showed that the treatment of PC3 and LNCaP prostate cancer cells with Taxotere, Furtulon, or lower concentration of Taxotere plus Furtulon resulted in dose and time-dependent inhibition of cell proliferation (Figure 1 ), demonstrating the inhibitory effect of Taxotere and Furtulon on the growth of PC3 and LNCaP prostate cancer cells. Regulation of mRNA expression by Taxotere and Furtulon treatment Microarray analysis showed that the alterations of gene expression were occurred as early as 6 hours of Taxotere and/or Furtulon treatment, and were more evident with longer treatment (Table 2 and 3 ). Clustering analysis based on gene function showed down-regulation of some genes for cell proliferation and cell cycle progression (cyclin A, cyclin F, CDC2, CDK2, etc), transcription factors (transcription factor A, ATF5, TAF1131L, FOXM1, etc), and oncogenesis (GRO oncogene, BRCA1 associated RING domain, tumor-associated nuclear protein p120, etc) in Taxotere and/or Furtulon treated prostate cancer cells (Table 2 and 3 ). In contrast, Taxotere and/or Furtulon up-regulated some genes that are related to the induction of apoptosis (GADD45A, GADD45B, etc), cell cycle arrest (p21 CIP1 , VDUP1, BTG, etc), and tumor suppression (Table 2 and 3 ). Combination treatment with Taxotere and Furtulon also altered expression of some genes (CDC27, CDK9, p18, IKKα, etc) that showed no change in mono-treatment (Table 4 and 5 ), suggesting the synergic effects of combination treatment on some genes. Taxotere and Furtulon also up-regulated some genes (S-100P, ALDH1A3, casein kinase, annexin, etc) responsible for chemotherapeutic resistance, suggesting the induction of cancer cell resistance to these agents (Table 2 and 3 ). Taxotere and Furtulon also showed differential effects on PC3 cells with alteration of metastasis-related genes and on LNCaP cells with down-regulation of survivin, cyclin B & E, CDC2, CDC25, and specifically AR by Furtulon, suggesting their effects mediated by both AR-independent and dependent pathways (Table 2 and 3 ). Target verification by real-time RT-PCR and western blot To verify the alterations of gene expression at the mRNA level, which appeared on the microarray, we chose representative genes with varying expression profiles for real-time RT-PCR and Western Blot analysis. The results of real-time RT-PCR for these selected genes were in direct agreement with the microarray data (Figure 2 ). The same alternations of gene expression were observed by real-time RT-PCR analysis, although the fold change in the expression level was not exactly same between these two different analytical methods. The results of Western Blot analysis were also in direct agreement with the microarray and real-time RT-PCR data (Figure 3 and our earlier report [ 11 ]). These results support the findings obtained from microarray experiments. Discussion It has been known that Taxotere binds to microtubules while Capecitabine is incorporated into DNA, inhibiting the bioprocess in cancer cells [ 4 , 13 ]. However, the precise molecular mechanisms for inhibiting cancer cell growth by Taxotere and/or Capecitabine have not been fully elucidated. From gene expression profiles of Taxotere and/or Capecitabine treated prostate cancer cells, we found that these chemotherapeutic agents caused alterations in the expression of many genes related to the control of cell proliferation, apoptosis, transcription, translation, cell signaling, oncogenesis, and angiogenesis (Figure 4 ), although the cellular target of Taxotere or Capecitabine appears to be different. It has been well known that CDCs regulate the molecules related to the cell cycle initiation and progression and that cyclins associate with cyclin-dependent protein kinases (CDKs) and CDCs to control the process of cell cycle [ 23 , 24 ]. The CDK inhibitors including p21 WAF1 , p16 INK4A , and p18 INK4C have been demonstrated to arrest the cell cycle and inhibit the growth of cancer cells [ 23 , 24 ]. Our results showed that Cyclins (cyclin A2, cyclin E2, cyclin F, cyclin B1), CDK2, CDC2, and other cell growth promotion genes (pescadillo, spermidine synthase, mitotin) [ 25 - 27 ] were down-regulated in Taxotere and/or Furtulon treated prostate cancer cells, while CDK inhibitor p21 WAF1 and other growth inhibitor genes (BTG2, VDUP1, anti-proliferative B-cell translocation gene 1) [ 28 , 29 ] were up-regulated, suggesting that Taxotere and/or Furtulon inhibited the growth of prostate cancer cells through the arrest of cell cycle and the inhibition of cell proliferation (Figure 4 ). The down-regulation of CDC27, CDK9, EGF, and FGF12B, and up-regulation of p16 INK4A and p18 INK4C were also observed in combination treatment but not in mono-treatment, suggesting the synergic effect of combination treatment. These observations are novel in Taxotere and/or Furtulon treated prostate cancer cells. Induction of apoptosis by chemotherapeutic agents also leads to the inhibition of cancer cell growth. It has been reported that Taxotere is able to induce apoptosis by caspase-3 dependent or independent cell death mechanism [ 30 ]. Capecitabine may induce apoptosis through Fas/FasL or Bax/Bcl-2 pathway [ 31 , 32 ]. From gene expression profile, we found that Taxotere and/or Furtulon increased level of growth arrest and DNA-damage-inducible alpha (GADD45A), GADD45B, p53 regulated PA26 nuclear protein (PA26), and p53-induced protein 11 (PIG11), all of which are related to the induction of apoptotic processes. GADD45A and GADD45B have been known to promote apoptosis and regulate G2/M arrest [ 33 ]. PA26 is a target of the p53 tumor suppressor and a member of the GADD family with the properties of inducing apoptosis [ 34 ]. PIG11 as a downstream target of p53 is also involved in the apoptotic processes [ 35 ]. The combination treatment also showed down-regulation of negative regulator of programmed cell death ICH-1S and Bcl-2-associated transcription factor, which was not occurred in mono-treatment. The induction of apoptosis mediated by GADD45A, GADD45B, PA25, and PIG11 could be another molecular mechanism by which Taxotere and/or Furtulon inhibit the growth of prostate cancer cells. We also found that Taxotere and/or Furtulon inhibited the expression of transcription factors (FOXM1, ATF5, TFAM, TAFII31L), translation factors (EIF1A, EIF5A), oncogene (GRO1, GRO3, BRCA1-associated protein, tumor-associated nuclear protein p120), and heat shock protein, and up-regulated the genes for differentiation (prostate differentiation factor). These results are novel, and suggest the beneficial effects of Taxotere and/or Furtulon on the inhibition of cancer cell growth and oncogenesis. It is important to note that Taxotere and/or Furtulon also up-regulated the expression of some genes which are known to induce cell resistance to chemotherapeutic agents and to favor cell survival. Among these genes, calcium-binding protein S100P has been found to be highly expressed in cells which develop acquired resistance to anti-tumor agents [ 36 ]. The overexpression of aldehyde dehydrogenase 1 (ALDH1) has also been detected solely in classical multidrug resistance cancer cells [ 37 , 38 ]. It has been reported that Annexin-I, casein kinase 1, and cisplatin-resistance associated protein expressions modulate drug resistance in tumor cells [ 39 , 40 ]. The up-regulation of these molecules by Taxotere and/or Furtulon could induce cell resistance to chemotherapeutic agents. Also, Taxotere and/or Furtulon were found to up-regulate the expression of Notch 3, angiopoietin, activating transcription factor 3, which could favor cell survival [ 41 - 43 ]. Further in depth mechanistic studies are needed to address these issues. The investigation on overcoming these unbeneficial effects with other agents must be devised, which is ongoing in our laboratory. Taxotere showed no effect on AR expression while Furtulon down-regulated AR expression in LNCaP cells, suggesting that the combination could be superior in AR-positive cells. The genes altered by Taxotere and/or Furtulon with respect to the control of cell growth, apoptosis, transcription, oncogenesis, and metastasis in androgen insensitive PC3 cells are different from that in androgen sensitive LNCaP cells, suggesting that the effects of Taxotere and Furtulon may be mediated by both AR-dependent and independent signaling pathways. We observed up-regulation of tissue inhibitor of metalloproteinase 1 (TIMP1), TIMP2, and protease inhibitor 3 in Taxotere and/or Furtulon treated PC3 cells, suggesting that Taxotere and/or Furtulon may exert anti-metastatic effect. However, we also observed increase in the expression of MMP1, MMP9, cathepsin B, uPA, and tPA in Taxotere and Furtulon treated PC3 cells, therefore, more experimental studies are needed to reveal the overall effect of Taxotere and Furtulon on metastatic processes. These results were not observed in androgen sensitive LNCaP cells, suggesting difference in effects that could be mediated through different cell signal transduction pathways. Conclusions In conclusion, Taxotere and/or Furtulon directly and indirectly caused changes in the expression of many genes that are critically involved in the control of cell proliferation, apoptosis, transcription, translation, oncogenesis, angiogenesis, metastasis, and drug resistance (Figure 4 ). These findings could provide molecular information for further investigation on the mechanisms by which Taxotere and Furtulon exerts their pleiotropic effects on prostate cancer cells. These results could also be important in devising mechanism-based targeted therapeutic strategies for prostate cancer, especially in devising combination therapy for drug resistant prostate cancers. However, further in-depth investigations are needed in order to establish cause and effect relationships between these altered genes and therapeutic response in prostate cancer cells. Competing interests The author(s) declare that they have no competing interests. Authors' contributions FHS designed the study and prepared the manuscript. YL carried out cell growth inhibition, microarray and Western Blot analysis and drafted the manuscript. MH participated in the design of the study. RL and SHS carried out real-time PCR. JE prepared Furtulon reagent. All authors read and approved the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548501.xml

548273

cAMP response element binding protein (CREB) activates transcription via two distinct genetic elements of the human glucose-6-phosphatase gene

Background The enzyme glucose-6-phosphatase catalyzes the dephosphorylation of glucose-6-phosphatase to glucose, the final step in the gluconeogenic and glycogenolytic pathways. Expression of the glucose-6-phosphatase gene is induced by glucocorticoids and elevated levels of intracellular cAMP. The effect of cAMP in regulating glucose-6-phosphatase gene transcription was corroborated by the identification of two genetic motifs CRE1 and CRE2 in the human and murine glucose-6-phosphatase gene promoter that resemble cAMP response elements (CRE). Results The cAMP response element is a point of convergence for many extracellular and intracellular signals, including cAMP, calcium, and neurotrophins. The major CRE binding protein CREB, a member of the basic region leucine zipper (bZIP) family of transcription factors, requires phosphorylation to become a biologically active transcriptional activator. Since unphosphorylated CREB is transcriptionally silent simple overexpression studies cannot be performed to test the biological role of CRE-like sequences of the glucose-6-phosphatase gene. The use of a constitutively active CREB2/CREB fusion protein allowed us to uncouple the investigation of target genes of CREB from the variety of signaling pathways that lead to an activation of CREB. Here, we show that this constitutively active CREB2/CREB fusion protein strikingly enhanced reporter gene transcription mediated by either CRE1 or CRE2 derived from the glucose-6-phosphatase gene. Likewise, reporter gene transcription was enhanced following expression of the catalytic subunit of cAMP-dependent protein kinase (PKA) in the nucleus of transfected cells. In contrast, activating transcription factor 2 (ATF2), known to compete with CREB for binding to the canonical CRE sequence 5'-TGACGTCA-3', did not transactivate reporter genes containing CRE1, CRE2, or both CREs derived from the glucose-6-phosphatase gene. Conclusions Using a constitutively active CREB2/CREB fusion protein and a mutant of the PKA catalytic subunit that is targeted to the nucleus, we have shown that the glucose-6-phosphatase gene has two distinct genetic elements that function as bona fide CRE. This study further shows that the expression vectors encoding C2/CREB and catalytic subunit of PKA are valuable tools for the study of CREB-mediated gene transcription and the biological functions of CREB.

Background The glucose-6-phosphatase system consists of the glucose-6-phosphate catalytic subunit (EC 3.1.3.9), embedded in the membrane of the endoplasmic reticulum (ER) via nine transmembrane domains, and the membrane spanning translocases, responsible in carrying either the substrate into the ER or the product from the ER [ 1 ]. Transport of substrate and product is necessary due to the orientation of the active site of the glucose-6-phosphatase enzyme towards the luminal side of the ER. Glucose-6-phosphate, the end product of both gluconeogenesis and glycogenolysis in the liver, is hydrolyzed by the glucose-6-phosphatase system allowing the liberation of glucose into the circulation. Thus, glucose-6-phosphatase plays a key role in the homeostasis of blood glucose. Mutations in the gene encoding the catalytic subunit of glucose-6-phosphatase are responsible for the development of glycogen storage disease type 1, also known as Gierke disease [ 2 ]. In animal models of diabetes mellitus, glucose-6-phosphatase activity is increased along with mRNA and protein levels [ 3 ]. Accordingly, inhibitors of glucose-6-phosphatase or the G6PT transporter are used for the treatment of type 2 diabetes. The glucose-6-phosphatase encoding gene is regulated by a variety of extracellular signaling molecules, including glucose, insulin, glucocorticoids, and cAMP. Insulin decreases the level of glucose-6-phosphatase mRNA in the liver and in hepatoma cells and three functionally distinct insulin response elements have been identified [ 4 ]. Glucocorticoids elevate glucose-6-phosphatase promoter activity mediated by a glucocorticoid response element and hepatocyte nuclear factor 1 [ 5 ]. Conflicting results were published concerning the stimulation of glucose-6-phosphatase promoter activity by elevated intracellular cAMP concentrations. While dibutyryl cAMP alone did not significantly stimulate transcription of a luciferase reporter gene under control of 1.2 kb of the human glucose-6-phosphatase promoter in H4IIIE hepatoma cells [ 6 ], a later report by the same group described a ≈ 2-fold stimulation of glucose-6-phosphatase promoter activity with dibutyryl cAMP [ 7 ]. The sequence from -161 to -152 of the human glucose-6-phosphatase promoter, including the motif 5'-TTTACGTAA-3', was proposed to mediate the effect of dibutyryl cAMP on reporter gene transcription [ 7 ]. In HepG2 cells a glucose-6-phosphatase promoter/chloramphenicol acetyltransferase reporter gene showed a 2 to 3-fold enhancement in transcription following stimulation of the cells with dibutyryl cAMP [ 8 ]. Here, the sequence from -136 to -129 of the human glucose-6-phosphatase gene, including the motif 5'-TTGCATCAA-3', was proposed to be essential to couple dibutyryl cAMP stimulation with enhanced reporter gene transcription [ 8 ]. The most important and best characterized protein that connects an elevation of intracellular cAMP concentrations with enhanced transcription of selected genes is the CRE binding protein CREB, a basic region leucine zipper protein. CREB plays an essential role in the regulation of glucose-6-phosphatase gene transcription in the liver, as exemplified by the fact that transgenic mice expressing a dominant-negative CREB mutant in the liver show reduced mRNA levels of glucose-6-phosphatase [ 9 ]. CREB not only mediates stimulus-transcription coupling of the cAMP signaling pathway but functions as a point of convergence of many other signaling molecules involving calcium, neurotrophins, tumor promoters, and growth factors [ 10 ]. CREB is inactive in the dephosphorylated state and turns into an activator upon phosphorylation. The key enzyme leading to CREB activation is the cAMP-dependent protein kinase (PKA), but CREB serves also as a substrate for calcium/calmodulin-dependent protein kinase IV and the mitogen-and stress-activated kinases MSK1 and 2 [ 11 , 12 ]. To study CREB regulated gene transcription, a signaling cascade needs to be activated that leads to phosphorylation and activation of CREB. This can be accomplished by adding cAMP analogues such as dibutyryl cAMP that passes the plasma membrane or via a direct activation of adenylate cyclase by forskolin. This experimental setting is problematic in that high levels of phosphodiesterase enzymes in the cells may immediately hydrolyze cAMP, with the result that no cAMP-mediated gene transcription can be monitored [ 13 ]. Moreover, elevated levels of cAMP may additionally activate the EPAC/Rap1 pathway, making it somewhat difficult to attribute the effect of cAMP solely to the cAMP/PKA signaling pathway. Many functions of cAMP previously attributed to PKA may be in fact the result of EPAC activation [ 14 ]. Instead of increasing the intracellular levels of cAMP some investigators employed an overexpression strategy using the catalytic subunit of cAMP-dependent protein kinase to analyze cAMP regulated genes. This approach has the advantage that by excluding a parallel signaling cascade via the cAMP-inducible nucleotide exchange factor EPAC, only the biological outcome of cAMP-dependent protein kinase activation is studied. However, the translocation of the catalytic subunit of cAMP-dependent protein kinase into the nucleus is not very efficient and may rely solely on diffusion [ 15 ]. Thus high amounts of expression vector in the range of 1 to 5 μg/plate are often transfected [ 16 - 18 ] in order to observe an effect on gene transcription. Naturally, these high amounts of expressed catalytic subunit are far away from the physiological concentrations within the cells. Here, we report on the regulation of glucose-6-phosphatase gene transcription by CREB and PKA. Using a constitutively active CREB2/CREB fusion protein, that does not require phosphorylation for activation, and a nuclear-targeted mutant of the catalytic subunit of PKA, we show that CREB activates transcription of glucose-6-phosphatase promoter/luciferase reporter genes via two distinct genetic elements. In contrast, the bZIP protein ATF2, known to compete with CREB for binding to the canonical CRE, did not transactivate these reporter genes containing CRE1, CRE2, or both CREs derived from the glucose-6-phosphatase gene. Results The proximal region of the human glucose-6-phosphatase gene contains two cyclic AMP response element like motifs Fig. 1A shows part of the proximal region of the human glucose-6-phosphatase gene. As indicated, two CRE-like sequences are present, resembling the canonical CRE sequence 5'-TGACGTCA-3'. The distal CRE-like site (CRE1) and the proximal CRE-like site (CRE2) differ on two or three positions, respectively, in comparison to the canonical CRE. To investigate the biological function of these elements we constructed a battery of reporter plasmids containing both CRE-like elements or only one of them (Fig. 1B,C ). Mutations were introduced into CRE1 and CRE2 sites to prevent CREB binding [ 19 ]. As a control, we generated a reporter plasmid containing two copies of a CRE/AP1-like element derived from the human tumor necrosis factor (TNF) α gene promoter (Fig. 1D ). The TNFα gene belongs to the target genes of CREB [ 10 ]. Additionally, the reporter genes contained a TATA box derived from the HIV long terminal repeat, an initiator element from the adenovirus major late promoter and the luciferase open reading frame. A constitutively active CREB2/CREB fusion protein transactivates reporter genes with either CRE1 or CRE2 of the glucose-6-phosphatase gene in their regulatory regions The fact that unphosphorylated CREB is transcriptionally silent excludes a simple overexpression strategy to analyze the biological impact of the CRE-like sequences within the glucose-6-phosphatase gene. The use of natural inducers of CREB such as glucagon or epinephrine, that all elevate the cAMP concentration in the cells, also triggers pleiotropic responses by activating the cAMP-dependent protein kinase and the EPAC/Rap1 pathway, so that the biological outcome cannot be attributed solely to CREB activation. We therefore designed a constitutively active CREB2/CREB fusion protein to uncouple the investigation of glucose-6-phosphatase gene transcription from signaling pathways in the cell. Fig. 2A shows the structural domains of the basic region leucine zipper (bZIP) transcription factors CREB and CREB2. Both proteins contain a bZIP domain on their C-termini responsible for dimerization and DNA-binding. The N-termini of CREB and CREB2 contain activation domains. While the activation domain of CREB2 is constitutively active and transferable to heterologous DNA-binding domains [ 20 ], the major activation domain of CREB is controlled by phosphorylation. We expressed the bZIP domain of CREB, which is responsible for DNA binding and dimerization, as a fusion protein with the constitutively active transcriptional activation domain of CREB2, generating the chimeric transcription factor C2/CREB (Fig. 2A ). The C2/CREB fusion protein contains additionally an immunological tag used for the detection of the protein. Proteins derived from nuclear extracts of HepG2 human hepatoma cells (mock) or HepG2 cells transfected with an expression vector encoding C2/CREB were fractionated by SDS-PAGE. The fusion protein was identified by Western Blot analysis using antibodies targeting the FLAG epitope. Fig. 2B shows that the CREB2/CREB fusion protein was synthesized as expected. To study the regulation of the glucose-6-phosphatase promoter/luciferase reporter genes by C2/CREB, we decided to use the human hepatoma cell line HepG2, as it has been reported that activation of the cAMP signaling pathway stimulates glucose-6-phosphatase gene transcription in these cells [ 17 , 21 , 22 ]. HepG2 cells were transfected with one of the luciferase reporter plasmids pG6PCRE1/CRE2luc, pG6PCRE1mut/CRE2luc, pG6PCRE1/CRE2mutluc, pG6PCRE1luc or pG6PCRE2luc together with either the "empty" expression vector pCMV5 (denoted "-") or an expression vector encoding C2/CREB (plasmid pCMV-FLAG-C2/CREB) (denoted "+"). As a control, we transfected the pTNFα(CRE/AP1) 2 luc reporter plasmid that contains two copies of the composite CRE/AP1 element of the TNFα promoter. We transfected additionally plasmid pRSVβ, encoding β-galactosidase under the control of the Rous sarcoma virus long-terminal repeat, to correct for variations in transfection efficiencies. Luciferase activities were normalized for transfection efficiency by dividing luciferase light units by β-galactosidase activities. The results of the transfection experiments are depicted in Fig. 3 . Transcription of the pG6PCRE1/CRE2luc reporter gene, that contains both CRE-like sequences in its regulatory region, was strongly induced following expression of C2/CREB (Fig. 3A , upper panel). Mutation of CRE1 or CRE2 did not abolish the transactivation potential of C2/CREB (Fig. 3A , middle, bottom panel), indicating that both CRE-like sequences function independently as bona fide CREs. This conclusion was confirmed by transfection experiments of reporter plasmids having only one of the CRE-like sequences in the regulatory regions. Fig. 3B shows that both reporter genes pG6PCRE1luc and pG6PCRE2luc were transactivated by C2/CREB. Finally, C2/CREB also transactivated the pTNFα(CRE/AP1) 2 luc reporter gene via the CRE/AP1 element (Fig. 3C ). We conclude that both CRE1 and CRE2 motifs of the glucose-6-phosphatase gene promoter function as bona fide CREs. Expression of a nuclear-targeted catalytic subunit of cAMP-dependent protein kinase The structural domains of the catalytic subunit of cAMP-dependent protein kinase (ATP:protein phosphotransferase, EC 2.7.1.37) are depicted in Fig. 4A . The protein is myristylated on the N-terminus, but myristylation is nonessential for enzyme activation [ 23 ]. We modified the N- terminal region of the protein by adding a nuclear targeting signal derived from the SV40 large T antigen (NLS) and a FLAG epitope to facilitate immunological detection. This mutant termed NLSCα should be sorted to the nuclear compartment, due to the presence of the NLS. HepG2 cells were transfected with an expression vector encoding NLSCα. Nuclear extracts of these cells were fractionated by SDS-PAGE and analyzed by Western blotting. Fig. 4B shows that the modified catalytic subunit of cAMP-dependent protein kinase was synthesized as expected. To test the biological activity of NLSCα, in comparison to the wild-type form of the catalytic subunit (Cα), we measured the activity of the phosphorylation-regulated transcription factor CREB using a fusion protein consisting of the GAL4 DNA-binding domain fused to the kinase inducible activation domain of CREB. The modular structure of the GAL4-CREB fusion protein is depicted in Fig. 4C . Transcriptional activation was monitored by co-transfection of the reporter plasmid pUAS 5 luc that contains five copies of the GAL4 binding site termed "upstream activating sequence" (UAS) upstream of a luciferase reporter gene (Fig. 4C ). Since mammalian cells do not express transcription factors that bind to the UAS, this system directly measures the effect of NLSCα or Cα on the transcriptional activation potential of CREB. The reporter plasmid pUAS 5 luc and the expression plasmid that encodes GAL4-CREB were transfected into HepG2 cells together with either an "empty" expression vector (denoted "-") or expression vectors encoding Cα or NLSCα. Transfection efficiency was monitored by co-transfecting pRSVβ. As a control, plasmid pM1 encoding only the DNA binding domain of GAL4 (GAL4 DBD ) was transfected. Cell extracts were prepared forty-eight hours later and β-galactosidase and luciferase activities were determined. Fig. 4C shows that the modified form of PKA catalytic subunit (NLSCα) was highly potent in stimulating the transcriptional activation potential of CREB. The activation was on the order of 60-fold. In contrast, no transcriptional activation was observed following overexpression of the wild-type form of the catalytic subunit. Next, we tested the biological activity of NLSCα with regard to transcription of the glucose-6-phosphatase promoter/luciferase reporter genes. In the experiments, we additionally overexpressed wild-type CREB, the CREBS133A containing a serine to alanine mutation at position 133, and K-CREB, a CREB mutant containing the point mutation R286L within the basic DNA-binding domain, impairing DNA-binding [ 24 ]. Transfection of expression vectors encoding the wild-type form of CREB, CREBS133A, or K-CREB did not significantly change basal transcription of the reporter genes. However, transfection of an NLSCα expression vector strongly stimulated transcription of the glucose-6-phosphatase promoter/luciferase reporter genes, containing both or one of the CRE-like sequences of the glucose-6-phosphatase promoter (Fig. 5A,B ). These results confirm the previous observations, obtained via overexpression of C2/CREB, that both CRE motifs of the glucose-6-phosphatase gene function as independent genetic elements responsive to couple elevated cAMP and PKA levels with enhanced glucose-6-phosphatase gene transcription. Similarly, coexpression of wild-type CREB and NLSCα stimulated reporter gene transcription mediated by the CRE/AP1 element derived from the TNFα gene (Fig. 5C ). Surprisingly, we still detected reporter gene activation following coexpression of NLSCα with CREBS133A, lacking the major PKA phosphorylation site. Compared to the coexpression experiments of NLSCα with the wild-type form of CREB, the reduced activation of reporter gene transcription in coexpression experiments of NLSCα with CREBS133A indicates that NLSCα catalyzed phosphorylation of serine residue 133 of CREB is important for reporter gene transcription. The fact that CREBS133A is still able to transactivate the reporter genes following expression of NLSCα suggests that NLSCα triggers further phosphorylation reactions leading to enhanced transcription via the CRE-like sequences within the glucose-6-phosphatase gene. In contrast, expression of K-CREB is transcriptionally inactive, in the presence or absence of NLSCα, indicating that DNA-binding is a prerequisite for CREB and CREBS133A to transactivate the reporter genes. Biological activity of a constitutively active CREB2/ATF2 fusion protein on glucose-6-phosphatase promoter/luciferase reporter genes The transcription factor ATF2, a substrate of stress-activated protein kinases, has been reported to bind to the classical cAMP responsive element (CRE) 5'-TGACGTCA-3' [ 25 ]. Recently, we confirmed this observation showing that ATF2 and CREB compete for binding to the CRE of the secretogranin II gene [ 26 ]. ATF2 was proposed to also bind with high affinity to the related DNA target sequence 5'-TTACGTAA-3' [ 19 ], which is identical to the CRE1 of the glucose-6-phosphatase gene promoter. We tested the biological activity of ATF2 on reporter genes containing sequences of the glucose-6-phosphatase promoter region. As a control, we used a reporter gene containing two copies of the CRE/AP-1 element of the TNFα gene. ATF2 has been shown to stimulate TNFα promoter activity [ 27 ]. The unphosphorylated form of ATF2 is transcriptionally inactive, due to an inhibitory intramolecular interaction between the activation domain of ATF2 and the bZIP domain of ATF2 [ 28 ]. We therefore expressed a constitutively active CREB2/ATF2 fusion protein to investigate whether ATF2 functions as a transactivator for reporter genes containing the CRE1 and/or CRE2 sequences of the glucose-6-phosphatase gene in the regulatory region. The domain structure of this C2/ATF2 fusion protein is depicted in Fig. 6A . HepG2 cells were transfected with one of the reporter plasmids (pG6PCRE1/CRE2luc, pG6PCRE1mut/CRE2luc, pG6PCRE1/CRE2mutluc, pG6PCRE1luc, and pG6PCRE2luc), the internal standard plasmid pRSVβ, and an expression vector encoding C2/ATF2. As a control, we analyzed the reporter plasmid pTNFα(CRE/AP1) 2 luc. The results show that C2/ATF2 was unable to transactivate the reporter genes containing CRE1, CRE2, or both CRE-like sequences derived from the glucose-6-phosphatase gene in the regulatory region, indicating that glucose-6-phosphatase gene transcription is not regulated by ATF2 via the two CRE-like sequences. In contrast, C2/ATF2 strongly activated transcription of the pTNFα(CRE/AP1) 2 luc reporter gene, confirming that TNFα gene transcription is regulated by ATF2. Discussion The gene encoding glucose-6-phosphatase is regulated by increased levels of intracellular cAMP, but the regulatory sites responsible for cAMP-induced gene transcription are still a matter of controversy. The objective of this study was to characterize the genetic elements that function as cis -acting sites for transactivation by CREB and that also respond to activated PKA in the nucleus. Two distinct genetic elements had been suggested to couple enhanced levels of cAMP and elevated PKA activity with increased glucose-6-phosphatase gene transcription [ 8 , 17 ] and we intended to clarify which of these elements is required for transactivation of the glucose-6-phosphatase gene by CREB. The reason for the differences in the published reports concerning cAMP-regulated glucose-6-phosphatase gene transcription can be explained by the technical problems that occurs following elevation of the intracellular cAMP concentration, as outlined in the introduction section. To solve these problems, we prefered to measure "transcriptional activation" instead of "transcription factor/DNA-binding" because although DNA-binding is required for a subsequent transcriptional activation by CREB, an enhanced binding activity of a transcription factor to DNA, monitored by an in vitro binding assay, does not necessarily prove an enhanced transcriptional activation potential of this protein [ 29 ]. In addition, we expressed a constitutively active CREB2/CREB fusion protein that was highly active in transactivating CREB-responsive target genes. Using this strategy we avoided the use of dibutyryl cAMP or forskolin, that may trigger other biological responses. Furthermore, only nanomolar levels of the expression vector encoding C2/CREB were required to show that both CRE-like sequences (CRE1 and CRE2) within the glucose-6-phosphatase gene function as target sites for an active CREB. Thus, both the reported genetic elements, the sequence from -161 to -152 [ 7 ] and the sequence from -136 to -129 [ 8 ] of the human glucose-6-phosphatase promoter, mediated reporter gene transactivation by CREB. The fact that the glucose-6-phosphatase gene contains two distinct genetic elements for the regulation by CREB is not surprising as multiple CREs have been found in other genes encoding for instance ICER, MKP-1, Nur77 or Egr-4 [ 30 ]. Using an overexpression strategy for the catalytic subunit of PKA, it has been reported that PKA directly stimulates reporter gene transcription under control of CRE1, whereas CRE2 was unable to connect elevated PKA activity with enhanced CRE2-controlled reporter gene transcription [ 17 ]. The authors concluded that only the CRE1, but not the CRE2 functions as bona fide CRE [ 17 ]. In this study a concentration of 5 μg of expression vector encoding the catalytic subunit of PKA was used. In the past, we also employed this overexpression strategy for the catalytic subunit of PKA in the laboratory [ 29 , 31 ] and observed that high levels of expression vectors are required to monitor an effect on gene transcription. We also observed that transcription of the reference gene encoding β-galactosidase under control of the Rous sarcoma virus long terminal repeat was changed following overexpression of the catalytic subunit, although no CRE has been mapped within this promoter/enhancer region. This fact indicates that transfection of micromolar levels of expression vectors encoding the catalytic subunit of PKA disturbs the machinery of transcription and the data obtained by this approach may not always depict the real picture. Likewise, it has been known for many years that expression of the strong transcriptional activator VP16 induces transcriptional repression within the cells, due to squelching. In the study reported here, expression of a modified catalytic subunit that had a nuclear targeting signal, activated reporter gene transcription very efficiently following transfection of nanomolar amounts of the expression vector. In fact, we used a 50-fold lower amount of expression vector compared to the study by Streeper et al. [ 17 ]. Transfection of nanomolar concentrations of an expression plasmid encoding the wild-type catalytic subunit of PKA did not show any effect on reporter gene transcription. These experiments, involving an overexpression of NLSCα, showed an enhanced transcription of reporter genes having either an intact CRE1 or CRE2 in its regulatory region, indicating that both CRE1 and CRE2 functioned as bona fide CRE. We also observed an activation of glucose-6-phosphatase promoter/luciferase reporter gene transcription following coexpression of CREBS133A with NLSCα. Phosphorylation of the serine 133 residue of CREB is the predominant mechanism to enhance the transcriptional activation potential of CREB via recruitment of the coactivator CREB binding protein (CBP) and its paralogue p300 to the promoter. Accordingly, we observed a clear reduction of reporter gene transcription in coexpression studies of NLSCα together with CREBS133A instead of wild-type CREB. However, expression of CREBS133A still contributed to reporter gene transcription in the presence of NLSCα, suggesting that either other residues of CREBS133A or other promoter-bound proteins involved in the regulation of CREB-mediated transcription are phosphorylated. PKA can phosphorylate CBP directly or other components of the transcriptional machinery downstream from CREB [ 32 , 33 ]. Nonetheless, the experiments confirmed that both CRE-like sequences within the glucose-6-phosphatase gene function as genetic elements to mediate transactivation via CREB. In contrast to the expression experiments involving C2/CREB, we did not detect an effect of C2/ATF2, a constitutively active CREB2/ATF2 fusion protein, on glucose-6-phosphatase promoter/luciferase reporter gene transcription, despite the fact that the sequence 5'-TTACGTAA-3', which is identical to the CRE1 site of the glucose-6-phosphatase promoter, has been shown to function as a high affinity binding site for ATF2 in vitro [ 19 ]. This discrepancy supports the view that transcriptional bioassays and not in vitro DNA-protein binding experiments describe the biological activities of transcription factors. The CRE-like sequences CRE1 and CRE2 of the glucose-6-phosphatase gene are therefore strikingly different from the CRE/AP1 site of the TNFα gene, that is strongly activated by C2/ATF2. Recently, we showed that ATF2 is able to specifically transactivate CRE-containing genes and we identified the secretogranin II gene as a target gene for ATF2 [ 26 ]. We also showed that C2/ATF2 only marginally enhanced transcription of a reporter gene carrying four copies of the c-Fos CRE in its regulatory region, while transcription regulated by the tyrosine hydroxylase promoter was not upregulated at all. Thus, ATF2 distinguishs between different CRE-containing genes. The biological activity of ATF2 is regulated by stress-activated protein kinases and ATF2 is thought to play an important role in the cellular stress response. Our data sheds light on the fact that expression of glucose-6-phosphatase is not connected – via ATF2 – to the cellular stress response, in contrast to the TNFα gene. Conclusions We have shown that there are two distinct CREB-responsive sites in the glucose-6-phosphatase gene promoter that are responding to either a constitutively active CREB or elevated concentrations of the catalytic subunit of cAMP-dependent protein kinase in the nucleus. This study further shows that the expression vectors encoding C2/CREB and NLSCα are valuable tools for the investigation of CREB-mediated gene transcription and the biological functions of CREB. CREB is a key molecule in neuronal survival, as demonstrated by the fact that a dominant-negative A-CREB induced apoptosis in sympathetic neurons grown in NGF [ 34 ]. The C2/CREB fusion protein, that directs CREB-mediated gene transcription in the absence of PKA activation, can be used in gain-of-function experiments to investigate the cytoprotective activity of CREB on the molecular level. Most importantly, the C2/CREB protein can be used to identify anti-apoptotic genes that are controlled by CREB. Likewise, expression of NLSCα will permit the analysis of PKA-regulated gene transcription, without influencing other functions of PKA in the cytosol. Methods Reporter constructs The glucose-6-phosphatase promoter/luciferase reporter genes pG6PCRE1/CRE2luc, pG6PCRE1mut/CRE2luc, and pG6PCRE1/CRE2mutluc were constructed by the insertion of the annealed oligonucleotides depicted in Fig. 1B with KpnI/XhoI cohesive ends into the KpnI/XhoI sites of plasmid pHIVTATAluc [ 35 ]. The glucose-6-phosphatase promoter/luciferase reporter genes pG6PCRE1luc and pG6PCRE2luc that contain either the CRE1 or the CRE2 sequence derived from the human glucose-6-phosphatase gene were constructed by the insertion of the annealed oligonucleotides depicted in Fig. 1C with XhoI/SalI cohesive ends into the XhoI/SalI sites of plasmid pHIVTATA-CAT [ 36 ]. These plasmids termed pG6PCRE1CAT and pG6PCRE2CAT were digested with XbaI, filled in with the Klenow fragment of DNA polymerase I, recut with XhoI and cloned into plasmid pGL3-Basic (Promega). The reporter gene pTNFα(CRE/AP-1) 2 luc, that contains two copies of the CRE/AP1 element of the human tumor necrosis factor α gene, was generated in a similar way with annealed oligonucleotides encompassing the sequence depicted in Fig. 1D . Plasmid pUAS 5 luc containing the luciferase reporter gene, a TATA box derived from the HIV long terminal repeat, an initiator element from the adenovirus major late promoter, and five binding sites for GAL4 (termed 'upstream activating sequence', UAS) has been described [ 35 ]. Expression vectors The expression vector pCMV-FLAG-C2/CREB, encoding a constitutively active CREB2/CREB chimera, has been described [ 26 ]. The CREB2/CREB fusion protein consists of the amino-terminal 187 amino acids from CREB2, encompassing the phosphorylation-independent transcriptional activation domain, and amino acids 182 to 326 of CREB, including the bZIP domain. Expression vectors encoding CREB, CREBS133A, and K-CREB were kind gifts of Wilhart Knepel and Elke Oetjen, Department of Molecular Pharmacology, University of Göttingen, Germany. The expression vector encoding a constitutively active CREB2/ATF2 fusion protein termed C2/ATF2 has been described [ 27 ]. The GAL4 expression plasmid pFA2CREB was purchased from Stratagene. The fusion protein encodes the transcriptional activation sequence of CREB (amino acids 1–281) fused to the DNA-binding domain of GAL4. An expression vector encoding the catalytic subunit of PKA (pCMVCα) was a kind gift of Michael Uhler from the University of Michigan, Ann Arbor [ 37 ]. The construction of the NLSCα encoding expression vector has been described [ 26 ]. NLSCα has a triple FLAG-tag and a nuclear localization signal on the N-terminus, followed by amino acids 19 to 351 of Cα. The expression vector pRSVβ, encoding β-galactosidase under the control of the Rous sarcoma virus long terminal repeat, has been described [ 29 ]. Cell culture, transfections, and reporter gene assays Human HepG2 hepatoma cells were cultured and transfected as described [ 38 ]. The amounts of expression vectors transfected are indicated in the figure legends. The luciferase reporter plasmids (1 μg) and the internal reference plasmid pRSVβ were transfected into cells grown on 60 mm plates. Lysates were prepared forty-eight hours later using cell culture lysis buffer (Promega) and β-galactosidase and luciferase activities were measured as described [ 35 ]. Western Blots Nuclear extracts were prepared as described [ 39 ]. 20 μg of nuclear proteins were separated by SDS-PAGE and the blot was incubated with the M2 monoclonal antibody directed against the FLAG epitope (Sigma, # F3165). Blots were developed with a horseradish peroxidase conjugated anti-mouse secondary antibody and ECL (Amersham, Freiburg, Germany). Abbreviations ATF activating transcription factor bZIP basic region leucine zipper CRE cAMP response element CREB cAMP response element binding protein G6P glucose-6-phosphatase PKA protein kinase A TNF tumor necrosis factor Authors' contributions GT designed the study, generated the reporter plasmids and the expression vectors encoding C2/CREB, C2/ATF2, and NLSCα, performed part of the transfection experiments and drafted the manuscript. JAS performed the transfection experiments depicted in Figs. 3 , 4C , and 5 . LS performed the C2/ATF2 transfection experiments depicted in Figs. 6 . All authors read and approved the manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548273.xml

524356

Significance of MDR1 and multiple drug resistance in refractory human epileptic brain

Background The multiple drug resistance protein (MDR1/P-glycoprotein) is overexpressed in glia and blood-brain barrier (BBB) endothelium in drug refractory human epileptic tissue. Since various antiepileptic drugs (AEDs) can act as substrates for MDR1, the enhanced expression/function of this protein may increase their active extrusion from the brain, resulting in decreased responsiveness to AEDs. Methods Human drug resistant epileptic brain tissues were collected after surgical resection. Astrocyte cell cultures were established from these tissues, and commercially available normal human astrocytes were used as controls. Uptake of fluorescent doxorubicin and radioactive-labeled Phenytoin was measured in the two cell populations, and the effect of MDR1 blockers was evaluated. Frozen human epileptic brain tissue slices were double immunostained to locate MDR1 in neurons and glia. Other slices were exposed to toxic concentrations of Phenytoin to study cell viability in the presence or absence of a specific MDR1 blocker. Results MDR1 was overexpressed in blood vessels, astrocytes and neurons in human epileptic drug-resistant brain. In addition, MDR1-mediated cellular drug extrusion was increased in human 'epileptic' astrocytes compared to 'normal' ones. Concomitantly, cell viability in the presence of cytotoxic compounds was increased. Conclusions Overexpression of MDR1 in different cell types in drug-resistant epileptic human brain leads to functional alterations, not all of which are linked to drug pharmacokinetics. In particular, the modulation of glioneuronal MDR1 function in epileptic brain in the presence of toxic concentrations of xenobiotics may constitute a novel cytoprotective mechanism.

Background Failure to respond to therapeutic concentrations of antiepileptic drugs (AEDs) is the usual basis for defining multiple drug resistant epilepsy, but the mechanisms underlying resistance to AEDs are still largely unknown. It is generally believed to be a multifactorial phenomenon, depending on both pharmacodynamic and pharmacokinetic mechanisms. The electrical and synaptic properties of neurons in epileptic human tissue are reported to undergo changes that may result in decreased susceptibility to some AEDs [ 1 ]. Multidrug resistance in epilepsy may also result from inadequate intraparenchymal AED concentrations due to poor penetration of the blood-brain barrier (BBB). Recent findings suggest that the molecular mechanisms of clinically defined multiple drug resistance involve drug-efflux transporters such as the ATP-binding cassette subfamily B member 1 (ABCB1), also known as MDR1 or P-glycoprotein (P-gp) [ 2 - 4 ]. Specifically, in epileptic brain, genes associated with multiple drug resistance are overexpressed in the endothelial cells that constitute the BBB [ 5 , 6 ]. MDR1 is also overexpressed in neurons [ 4 , 7 ] and astrocytes (in which MDR1 is normally not measurable) from lesions associated with active epileptogenic foci in human and rodent brain [ 8 - 10 ]. The link between MDR1 overexpression and drug resistance in epilepsy is still poorly understood [ 6 , 11 ]. Thus, while localization of the drug extrusion pump in the BBB is consistent with reduced penetration of AEDs into the CNS, it is not known if or how the presence of MDR1 in the parenchyma affects drug delivery and distribution, or whether it is involved in different cellular functions. Evidence from several groups suggests that MDR1 diminishes the apoptotic response induced by growth factor withdrawal [ 12 ], decreases complement-mediated cytotoxicity ([ 13 ] and impairs the activation of caspase-dependent cell death pathways [ 14 , 15 ]. Reports indicate that all these events occur in epileptic tissue, and they are thought to be at least partly responsible for seizure-associated neuronal cell death. In addition, we recently found that epileptic astrocytes that overexpress MDR1 are devoid of p53, a proapoptotic factor [ 8 ]. In this study we have evaluated the relationship between neuronal and astrocytic MDR1 expression and the capacity of the cells to survive cytotoxic insults, using drug resistant epileptic human specimens. Our results suggest that overexpression of MDR1 leads to functional alterations in the CNS that may be linked to both drug pharmacokinetics and neuroglial survival in injured brain. Methods Human tissue Human subjects were used as donors of cortical tissue samples. The investigation conformed to the principles outlined in the Declaration of Helsinki. Patient consent was obtained as per Institutional Review Board instructions before collection of the specimens. All the experiments involved small portions of human neocortical or hippocampal tissue, which were excised for therapeutic reasons from patients with pharmacoresistant epilepsy (see Table 1 for patient identification). Temporal lobe tissue was taken from the inferior or middle temporal gyrus during standard temporal lobectomy; frontal or parietal lobe samples were chosen from the most epileptogenic areas, as determined by chronic subdural grid or intraoperative electrocorticographic (EEG) recordings. Handling of the excised tissue depended on the experiment to be conducted, as described below. Cell isolation and primary cultures Astrocyte cell cultures were established, as described by Marroni et al. [ 8 , 9 , 16 ], from cerebral cortical tissue obtained during temporal lobectomies (n = 3) conducted to relieve medically intractable seizures (see Table 1 for details on patients, ID# 12, 13, 14). The cells were passaged up to three times before use. MDR1 expression was not affected by the number of passages [ 8 , 9 , 16 ]. Commercially available normal human astrocytes were used as controls (ACBRI 371, Applied Cell Biology Research Institute, Kirkland, WA, USA, and Clonetics, Biowhitaker, Walkersville, MD, USA). Immunohistochemistry To investigate MDR1 protein expression and cellular distribution in human epileptic tissue, slide-mounted sections (10 μm thickness) from frozen brain tissue (Table 1 , patient ID# 1–11; 3 slices per patient) were double immunostained as previously described [ 8 , 9 , 16 ]. The primary antibodies used were mouse monoclonal anti-P-Glycoprotein (C494) (1:40, Calbiochem-Novabiochem Corporation, San Diego, CA, USA), human polyclonal anti-P-Glycoprotein (1:100, Calbiochem-Novabiochem Corporation, San Diego, CA, USA), mouse monoclonal anti-neuronal nuclei (NeuN) (1:500, Chemicon International, Temecula, CA, USA), rabbit polyclonal anti-neurofilament (NF) (1:200, Chemicon International, Temecula, CA, USA) and rabbit anti-cow glial fibrillary acidic protein (GFAP) (1:200, DAKO Corporation, Carpinteria, CA, USA). Secondary antibodies were chosen according to the primary antibody hosts: Texas red dye-conjugated affinipure donkey anti-mouse IgG (1:50, Jackson Immunoresearch Laboratories Inc., West Grove, PA, USA), and Fluorescein isothiocyanate (FITC)-conjugated affinipure donkey anti-mouse and anti-rabbit IgG (1:200, Jackson Immunoresearch Laboratories Inc., West Grove, PA, USA). Sections were coverslipped on glass slides using Vectashield mounting medium with DAPI (Vector, Burlingame, CA, USA) and analyzed by fluorescent microscopy. Cell counting For quantitative evaluation of neurons and astrocytes expressing MDR1 in epileptic tissue, we chose three 1600 μm 2 fields at random in each tissue slice (3 slices per patient, n = 11 patients; see Table 1 , patient ID# 1–11). Within each field, NeuN or GFAP positive cells co-expressing MDR1 were counted using NIH Software and Photoshop 6 (Adobe), and the number was expressed as a fraction of the total number of neurons and astrocytes, respectively, in that field. Numbers for each field were then averaged for every slice, and the mean value for each patient is given in Fig. 1B,1D . Doxorubicin uptake Astrocytes from epileptic tissue (Table 1 , patient ID# 12, 13, 14) or commercially available control human astrocytes were cultured in 8-well chamber-slides. At confluence, the cells (approximately 8 × 10 4 per well) were treated overnight with 1μM XR9576, a specific MDR1 blocker [ 17 , 18 ], or for 2 h with 50 μM verapamil, a non-specific MDR1 blocker [ 19 , 20 ]. Further sets of astrocytes from epileptic tissue and controls were incubated as above but in the absence of blockers. Fluorescent red doxorubicin [ 21 ] was then added to all the cultures for 1, 2, 3, 5, 7 and 24 h. Incubation was stopped by rinsing twice with PBS and the cells were fixed overnight at 4°C in 4% formalin in PBS (pH 7.4). They were mounted and coverslipped the next day using Vectashield mounting medium. Doxorubicin uptake was analyzed by fluorescent microscopy (Leica Leitz DM-RXE, Wetzlar, Germany) using NIH Image Software. For each well, four fields (1600 μm 2 ) were randomly analyzed and optical density measurements were averaged to obtain representative data. To quantify the doxorubicin in the cells, the fluorescence in each sample was compared with standard solutions of doxorubicin (10 nM-10 μM). Data were analyzed using Origin Lab 7 software. 14 C-Phenytoin uptake 14 C-Phenytoin uptake was measured in cultured astrocytes from epileptic tissue (Table 1 , patient ID# 12, 13, 14) and in control astrocytes, as described by Meyer et al. [ 22 ]. The cells were cultured in 30-mm Petri dishes (approximately 10 6 cells) and incubated at 37°C with 1 ml of T3 cell buffer (Tris-HCl 50 mM, NaCl 120 mM, KCl 50 mM, pH 7.4) containing 10 μM 14 C-Phenytoin (Specific activity = 9.57 μCi/μmol). The incubation was terminated after 10 s, 30 s, 1 min, 5 min or 10 min by adding 1 ml of ice-cold T3 buffer. The cells were washed with 1 ml of ice-cold T3 buffer and solubilized with 1% (w/v) Triton X-100 for 1 h at 37°C. Intracellular radioactivity was measured using a scintillation cocktail (Packard Ultima Gold, ECN, Costa Mesa, CA, USA). Acute in vitro toxicity Brain slices (500 μm) were cut using a vibratome from dysplastic human cortex and temporal lobe epilepsy brain specimens (Table 1 , patient ID# 9,10,11). Slices (n = 6 per patient) were maintained in vitro as previously described [ 23 , 24 ]. A further set of 6 slices was obtained from 3 naïve rats as previously described [ 25 ]. Slices (n = 2 per patient per experimental condition) were incubated for 2 h in artificial CSF containing, in mM: NaCl 124; KCl 3; CaCl 2 1; MgCl 2 1.4; NaHCO 3 - 26; KH 2 PO 4 1.25; glucose 10, with or without 3 μM XR9576. A toxic concentration (375 μM) of Phenytoin (PHE) was then added for 5 h [ 22 ]. An additional set of slices was kept for 7 h in artificial CSF only, and electrophysiological measurement of activity was used to assess tissue viability throughout the experiment. Live/Dead Viability/Cytotoxicity solution (Molecular Probes L3224, Eugene, OR, USA) was used to quantify cell death or survival; EthD-1, a component of the solution, enters cell nuclei through damaged membranes, producing a bright red fluorescence in dead cells. The 2 mM EthD-1 stock solution (20 μl) was added to 10 ml of 1X sterile PBS. At the end of the experiment, each slice was washed with PBS and incubated with 2 ml of the above solution for 10 min under continuous oxygenation. The slices were then fixed in 4% formalin in PBS overnight at 4°C and cryoprotected by 30% sucrose. Three 20 μm slices were cryosectioned from each 500 μm slice and analyzed by fluorescent microscopy. DAPI was used to assess the total number of cells per slice, and GFAP and NeuN were used to identify glia and neurons (see 'Immunohistochemistry'). The number of EthD-1 positive cells was reckoned in 3 randomly chosen fields (1600 μm 2 ) within each slice. These values were averaged for each slice and expressed as percentages of the average number of DAPI-positive cells in the chosen fields. Evaluation of glioneuronal damage To evaluate the relationship between MDR1 positive cells and nuclear morphology (DAPI staining), 10 μm slices from 11 patients (Table 1 , patient ID# 1–11) were obtained as described in 'Immunocytochemistry'. Small and condensed nuclei indicate apoptosis or irreversible cell damage, while large nuclei with diffuse DNA (DAPI) staining are typical of healthy cells. MDR1 positive cells with diffuse DNA staining (healthy cells) were counted over three randomly chosen fields in each slice and analyzed using NIH Software and Photoshop. Numbers for each field were averaged for every slice; the corresponding mean value is given in figure 3B . Statistical analysis of data Data were expressed as means ± SEM. The level of significance between means was estimated by ANOVA (Origin 6.0. Microcal). Differences with p < 0.05 were considered significant. Results Results were obtained from 14 patients (50% female) affected by intractable seizures. The mean patient age was 16.4 years (range 8 months – 49 years; SE = 4.6). For details, see Table 1 . MDR1 expression in human epileptic brain tissue Figure 1 shows representative micrographs of MDR1 expression in endothelial, glial and neuronal cells from epileptic human brain specimens (Table 1 , patient ID# 1–11). The findings are consistent with previous work demonstrating that MDR1 expression in multiple drug resistant brains is confined to the cortical lesion site [ 4 , 7 - 9 , 16 ]; this was consistently observed in all epileptic samples in the present study. Thus, relatively normal brain distant from the dysplastic tissue can be used as "control". Abundant MDR1 immunopositive endothelial cells ( thin arrows in A ) were observed in epileptic cortex, as well as immunopositive parenchymal and perivascular astrocytes ( arrowheads in A ) double-labeled with GFAP, a specific glial marker. Out of 107 GFAP-positive astrocytes, 91 (85%) showed MDR1 staining (Figure 1B ). We also examined MDR1 expression in neurons from epileptic tissue, identified by NeuN and NF immunoreactivity (Fig. 1C ). Approximately 169 out of 264 NeuN-positive neurons (64%) showed MDR1 staining (Figure 1D ). Role of MDR1 expression in drug extrusion by astrocytes To determine whether MDR1 expression in 'epileptic' astrocytes results in enhanced extrusion of xenobiotics compared to 'normal' astrocytes, we measured the cellular uptake of PHE and doxorubicin, two established substrates of P-glycoprotein [ 19 , 20 , 26 , 27 ] (Table 1 , patient ID# 12, 13, 14; Fig. 2 ). We used 1 μM red fluorescent doxorubicin [ 21 ] and visualized its cellular distribution by fluorescent microscopy (Fig. 2A ). We previously showed by western blot analysis that 'normal' astrocytes have lower levels of P-gp than 'epileptic' astrocytes [ 8 , 9 ]. Doxorubicin uptake was reduced in astrocytes from epileptic tissue compared to control astrocytes (p < 0.05). This effect was abolished by the specific blocker XR9576 (1 μM) [ 17 , 18 ], or by the less specific antagonist verapamil (50 μM), indicating active MDR1-mediated extrusion of doxorubicin by epileptic glia. Fig. 2B shows that uptake of 10 μM 14 C-Phenytoin was significantly lower in astrocytes from epileptic tissue than in control astrocytes, and the difference was maximal after 10 min incubation with the AED (p < 0.05). The difference was abolished in the presence of 1 μM XR9576, indicating MDR1-mediated efflux of PHE in epileptic astrocytes. The MDR1 blockers did not affect doxorubicin or Phenytoin uptake in control astrocytes (not shown). Neither Phenytoin nor doxorubicin treatment induced toxicity in cell cultures under our experimental conditions. MDR1 and cell survival MDR1 is involved in detoxification mechanisms that protect cells from xenobiotics [ 28 ], apoptosis [ 12 ] and other cellular stresses [ 13 - 15 ]. To investigate whether MDR1 expression in astrocytes and neurons from epileptic tissue enhances survival of toxic or injurious events, we exposed neocortical slices from human epileptic (Table 1 , patient ID# 9, 10, 11) and rat brain to concentrations of PHE higher than those reported to induce toxicity in cultured rat astrocytes [ 22 ]. Bar histograms (Figure 3A ) indicate that slices of human epileptic cortex exposed for 5 h to 375 μM PHE showed no cellular toxicity, as assessed by co-localization of the nuclear marker DAPI with EthD-1, a marker of cell damage. Conversely, significant cell loss (p < 0.05) was observed in rat brain slices under these experimental conditions. When MDR1 activity was blocked by 3 μM XR9576, PHE induced cell damage in the human epileptic cortex to an extent similar to that in rat brain tissue. Neurons and astrocytes, identified using NeuN and GFAP respectively, were similarly affected by PHE (Fig. 3A ). The MDR1 inhibitor XR9576 alone (3μM) did not affect tissue viability (data not shown). Figure 3B shows a significant positive correlation (R = 0.4, p < 0.006) between MDR1 expression and cell integrity, as measured by morphological evaluation of nuclear condensation (DAPI staining) [ 29 , 30 ]. Quantification of the cells retaining nuclear DNA integrity showed that no DNA damage occurred in 85% of glia and 66% of neurons expressing MDR1, suggesting a novel role for MDR1 in protecting neurons and glia from toxic agents. Discussion The main finding was that MDR1 might have different roles depending on the location in the brain where it is expressed. Thus, in addition to overexpression in endothelial cells, which is likely to affect the penetration of AEDs into the brain, MDR1 is also overexpressed in the parenchyma, where it might have a cytoprotective role, extruding otherwise toxic concentrations of xenobiotics from the intracellular compartment. MDR1 expression in epileptic brain In agreement with previous findings, we report here that MDR1 is highly expressed in vessels of the BBB and in parenchymal cells (histochemically identified as astrocytes and neurons) in drug refractory epilepsy of different etiologies [ 4 , 7 - 9 , 11 , 27 ]. Pharmacological evidence suggests that MDR1 overexpression in blood vessels of the BBB has the crucial role of extruding AEDs from brain to blood [ 19 , 20 , 27 , 31 ], and this phenomenon may contribute to failure of antiepiletic treatments. In this study, we focused on the effects of MDR1 expression in parenchymal neurons and astrocytes from pharmacoresistant epileptic tissue. MDR1 expression in astrocytes: drug uptake studies We found that PHE and doxorubicin uptake by astrocytes from human epileptic tissue is reduced in comparison with normal astrocytes. The difference is abolished when MDR1 function is blocked, indicating that these drugs are efficiently extruded by "epileptic" astrocytes by a P-gp mediated mechanism. We suggest that this might contribute to decreasing the AED brain levels only if it occurs in perivascular astrocytes. Thus, perivascular astrocytes impinging on blood vessels may act as an additional barrier to drug penetration into the brain in regions where BBB permeability is transiently altered, for example during epileptic activity [ 32 , 33 ] (Fig. 4 ). In contrast, MDR1 overexpression by parenchymal glia would result in enhanced extrusion of substrates into the brain extracellular space; this is not compatible with a decrease in the concentration of AEDs at their neuronal targets, suggesting that it subserves a different function (see below). MDR1 expression in neurons and parenchymal glia: relationship to cell survival We show in this paper that MDR1 is expressed in immunocytochemically identifiable neurons and astrocytes in brain slices from refractory human epilepsy patients. We have previously reported that basic apoptotic mechanisms may be defective in glia from epileptic tissue, since the pro-apoptotic proteins p53 and p21 could not be detected in "epileptic" astrocytes [ 8 , 9 , 16 ]. This evidence, together with overexpression of MDR1, suggests that "epileptic" astrocytes have gained a distinct survival advantage. This is supported by the marked enhancement of Phenytoin cytotoxicity in both glia and neurons when MDR1 function is blocked. The conclusion is consistent with the findings of Bittigau et al. [ 34 ] that various AEDs induce apoptotic neuronal cell death in developing naïve rat brain at plasma concentrations relevant to seizure control in humans; activators of MDR1 transcription prevented these effects. Finally, we found a positive correlation between neuronal and astrocytic expression of MDR1 and lack of nuclear condensation, a marker of apoptosis and irreversible cell damage. Thus, expression of MDR1 in glia and neurons may protect the cells against toxic xenobiotics or against endogenous compounds that enter the brain in pathological conditions. Conclusions In conclusion, as summarized in figure 4 , our findings indicate a possible new function for MDR1. In normal brain, MDR1 operates at the blood-brain barrier, regulating the plasma/brain exchange of xenobiotics. In epileptic brain, the levels of astrocytic, neuronal and endothelial MDR1 are abnormal, possibly leading to altered brain penetration/distribution of drugs. Perivascular astrocytes may also contribute to this phenomenon. In addition to its drug extrusion effect at the BBB, which may be relevant for pharmacoresistance in epilepsy, this protein may have a role in neuroglial survival under hostile conditions such as those occurring in epileptic brain. The overexpression of multiple drug resistance could be the consequence of an altered mechanism of apoptotic cell death; this hypothesis is supported by the finding that changes in multiple drug resistance gene expression correlate with negative regulation of p53 and other pro-apoptotic genes [ 8 ]. Abbreviations AEDs: anti-epileptic drugs BBB: blood-brain barrier DOX: doxorubicin GFAP: glial fibrillary acidic protein MDR1: multiple drug resistance protein NeuN: neuronal nuclei NF: neurofilament PHE: Phenytoin Competing interests The authors declare that they have no competing interests. Author's contributions NM and LC carried out the pharmacological experiments and data analysis. KLH and KMK obtained surgical resections, established astrocytic cell cultures and performed the immunohistochemical and cell counting studies. GM performed the acute in vitro toxicity experiments. WB provided the surgical specimens. GD provided additional input and editing. AV and DJ designed and co-ordinated some of the experiments. They also contributed equally to the preparation of the manuscript. DJ, the PI of the study, co-ordinated and supervised most of the personnel involved in this project. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524356.xml

549186

Parental educational level and cardiovascular disease risk factors in schoolchildren in large urban areas of Turkey: Directions for public health policy

Background It is widely accepted that the development of atherosclerosis starts at an early age. However, there are very few studies evaluating the prevalence of the common clinical and behavioral cardiovascular disease (CVD) risk factors among children, especially in developing countries. The aim of the present cross-sectional survey was to evaluate the distribution of blood lipid profile and various behavioral (i.e. dietary habits, physical activity status) factors related to CVD risk and its relationships to paternal (PEL) and maternal educational level (MEL) among primary schoolchildren in Turkey. Methods In three major metropolises in Turkey (Istanbul, Ankara and Izmir), a random sample of 1044 children aged 12 and 13 years old was examined. ANOVA was applied to evaluate the tested hypothesis, after correcting for multiple comparisons (Tukey correction). Results After controlling for energy and fat intake, physical activity status and Body Mass Index (BMI), it was found that mostly PEL had a significant positive effect for most of the subgroups examined (Lower vs. Higher and Medium vs. Higher) on TC and HDL-cholesterol and a negative effect on TC/HDL ratio for both genders. Furthermore, both boys and girls with higher PEL and MEL were found to have higher energy intake derived from fat and protein than their counterparts with Medium and Lower PEL and MEL, while the opposite was observed for the percentage of energy derived from carbohydrates. Conclusions Our study provides indications for a possible association between an adverse lipid profile, certain dietary patterns and Higher PEL and MEL among schoolchildren in Turkey. These findings underline the possible role of social status, indicated by the degree of education of both parents, in developing certain health behaviors and health indices among Turkish children and provide some guidance for Public Health Policy.

Background Cardiovascular disease (CVD) is the primary cause of mortality in developed countries and generates a major burden of morbidity throughout life [ 1 , 2 ]. Additionally, CVD emergencies have become the leading cause of death in developing countries as well [ 3 , 4 ]. Prospective and retrospective studies have shown that CVD risk factors, namely obesity, the lipid profile, unhealthy diets and sedentary lifestyle, have their roots in childhood and tend to track into adulthood [ 5 - 8 ]. Therefore, early identification and understanding of behavioral and physiological variables related to CVD are essential, so that appropriate interventions can be targeted to children, minimizing the risk of developing the disease later in life. The continuing modernization and technological advancement of the developing world has brought about rapid lifestyle changes (e.g. fast-food culture, caloric dense diets, sedentary lifestyle), which are known to have a major impact on the development of CVD and other chronic diseases [ 8 , 9 ]. Many investigators have pointed out social and economic status differences, indicated either by the type of school attended (i.e. public vs. private), by the region of residence (i.e. rural vs. urban), or by the level of parental education (primary school vs. University) as important determinants that appear to influence the prevalence of CVD risk factors in both developed and developing world [ 10 - 12 ]. Currently conducted cross-sectional studies have indicated that various socioeconomic factors are inversely associated with CVD morbidity and mortality in developed countries [ 11 , 13 ], while in developing countries, existing data on socioeconomic conditions and the prevalence of CVD are inconsistent [ 14 - 16 ]. The way in which parental educational level seems to exert its effect on health indices and health behaviors differentiates and therefore, should not be considered constant over time and geographic regions [ 17 ]. Together health behaviors with social and economic status indices are the outcome of a dynamic and interactive process among several parameters such as education, income, wealth, local culture, beliefs and practices and since none of those parameters stay constant over time and region so do these indices and the way they influence health [ 18 , 19 ]. This observation indicates the need for periodic assessment of these conditions in each population in order to have a clear picture of the social and economic variables and their effects over the populations' health which could effectively guide Public Health Policy in achieving "health for all" and social justice [ 18 ]. As far as Turkey is concerned, the available data on the possible effect of parental educational level on the clustering of CVD risk factors among children are limited. Turkey, which has experienced rapid urbanization and industrialization in the past few decades, is a middle-income country, located between Europe and Asia and bordering the Mediterranean, Aegean and Black Seas. The purpose of the present study was to examine the influence of PEL and MEL on CVD risk factors, namely the lipid profile and behavioral indices, among children of Turkey. Expanding our knowledge and understanding on the interrelationship of all these parameters could effectively guide Public Health Policy in early prevention of CVD risk factors in childhood. Methods Sampling The first phase of this cross-sectional study took place during March-May 2001 and the second phase during January-May 2002. The study population consisted of primary schoolchildren aged 12 and 13 years old, living in three urban areas of Turkey, namely Istanbul, Ankara and Izmir. Out of totally 1320 6 th grade primary school children registered in the selected schools, 1044 pupils (510 from Istanbul, 289 from Ankara, 245 from Izmir, 49.2% males, 50.8% females) were finally examined. Inclusion of subjects was on a voluntary basis; prior to acceptance, children's parents or guardians were fully informed about the objectives and methods of the study and were asked to sign a consent form. The study population was selected from nine primary schools (three from each city) using the multi-stage sampling method. Schools were selected taking into consideration the available records of the Ministry of National Education and the National Statistical Center of Turkey, in an attempt to obtain a representative sample from the overall population. In the case of Ankara and Izmir data were collected from three public schools respectively, while in the case of Istanbul sample came from one private and two public schools. All children in the same class were invited to participate in the study to avoid ethical problems. Approval to conduct this survey was granted by the Ethical Committee of Marmara University and the Turkish Ministry of Education. Data collection Baseline data was collected during face-to-face interviews with children by a team of trained personnel. The data collected from the children consisted of physiological indices, namely biochemical and behavioral indices, such as dietary habits, estimation of energy and nutrient intake and physical activity assessment. These data are presented in more detail below. Demographic characteristics A coded questionnaire was developed and administered to the study participants, in order to obtain information on socioeconomic conditions and their personal characteristics, such as family size, parental academic qualifications, occupation and car ownership. Both Paternal (PEL) and Maternal Educational Level (MEL) were classified into three categories. "Lower" PEL and MEL corresponded to illiterate, literate with no formal education and to primary school graduates. "Medium" PEL and MEL was attributed to those parents, who had completed junior high school or high school, whereas the category of "Higher" PEL and MEL included parents who possessed a college or a university diploma. Biochemical indices Early morning venous blood samples were obtained from each child for biochemical screening tests, following a 12-h overnight fast. Professional staff performed venipuncture, using vacutainers to obtain 10 ml of whole blood. Blood was centrifuged for plasma separation at the local University Hospital and 1.5 ml aliquots were pipetted into plastic Eppendorf tubes and stored at -80°C. When blood collection from each city was completed, all samples were sent to Marmara University, Faculty of Health Education, where the actual biochemical analysis took place. Total cholesterol (TC) was determined using Allain's method [ 20 ], while Fossati's method was used for triglycerides (TG) determination [ 21 ]. High Density Lipoprotein-cholesterol (HDL-C) was measured by the heparin-manganese precipitation method. Low Density Lipoprotein-cholesterol (LDL-C) was calculated as follows: LDL-C = TC - (HDL-C + TG/5) [ 22 ]. Anthropometrical measurements Body weight was measured using a digital scale (Seca) with an accuracy of ± 100 g. Subjects were weighed without shoes, in the minimum clothing possible, i.e. underwear. Standing height was measured without shoes to the nearest 0.5 cm with the use of a commercial stadiometer, with the shoulders in relaxed position and arms hanging freely [ 23 ]. Body Mass Index (BMI) was calculated by dividing weight (kg) by height squared (m 2 ). Dietary assessment Food consumption of children was assessed by the 24-hour recall technique on three consecutive days. Dietary data was collected from children during a face-to-face interview with a trained dietician. Dieticians were trained as a group to minimize inter-observer variation. During the interview, food models and photos of common Turkish dishes of various portions, as well as household cups and measures were used to define amounts, in order to obtain as accurate information as possible, regarding the type and amount of food and beverages consumed during the previous day. Macronutrient and micronutrient intake were calculated using the food database available in Marmara University, Faculty of Health Education. This database contains Turkish food composition tables for all food, including cooked Turkish dishes. Information on processed foods was obtained from food companies and national as well as international fast food chains. Physical activity assessment Physical activity during school hours and leisure time was assessed using a standardized physical activity diary completed by the children for two consecutive weekdays and one weekend day. A member of the research team crosschecked diary information during daily interviews. The diary was constructed for searching various physical activities ranging from mild to vigorous [ 24 ], while the time spent for each type of activity was recorded in hours. Activities were classified into two groups, namely Sedentary and Light Activities (< 4 METs) and Moderate to Vigorous Physical Activities (MVPA) (> 4 METs). Typical activities in the Sedentary and Light category were watching TV, board and computer games, studying and extra curricular classes (e.g. music, language), etc. The MVPA category included activities such as walking, bicycling, rhythmic-gymnastics, dancing, basketball, soccer, athletics, tennis, swimming, running up and down, jumping rope and general participation in active outdoors games. Given the age group, MVPA was defined as continuous vigorous activity causing sweating and heavy breathing for periods longer than 15 minutes, but with occasional breaks in intensity, rather than the strict aerobic definition of 20 continuous minutes appropriate for adults. Statistical analysis Descriptive statistics of continuous variables were expressed as the Mean ± standard deviation (s.d.), as well as median and 25 th and 75 th percentile for not normally distributed variables. The 2- sample Z test was used to compare proportions regarding values above normal range among the different categories of parental education for both genders. One- way analysis of variance (ANOVA) with a Tukey post- hoc analysis and multivariate analysis of variance (MANOVA) controlling for certain confounding factors, was conducted to determine whether differences derived from the comparison of continuous variables among the educational categories were statistically significant. For not normally distributed variables the non- parametric univariate Mann-Whitney test was used to verify the statistical significance of the existing differences among the three educational categories. All analyses were conducted using the SPSS 10.0 statistical software package for Windows. In all analyses a 5% significance level was used. Results Table 1 summarizes the mean serum lipid values determined for the study participants by gender and parental educational level, as well as the significant differences observed among the three educational level categories. According to these findings Higher PEL and MEL boys and girls were found to have significantly higher mean levels of TC and HDL-C, but lower mean TC to HDL-C ratio, compared to their Medium and Lower PEL and MEL peers. However, after adjusting for sex, energy intake, fat intake, BMI and MVPA some of the differences presented in Table 1 lost their significance (data not shown). More specifically, when PEL was used as a grouping variable HDL-C remained higher for Higher PEL boys and girls, compared to those of Medium PEL (p = 0.010 for boys and p < 0.001 for girls) as well as for Higher PEL girls compared to those of Lower PEL (p < 0.001). A similar observation applied for TC to HDL-C ratio, which remained lower for Lower PEL boys and girls, compared to Medium (p = 0.021 for boys and girls) and to Lower PEL ones (p = 0.022 for boys and p = 0.005 for girls). When mean serum lipid levels where compared on the basis of MEL categories, most of the differences observed in Table 1 lost their significance, since only HDL-C and LDL-C remained higher for Higher MEL boys, compared to Medium (p = 0.018) and to Lower MEL ones (p = 0.020) respectively. Table 1 Biochemical indices by gender and parental educational level Father's educational level (PEL) BOYS Serum Lipids Lower (n = 174) Medium (n = 264) Higher (n = 95) p-value (ANOVA) TC (mg/dl) 159.7 ± 30.2 167.6 ± 31.6 a 169.3 ± 30.9 b 0.007 HDL-C* (mg/dl) 53.8 ± 31.9 54.2 ± 14.7 60.3 ± 15.2 b,c <0.001 LDL-C cholesterol (mg/dl) 90.3 ± 27.2 95.6 ± 30.0 96.2 ± 29.7 0.930 Triglycerides* (mg/dl)] 90.6 ± 49.0 85.3 ± 38.0 85.4 ± 43.4 0.568 TC/ HDL-C ratio 3.23 ± 0.85 3.28 ± 0.98 2.96 ± 0.86 b,c 0.011 Mother's educational level (MEL) Serum Lipids Lower (n = 276) Medium (n = 209) Higher (n = 46) p-value (ANOVA) TC (mg/dl) 162.3 ± 32.1 168.9 ± 29.4 a 168.8 ± 34.0 0.036 HDL-C* (mg/dl) 54.0 ± 15.9 53.7 ± 13.5 60.0 ± 18.0 b,c 0.033 LDL-C cholesterol (mg/dl) 90.9 ± 28.6 97.2 ± 29.1 a 99.6 ± 31.8 0.017 Triglycerides* (mg/dl)] 88.0 ± 45.2 87.2 ± 42.4 85.0 ± 39.1 0.851 TC/ HDL-C ratio 3.16 ± 0.83 3.34 ± 1.01 3.02 ± 0.97 0.023 Father's educational level (PEL) GIRLS Serum Lipids Lower (n = 164) Medium (n = 281) Higher (n = 66) p-value (ANOVA) TC (mg/dl) 168.9 ± 29.8 172.0 ± 34.2 175.5 ± 30.5 0.308 HDL-C* (mg/dl) 51.1 ± 11.5 52.9 ± 12.5 61.5 ± 20.0 b,c <0.001 LDL-C cholesterol (mg/dl) 103.2 ± 85.4 99.9 ± 32.8 95.5 ± 31.9 0.609 Triglycerides* (mg/dl)] 95.3 ± 42.3 95.5 ± 49.6 98.1 ± 45.9 0.879 TC/ HDL-C ratio 3.44 ± 0.84 3.40 ± 0.92 3.06 ± 0.89 b,c 0.008 Mother's educational level (MEL) Serum Lipids Lower (n = 259) Medium (n = 227) Higher (n = 27) p-value (ANOVA) TC (mg/dl) 168.4 ± 32.2 174.2 ± 32.7 179.7 ± 28.1 b 0.046 HDL-C* (mg/dl) 51.6 ± 13.9 54.9 ± 12.8 a 56.2 ± 13.7 b 0.002 LDL-C cholesterol (mg/dl) 100.6 ± 72.0 99.8 ± 33.0 107.4 ± 26.3 0.799 Triglycerides* (mg/dl)] 95.2 ± 42.1 97.4 ± 51.9 81.5 ± 35.1 0.194 TC/ HDL-C ratio 3.41 ± 0.85 3.32 ± 0.93 3.38 ± 0.95 0.488 a p < 0.05 Medium SES vs. Low SES b p < 0.05 High SES vs. Low SES c p < 0.05 High SES vs. Medium SES * Parameter was log transformed. Table 2 presents the proportions of children with serum lipid levels above the cut-off points, proposed by the NCEP [ 25 ]. According to these findings the prevalence of children found to be in "borderline or high risk" according to their TC and LDL-C serum levels, was significantly higher in Higher PEL and MEL boys and girls, compared to their Medium and Lower PEL and MEL peers, respectively. Furthermore, the prevalence of children with HDL-C levels below 35 mg/dl was found to be lower in Higher PEL boys, compared to those of Medium and Lower PEL. Table 2 Percentage of children characterized as "borderline or high risk" by gender and parental educational level. Father's educational level (PEL) BOYS GIRLS Borderline or High Risk Lower % (n = 174) Medium % (n = 264) Higher % (n = 95) Lower % (n = 164) Medium % (n = 281) Higher % (n = 66) TC ≥ 170 32.0 43.6 a 49.5 b 45.4 50.2 60.6 b LDL-C ≥ 110 21.4 28.0 25.3 33.2 37.7 30.3 HDL-C<35 7.8 5.7 0.0 b,c 6.1 4.5 3.0 TC/ HDL-C>4.5 9.7 9.1 5.3 10.7 10.9 9.1 Mother's educational level (MEL) Borderline or High Risk Lower % (n = 276) Medium % (n = 209) Higher % (n = 46) Lower % (n = 259) Medium % (n = 227) Higher % (n = 27) TC ≥ 170 36.2 45.7 a 50.0 45.1 54.8 a 70.4 b LDL-C ≥ 110 21.2 29.8 a 32.6 32.5 38.0 40.7 HDL-C<35 5.8 6.5 0.0 6.8 4.9 5.7 TC/HDL-C>4.5 7.4 11.0 8.7 8.5 12.2 14.8 * Lipids cut-off points in mg/ dl (according to NCEP) a p < 0.05 Medium SES vs. Low SES b p < 0.05 High SES vs. Low SES c p < 0.05 High SES vs. Medium SES Dietary and physical activity characteristics of the under study population are reported in Tables 3 and 4 . Both boys and girls with Higher PEL were found to have higher intakes of energy derived from fat and protein than Lower and Medium PEL boys (p = 0.002 and p = 0.021) and girls (p = 0.007 and p = 0.010) respectively. Regarding boys of Higher MEL, they were found to have lower total energy intake (p = 0.005) but higher percentages of energy derived from fat (p = 0.002) and protein (p = 0.001), compared to boys of Lower and Medium MEL. The percentage of energy derived from fat was also found to be higher for Medium and Higher MEL girls, compared to those of Lower MEL (p = 0.001). On the other hand the contribution of carbohydrates to the total daily energy intake was found to be higher for Lower MEL boys (p = 0.025) and girls (p = 0.001), compared to Medium and Higher MEL children. With the exception of MVPA, which was found to be higher only for Medium MEL boys, compared to Lower MEL ones (P < 0.05), no other statistically significant differences were observed. Table 3 Energy and macronutrient intakes by gender and parental educational level. Father's educational level (PEL) BOYS Dietary Indices Lower (n = 171) Medium (n = 266) Higher (n = 99) p-value (ANOVA) Energy Intake* (Kcal/ day) 2031.4 ± 556.1 2017.7 ± 603.5 1952.3 ± 642.9 0.373 Fat Intake (%) 29.7 ± 7.2 30.2 ± 6.3 32.4 ± 5.5 b,c 0.002 Protein Intake (%) 13.3 ± 2.3 13.3 ± 2.6 14.1 ± 2.5 b,c 0.021 CHO Intake (%) 40.0 ± 10.9 39.3 ± 10.5 39.9 ± 9.3 0.739 Mother's educational level (MEL) Dietary Indices Lower (n = 274) Medium (n = 213) Higher (n = 46) p-value (ANOVA) Energy Intake* (Kcal/ day) 2000.7 ± 546.9 2059.6 ± 650.8 1754.8 ± 557.5 b,c 0.005 Fat Intake (%) 29.7 ± 6.6 30.6 ± 6.4 33.2 ± 6.9 b,c 0.002 Protein Intake (%) 13.3 ± 2.4 13.4 ± 2.6 14.8 ± 2.8 b,c 0.001 CHO Intake (%) 41.1 ± 8.6 40.6 ± 10.7 38.4 ± 10.2 b 0.025 Father's educational level (PEL) GIRLS Dietary Indices Lower (n = 157) Medium (n = 278) Higher (n = 73) p-value (ANOVA) Energy Intake* (Kcal/ day) 1913.3 ± 562.9 1926.8 ± 585.8 1796,0 ± 413,0 0.414 Fat Intake (%) 30.5 ± 6.7 31.0 ± 6.3 33,2 ± 6,4 b,c 0.007 Protein Intake (%) 12.9 ± 2.6 12.9 ± 2.5 13,8 ± 2,2 b,c 0.010 CHO Intake (%) 38.4 ± 11.5 36.7 ± 10.4 36,6 ± 7,9 0.189 Mother's educational level (MEL) Dietary Indices Lower (n = 260) Medium (n = 220) Higher (n = 31) p-value (ANOVA) Energy Intake* (Kcal/ day) 1889,3 ± 554,7 1915.5 ± 533.4 1780.2 ± 609.5 0.404 Fat Intake (%) 30,2 ± 6,5 31.9 ± 6.5 a 33.4 ± 6.5 b 0.001 Protein Intake (%) 12,8 ± 2,4 13.0 ± 2.6 13.6 ± 2.3 0.260 CHO Intake (%) 38,8 ± 11,2 35.7 ± 9.6 a 35.4 ± 10.0 0.001 a p < 0.05 Medium SES vs. Low SES, b p < 0.05 High SES vs. Low SES, c p < 0.05 High SES vs. Medium SES * Parameter was log transformed Table 4 Physical activity by gender and parental educational level. Father's educational level (PEL) BOYS GIRLS Physical Activity Lower (n = 174) Medium (n = 267) Higher (n = 94) Lower (n = 157) Medium (n = 281) Higher (n = 71) MVPA (hours/week) Median (25th – 75th percentile) 6.7 (2.3–11.1) 7.0 (3.5–11.6) 7.0 (3.2–11.6) 2.3 (0.0–7.0) 3.5 (0.0–7.0) 1.2 (0.0–7.0) Mean ± SD 7.3 ± 6.4 7.9 ± 6.1 7.6 ± 6.4 4.1 ± 4.4 4.2 ± 4.5 4.0 ± 5.3 Mother's educational level (MEL) Physical Activity Lower (n = 273) Medium (n = 214) Higher (n = 45) Lower (n = 260) Medium (n = 222) Higher (n = 30) MVPA (hours/week) Median (25th – 75th percentile) 5.8 (2.3–10.5) 8.1 a (3.5–11.6) 5.8 (2.3–8.1) 2.3 (0.0–7.0) 3.5 (0.0–7.0) 0.6 (0.0–5.2) Mean ± SD 7.2 ± 6.4 8.2 ± 6.1 a 6.7 ± 5.8 3.8 ± 4.3 4.4 ± 4.8 3.5 ± 5.0 a p < 0.05 Medium SES vs. Low SES (Mann Whitney test). Discussion Despite the traditional notions of CVD as a Western disease of "affluence", more than three quarters of global CVD mortality now occurs in middle- and lower- income nations (WHO, 2001) [ 26 ]. Furthermore, progression of CVD as well as the risk factors and health behaviors leading to this medical condition seem to be related with various social and economic indices for both adults and children [ 11 , 27 ]. However, prevention seems to be the most effective way to treat CVD and since the roots of the disease are located in childhood, early detection and management of the related risk factors should begin at that age [ 27 , 28 ]. According to the findings of the current study there was a higher prevalence of "borderline or high risk", with respect to TC levels, for children with Medium and Higher parental educational level (Table 2 ). Several other investigators, who have conducted analogous epidemiological surveys in other developing nations, such as in Philippines, Costa Rica and Iran [ 29 - 31 ], have confirmed these alarmingly high prevalence rates among children living under the supervision and care of better educated parents. Furthermore, in contrast with Western populations where TC levels are often inversely associated with income and education [ 32 , 33 ], but in consistency with several other studies conducted in recent years in Turkey the findings of the current one indicated a positive effect of parental education on serum TC levels [ 34 , 35 ]. The same trend applied for HDL-C levels since, in consistency to the findings of Mahley et al. (2001) [ 34 ] for upper socioeconomic status prepubescent subjects, the present study showed that children with Medium and Higher PEL and MEL were found to have higher serum levels of this lipoprotein than those with Lower PEL and MEL. The effect of diet on serum lipids and lipoproteins has been extensively studied in various clinical trials and epidemiological studies, and although dietary content is clearly an important determinant, several environmental and metabolic factors intervene to modulate the dietary effect (36). Previously conducted surveys in urban regions of Turkey have provided strong evidence that children coming from more affluent and/or well-educated families consumed diets higher in animal protein and fat, while the typical diet of less privileged children was rich in carbohydrates, including cereal grains and sugar (37). These findings, regarding the macronutrient content of the consumed diet, are in consistency with the current study, since children from Medium and Higher PEL and MEL were found to consume higher percentage of energy derived from fat and protein, while the opposite was observed for carbohydrate intake (Table 3 ). Based on the existing literature (38) we could safely speculate that higher intake of fat, especially saturated, and of protein, originating mostly from animal products, could explain the more unfavorable lipidemic profile, with respect to TC and LDL-C serum levels, among children with Higher PEL and MEL. Additionally the lower HDL-C levels and the higher TC to HDL-C ratio reported for children of Lower PEL and MEL could probably be attributed to the higher ratio of carbohydrates constituting their diet. Indeed the percentage of energy derived from carbohydrates has been inversely associated with HDL-C levels according to the results from several other studies [ 39 , 34 ]. Regarding physical activity, it is well established that higher levels of physical activity are associated with higher HDL-C and lower TC to HDL-C ratio values in both children and adolescents [ 40 - 42 ]. However the current study, in agreement to similar findings of Mahley et al (2001) [ 34 ], has not revealed any significant differences with respect to MVPA among children of different PEL and MEL. Still, of great interest are the differences observed in physical activity levels between genders. The present study in consistency with many other studies conducted in the developed world [ 43 - 45 ] revealed that boys were found to devote more time on MVPA than girls. This gender differentiation in physical activity levels should not be attributed to physiological differences between the two sexes but to social and cultural beliefs of parents and teachers as to the types of activities appropriate for boys and girls. Family and society have been shown to influence the level and type of physical activity, girls are engaged in and therefore may determine the lifetime habits with respect to habitual physical activity [ 45 ]. The findings of the current study are in agreement with previous surveys conducted in Turkey indicating that parental education, either as a single estimate of social status or combined with other social and/or economic indices, is a "reliable index" of socioeconomic status [ 34 , 46 ]. Both indices (PEL and MEL) utilized in the present study provided similar findings regarding the differences on children's serum lipid levels. These could probably be attributed to dietary differences, since no significant differences were reported for physical activity. Furthermore, higher percentage of energy derived from fat and protein recorded for Higher PEL and MEL children, compared to those with Lower PEL and MEL, in conjunction with the opposite finding regarding carbohydrates, is in agreement with previous studies indicating "limited" access to food by children of lower socioeconomic status [ 46 ]. These findings may provide some indication that less privileged children are not only under a greater risk of undernourishment but furthermore, by having a higher ratio of TC to HDL-C, that they combine, at the same time, a high prevalence of a health risk index of affluence. However, due to its cross sectional design the present study could not establish causal-effect relationships, but only generate hypothesis about the possible role of parental education on certain CVD risk factors. Another limitation that has to be considered is the fact that the study's focus was on certain biochemical and behavioral indices, since it has not included other important CVD risk factors such as fasting glucose, insulin, blood pressure and smoking habits. The underlying influence of socio-economic status on health status is far more complex than diet and physical activity alone, since many other factors interact in producing the differences observed for serum lipids or other health indices. Based on the data provided by the current study it could be concluded that the differences regarding the lipidemic profile of children with different parental educational level could possibly be attributed to the differences observed on their dietary and macronutrient intake. Any attempt for the development and implementation of a health and education programme should consider these findings and set priorities accordingly. However further research is needed in order to develop an effective index for assessing socioeconomic level, as well as in better understanding its multidimensional role on public health. All these will help the public health authorities to develop effective strategies, which will efficiently tackle these health issues early in life. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549186.xml

539307

Neural network analysis in pharmacogenetics of mood disorders

Background The increasing number of available genotypes for genetic studies in humans requires more advanced techniques of analysis. We previously reported significant univariate associations between gene polymorphisms and antidepressant response in mood disorders. However the combined analysis of multiple gene polymorphisms and clinical variables requires the use of non linear methods. Methods In the present study we tested a neural network strategy for a combined analysis of two gene polymorphisms. A Multi Layer Perceptron model showed the best performance and was therefore selected over the other networks. One hundred and twenty one depressed inpatients treated with fluvoxamine in the context of previously reported pharmacogenetic studies were included. The polymorphism in the transcriptional control region upstream of the 5HTT coding sequence (SERTPR) and in the Tryptophan Hydroxylase (TPH) gene were analysed simultaneously. Results A multi layer perceptron network composed by 1 hidden layer with 7 nodes was chosen. 77.5 % of responders and 51.2% of non responders were correctly classified (ROC area = 0.731 – empirical p value = 0.0082). Finally, we performed a comparison with traditional techniques. A discriminant function analysis correctly classified 34.1 % of responders and 68.1 % of non responders (F = 8.16 p = 0.0005). Conclusions Overall, our findings suggest that neural networks may be a valid technique for the analysis of gene polymorphisms in pharmacogenetic studies. The complex interactions modelled through NN may be eventually applied at the clinical level for the individualized therapy.

Background The increasing number of available genotypes for genetic studies in humans requires more advanced techniques of analysis [ 1 ]. Moreover, genes interact in a complex way, with some gene variants acting additively with others, in a multiplicative way or with a compensatory effect [ 2 , 3 ]. Traditional statistical techniques are not appropriate for detecting such effects [ 4 ], because they rely on the basic assumption of linear combinations only [ 5 ]. Investigation in multifactorial disorders in fact evidenced that non linear interactions are not detected by traditional regression analyses [ 6 ]. In particular, psychiatric disorders are characterized by a non mendelian, multifactorial genetic contribution with a number of susceptibility genes interacting with each other [ 7 , 8 ]. In the process of disentangling the contribution of environment versus genes, it has been recently suggested to focus on endophenotypes instead of psychiatric syndromes as a whole [ 9 , 10 ]. One interesting endophenotype is drug response, a field that gained much attention due to the possible clinical applications, ranging from individualized therapy to new drug development [ 11 - 14 ]. However, notwithstanding the promising results observed in the pharmacogenetic field, no single major effect gene was identified, but a variable number of polymorphisms in various genes are supposedly involved in modulating the response and/or side effects to drugs [ 15 - 20 ]. Since our initial study [ 21 ] we investigated the short term response to Selective Serotonin Reuptake Inhibitors (SSRIs) and a number of candidate genes, observing both positive and negative associations [ 22 ]. However, both the increasing number of genes associated with response and the limitations of traditional methods of analysis are factors requiring the use of new techniques of analysis that more closely resemble to the underlying biological process, i.e. that allows for non-linear interactions. Neural networks (NN) have been proposed for such studies [ 1 , 23 , 24 ]. The main advantage of neural networks is that complex non-linear relationships can be modelled, potentially incorporating high-order interactions between predictive variables. This is of particular importance in a complex phenotype such as antidepressant response [ 22 , 25 ]. NN have been used in other fields of medicine, for example to predict cyclosporine dosage in patients after kidney transplantation [ 26 ], perspective outcome in AIDS research [ 27 ] but also in a genetic analysis in heart disease analysing 10 candidate genes simultaneously [ 28 ]. More complex models including gene-environment interactions have been developed [ 29 ]. In fact, neural networks proved to outperform single marker association tests, particularly in the case of a complex mode of inheritance or where multiple mutations result in more than one haplotype associated with the disease [ 25 , 30 , 31 ]. In the present paper we have re-analysed our sample where polymorphism in the transcriptional control region upstream of the 5HTT coding sequence (SERTPR) and in the Tryptophan Hydroxylase (TPH) gene were analized [ 32 ], in that paper we observed an association of both polymorphisms with drug response but we could not evaluate their possible non linear interactions. In the present paper we had the aim of evaluating the validity of NN models and of comparing them with traditional statistical techniques (multiple regression and discriminant function analysis). Methods Sample The sample was already described in the original paper [ 32 ]. Briefly, two hundred and seventeen depressed inpatients were included in this study (age = 52.11 ± 12.04; onset = 37.97 ± 12.16; female/male: 144/73; bipolars: delusional/non delusional = 40/33, major depressives: delusional/non delusional = 71/73). All patients were evaluated at baseline and weekly thereafter until the sixth week using the 21-item Hamilton Rating Scale for Depression (HAM-D-21) [ 33 ] administered by trained senior psychiatrists blind to genetic data and to treatment (fluvoxamine 300 mg daily from day 8 plus pindolol 7.5 mg to one third of the sample). A decrease in HAM-D scores to 8 or less was considered the response criterion. After the procedure had been fully explained to all subjects, informed consent was obtained. Plasma fluvoxamine levels were determined by high-performance liquid chromatography after 2 weeks of stable 300 mg daily dose [ 34 ]. Nine patients with extreme plasma levels (more than 2 standard deviations) were removed from the study in order to avoid biases due to side effects that are present at high doses, also subjects with plasma levels below 20 ng/ml were excluded as this may indicate non compliance, but no cases with such low doses were observed. The influence of both SERTPR and TPH polymorphisms was limited to subjects not taking pindolol [ 32 ] therefore we included in the present study the 121 subjects including fluvoxamine alone (81 responders/40 non responders). DNA analysis was performed as described in the original paper [ 32 ]. Review of the models used Multilayer Perceptrons This is one of the most popular network architecture in use today, though relatively recent [ 35 ]. In MLP the units each perform a biased weighted sum of their inputs and pass this activation level through a transfer function to produce their output, and the units are arranged in a layered feedforward topology. The first step of the analysis is the choice of the number of layers and nodes. This is performed searching for a minimum in the error/performance hyperplane. Once the number of layers, and number of units in each layer, have been selected, the weight and threshold of the network must be set so as to minimize the prediction error made by the network. This is the role of the training algorithms. The best-known example of a neural network training algorithm is back propagation. In back propagation, the gradient vector of the error surface is calculated and used to decrease the error. A sequence of such moves (slowing as we near the bottom – epochs) will eventually find a minimum. A large number of epochs with no further improvement in the performance suggests that the optimum set of weights has been reached. Linear Networks Originally developed about 60 years ago by Fisher [ 36 ], in classification, the hyperplane is positioned to divide the two classes (a linear discriminant function) while in regression, it is positioned to pass through the data. A linear model is typically represented using an NxN matrix and an Nx1 bias vector. The linear network provides a good benchmark against which to compare the performance of your neural networks. Radial Basis Function Networks In a radial basis function network the response surface of a single radial unit is a Gaussian (bell-shaped) function, peaked at the center, and descending outwards. RBF networks have advantages and disadvantages over MLPs. First, they can model any non-linear function using a single hidden layer, which removes some design-decisions about numbers of layers. Second, the simple linear transformation in the output layer can be optimized fully using traditional linear modelling techniques, which are fast and do not suffer from problems such as local minima which plague MLP training techniques. However the clumpy approach also implies that RBFs are not inclined to extrapolate beyond known data: the response drops off rapidly towards zero if data points far from the training data are used, therefore they are less reliable for clinical samples such our one. Detailed review of the models are reported elsewhere [ 24 , 37 ]. Model development and selection An "intent-to-treat" analysis was carried out for all patients who had a baseline assessment and at least 1 assessment after randomization, with the last observation carried forward on the HAM-D. For the current application the inputs to the first layer of the neural network consist of SERTPR and TPH genotypes while the target outputs consist of response status. The network is then trained to attempt to predict response from genotypes. Each node of the input layer of the network is set to a value representing the genotype of each polymorphism. For each polymorphism and for each subject this value is set to genotypes aa, ab or bb. If a marker genotype is missing then the input is assigned a value equal to the average of the values for all subjects in the dataset, however no missing data were present in our sample. The target output for the network is set to 1 or 2 depending on whether the subject is responding or not. The best network was selected on the basis of its discriminating error and performance, positive and negative predictive values were also reported for each model. This last was expressed as area under the Receiving Operator Characteristic (ROC) Curve. The area under a ROC curve ranges from zero to one, with values close to unity indicating better predictive power; an area of 0.5 indicates that the model is not predicting better than a random choice. However, one major problem of NN analyses is to establish if the prediction from genotypes is greater than would be expected by chance. If the whole sample is used for training, the network will to some extent "learn to recognise" particular features of each member of the dataset and can use these to predict response in a way which may not reflect any general association between marker genotypes and disease. Generally, this problem is faced by a set of strategies: dividing the dataset (50:50, 80:20...), Jackknife, bootstrapping, cross-validation and so on. However those methods present some disadvantages, in particular if only a part of the data is used to train the network this leads to a loss of power given that subjects in the validating part have different patterns of association between genotypes and drug response. In order to remedy these problems, in the case of MLP, it has been suggested to perform both training and testing on the entire dataset. The statistical significance of any observed association between outputs and affection status can be estimated using a permutation test [ 25 ]. Once the network was defined, a statistic, denoted T, is calculated to compare the outputs for responders and non responders in the same way as an unpaired t statistic, although the statistic is not expected to follow a t distribution under the null hypothesis. Instead, in order to estimate statistical significance a permutation procedure is performed. A large number of replicate data sets are generated from the original data and the obtained network model by randomly permuting genotypes with respect to affection status. For each of these replicate data sets we can then train and test the data set as before, each time calculating T. Since each permuted data set will have only random association between genotype and affection status we obtain N values of T which provide a distribution of T under the null hypothesis. We count the number of times any of these values exceeds the value of T we obtained for the real dataset and denote this number R. Then (R + 1)/(N + 1) provides an unbiased estimate of the statistical significance of the association between genotype and affection status in the real dataset. In order to estimate a p-value of alpha, one should carry out approximately 10/alpha replicates. Typically, in order to detect association at a significance of 0.01 one would perform 1000 replicates (including the real dataset and 999 permuted datasets). In the case of the present paper we performed 10000 replicates. Multiple regression and discriminant function analyses were performed to compare the results obtained with the NN strategy with traditional techniques. Responder status was the dependent variable with SERTPR and TPH as independent variables. Genotypes were scored in the following way according to the hypothesis of codominance (SERPR*l/l = 1, SERPR*l/s = 2, SERPR*s/s = 2, TPH*C/C = 1, TPH*C/A = 2, TPH*A/A = 2). Calculations for the NN selection were performed using STATSOFT (Kernel release 5.5 A). Evaluation was performed using the NNPERM package [ 31 ]. Results MLP showed the best performance and was therefore selected over the other networks (see table 1 ). The MLP selected over the other models on the basis of error and performance was composed by 1 hidden layer with 7 nodes (Figure 1 ) after testing about 150 different MLP models. The network showed a very good basic performance (Error 0.430, Performance 0.685). Table 1 Comparison of NN models. PPV = Predictive Positive Value, NPV = Predictive Negative Value, ROC = area under the ROC curve. Network type Error Performance sensitivity specificity PPV NPV ROC Youden's J Linear 0.447 0.636 67.12 56.34 75.97 45.46 0.687 0.23 RBF 0.449 0.691 85.61 35.21 73.10 54.35 0.664 0.21 MLP (1 – 7) 0.439 0.682 77.50 51.20 76.35 52.17 0.731 0.28 Figure 1 MLP composed by 1 hidden layer with 7 nodes used for the analysis. After, we trained the network with the back propagation algorithm. Initially we used a learning rate of 0.1 (momentum 0.3, noise set to 0), after 5000 epochs we reduced it to 0.01 but after 5000 further epochs we observed no improvement and therefore we finished the selection process and retained the network. Both polymorphisms contributed substantially to the model (SERTPR error= 0.532, ratio = 1.21; TPH error= 0.450, ratio = 1.02). This was expected since both markers were individually associated with response. In detail single marker significance, calculated as simple allelic chi-square, was p = 0.00058 for SERTPR and p = 0.025 for TPH. The classification of subjects in responders and non responders was 77.5 % for responders and 51.2% for non responders. Classification may vary depending from the selected threshold, therefore the area under the ROC curve is a better indicator of performance, in this case the area was 0.731. We also evaluated the predictive power of the network with the SERTPR polymorphism only, in this case the area under the ROC curve was 0.698. We may therefore observe that the add of TPH polymorphism increases the predictive power of the system. In order to evaluate the significance of the network we applied a permutation test with 10000 replicates. The t statistic for the network was 4.35, it was achieved in 81 out of 10000 simulations yielding a network p-value = (81+1)/(10000+1) = 0.0082. Finally, we performed a comparison with traditional techniques. A multiple regression analysis showed a significant correlation (p = 0.0004) with a variance explained of 12.5%. The discriminant function analysis correctly classified 34.1 % of responders and 68.1 % of non responders (F = 8.16 p = 0.0005). Following, we tested the possible impact of clinical variables on response. We included in the model the following variables: Age, age at onset, sex, education, diagnosis, presence of delusional features, recurrence index (defined as number of episodes per year), pindolol augmentation and baseline HAM-D. With those variables no satisfactory network was identified. They were therefore not considered as possible confounding factors in the genetic analysis. Discussion This paper reports the first attempt to use NN in pharmacogenetic analyses. We applied this technique to short term antidepressant response in mood disorders. Our analyses suggest that MLP network is the most appropriate for this kind of data, in according with previous observations [ 25 ]. The growing number of polymorphisms (about 3.000.000) and the growth of simultaneous techniques such as gene arrays ask for appropriate techniques of analysis. Traditional ones have strong limitations not allowing for non linear interactions and the risk of overfitting in the case of multiple polymorphisms analysed in necessary limited sample sizes. We observed that a relatively simple MLP NN is able to predict response in a way comparable to traditional techniques. The lack of non linear interactions in the simple model we analysed [ 32 ] explains why did not observe a marked superiority of NN over traditional analyses. However the most promising result of the strategy we tested in the present paper is the possibility to add a large number of polymorphism to the network and to evaluate the improvement in the prediction, showed by the area under the ROC curve. Moreover the significance of the network can be evaluated with the permutation test [ 25 , 31 ]. Moreover the MLP model we used is quite parsimonious in terms of parameters used (2 input variables, 1 output variable and 1 hidden layer with 7 nodes). Further developments of this strategy are the inclusion of more detailed information on the phenotypic side. The classification results we obtained are not sufficient in clinical terms were in particular much higher specificities are needed in order to recognize in advance non responders. To reach this target we should consider that we previously observed that some polymorphism influence only part of the whole depressive symptomatology [ 38 ]. Further clinical variables should also be considered as reported to influence the short term antidepressant outcome [ 39 ], even if previous NN studies failed to identify clinical predictors of antidepressant response [ 40 ]. Our analyses are in line with this view, in fact the clinical variables we analysed were not significantly associated with outcome. The relatively small sample we used does not guarantee against a possible overfitting phenomenon, therefore enlargement of the sample is a priority. Moreover we used the same sample for testing and validating our result, this is not a standard technique [ 41 ], this problem is usually faced by a set of strategies such as dividing the dataset, Jackknife, bootstrapping, cross-validation and so on. However those methods present some disadvantages, in particular if only a part of the data is used to train the network this leads to a loss of power in the case that subjects in the validating part have different patterns of association between genotypes and drug response. Therefore in the present paper we performed both training and testing on the entire dataset with the use of a permutation test to validate the results [ 25 ]. Another limitation of the present paper is that we compared NN with multiple regression only, other techniques could be tested as well such as set association [ 30 ], multifactor dimensionality reduction [ 42 ], and logic regression [ 43 ]. Differences in allele frequency for different populations have been reported [ 44 ]. However our sample was composed of subjects mainly collected in the North of Italy with Italian antecedents for at least two generations, though genetic heterogeneity have been evidenced for some isolate populations (such as Sardinia, not included in our sample) Italy is characterized by a substantial genetic homogeneity [ 45 ]. Another caveat is linked to the characteristics of our sample. In fact the Center for Mood Disorders of San Raffaele Hospital is a tertiary structure, therefore we cannot exclude a potential selection bias associated with illness severity and possible extension to outpatients or drug abusers are not warranted [ 46 ]. Conclusions Overall, our findings suggest that neural networks may be a valid technique for the analysis of gene polymorphisms in pharmacogenetic studies. The complex interactions modelled through NN may be eventually applied at the clinical level for the individualized therapy [ 47 ]. Competing interests The author(s) declare that they have no competing interests. Authors' contributions AS conceived the study, drafted the manuscript and participated in the design of the study and performed the statistical analysis. ES participated in its design and coordination. All authors read and approved the final manuscript Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539307.xml

534810

Phylogeography and Genetic Ancestry of Tigers (Panthera tigris)

Eight traditional subspecies of tiger (Panthera tigris), of which three recently became extinct, are commonly recognized on the basis of geographic isolation and morphological characteristics. To investigate the species' evolutionary history and to establish objective methods for subspecies recognition, voucher specimens of blood, skin, hair, and/or skin biopsies from 134 tigers with verified geographic origins or heritage across the whole distribution range were examined for three molecular markers: (1) 4.0 kb of mitochondrial DNA (mtDNA) sequence; (2) allele variation in the nuclear major histocompatibility complex class II DRB gene; and (3) composite nuclear microsatellite genotypes based on 30 loci. Relatively low genetic variation with mtDNA, DRB, and microsatellite loci was found, but significant population subdivision was nonetheless apparent among five living subspecies. In addition, a distinct partition of the Indochinese subspecies P. t. corbetti into northern Indochinese and Malayan Peninsula populations was discovered. Population genetic structure would suggest recognition of six taxonomic units or subspecies: (1) Amur tiger P. t. altaica; (2) northern Indochinese tiger P. t. corbetti; (3) South China tiger P. t. amoyensis; (4) Malayan tiger P. t. jacksoni , named for the tiger conservationist Peter Jackson; (5) Sumatran tiger P. t. sumatrae; and (6) Bengal tiger P. t. tigris . The proposed South China tiger lineage is tentative due to limited sampling. The age of the most recent common ancestor for tiger mtDNA was estimated to be 72,000–108,000 y, relatively younger than some other Panthera species. A combination of population expansions, reduced gene flow, and genetic drift following the last genetic diminution, and the recent anthropogenic range contraction, have led to the distinct genetic partitions. These results provide an explicit basis for subspecies recognition and will lead to the improved management and conservation of these recently isolated but distinct geographic populations of tigers.

Introduction The tiger (Panthera tigris) is the largest felid species and a widely recognized symbol of wildlife conservation. Historically tigers inhabited much of Asia, including the regions between the Caspian and Aral Seas, southeastern Russia, and the Sunda islands ( Mazak 1981 ; Hemmer 1987 ; Herrington 1987 ). Since the early 1900s, however, habitat loss, fragmentation, and human persecution have reduced tiger populations from probably over 100,000 in 1900 to fewer than 7,000 free-ranging individuals ( Nowell and Jackson 1996 ; Dinerstein et al. 1997 ; Kitchener and Dugmore 2000 ). Most populations consist of less than 120 animals, increasing the risk of local extirpation due to demographic and genetic factors ( Smith and McDougal 1991 ; Dinerstein et al. 1997 ). There are eight generally accepted tiger subspecies in accordance with their geographic distribution ( Figure 1 ). Bali (P. t. balica), Caspian (P. t. virgata), and Javan ( P. t. sondaica ) tiger subspecies were eradicated by the 1940s, 1970s, and 1980s respectively ( Nowell and Jackson 1996 ). Today an estimated 3,200–4,500 Indian or Bengal tigers (P. t. tigris) exist in Bangladesh, Bhutan, western China, India, western Myanmar, and Nepal ( Seidensticker et al. 1999 ). Fewer than 500 Amur or Siberian tigers (P. t. altaica) survive in eastern Russia, northeastern China, and Korea ( Matyushkin et al. 1999 ; Miquelle and Pikunov 2003 ), while approximately 50 Amoy or South China tigers (P. t. amoyensis) now exist in captivity only ( Tilson et al. 2004 ). An estimated 400–500 Sumatran tigers (P. t. sumatrae) occur in Sumatra ( Seidensticker et al. 1999 ); and 1,200–1,800 Indochinese tigers (P. t. corbetti) live in Cambodia, China, Laos, Malaysia, east Myanmar, Thailand, and Vietnam ( Seidensticker et al. 1999 ) ( Figure 1 ). Figure 1 Historic and Current Geographic Distribution of Tigers Corresponding to the Eight Traditional Subspecies Designation Geographic origin of samples and sample size (circles or squares) from each location are indicated (see Table 3 for sources). Three-letter codes (TIG, ALT, etc.) are indicated subspecies abbreviations. Dotted lines are approximate boundaries between tiger subspecies studied here. The Isthmus of Kra divides the traditional Indochinese tigers into the northern Indochinese tigers P. t. corbetti I and the Malayan tigers P. t. corbetti II based on the present study. We propose the Malayan tiger subspecies, COR II, be named P. t. jacksoni, to honor Peter Jackson, the former Chair of the IUCN's Cat Specialist Group who has contributed significantly to worldwide tiger conservation. Table 3 Samples of Panthera tigris Used in the Study a Birth Status of each tiger: W, wild-born, C, captive-born; U, status unknown b Identification number of tiger individuals as they are listed in the database at the Laboratory of Genomic Divesity, National Cancer Institute, Frederick, Maryland, United States c MtDNA haplotype assigned to each sample sequenced in the study d MHC ClassII DRB allele genotypes e Samples of pelt or hair f Red samples represent samples with microsatellite data from 30 loci g Tigers individuals classified as South China tiger originally Subspecies of tigers are traditionally defined by body size, skull characters, pelage coloration, and striping patterns ( Mazak 1981 ; Herrington 1987 ). It is generally believed that the largest tigers occur in the Russian Far East, and the smallest are found in the Sunda Islands. The shape of the occiput in the skull is characteristically narrow in the Javan and Bali tigers and much broader in Caspian tigers ( Mazak 1996 ). However, the adequacy of these traditional subspecies designations is tentative at best, since morphological distinctions in many cases have been based on a few specimens, and because subsequent studies have failed to affirm these distinctions. Herrington (1987) and Kitchener (1999) have revealed a wide range of morphological variations within the subspecies and, to some extent, overlapping among the subspecies. A previous molecular genetic assessment of 28 tigers has indicated a low level of genetic variation, revealing little evidence for subspecies distinctiveness ( Wentzel et al. 1999 ). Moreover, ecological analyses of tiger habitat ( Kitchener and Dugmore 2000 ) indicate that there have been few geographic barriers (e.g., mountain ranges and deserts) to migration and gene flow that would have been sufficient for subspecies isolation. One ecology-based conservation approach emphasizes protection of about 160 continuous habitat patches or tiger conservation units regardless of subspecies designation ( Dinerstein et al. 1997 ). Although this strategy may be desirable, optimal tiger conservation may also require additional interventions such as establishing corridors and buffer zones and/or implementing reintroduction programs ( Tilson et al. 2001 ). To this end, an assessment of population genetic structure of living tigers interpreted in the context of traditional intraspecific taxonomy and the species' evolutionary history would benefit both in situ and ex situ conservation management design. Molecular genetic markers have been increasingly applied to assess genetic partitions among geographically isolated populations, to define the evolutionary significant unit below the species level for conservation management purposes, and to revise the traditional species and subspecies designations ( Avise and Ball 1990 ; Moritz 1994 ; Fraser and Bernatchez 2001 ). Subspecies recognition is particularly relevant for tigers, because the current conservation strategy for this species has been inextricably bound to knowledge of its subspecific taxonomy. In this study we adhere to the subspecies concept as defined by Avise and Ball (1990) and O'Brien and Mayr (1991) , to include populations below the species level that share a distinct geographic distribution, a group of phylogenetically concordant characters, and a unique natural history relative to other subdivisions of the species. Here we attempt to overcome several factors that have complicated previous efforts to fully describe patterns of genetic variation in tigers. Foremost among these has been the limited sample size of “voucher specimens” (defined as individuals that were verified as wild-born from a specific geographic locale or captive-born from geographically verified wild-born parents). In addition, the presence of Numt, a nuclear pseudogene insertion of cytoplasmic mitochondrial DNA (mtDNA) in tiger autosomes ( Lopez et al. 1994 ; Johnson et al. 1996 ; Cracraft et al. 1998 ; J. H. Kim, A. Antunes, S.-J. Luo, J. Menninger, W. G. Nash, et al., personal communication) has made it difficult to utilize universal mammalian primer sets for mitochondrial genes, because they will coamplify Numt. Furthermore, paucity of genetic diversity across tigers, especially in mtDNA ( Wentzel et al. 1999 ), has made it necessary to sequence a large portion of the mtDNA genome and to assess genetic variation in multiple rapidly evolving microsatellite loci. To establish proper biological reference specimens, samples from 134 tigers of known geographic origin were collected. Three genetic markers were examined: (1) 4 kb of mtDNA sequence derived from primer pairs that excluded Numt amplification, (2) allele variation in the major histocompatibility complex (MHC) DRB gene; and (3) allele size variation of 30 hypervariable short tandem repeat loci or microsatellites. Observed patterns of population genetic variation replicated with different gene families form the basis of interpretation of the tiger's evolutionary history and recommendations for its management. Results Phylogenetic Analysis of mtDNA and Microsatellites Mitochondrial gene fragments were amplified and sequenced from DNA extracted from 72 blood or tissue specimens using 10 cytoplasmic mitochondria (Cymt)-specific primer pairs ( Figure 2 and Table 1 ). The fragments were concatenated in a 4,078-bp contiguous sequence. Additional mtDNA sequences were generated from 28 historical samples (pelt or hair) by amplifying shorter fragments (less than 400 bp) targeting selected variable sites to determine their similarity to the previously characterized haplotypes. Combined mtDNA sequences were obtained from 100 tigers from Russian Far East ( n = 13), south China ( n = 4), northern Indochina ( n = 30), Malayan Peninsula ( n = 22), Sumatra ( n = 16), and the Indian subcontinent ( n = 15). The mtDNA sequences specified 54 variable sites defining 25 haplotypes ( Table 2 ). Thirty of the polymorphisms were observed in more than one individual and were thus phylogenetically informative ( Table 2 ), and 29 of the 30 changes were transitions. Figure 2 Schematic of P. tigris mtDNA The position of PCR primers used for amplification of Cymt specific sequences and alignment of the homologous Numt sequence (outer, dashed line) in tiger mitochondria. Fifteen Cymt-specific primer sets spanning 6,026 bp of mtDNA were designed and screened for polymorphism in tigers (inner, solid line). Five indicated segments showed no variation among fifteen tigers that represented five traditional subspecies and therefore were excluded from further analysis. The ten variable segments (4,078 bp) were amplified in 100 tiger individuals. Primer sequences are listed in Table 1 . Diamonds indicate polymorphic mtDNA segments; brackets indicate monomorphic mtDNA segments among tigers that were excluded from phylogenetic analysis. Table 1 PCR Primers Specific for Cytoplasmic Mitochondrial DNA Sequences a Primers are listed in the 5′-to-3′ direction b PCR products amplified using these primer sets show no variation among all samples Table 2 Haplotypes and Variable Sites in Combined Analysis of 4,078 bp of Tiger (P.tigris) mtDNA Sequences a Nucleotide positions correspond to the complete reference Felis catus mtDNA sequence ( Lopez et al. 1996 ) b Subspecies abbreviation code as in Figure 1. Base pairs identical to haplotype ALT are indicated by a dash c Number of individuals with each haplotype. Individual tiger mtDNA haplotypes are listed in Table 3 d Red nucleotides are subspecies-specific sites e COR1/AMO3 is a haplotype shared by 21 tigers that are initially designated as COR and one AMO (text and Table 3 ) Table 3 Continued Table 3 Continued Phylogenetic analyses of the mtDNA haplotypes using maximum parsimony (MP), minimum evolution (ME), and maximum likelihood (ML) approaches produced congruent topologies that defined major geographic partitions ( Figure 3 A). Eight haplotypes (SUM1 to SUM8) generated from 16 Sumatran tigers (P. t. sumatrae) formed a monophyletic group (80% MP, 70% ME, and 66% ML bootstrap support). A second monophyletic cluster of six haplotypes (TIG1 to TIG6) from 15 Bengal tigers (P. t. tigris) also received high bootstrap support (93% MP, 82% ME, and 90% ML). The rest of the mainland Asian haplotypes grouped together and partitioned into three distinct geographic groups: (1) a genetically invariant Amur tiger lineage (P. t. altaica) represented by a single haplotype in 13 individuals, (2) a northern Indochinese lineage ( P. t. corbetti I) of individuals originating from south China to the Indochinese countries north of the Isthmus of Kra, and (3) a paraphyletic assembly of haplotypes from tigers from Malayan Peninsula ( P. t. corbetti II). Support for subdividing the conventional Indochinese subspecies of tigers P. t. corbetti into two clusters was high (bootstrap values for P. t. corbetti I were 94% MP, 96% ME, and 94% ML). The COR1/AMO3 haplotype, represented by 22 individuals from Vietnam ( n = 2), Cambodia ( n = 14), northeast Thailand ( n = 5), and south China ( n = 1), was the only haplotype found in two classical subspecies lineages ( P. t. amoyensis and P. t. corbetti ) ( Table 2 ). Figure 3 Phylogenetic Relationships among Tigers from mtDNA Haplotypes (A) Phylogenetic relationships based on MP among the tiger mtDNA haplotypes from the combined 4,078 bp mitochondrial sequence ( Table 2 ). Branches of the same color represent haplotypes of the same subspecies. Trees derived from ME and ML analyses have identical topologies. Numbers above branches represent bootstrap support from 100 replicates using the MP method, followed by bootstrap values using the ME-ML analyses (only those over 70% are indicated). Numbers below branches show number of MP steps per number of homoplasies from a strict consensus tree. Numbers in parentheses represent numbers of individuals sharing the same haplotype. MP analysis using heuristic search and tree-bisection-reconnection branch-swapping approach results in two equally most-parsimonious trees and the one resembling the ME and ML trees is shown here (tree length = 60 steps; CI = 0.900). The ME tree is constructed with PAUP using Kimura two-parameter distances (transition to transversion ratio = 2) and NJ algorithm followed by branch-swapping procedure (ME = 0.0142). The ML approach is performed using a TrN (Tamura-Nei) +I (with proportion of invariable sites) model, and all nodes of the ML tree were significant (a consensus of 100 trees, –Ln likelihood = 5987.09). (B) Statistical parsimony network of tiger mtDNA haplotypes based on 4,078 mtDNA sequences constructed using the TCS program ( Clement et al. 2000 ). The area of the circle is approximately proportional to the haplotype frequency, and the length of connecting lines is proportional to the exact nucleotide differences between haplotypes with each unit representing one nucleotide substitution. Missing haplotypes in the network are represented by dots. Haplotype codes and the number of individuals (in parentheses) with each haplotype are shown (see Table 2 ). Voucher samples of five captive tigers collected in China, designated South China subspecies P. t. amoyensis, fell into two very distinct phylogenetic origins . Two tigers from the Suzhou Zoo (Pti-217 and Pti-218; Table 3 ) carried the COR1/AMO3 haplotype, and the third (Pti-222) contained haplotype AMO2, which differed by a single nucleotide substitution from COR1/AMO3 ( Table 2 ). The two South China tiger haplotypes grouped phylogenetically with the northern Indochinese P. t. corbetti I haplotypes (COR1–COR3) in all phylogenetic analyses ( Figure 3 A and 3 B), and likely indicate that the maternal (mitochondrial) lineages of these tigers derived from individuals from the P. t. corbetti I phylogenetic lineages. In contrast, two P. t. amoyensis tigers (Pti-219 and Pti-220) from the Chongqing Zoo collection had a haplotype (AMO1) that formed a separate lineage that was ten nucleotide substitutions from its nearest sequence (Sumatran; Figure 3 B and Table 2 ). If affirmed by larger sampling, this lineage would reflect a unique P. t. amoyensis genetic haplotype. A statistical parsimony network of the tiger mtDNA sequences provided additional analytical support for the differentiation of P. t. sumatrae, P. t. tigris, P. t. altaica, P. t. corbetti I, P. t. corbetti II, and P. t. amoyensis (AMO1 only) ( Figure 3 B). Haplotypes from the same geographic group tended to be interrelated, and intergroup distances among haplotypes were generally larger than branch lengths within each group (1–4 bp). The exceptions were two lineages within the Malayan P. t. corbetti II cluster that were separated by 7 bp, which may be a result of the existence of further population substructure or, alternatively, of limited sampling in the region. Each of the six tiger subspecies groups was connected to other groups in close but not exact correspondence to their geographic location. For instance, P. t. altaica was the sister taxon to P. t. corbetti I which was connected to P. t. corbetti II. P. t. sumatrae haplotypes were linked to P. t. tigris by 7 bp and to P. t. amoyensis by 10 bp. Nonetheless, the phylogenetic relationships among the subspecies were not resolved to a robust hierarchy, and therefore were consistent with a contemporaneous divergence of extant phylogeographic lineages. Composite genotypes from 30 felid-specific microsatellite loci ( Menotti-Raymond et al. 1999 ) were obtained in 113 tiger samples. Neighbor joining (NJ) analyses of individual tiger genotypes based on the proportion of shared allele (Dps) and kinship coefficient (Dkf) genetic distances produced concordant topologies ( Figures 4 and S1 ) that lend support to the same phylogeographic population subdivisions observed in the mtDNA analysis. Tigers from Sumatra (P. t. sumatrae) formed a monophyletic clade with 97% bootstrap support, and Amur tigers (P. t. altaica) grouped with 76% bootstrap support. The remaining tiger genotypes partitioned into two weakly supported monophyletic lineages (Indian Subcontinent P. t. tigris and Malayan Peninsula P. t. corbetti II) and a paraphyletic assemblage of northern Indochinese P. t. corbetti I. For example, three individuals from Thailand (Pti-296, Pti-297, and Pti-301) clustered with samples from the India subcontinent, blurring the distinction between P. t. corbetti I and P. t. tigris. The three South China tigers from the Suzhou Zoo, China, that had clustered with P. t. corbetti I by mtDNA (Pti-217, Pti-218, and Pti-222) also associated more closely with P. t. corbetti I from northern Indochina by microsatellite analysis ( Figure 4 ). The two distinct (by mtDNA) P. t. amoyensis individuals (Pti-219 and Pti-220) from the Chongqing Zoo, China, likewise formed a distinct lineage in the microsatellite analysis ( Figure 4 ). Figure 4 Phylogenetic Relationships among the Individual Tigers from Composite Microsatellite Genotypes of 30 Loci Branches of the same color represent tiger individuals of the same subspecies. The NJ tree, which is based on Dps and Dkf with the (1 – ps/kf) option in MICROSAT ( Minch et al. 1995 ), generated similar topologies, and only the Dps tree is shown here. Numbers are individual Pti codes ( Table 3 ). Bootstrap values over 50% are shown on the divergence node. Population Subdivision Analysis To quantify the extent of population differentiation in modern tigers, we evaluated four different geographic subdivision scenarios and compared them on the basis of analysis of molecular variance (AMOVA) with both mtDNA haplotypes and microsatellite genotypes ( Table 4 ). P. t. amoyensis individuals (Pti-219 and Pti-220) were excluded in this subdivision analysis due to our small sample size. In our first hypothesis, two groups were considered: the P. t. sumatrae island population and all contemporary mainland populations ( P. t. altaica, P. t. corbetti I, P. t. corbetti II, P. t. tigris) . This recently proposed model ( Cracraft et al. 1998 ; Kitchener 1999 ; Kitchener and Dugmore 2000 ) presumes continuous habitat distribution on the mainland. The second scenario considered tigers as three groups: the Sumatran population (P. t. sumatrae), the Amur tigers (P. t. altaica), which presently are isolated from other tiger populations by more than their maximum known dispersal distance ( Mazak 1996 ), and a group of the other mainland tigers subspecies. The third hypothesis followed the division of the four traditional subspecies: (1) Amur tigers (P. t. altaica), (2) P. t. corbetti, including Indochina and part of south China, (3) Bengal tigers (P. t. tigris), and (4) Sumatran tigers (P. t. sumatrae). The fourth scenario, based on the results of the mtDNA phylogenetic analyses (see Figure 3 ) and the hypothesis that the Isthmus of Kra may serve as a potential geographic barrier ( Kitchener 1999 ), further subdivided classical P. t. corbetti into the northern Indochina region P. t. corbetti I and the Malayan Peninsula P. t. corbetti II, resulting in five groups. The AMOVA results for each of the four scenarios are presented in Table 4 . Table 4 Measures of Geographic Subdivision Based on AMOVA with MtDNA and Microsatellite Data a Population subdivision scenarios are described in the text b Subspecies was grouped by brackets into populations for the analysis For both mitochondrial haplotype and microsatellite data, the five-group scenario yielded the highest F st (for mtDNA, defined as the proportion of total genetic variation that is attributable to genetic differences between populations) and R st (for microsatellites, an F st analogy suited for the stepwise mutation model that applies to microsatellite data) values. Under this model, 31% of the microsatellite variation discriminated between the five groups, while the balance, 69%, occurred within each group. For mtDNA the F st was very high (0.838), indicating that 84% of the variation was partitioned among the different phylogeographic subspecies. Each of the five subspecies showed highly significant population genetic differentiation ( p < 0.0001) by pairwise F st and R st with 10,000 permutations ( Table 5 ). The contrast between the mtDNA and microsatellite genetic variation probably reflects the difference in the effective population size assessed by these two different markers and/or, to some extent, the intersexual differences in dispersal. Table 5 Measures of Pairwise Comparisons in Tigers Based on AMOVA with mtDNA and Microsatellite Data Population pairwise F st estimates under the five-group scenario using the combined data from the mitochondrial regions and Kimura two-parameter are below the diagonal; R st estimates using data from 30 microsatellite loci are above the diagonal. All populations are significantly different ( p < 0.0001) by F st values based on mitochondrial data or R st values based on microsatellite data An alternative analysis of the combined microsatellite and mitochondrial haplotype data using a Bayesian approach ( Figure S2 and Table S1 ) as implemented in the program STRUCTURE ( Pritchard et al. 2000 ) supported the partitioning of P. t. altaica, P. t. sumatrae, P. t. tigris, and P. t. corbetti II, but further split the 33 P. t. corbetti I individuals into three distinctive population groups: (1) four tigers from China and Vietnam; (2) nine tigers from Cambodia; and (3) 20 tigers from Cambodia and northern Thailand ( K = 7, Pr [ K ] = 0.993). In this scenario, most individuals were assigned to a cluster with high probability ( q > 0.90), indicating very low level of gene flow between the groups. However, because this additional substructure within P. t. corbetti I had little geographic or ecological basis, and because AMOVA analysis based on this population subdivision resulted in lower F st and R st values than that in the five-group scenario (unpublished data), the distinction was not considered to be a consistent basis for subspecies classification and may reflect additional population differentiation within a subspecies. Genetic Variation in Tigers Quantitative estimates of mtDNA diversity in tigers with comparable estimates from selected felid species demonstrated that overall, tigers had moderate levels of mtDNA diversity ( Table 6 ), substantially less than leopards (P. pardus) ( Uphyrkina et al. 2001 ), Geoffroy's cat (Oncifelis geoffroyi), Pampas cat (O. colocolo), or tigrina (Leopardus tigrinus) ( Johnson et al. 1999 ), but comparable to pumas (Puma concolor) ( Culver et al. 2000 ) in percent variable sites, mean pairwise distance among individuals, and average nucleotide diversity. Four tiger subspecies ( P. t. tigris, P. t. sumatrae, P. t. corbetti I, and P. t. corbetti II) showed moderate nucleotide diversity (π), ranging from 0.0001 to 0.0070 ( Table 6 ). The P. t. altaia sampling of 13 individuals showed no mtDNA haplotype variation. Of the five individuals originally designated as P. t. amoyensis, three were genetically indistinguishable from P. t. corbetti I, resulting in an inadequate sample size for a meaningful estimation of population variation. Table 6 Estimates of Molecular Genetic Variation from Combined MtDNA Sequences (4,078 bp) a Fifteen tigers were screened in a 6,026 bp mtDNA segment, and 1,948 bp was excluded in the following large-scale sampling because of lack of variation b From a combined analysis of mtDNA ND5 (611 bp) and CR (116 bp) ( Uphyrkina et al. 2001 ) c From a combined analysis of mtDNA 16S (364 bp), ATP8 (191 bp), and ND5 (318 bp) ( Johnson et al. 1999 ) d From a combined analysis of mtDNA 16S (382 bp), ATP8 (191 bp), and ND5 (318 bp) ( Culver et al. 2000 ) Parameters of microsatellite variation have been shown to provide sensitive measures of historic demographic perturbations in felid and other species ( Driscoll et al. 2002 ). Estimates of heterozygosity, average numbers of allele per locus, microsatellite variance in allele size, and allele size range in tigers were comparable to other felid species such as jaguar, leopard, puma, lions, and cheetahs across the same microsatellite loci ( n = 17) ( Table 7 ). After Bonferroni correction, eight of the 30 loci were significantly out of Hardy-Weinberg equilibrium in P. t. corbetti I ( p < 0.00167), possibly reflecting further population subdivision in this region. Expected heterozygosity in tigers ranged from 0.456 in P. t. altaica to 0.670 in P. t. corbetti I ( Table 7 ). Average microsatellite variance was highest in P. t. tigris (4.94) and P. t. corbetti I (3.58) and lowest in P. t. altaica (1.93). Table 7 Genetic Variation across 30 Microsatellite Loci in Tiger Subspecies Included are values describing genetic variation across 30 microsatellite loci in the six revised tiger subspecies, and a comparison with other Felidae species across the same 17 loci. Estimates of microsatellite diversity are calculated across a subset of microsatellite loci used in previous studies ( Driscoll et al. 2002 ; Uphyrkina et al. 2001 ; Eizirik et al. 2001 ) All six phylogeographic subspecies groups showed population-specific alleles that tended to represent the extreme sizes of allele distributions ( Table 8 ). Of the 49 private alleles, 26 were either the largest or smallest size class among all tigers, and 38 were either the smallest or the largest for a specific subspecies, thus supporting a recent derivation. Frequencies of such private alleles were low in each population, from 1.5% of total allele numbers in P. t. amoyensis to 14.6% in P. t. corbetti I ( Table 8 ) . In addition, P. t. corbetti I had the highest average number of alleles per locus, the highest average allele size range per locus, and the most continuous and heterogeneous allele size distribution among all subspecies groups. Table 8 Diagnostic Characters and Habitat of the Six Phylogeographic Tiger Groups or Subspecies a See Table 2 for mtDNA nucleotide coordinates b Possibly extinct in the wild ( Tilson et al. 2004 ) ND, no data Major Histocompatibility Complex— DRB Gene Variation The most polymorphic gene complex in all mammals is the MHC. This critical region for immunological recognition of infectious agents has 147 genes in the domestic cat, including three functional class II DRB genes on chromosome B3 ( Yuhki et al. 2003 ). DRB gene homologs were amplified from DNA extracted from 21 tigers and screened for sequence diversity using single strand conformational polymorphism (SSCP). There were a total of seven electrophoretic allele variants (A–G). This is a relatively low MHC- DR diversity compared to human and domestic cat, which possess 126 and 63 DRB alleles, respectively, for the same gene segment in samplings of 251 humans and 37 cats, respectively ( Yuhki and O'Brien 1997 ; Bodmer et al. 1999 ). Despite this reduced DRB variation among tigers, there was detectable population differentiation. Three mainland subspecies P. t. tigris ( n = 1), P. t. altaica ( n = 5), and P. t. corbetti I ( n = 2) were genetically identical for DRB -A allele sequence. Three additional DRB alleles (B, C, and D) were found only in P. t. corbetti II ( n = 2), while three others (E, F, and G) were unique to P. t. sumatrae ( n = 11) ( Tables 3 and 8 ) ( Wentzel et al. 1999 ). Estimation of the Coalescence Time of Genetic Variations in Tigers The mtDNA sequence divergences in a combined data set of 3,217 bp, of which homologous sequences from the tiger and leopard were both determined (see Materials and Methods ), were used to estimate coalescence time for extant tiger mtDNA lineages and its 95% confidence interval (CI: ± two standard errors) based on a linearized tree method ( Takezaki et al. 1995 ). Neither the two-cluster nor the branch-length molecular clock test revealed significant rate heterogeneity among tiger sequences (confidence probability less than 95%), suggesting that the divergence of the mtDNA sequences were compatible with a molecular clock hypothesis. Thus, all sequences were used to construct a linearized tree using the NJ tree algorithm with Kimura two-parameter distances. Assuming a divergence time for leopards and tigers of 2 MY, there were an estimated 2.29 × 10 –8 substitutions per site per y, or one substitution every 14,000 y in the segment examined. According to this rate, the estimated coalescence time of mtDNA variation for extant tiger lineages was 72,000 y (95% CI = 39,000–104,000 y). An older fossil record calibration of 3 MY for the separation of leopards and tigers produced a rate of 1.53 × 10 –8 substitutions per site per y, or one substitution every 20,000 y. According to this substitution rate, mtDNA diversity of modern tigers originated about 108,000 y (95% CI = 59,000–157,000 y) ago. Based on either calibration, the Amur tigers probably experienced a genetic reduction or founder event more recently (less than 20,000 y), as no variation was detected within the population. The estimate of microsatellite variance in average allele repeat-size can also be used as a surrogate for evolutionary time based on the rate of allele range reconstitution subsequent to a severe founder effect ( Driscoll et al. 2002 ). Using a standard curve for the relationship of microsatellite variance to elapsed time (see Figure 4 in Driscoll et al. [2002] ), the variance for all tigers converged to 19,000 y ago. The age of different subspecies, based on populations for which we had an adequate sample size ( n > 15), ranged from 9,900 y in Amur tigers P. t. altaica to 18,437 y in northern Indochinese tigers P. t. corbetti I. We estimated the historic population size required to sustain the level of mitochondrial genetic variation under the assumption of neutrality of substitution and mutation-drift equilibrium ( Kimura 1955 ; Nei 1987 ), where the population parameter θ = 2 N e μT, and N e is the long-term effective female population size, μ the substitution rate per site per year, and T the generation time. From a coalescent-based simulation of the mitochondrial sequences, the average estimate of θ was 0.00255 per nucleotide site, with a 95% CI from 0.00147 to 0.00417. With the substitution rate calibrated from this study (1.91 × 10 –8 bp –1 y –1 ) and an average generation time of 5 y for tigers ( Smith and McDougal 1991 ), the historical effective population size is an estimated 13,350 females (95% CI = 7,700–21,830). Discussion Overall, tigers displayed moderate levels of molecular genetic variation in mtDNA and DRB sequences compared with other mammalian species, consistent with previous allozyme studies ( O'Brien et al. 1987 ). There was a variable site every 75 bp, with 54 sites in the more variable 4-kb segment and one variable site every 112 bp in the larger 6,026-bp segment (see Materials and Methods ). This value was less than what was observed in leopards in a smaller portion of mtDNA (one variable site every 15 bp in 727 bp of the gene encoding NADH dehydrogenase subunit 5, called ND5, and that for the control region, called CR; 34 haplotypes were found) ( Uphyrkina et al. 2001 ). MHC class-II DRB gene variation was also low relative to human and domestic cat ( Yuhki and O'Brien 1997 ; Bodmer et al. 1999 ). By contrast, estimates of tiger microsatellite variability were more similar to those of other felid species ( Table 7 ) ( Culver et al. 2000 ; Eizirik et al. 2001 ; Uphyrkina et al. 2001 ; Driscoll et al. 2002 ). The oldest tiger fossils, around two million y (MY) old, are from northern China and Java ( Hemmer 1987 ). By the late Pliocene and early Pleistocene tigers were widely distributed in eastern Asia. However, Pleistocene glacial and interglacial fluctuations and other geological events probably caused repeated geographic restrictions and expansions ( Hemmer 1987 ; Kitchener 1999 ; Kitchener and Dugmore 2000 ). We estimated the most recent common ancestor for tiger mtDNA haplotypes was 72,000–108,000 y ago, with a lower and upper bound of 39,000 y and 157,000 y, respectively. This estimate is much earlier than that derived for the leopard, which is considered to have originated in Africa 470,000–825,000 y ago and to have arrived in Asia 170,000–300,000 y ago ( Uphyrkina et al. 2001 ). Likewise, extant jaguar (Panthera onca) lineages diverged approximately 280,000–510,000 y ago ( Eizirik et al. 2001 ). Our coalescence estimate for tigers corresponds roughly with the catastrophic eruption of Toba in Sumatra around 73,500 y ago ( Rampino and Self 1992 ), which has been linked to the Late Pleistocene bottleneck in human evolution ( Ambrose 1998 ) and to a major northward dispersal event in the Asian elephants ( Fleischer et al. 2001 ). Based on the subspecies definition of O'Brien and Mayr (1991) and Avise and Ball (1990) , our data suggest that there are at least five and possibly six tiger subspecies: Amur tigers (P. t. altaica); northern Indochinese tigers ( P. t. corbetti I); southern Indochinese tigers ( P. t. corbetti II), which are confined to the Malayan Peninsula; Sumatran tigers (P. t. sumatrae); Bengal tigers (P. t. tigris); and, if its uniqueness is affirmed by more extensive sampling, South China tiger (P. t. amoyensis) . These conclusions are based on significant genetic structure among tigers from these different geographic regions with the MHC, mtDNA, and microsatellite data, and extremely limited gene flow as shown by disjunct distributions of genetic variation (unique mtDNA haplotypes and microsatellite alleles) and the high mtDNA F st and microsatellite R st values. In addition, each subspecies has an allopatric geographical distribution (see Figure 1 ) and differential natural history ( Kitchener 1999 ; Seidensticker et al. 1999 ). The hypothesis that tiger population structure reflects recent (less than 10,000 y ago), human-induced population fragmentation and random lineage loss from a single panmictic population is not supported by the strong geographical partitioning of the mitochondrial lineages or by differences in measures of nucleotide diversity within each subspecies. Mismatch analysis ( Rogers and Harpending 1992 ) of pairwise differences among all tiger mtDNA haplotypes also revealed a multimodal distribution significantly different from a Poisson expectation, indicating the existence of several highly divergent populations (unpublished data). It is plausible that tiger populations (subspecies) differentiated through the combined effects of genetic drift in isolated populations and local adaptation to rapidly changing habitats across the tiger range during the Holocene ( Lister 2004 ). For example, Sumatran tigers currently occupy tropical moist forests, and Bengal tigers range from tropical dry forests, terai forests, and tall grasslands to the Himalayan foothills. However, we cannot rule out the possibility that some of the current population subdivision, particularly in the case of the divergence of P. t. altaica and P. t. amoyensis/P. t. corbetti I, could be related to the disruption of an isolation-by-distance pattern caused by the recent extinction of intermediate populations; this hypothesis can be tested only when a larger geographic sampling is available. The differences in molecular genetic patterns among the six hypothesized subspecies are dramatic ( Table 8 ). Further, the results lend support to the hypothesis that the Pleistocene centrum of tiger radiation is located within northern Indochina and southern China. Modern P. t. corbetti I has a large number of mtDNA diagnostic sites (three), the largest number of unique microsatellite alleles (19 out of 130), and the highest overall microsatellite diversity ( Tables 7 and 8 ). In addition, no microsatellite allele at any locus occurred with a frequency higher than 81%. The observed allele size distribution in P. t. corbetti I was generally continuous for most loci (there were fewer allele size gaps compared to other subspecies), evidence of a fairly stable demographic history, and alleles found in the other subspecies were almost always a subset of those found in P. t. corbetti I. Additional sampling of modern and/or historic samples could reveal additional structure (putative subspecies) in the P. t. corbetti I region (see Figure 1 ), as there were several microsatellite loci out of Hardy-Weinberg equilibrium, and the Bayesian population structure analysis identified possible substructure within P. t. corbetti I ( Figure S2 ). The ultimate classification of tigers of the southern China and northern Indochina region is further complicated by the poor definition of the geographic boundary between P. t. corbetti I and P. t. amoyensis, and because the South China tiger subspecies is represented only by captive-born animals of imprecise origin. One of the two phylogenetic lineages in this captive population (Pti-217, Pti-218, and Pti-222) was indistinguishable from northern Indochinese tigers (see Figures 3 and 4 ), perhaps as a consequence of introgression of the northern Indochinese tigers into the Chinese captive population or a more-northern distribution of the Indochinese tigers than had previously been recognized. A comprehensive morphological and genetic assessment of the captive population (around 50 individuals) ( Tilson et al. 2004 ), of historic samples, and of additional wild tigers from southern China, in the context of subspecies patterns seen here would be useful to resolve remaining uncertainties and to inform in situ and ex situ management strategies. By contrast, the other subspecies delineations are better defined. To the north, Amur tigers, presently an isolated population of fewer than 500 individuals, are confined almost entirely to the Russian Far East ( Matyushkin et al. 1999 ). They display low genetic diversity in comparison to other subspecies, with a single mtDNA haplotype that is likely derived from P. t. corbetti I Indochinese tigers ( Figure 3 A). The Amur tiger genetic variability may have been reduced during a post-ice age colonization of the region around 9,000 y ago and/or during the early 20th century when an estimated 20–30 tigers survived intense human persecution ( Kaplanov 1948 ). In Indochina, the genetic distinction between P. t. corbetti I and P. t. corbetti II (pairwise mtDNA F st = 0.797 and microsatellite R st = 0.225, p < 0.0001; P. t. corbetti II is characterized by three unique microsatellite alleles and five subspecies-specific mtDNA haplotypes [ Table 8 ]) supports the hypothesis that the Isthmus of Kra has been an ecological barrier restricting gene flow between tigers in Malaya Peninsula and mainland Southeast Asia. Previous biogeography studies have placed numerous species and subspecies boundaries of mammals ( Corbett and Hill 1993 ; Tosi et al. 2002 ), birds ( Hughes et al. 2003 ), and plants ( Woodruff 2003 ) near the Isthmus of Kra, making it a significant biogeographical transition between Indochina and Sundaic regions. The isolation of Sumatran tigers from mainland populations is supported by multiple unique characters, including two diagnostic mtDNA nucleotide sites, eight mtDNA haplotypes, and 11 (of 108) unique microsatellite alleles ( Table 8 ). Cracraft et al. (1998) and Hendrickson et al. (2000) also described genetic variation distinguishing Sumatran tigers from other tiger subspecies. The relatively high genetic variability and phylogenetic distinctiveness of Sumatran tigers suggest a historically large effective population size followed by highly restricted gene flow between the island and other populations. The Bengal tigers are defined by three distinct mitochondrial nucleotide sites and 12 unique microsatellite alleles. The pattern of genetic variation in the Bengal tiger corresponds to the premise that tigers arrived in India approximately 12,000 y ago ( Kitchener and Dugmore 2000 ). This recent history of tigers in the Indian subcontinent is consistent with the lack of tiger fossils from India prior to the late Pleistocene and the absence of tigers from Sri Lanka, which was separated from the subcontinent by rising sea levels in the early Holocene. Similar biogeographical boundaries to those separating the six tiger subspecies have been proposed in other species including leopard ( Uphyrkina et al. 2001 ), Asian elephant ( Fleischer et al. 2001 ), and rodents ( Gorog et al. 2004 ), but warrant further study to determine their importance as recent barriers to gene flow for large mammals in Asia. Our results have several implications for tiger conservation. Management strategies for the tiger, both in situ and ex situ, have been historically influenced by perceptions of its geographical variation and subspecific taxonomy ( Maguire and Lacy 1990 ; Seidensticker et al. 1999 ), and several captive tiger breeding programs have attempted to maintain purebred lines ( Foose 1987 ; Maguire and Lacy 1990 ). Our data suggest, however, that while supporting and refining most existing (and extant) tiger subspecies designations, there is additional substructure within some subspecies that should be considered when formulating management strategies for captive animals or when considering the maintenance of sufficiently large and interconnected wild populations. Specifically, the distinctiveness of tigers from Malayan Peninsula is comparable to differences among other recognized and separately managed subspecies. To be consistent, the Malayan subspecies should also be managed as such unless inbreeding depression has become an issue due to declined genetic variability. Since the current type specimen for P. t. corbetti is located in northern Vietnam ( Mazak 1968 ), and no prior name has been given to the southern populations, we propose the newly defined tiger subspecies from Malayan Peninsula be designated P. t. jacksoni, to honor the contributions of Peter Jackson, the former Chair of the the World Conservation Union (IUCN) Cat Specialist Group, who tirelessly labored for more than 40 y on behalf of tiger conservation. We designate the type specimen of the Malayan tigers to Pti-163 from the Zoo Melaka, Malaysia, and the taxonomic diagnosis will be described elsewhere. The present status of tigers from northern Indochina and from Malayan Peninsula is uncertain, urging more extensive study and conservation. Our results also show that, although modern tigers have a relatively young history, ecological, demographic, and biogeographic factors have led to recognizable subdivisions among otherwise closely related populations. We therefore might expect that more extensive geographic sampling would reveal additional phylogenetic divisions among populations, especially in the Indian Subcontinent and the Indochina bioregions, or alternatively, would blur the apparent phylogenetic subdivisions and reveal a clinal distribution of genetic variation across different subspecies. Further sampling of modern and historic specimens will also help clarify whether the patterns we have observed are attributable to the recent substantial population decline throughout the range in tigers, or whether the observed differentiations among tigers occurred earlier. Materials and Methods Samples. A total of 134 tiger individuals were sampled throughout the distribution range (see Figure 1 and Table 3 ). Of these, 100 were verified as either wild-born from a specific geographic locale or captive-born from geographically verified wild-born parents. An additional 34 individuals were of reasonably certain geographic origin and were used to complement estimated levels of molecular genetic variation in tigers. Individuals were labeled with traditional subspecies classifications based on their geographical origin following Mazak (1996) . Genomic DNA from blood or primary skin fibroblast cell culture was isolated using a standard proteinase K digestion and phenol-chloroform extraction procedure ( Sambrook et al. 1989 ). DNA was isolated from dry skin using guanidine thiocyanate ( Boom et al. 1990 ) and silica-based purification ( Hoss and Paabo 1993 ). DNA from hair was obtained by a modification of the previously described chelex method ( Higuchi et al. 1988 ). Analysis of historical samples was carried out with strict precautions at an isolated laboratory specializing in work with ancient DNA and was independently repeated to exclude possible contamination from any high-copy DNA source ( Hofreiter et al. 2001 ). Mitochondrial DNA analysis. Analyses of mtDNA in tigers and in other Panthera species is complicated by the presence of a large 12.8 kb nuclear mtDNA fragment that transposed to chromosome F2 in an ancestral Panthera species approximately 3 MYA ( Johnson et al. 1996 ; Lopez et al. 1996 ; Cracraft et al. 1998 ; J. H. Kim, A. Antunes, S.-J. Luo, J. Menninger, W. G. Nash, et al., personal communication). Although the Numt and Cymt DNA sequences have diverged, primers designed from conserved regions often coamplify both copies. Fifteen Cymt-specific primer sets (see Figure 2 and Table 1 ) were designed on the basis of sequence differences from the alignments of the complete tiger Numt and the homologous 12.8-kb Cymt sequences (J. H. Kim, A. Antunes, S.-J. Luo, J. Menninger, W. G. Nash, et al., personal communication). These Cymt primers amplified a total of 6,026 bp of sequence, spanning ten mitochondrial gene segments, including NADH dehydrogenase subunits 1, 2, 5, and 6 ( ND1 , ND2 , ND5, and ND6 ), cytochrome B (CytB), control region (CR), 12S rRNA (12S), cytochrome C oxidase subunits I and II (COI and COII), and ATPase8 (ATP8) (see Figure 2 ). The primer sets were tested in 15 individuals representing tigers from all five traditional subspecies. Five segments that revealed no variation in the pilot screening were excluded from further analysis (see Figure 2 ). PCR products were amplified from 50 ng of genomic DNA in a 25 μL reaction system containing 2.0 mM MgCl 2 , 1.0 mM dNTPs, 0.25 units of AmpliTaq Gold DNA polymerase (Applied Biosystems, Foster City, California, United States), and 1× PCR buffer II; the amplification protocol was: denaturation 10 min at 95 °C, a touch-down cycle of 95 °C for 30 s, 52 °C for 30 s decreased by 1 °C in the next cycle for 10 cycles, 72 °C for 45 s, then 35 amplification cycles of 95 °C for 30 s, 52 °C for 30 s, and 72 °C for 45 s, followed by an extension of 10 min at 72 °C. PCR products were purified using Microcon PCR filters (Millipore, Billerica, Massachusetts, United States) and were directly sequenced in both directions using BigDye Terminator kits (Applied Biosystems) and run on an ABI 377 sequencing apparatus. Sequences were inspected using SEQUENCHER (Gene Codes, Ann Arbor, Michigan, United States), unambiguously aligned using Clustal-X ( Thompson et al. 1997 ), and visually inspected. Sequences for each mtDNA fragment were combined for a total evidence approach. Sequences were deposited in GenBank. Phylogenetic relationships among mtDNA haplotypes were assessed using three approaches implemented in PAUP ( Swofford 2001 ). An MP analysis was conducted using a heuristic search, with random additions of taxa and tree-bisection-reconnection branch swapping. The ME heuristic search approach consisted of NJ trees constructed from Kimura two-parameter distances followed by a branch-swapping procedure. ML analysis was done using the TrN (Tamura-Nei) +I (with proportion of invariable sites) model with the proportion of invariable sites set to 0.93, and the rate among sites equal, as estimated using MODELTEST 3.06 ( Posada and Crandall 1998 ). The reliability of the nodes in each of the analyses was assessed by 100 bootstrap iterations. A statistical parsimony network was constructed using TCS 1.13 ( Clement et al. 2000 ) to infer phylogeographic and potential ancestor-descendent relationships among haplotypes. Measures of population genetic variation, such as mean number of pairwise differences, gene diversity, and nucleotide diversity were estimated using ARLEQUIN 2.0 ( Schneider et al. 2000 ). The extent of geographic subdivision among populations was assessed by F st values (with Kimura two-parameter distance) using AMOVA as implemented in ARLEQUIN 2.0. Statistical significance was tested using 10,000 permutations. The approximate coalescence time of tigers was estimated with a linearized tree method as implemented in the program LINTRE ( Takezaki et al. 1995 ). This program constructs linearized NJ trees reestimating branch lengths under the molecular clock assumption and incorporates two tests for the assumption ( Takezaki et al. 1995 ). The mtDNA sequence divergence was based on the standard equation H = 2μ T , where H was the branch height in the linearized tree correlated to the average pairwise distance among haplotypes, μ the substitution rate, and T the divergence time. Since there was no comparable sequence from other Panthera species available, we generated from PCR and GenBank a chimera-homologous sequence of 3.2 kb (including fragments from mitochondrial genes ND1, ND2, ND5, ND6, CytB, 12S, and COI ) from three leopard individuals. The divergence time between leopard and tiger was used as a calibration point, and two fossil dates were chosen. Two million y was a commonly used lower bound for the Panthera lineage radiation and the date for the earliest reported tiger in the fossil record ( Hemmer 1987 ; O'Brien et al. 1987 ; Wayne et al. 1991 ). An earlier record of 3 million y was also chosen because leopard fossils have been reported from this time period ( Turner and Anton 1997 ). Domestic cat (Felis catus) was used as an outgroup. Coalescent-based simulation of the population parameter θ estimation and its 95% CI was conducted in DnaSP 4.0 ( Rozas et al. 2003 ) with 1,000 replicates, given that the mutations along the lineages followed a Poisson distribution. Microsatellite analysis Thirty polymorphic microsatellite loci (FCA005, FCA008, FCA032, FCA043, FCA044, FCA069, FCA077, FCA090, FCA091, FCA094, FCA096, FCA105, FCA123, FCA126, FCA129, FCA139, FCA161, FCA176, FCA201, FCA211, FCA212, FCA220, FCA229, FCA242, FCA290, FCA293, FCA304, FCA310, FCA391, and FCA441) originally designed in the domestic cat (F. catus) ( Menotti-Raymond et al. 1999 ) were amplified by PCR using fluorescently labeled primers under previously published conditions ( Menotti-Raymond et al. 1999 ). Two loci (FCA391 and FCA441) were tetranucleotide repeats, and the others were dinucleotides. All loci have been mapped in the domestic cat and located on 11 of the 19 chromosomes ( Menotti-Raymond et al. 1999 ; Menotti-Raymond et al. 2003 ). These microsatellites were in different linkage groups or at least 12 centimorgans apart in the domestic cat, except for FCA 211 and FCA 212 (4 centimorgans), and were likely to be in linkage equilibrium. The dye-labeled PCR products of the 30 microsatellite primer sets were pooled and diluted based on size range and fluorescent dye so that 2–4 loci could be multiplexed and subsequently analyzed by electrophoresis in an ABI 377 automated sequencer (Applied Biosystems). Patterns were scored and analyzed using GENESCAN 2.1 and GENOTYPER 2.5 software. Of 134 tiger samples, 113 were included in the microsatellite analysis. DNA samples from pelt or hair for which fewer than 20 of the loci amplified successfully were excluded from the analysis. Tests for genotypic linkage disequilibrium and deviations from Hardy-Weinberg equilibrium for each locus in each population were performed using GENEPOP web version of 3.1c ( http://wbiomed.curtin.edu.au/genepop/ ) ( Raymond and Rosset 1995 ). Measures of microsatellite genetic variation in terms of average observed heterozygosity and expected heterozygosity, average number of alleles per locus, average allele size range per locus, number of unique alleles, and average variance were estimated with MICROSAT ( Minch et al. 1995 ). Pairwise genetic distances among individual tigers were estimated based on Dps and Dkf with the [1 – ps/kf] option in MICROSAT and were used to construct NJ phylogenetic trees with the program NEIGHBOR in the PHYLIP 3.5 package ( Felsenstein 1989 ). Assessments of different geographic subdivision scenarios and population pairwise comparisons (using R st , sum of square size differences) were derived from ARLEQUIN 2.0. The statistical significance of R st values, sum of squared size differences, was tested with 10,000 permutations as implemented in ARLEQUIN. A Bayesian clustering method implemented in the program STRUCTURE ( Pritchard et al. 2000 ) was used to infer population structure based upon multilocus microsatellite genotype and sequence data. MtDNA was treated as a single haploid locus, and each observed haplotype was coded with a unique integer (e.g., 1, 2) for the first allele and the missing data symbol (e.g., -9) for the second. We calculated the probability of individual assignments to population clusters (K) without prior information of the origin of individuals. A series of tests was conducted using different numbers of population clusters to guide an empirical estimate of the number of identifiable populations, assuming an admixture model with correlated allele frequencies and with burn-in and replication values set at 50,000 and 10 6 , respectively. Each test yielded a log likelihood value of the data (Ln probability), the highest of which would indicate which test was closest to the actual number of genetically distinct populations. These tests also provided an alpha value, the measure of admixed individuals in the data set. Class II MHC. Allele variation in the nuclear MHC class II DRB gene was assessed in five Amur, two northern Indochinese, two Malayan, 11 Sumatran, and one Bengal tiger. Conserved PCR primers designed from the human DRB sequence were used to amplify homologous DRB sequences (of around 238 bp) in 21 tiger voucher DNAs with primers 61a (5′- CCGCTGCACTGTGAAGCT-3′) and 219a (5′- CCACACAGCACGTTTCTT-3′). Products were screened for polymorphisms using SSCP, a method that detects single-basepair substitutions in 100–300-bp DNA fragments. For SSCP, PCR products were mixed with a solution of 120 μl of formamide, 20 μl of TAMRA (Applied Biosystems) lane standard, and 20 μl of a blue dextran loading dye (from a stock solution of 50 mg/ml with 25 mM EDTA), then denatured at 95 °C for 3 min. The electrophoresis ran at 2,000 volts, 400 amps, and 25 watts in 1× TBE buffer through a 6% denaturing polyacrylamide gel (19.5:1 acrylamide:bis). The SSCP fragments were visualized by autoradiography, and alleles were scored by eye ( Yuhki and O'Brien 1997 ). Supporting Information Figure S1 Phylogenetic Relationships among the Individual Tigers from Composite Microsatellite Genotypes of 30 Loci Branches of the same color represent tiger individuals of the same classically named subspecies. NJ tree constructed based on kinship coefficient (Dkf) with the (1 – kf) option in MICROSAT ( Minch et al. 1995 ). Numbers are individual Pti codes ( Table 3 ). Bootstrap values over 50% are shown on divergence nodes. (108 KB DOC). Click here for additional data file. Figure S2 Bayesian Population Structure Analysis of 111 Tigers Data obtained from microsatellite genotype and mitochondrial haplotype data were analyzed using STRUCTURE ( Pritchard et al. 2000 ). Simulations were set at 50,000 burn-in period followed by 10 6 replicates. Each individual is represented by a thin vertical bar, which is partitioned into K colored segments that represent the individual affiliation to each of K clusters. Here shows the population structure when K = 7, which produced the highest probability among other choices of K . Three STRUCTURE runs produced almost identical individual affiliation. (62 KB DOC). Click here for additional data file. Table S1 Bayesian Clustering Analyses for Tiger Microsatellite and Mitochondrial Data (59 KB DOC). Click here for additional data file. Accession Numbers The GenBank ( http://www.ncbi.nlm.nih.gov/ ) accession numbers of the mtDNA fragments discussed in this paper are AY736559–AY736808.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534810.xml

350667

Genetic and Functional Diversification of Small RNA Pathways in Plants

Multicellular eukaryotes produce small RNA molecules (approximately 21–24 nucleotides) of two general types, microRNA (miRNA) and short interfering RNA (siRNA). They collectively function as sequence-specific guides to silence or regulate genes, transposons, and viruses and to modify chromatin and genome structure. Formation or activity of small RNAs requires factors belonging to gene families that encode DICER (or DICER-LIKE [DCL]) and ARGONAUTE proteins and, in the case of some siRNAs, RNA-dependent RNA polymerase (RDR) proteins. Unlike many animals, plants encode multiple DCL and RDR proteins. Using a series of insertion mutants of Arabidopsis thaliana , unique functions for three DCL proteins in miRNA (DCL1), endogenous siRNA (DCL3), and viral siRNA (DCL2) biogenesis were identified. One RDR protein (RDR2) was required for all endogenous siRNAs analyzed. The loss of endogenous siRNA in dcl3 and rdr2 mutants was associated with loss of heterochromatic marks and increased transcript accumulation at some loci. Defects in siRNA-generation activity in response to turnip crinkle virus in dcl2 mutant plants correlated with increased virus susceptibility. We conclude that proliferation and diversification of DCL and RDR genes during evolution of plants contributed to specialization of small RNA-directed pathways for development, chromatin structure, and defense.

Introduction Eukaryotic small RNAs of approximately 21–24 nucleotides function as guide molecules in a remarkably wide range of biological processes, including developmental timing and patterning, formation of heterochromatin, genome rearrangement, and antiviral defense ( Carrington and Ambros 2003 ; Finnegan and Matzke 2003 ; Lai 2003 ). They belong to at least two general classes, microRNA (miRNA) and short interfering RNA (siRNA). miRNAs (approximately 21–22 nucleotides) are found in plants and animals and are often phylogenically conserved within their respective kingdoms. They arise from non-protein-coding genes through formation of a precursor transcript followed by one or more nucleolytic processing steps ( Lai 2003 ). Part of the precursor adopts a fold-back structure that interacts with a multidomain RNaseIII-like enzyme termed DICER or DICER-LIKE (DCL1 in Arabidopsis ), which catalyzes accurate excision of the mature miRNA ( Denli and Hannon 2003 ). The miRNAs then associate with ribonucleoprotein complexes that function to negatively regulate target genes controlling a range of developmental events, such as timing of cell fate decisions, stem cell maintenance, apoptosis, organ morphogenesis and identity, and polarity ( Ambros 2003 ; Carrington and Ambros 2003 ). siRNAs are chemically similar to miRNAs, although in plants they typically range in size between 21 and 24 nucleotides ( Hamilton et al. 2002 ; Llave et al. 2002 a; Tang et al. 2003 ). They are associated with both post-transcriptional forms of RNA interference (RNAi) and transcriptional silencing involving chromatin modification ( Finnegan and Matzke 2003 ). siRNAs are processed from precursors containing extensive or exclusive double-stranded RNA (dsRNA) structure, such as transcripts containing inverted repeats or intermediates formed during RNA virus replication ( Hannon 2002 ). siRNA precursors can also be formed by the activity of one or more cellular RNA-dependent RNA polymerases (RdRp), as was shown genetically in several screens for RNA silencing-defective mutants ( Cogoni and Macino 1999 ; Dalmay et al. 2000 ; Mourrain et al. 2000 ; Smardon et al. 2000 ; Volpe et al. 2002 ). Arabidopsis plants contain at least three active RdRp genes, termed RDR1 , RDR2 , and RDR6 (also known as SDE1/SGS2 ) ( Dalmay et al. 2000 ; Mourrain et al. 2000 ; Yu et al. 2003 ). RDR6 is necessary for sense transgene-mediated RNAi, but not for silencing of constructs that encode transcripts with hairpins containing extensive dsRNA structure ( Dalmay et al. 2000 ; Mourrain et al. 2000 ; Beclin et al. 2002 ). In many animals, both miRNAs and siRNAs are formed by the activity of the same DICER enzyme ( Grishok et al. 2001 ; Hutvágner et al. 2001 ; Ketting et al. 2001 ; Knight and Bass 2001 ; Provost et al. 2002 ; Zhang et al. 2002 ; Myers et al. 2003 ), although in plants they are formed by distinct DCL activities ( Finnegan et al. 2003 ). Arabidopsis contains four DCL genes ( DCL1 to DCL4 ), only one of which ( DCL1 ) has been assigned a definitive function in small RNA biogenesis ( Park et al. 2002 ; Reinhart et al. 2002 ; Schauer et al. 2002 ). Biochemical data indicate, however, that multiple DCL activities or pathways catalyze formation of siRNAs of small-sized (approximately 21 nucleotides) and large-sized (approximately 24 nucleotides) classes ( Tang et al. 2003 ). Endogenous siRNAs in plants arise from many types of retroelements and transposons, other highly repeated sequences, pseudogenes, intergenic regions (IGRs), and a few expressed genes ( Hamilton et al. 2002 ; Llave et al. 2002 a; Mette et al. 2002 ). Exogenous siRNAs can arise from both sense and hairpin transcript-forming transgenes and by viruses ( Hamilton and Baulcombe 1999 ; Mette et al. 2000 ). Both siRNAs and miRNAs function post-transcriptionally to suppress or inactivate target RNAs. siRNAs guide sequence-specific nucleolytic activity of the RNA-induced silencing complex (RISC) to complementary target sequences ( Hannon 2002 ). Among other proteins, RISCs contain ARGONAUTE (AGO) family members that likely bind siRNAs or target sequences ( Carmell et al. 2002 ). In plants and insects, post-transcriptional RNAi serves as an adaptive antiviral defense response ( Waterhouse et al. 2001 ; Li et al. 2002 ). miRNAs are fully competent to guide nucleolytic function of RISC, provided that a target sequence with sufficient complementarity is available ( Hutvágner and Zamore 2002 ; Doench et al. 2003 ; Tang et al. 2003 ). Many plant miRNAs function as negative regulators through this cleavage-type mechanism ( Llave et al. 2002 b; Rhoades et al. 2002 ; Emery et al. 2003 ; Kasschau et al. 2003 ; Palatnik et al. 2003 ; Tang et al. 2003 ; Xie et al. 2003 ). In animals, the level of complementarity between target and miRNA sequences is generally low, which inhibits nucleolytic activity. Animal miRNAs suppress translation of target mRNAs ( Olsen and Ambros 1999 ; Reinhart et al. 2000 ). Some plant miRNAs may also function as translational suppressors ( Aukerman and Sakai 2003 ; Chen 2003 ). siRNAs also guide chromatin-based events that result in transcriptional silencing. Two lines of evidence support this view. First, in Schizosaccharomyces pombe and Arabidopsis , endogenous siRNAs from repeated sequences corresponding to centromeres, transposons, and retroelements are relatively abundant ( Llave et al. 2002 a; Mette et al. 2002 ; Reinhart and Bartel 2002 ). RNAi-related factors (DICER, RdRp, and AGO proteins) are required to maintain S. pombe centromeric repeats and nearby sequences in a transcriptionally inactive, heterochromatic state ( Hall et al. 2002 ; Volpe et al. 2002 ). Mutants that lose RNAi component activities lose heterochromatic marks, such as histone H3 methylation at the K9 position (H3K9), as well as centromere function ( Hall et al. 2002 ; Volpe et al. 2002 , 2003 ). In plants, AGO4 is necessary to maintain transcriptionally silent epialleles of SUPERMAN . The ago4 mutants lose both cytosine methylation, particularly at non-CpG positions, and H3K9 methylation at SUPERMAN and other constitutive heterochromatic sites (the Arabidopsis thaliana short interspersed element 1 [ AtSN1 ] locus) ( Zilberman et al. 2003 ). And, second, heterochromatin formation of nuclear DNA can be triggered, in a sequence-specific manner, by post-transcriptional silencing of cytoplasmic RNAs ( Jones et al. 1999 ; Aufsatz et al. 2002 ; Schramke and Allshire 2003 ). The RNA-directed DNA methylation (RdDM) signal transmitted from the cytoplasm to the nucleus is most likely siRNA. The prevailing view states that chromatin-based silencing guided by siRNAs serves, among other purposes, as a genome defense system to suppress mobile genetic elements or invasive DNA ( Dawe 2003 ; Schramke and Allshire 2003 ). Using a genetic approach, we show here the existence of three small RNA-generating pathways with unique requirements in Arabidopsis. Plants with point mutations or insertions in several members of the DCL and RDR gene families were examined. The data indicate that plants genetically diversified several factors involved in formation of functionally distinct small RNAs. Results Genetic Requirements for miRNA Formation At least two factors, DCL1 and HEN1 (HUA ENHANCER1), are involved in Arabidopsis miRNA formation. As shown for miR-171, miR-159 ( Figure 1 A), and several other miRNAs ( Park et al. 2002 ; Reinhart et al. 2002 ), mutants with dcl1 loss-of-function alleles lose most of their miRNA populations ( Figure 1 B). Plants with mutant hen1 alleles either lose miRNAs or the apparent size of miRNAs is increased by one or more nucleotides ( Park et al. 2002 ; Boutet et al. 2003 ) ( Figure 1 B). miRNA function to suppress target mRNAs is diminished in both dcl1 and hen1 mutants ( Boutet et al. 2003 ; Kasschau et al. 2003 ; Xie et al. 2003 ). To determine whether other DCL or RDR proteins are required for miRNA formation in Arabidopsis , miR-171 and miR-159 were analyzed in four new mutants. The dcl2-1 and dcl3-1 mutants contained T-DNA insertions in DCL2 (At3g03300) and DCL3 (At3g43920) genes, respectively ( Figure S1 ). In wild-type plants, DCL2 and DCL3 transcripts accumulated to detectable levels in inflorescence tissues, but not in leaves. The mutant dcl2-1 and dcl3-1 transcripts were not detected in either tissue type ( Figure S1 ). The rdr1-1 and rdr2-1 mutants contained T-DNA insertions in RDR1 (At1g14790) and RDR2 (At4g11130), respectively ( Figure S1 ). RDR1 and RDR2 transcripts accumulated in inflorescence tissue, but not leaves, of untreated wild-type plants ( Figure S1 ). The RDR1 transcript levels were elevated in salicylic acid (SA)-treated leaves, as shown previously ( Yu et al. 2003 ), but RDR2 transcript levels were not affected by SA ( Figure S1 ). Both rdr1-1 and rdr2-1 transcripts were below the detection limit in the corresponding mutant plants. In addition, a mutant containing an insertion in the RDR6 gene (also known as SDE1/SGS2 ; At3g49500) was analyzed in parallel with the rdr1 and rdr2 mutants. This rdr6-1 mutant displayed a weak virus-susceptibility phenotype that was consistent with previously reported sde1 and sgs2 mutants ( Mourrain et al. 2000 ; Dalmay et al. 2001 ). However, no differences in RDR6 transcript levels were detected between wild-type and rdr6-1 mutant plants (data not shown). Figure 1 Genetic Requirements for miRNA and Endogenous siRNA Generation (A) miRNA genes and selected loci corresponding to three siRNAs or siRNA populations. Cloned small RNA sequenc-es are shown as green (sense orientation relative to the genome) or red (antisense orientation) bars. Protein-coding and miRNA genes are indicated by blue arrowheads. From top to bottom: miR-171 and miR-159a loci; siRNA02 loci, with each siRNA02 sequence indicated by an asterisk and the inverted duplication shown by the gray arrows; cluster2 siRNA locus; a segment of chromosome III showing 10 5S rDNA repeats (blue indicates 5S rRNA, gray indicates spacer) containing the siRNA1003 sequence. (B) Small RNA blot assays for miR-171, miR-159, and endogenous siRNAs. Ethid-ium bromide-stained gels (prior to transfer) in the zone corresponding to tRNA and 5S RNA are shown at the bottom. Each mutant is presented in a panel with the corresponding wild-type control (Col-0 or La- er ). Accumulation of miR-171 and miR-159 was unaffected in the dcl2 and dcl3 mutants (see Figure 1 B). This was in contrast to the low level or shifted mobility of miR-171 and miR-159 in dcl1-7 and hen1-1 , respectively (see Figure 1 B). Similarly, accumulation of miR-171 and miR-159 was unaffected in rdr1 and rdr2 mutants. Composition of Endogenous siRNA Populations A library of cloned small RNAs from inflorescence tissues of Col-0 ecotype plants was partially sequenced and analyzed. Initial characterization of 125 of these sequences revealed that most of the clones corresponded to siRNA-like sequences ( Llave et al. 2002 a). A total of 1,368 distinct small RNAs, ranging in size between 20 and 26 nucleotides, were provisionally categorized here as siRNAs, with 24 nucleotides representing the most common size ( Figure 2 A; all sequences are available to view or download at http://cgrb.orst.edu/smallRNA/db/ ). The siRNA sequences were identified at 5,299 genomic loci ( Table S1 ). Approximately 27% of endogenous siRNAs derived from transposon or retroelement sequences in the sense or antisense polarity ( Figure 2 B). Centromeric and pericentromeric siRNAs were common, which was partly due to the prevalence of transposons and retroelements at these sites. Forty-five small RNAs of sense and antisense polarity arose from highly repeated 5S, 18S, and 25S rDNA. While it is likely that some rDNA-derived sequences resulted from nonspecific breakdown of highly abundant rRNAs, some had specific genetic requirements and properties that were consistent with functional siRNAs (see below). Thirty-one siRNAs came from sequences annotated as psuedogenes and 147 from hypothetical or predicted genes ( Figure 2 B). Only 28 were identified as originating from genes that are known to be expressed ( Figure 2 B). The remaining 816 sequences mapped to loci that were collectively labeled as an IGR sequence. The IGR-derived siRNAs arose from unique sequences adjacent to known genes, inverted duplications, satellites, and other repeated sequences, although many of these may actually correspond to transposon or retroelement sequences that were not recognized by the search programs. Figure 2 Endogenous siRNAs in Arabidopsis (A) Size distribution of endogenous siRNAs. (B) Distribution of distinct siRNAs in different sequence categories. (C) Density of siRNAs from highly repeated (mainly transposons and retroelements; the asterisk shows repeat sequences identified using RepeatMasker), 5S rDNA, and unique genomic sequence. The frequency of unique siRNAs arising from highly repeated sequences (mainly transposons and retroelements), 5S rDNA repeats, and nonrepetitive sequence was calculated ( Figure 2 C). siRNAs in the library occurred at a frequency of 2.42 per 100 kb repetitive DNA, which was approximately 2.4-fold higher that the frequency of siRNAs from nonrepetitive sequence (1.02 per 100 kb). Based on the number of repeats in the most current version of the Arabidopsis genome sequence, unique siRNAs corresponding to 5S rDNA were identified at a frequency of 7.55 per 100 kb. These data indicate that siRNAs arise more frequently from highly repeat genome sequences. Genetic Requirements for Endogenous siRNA Formation A set of four siRNAs or siRNA populations, representing the major categories identified in the library, were selected for genetic analysis. Twenty-six siRNAs corresponded to SINE retroelements, one of which ( AtSN1 ) was selected for detailed analysis. AtSN1 -derived siRNA formation requires AGO4 ( Zilberman et al. 2003 ) and SDE4 ( Hamilton et al. 2002 ). One siRNA (siRNA1003) originating from 5S rDNA was selected. The 5S rRNA genes occur in tandem arrays in chromosomes III, IV, and V, with the typical repeat unit (approximately 500 nucleotides) being composed of transcribed sequence (120 nucleotides) and flanking spacer sequences ( Cloix et al. 2002 ; Mathieu et al. 2003 ). The siRNA1003 sequence was identified in the sense orientation within the spacer sequence in 202 repeats in chromosome III and four repeats in chromosome V (see Figure 1 A). The cluster2 siRNA population from a 125-nucleotide IGR segment in chromosome I was represented by seven unique siRNAs in the library (see Figure 1 A). Finally, the siRNA02 sequence corresponded to two loci separated by approximately 2.1 kb in chromosome V. One locus occurred in an IGR sequence, and the other within a hypothetical gene (At5g56070) of unknown function. The two siRNA02 loci occur in sequences that correspond to arms of an inverted duplication (see Figure 1 A) ( Llave et al. 2002 a). The AtSN1 , cluster2, and siRNA02 probes detected populations that accumulated as 24-nucleotide RNAs, while the siRNA1003 probe detected a population containing 21- to 24-nucleotide species (see Figure 1 B). The abundance of each siRNA population was decreased in the dcl3-1 mutant, but not in the dcl1-7 or dcl2-1 mutants (see Figure 1 B). This was in strict contrast to miR-171, miR-159 (see Figure 1 B), and several other miRNAs tested (data not shown), which depended specifically on DCL1 . Interestingly, weak signals corresponding to siRNA02, AtSN1 siRNAs, and cluster2 siRNAs were detected in faster-migrating positions in the dcl3-1 mutant (see Figure 1 B). This may have resulted from exposure of siRNA precursors to alternate DCL activities in the absence of DCL3. Notably, both small and large siRNAs detected by the 5S rDNA-derived siRNA1003 probe were diminished in dcl3-1 plants. Each siRNA population was eliminated in the rdr2-1 mutant, but not in the rdr1-1 mutant (see Figure 1 B). In preliminary experiments, each siRNA population was unaffected by the rdr6-1 mutation, although these data should be interpreted cautiously because of the possibility that the rdr6-1 allele is weak (data not shown). The endogenous siRNA requirement for RDR2 contrasted with the miRNAs, which exhibited complete insensitivity to each of the rdr mutations tested (see Figure 1 B). These data genetically identify DCL3 and RDR2 as components of an endogenous siRNA generating system that differs functionally from the miRNA-generating apparatus. The HEN1 protein was implicated in post-transcriptional silencing of sense-, but not hairpin-forming, transgenes ( Boutet et al. 2003 ). We tested the requirement of HEN1 for endogenous siRNA formation using the hen1-1 mutant. Two of the siRNA populations, siRNA1003 and the AtSN1 -siRNAs, were reduced to undetectable levels in hen1-1 plants (see Figure 1 B). The siRNA02 and cluster2 siRNAs, on the other hand, reproducibly accumulated to higher levels in hen1-1 plants compared to wild-type La- er plants. Thus, each type of endogenous siRNA tested requires DCL3 and RDR2, but only the highly repeated 5S rDNA and retroelement-derived siRNAs require HEN1. In fact, the requirement for, or independence from, HEN1 was precisely the same as AGO4 at each of these loci (D. Zilberman and S. Jacobsen, unpublished data). Function of the Endogenous siRNA-Generating System Two previous studies showed that SDE4 and AGO4 are required for AtSN1 siRNA accumulation and methylation of cytosine positions at the AtSN1 locus ( Hamilton et al. 2002 ; Zilberman et al. 2003 ). In an ago4 mutant, loss of AtSN1 siRNA is associated with decreased histone H3K9 methylation ( Zilberman et al. 2003 ). Cytosine methylation and increased histone H3K9 methylation are hallmarks of transcriptionally silent and heterochromatic DNA in plants and other organisms, and siRNAs may recruit chromatin modification complexes to specific loci ( Grewal and Moazed 2003 ). To determine whether DCL3 and RDR2 catalyze formation of siRNAs that functionally interact with chromatin, cytosine methylation at AtSN1 and 5S rDNA loci and methylation of H3K9 and H3K4 positions in AtSN1 were examined in wild-type, dcl3-1 , and rdr2-1 plants. We also analyzed AtSN1 -derived transcript levels to determine whether the mutations affected expression of the locus. Consistent with previous reports ( Hamilton et al. 2002 ; Zilberman et al. 2003 ), bisulfite sequencing of AtSN1 genomic DNA revealed extensive CpG (72.0%), CpNpG (43.1%), and asymmetric CpHpH (16.3%) methylation in Col-0 wild-type plants ( Figure 3 A; Table S2 ). In the rdr2-1 mutant, CpNpG and CpHpH methylation was reduced to 24.6% and 4.5%, respectively. Only a slight reduction in CpG methylation was detected in rdr2-1 plants ( Figure 3 A). This methylation pattern was similar to that detected in mutants lacking CHROMOMETHYLASE3 ( cmt3-7 ; Figure 3 A), which is necessary for efficient methylation of AtSN1 at non-CpG sites, and in a mutant lacking AGO4 ( Zilberman et al. 2003 ). In the dcl3-1 mutant, however, cytosine methylation was decreased only at asymmetric sites, while CpG and CpNpG methylation was similar to that of wild-type plants ( Figure 3 A). Figure 3 Effects of Mutations on AtSN1 and 5S rDNA Chromatin Structure and Gene Expression (A) Analysis of CpG (left), CpNpG (center), and CpHpH (right) methylation in AtSN1 by bisulfite sequencing of genomic DNA. (B) Blot analysis of 5S rDNA digested with methylation-sensitive restriction enzymes HpaII (left) and MspI (right). HpaII is sensitive to CpG and CpNpG methylation, whereas MspI is sensitive to only CpNpG methylation. Methylation is indicated by the ascending ladder, which corresponds to 5S rDNA multimers (monomer = approximately 0.5 kb). Duplicate samples from each plant were analyzed. (C) ChIP assays using antibodies against dimethyl-histone H3K9 and dimethyl-histone H3K4. Genomic DNA associated with immunoprecipitated chromatin was analyzed by semiquantitative PCR with primer pairs specific for AtSN1 , retrotransposon reverse transcriptase (At4g03800) (internal control for H3K9 methylation), and PFK (At4g04040) (internal control for H3K4 methylation). The PCR products were quantitated and compared against the respective internal controls, and the relative H3K4 and H3K9 methylation levels were expressed relative to that in Col-0 (arbitrarily set to 1.00). (D) Detection of AtSN1 -specific transcripts by semiquantitative RT-PCR. Primers specific for PFK transcripts were used as the internal control. A parallel set of reactions without addition of reverse transcriptase (RT) was run as a quality control for genomic DNA contamination. The PCR products were normalized relative to PFK, and the expression levels were calculated relative to that in Col-0 (arbitrarily set to 1.00). Because of the number of 5S rDNA repeats, analysis of cytosine methylation was done using restriction enzymes HpaII or MspI and DNA blot assays. Sensitivity to HpaII indicates lack of methylation at CpG or CpNpG sites (or both), whereas sensitivity to MspI indicates lack of methylation at only CpNpG sites. In wild-type Col-0 and La- er plants, 5S rDNA loci were heavily methylated at CpG+CpNpG sites, as shown by detection of only high molecular weight forms using HpaII, and partially methylated at CpNpG as shown using MspI ( Figure 3 B). In rdr2-1 plants, methylation was partially lost at CpNpG sites (increased MspI sensitivity; Figure 3 B, lanes 15–16), although to a lesser degree than in cmt3-7 plants ( Figure 3 B, lanes 21–22). Methylation detected by HpaII sensitivity was partially lost in the rdr2-1 mutant ( Figure 3 B, lanes 3–4), which was most likely due to loss of CpG methylation. Loss of only CpNpG methylation in rdr2-1 plants would not account for the increased sensitivity to HpaII, as HpaII sensitivity in cmt3-7 plants (lacking nearly all CpNpG methylation) was unaffected ( Figure 3 B, lanes 9–10). Sensitivity of 5S rDNA sites to HpaII and MspI in dcl3-1 plants was only slightly increased ( Figure 3 B, lanes 5–6 and 17–18). In the ago4-1 mutant, CpG methylation was partially lost as revealed by increased sensitivity to HpaII ( Figure 3 B, lanes 11–12). Chromatin immunoprecipitation (ChIP) assays were used to detect changes in H3K4 and H3K9 methylation at AtSN1 in rdr2-1 and dcl3-1 mutant lines. Loci containing genes encoding a retrotransposon reverse transcriptase and phosphofructokinase β subunit (PFK) were used as positive controls for sequences associated primarily with K9- and K4-methylated histone H3, respectively ( Gendrel et al. 2002 ). At AtSN1 , decreased levels of histone H3K9 methylation were detected in both rdr2-1 and dcl3-1 mutants (see Figure 3 C). This was accompanied by a slight increase in H3K4 methylation (see Figure 3 C). The extent to which H3 methylation changed was greater in rdr2-1 relative to dcl3-1 plants. Little or no change in H3K4 and H3K9 methylation was detected at the control loci. In addition, no changes in H3K4 or H3K9 methylation were detected at AtSN1 in cmt3-7 plants (data not shown). The changes in H3 methylation shown here are similar to those at several heterochromatic or silenced loci in ago4 mutant plants ( Zilberman et al. 2003 ). The level of AtSN1 -derived transcripts was measured in rdr2-1 and dcl3-1 mutant plants and compared against the level of PFK transcript using semiquantitative RT-PCR. As shown in Figure 3 D, relatively low levels of AtSN1 transcripts were detected in wild-type Col-0 plants. However, the normalized level of AtSN1 transcripts was over 8- and 3-fold higher in rdr2-1 and dcl3-1 mutant plants, respectively, compared to wild-type plants. Therefore, loss of siRNA-forming capability correlated with loss of heterochromatic marks and elevated transcript levels at an endogenous locus that is normally silenced at the chromatin level. Given that RDR2, DCL3, and AGO4 are involved in chromatin-associated events and that HEN1 is required for accumulation of certain endogenous siRNAs associated with chromatin modification, it was hypothesized that each of these proteins accumulates in the nucleus. The presence of nuclear transport signals in each protein was tested by transient expression and analysis of green fluorescent protein (GFP) fusions in a heterologous plant, Nicotiana benthamiana , using an Agrobacterium infiltration assay. Subcellular accumulation sites for these proteins were compared to those of β-glucurodinase (GUS)–GFP (cytosolic control) and nuclear inclusion a protein (NIa)–GFP (nuclear control). The DCL3–GFP, HEN1–GFP, and GFP–AGO4 fusion proteins were detected exclusively in the nucleus ( Figure 4 ; Figure S2 ), indicating that DCL3, HEN1, and AGO4 possess independent nuclear transport capability. Subcellular localization experiments with RDR2–GFP and GFP–RDR2 fusion proteins, however, were inconclusive due to low expression levels and protein instability (data not shown). Figure 4 Subcellular Localization of GFP Fusion Proteins Pairwise presentation of confocal microscopic images showing GFP fluorescence (top) and DAPI fluorescence (bottom) in N. benthamiana expressing the indicated GFP fusion proteins. Arrowheads indicate the location of nuclei. Note that the GUS–GFP control protein accumulates in cytoplasm at the cell periphery and immediately surrounding nuclei, while the NIa–GFP control protein accumulates in nuclei. Scale bar = 25μm. Genetic Requirements for Virus-Derived siRNA Formation The involvement of DCL1, DCL2, and DCL3 in siRNA formation in response to infection by three dissimilar RNA viruses was tested using the dcl mutant series. Two of the viruses, a GFP-tagged version of turnip mosaic virus (TuMV–GFP) and turnip crinkle virus (TCV), infect Arabidopsis systemically and cause moderate to severe disease symptoms. The third virus, cucumber mosaic virus strain Y (CMV-Y), infects plants systemically, but causes only mild symptoms. Wild-type (Col-0 and La- er ) and mutant plants were inoculated on rosette leaves, and upper, noninoculated tissue (cauline leaves and inflorescences) was analyzed for virus-specific siRNAs at 7 and 14 d post-inoculation (dpi). Viral siRNAs were detected in systemic tissues from wild-type plants at both timepoints ( Figure 5 A– 5 C, lanes 3, 5, 10, and 13), with siRNA levels generally higher at 14 dpi. In TuMV- and CMV-infected dcl1-7 , dcl2-1 , and dcl3-1 mutant plants, siRNAs accumulated to levels that were similar to those in infected wild-type plants at 7 and 14 dpi (Figures 5 A and 5 B). TuMV and CMV titers and symptom phenotypes in the three mutants were indistinguishable from those in their respective parents (data not shown). Similarly, in TCV-infected dcl1-7 and dcl3-1 plants, viral siRNA levels, virus titer, and symptom severity were essentially the same as in wild-type plants ( Figure 5 C; Figure 6 A and 6 B; data not shown). Figure 5 Genetic Requirements for DCLs in Viral siRNA Generation Blot analysis of viral siRNA. Systemic tissue samples were analyzed at the indicated time points from parental and mutant lines that were infected with TuMV–GFP (A), CMV-Y (B), and TCV (C). RNA blots were analyzed using virus-specific probes to detect siRNAs. Ethidium bromide-stained gels in the zone corresponding to tRNA and 5S RNA are shown. Relative accumulation (RA) of siRNAs is indicated at the bottom of each panel, with the level measured in infected control plants (Col-0 or La- er , depending on the mutant) at 7 dpi arbitrarily set to 1.0. Figure 6 Altered Susceptibility to TCV Infection in dcl2-1 Mutant Plants (A) Noninfected control (left) and TCV-infected (right) Col-0, dcl2-1 , and dcl3-1 plants at 14 dpi. (B) TCV accumulation, as measured by ELISA, in the systemic tissues of infected wild-type and mutant plants at 7 dpi (open bars) and 14 dpi (filled bars). (C) Plant height (left), number of flowers/plant (center), and fresh weight of bolt tissue (right) were measured at 14 dpi in noninfected (open bars) and infected (filled bars) plants ( n = 9). In contrast, TCV-derived siRNAs accumulated to levels that were 5-fold lower in dcl2-1 plants compared to wild-type plants at 7 dpi (see Figure 5 C, lanes 10–11). This was a transient deficit, as TCV siRNA levels rebounded to near wild-type levels by 14 dpi (see Figure 5 C, lanes 13–14). The slow accumulation of siRNAs was not due to lack of TCV replication or movement in the tissues analyzed, as TCV titer in the dcl2-1 mutant was similar to (7 dpi) or significantly higher than ( p < 0.05, 14 dpi) the titers in wild-type plants ( Figure 6 B). Additionally, TCV-induced disease was more severe in dcl2-1 plants, as plant height, fresh weight of bolts, and number of flowers in infected dcl2-1 plants were each significantly ( p < 0.01 for plant height and flower number; p < 0.05 for weight of bolts) lower compared to infected wild-type plants ( Figure 6 A and 6 C). Therefore, DCL2 functions as a component of the antiviral silencing response in TCV-infected plants. The DCL2–GFP fusion protein accumulated predominantly in the nucleus of N. benthamiana cells in the transient assay system, although some cytosolic localization was also detected (see Figure 4 ). Thus, DCL1 ( Papp et al. 2003 ), DCL2, and DCL3 each have nuclear transport activity. Discussion Genetic Diversification of Small RNA-Generating Systems in Plants We show here that Arabidopsis has at least three systems to generate distinct classes of endogenous or virus-induced small RNAs and that these are associated with specialized regulatory or defensive functions. First, the miRNA-generating system requires DCL1, as shown previously ( Park et al. 2002 ; Reinhart et al. 2002 ), but none of the RDR proteins tested. In principle, there should be no requirement for an RDR activity during miRNA biogenesis, as the DCL1 substrate is formed directly as a result of DNA-based transcription. DCL1 likely functions in the nucleus ( Papp et al. 2003 ). It also functions, either directly or indirectly, with HEN1, which may confer substrate specificity, processing accuracy, or catalytic function. The second system requires DCL3 and RDR2 and generates endogenous siRNAs primarily of the large-sized (approximately 24 nucleotides) class. While DCL3 undoubtedly functions as the ribonuclease to process dsRNA precursors, RDR2 presumably functions as a polymerase to form dsRNA molecules de novo using templates resulting from transcription of DNA. At some loci, however, RDR2 may be unnecessary as a catalytic subunit, but rather contribute to the formation or stability of a complex that contains active DCL3. This could be the case at some sites, such as the siRNA02 locus, that contain inverted duplications and that may form transcripts with extensive dsRNA structure. Interestingly, accumulation of siRNAs specific to a hairpin construct was shown to be RdRp dependent in fission yeast ( Schramke and Allshire 2003 ). At some loci, this system appears to interface with AGO4, HEN1, and SDE4. The third system functions in antiviral defense and involves DCL2. Loss of this system was specifically detected in TCV-infected dcl2-1 plants, which exhibited delayed viral siRNA accumulation and increased susceptibility and sensitivity. However, there are several reasons to suspect that multiple antiviral, siRNA-generating systems exist. siRNAs triggered by TCV were not eliminated in dcl2-1 plants, but rather siRNA accumulation was delayed. Although this could be due to incomplete loss of DCL2 function in the mutant, it could also reflect the existence of secondary or redundant DCL activities. Among the three viruses tested, two were unaffected by the dcl2-1 mutation. This strongly implies the existence of one or more other siRNA-generating activities with unique or redundant antiviral specificity. Further, the DCL2-dependent system may have functions in addition to those associated with antiviral defense. The DCL2–GFP fusion protein was detected primarily in the nucleus, whereas TCV replicates and accumulates outside of the nucleus. Experiments to determine the genetic requirements for RDR1 and RDR2 during antiviral silencing against the three viruses were inconclusive, again possibly the result of functional redundancies or the presence of confounding viral RdRp activities ( Ahlquist 2002 ). Mourrain et al. (2000 ), on the other hand, showed that rdr6 ( sde1/sgs2 ) mutants were deficient in CMV-induced silencing. Additionally, Yu et al. (2003 ) showed that RDR1 contributed to defense against tobamoviruses. Tang et al. (2003 ) identified two siRNA-generating DCL activities in wheat-germ extracts. These were detected using dsRNA as a substrate. Although monocots contain a DCL gene family, the members do not correlate one-for-one with those in Arabidopsis (Z. Xie and J. Carrington, unpublished data). Further study is required to correlate the DCL activities from wheat germ with those in Arabidopsis. The degree of genetic diversification of the DCL family in plants is in contrast to the situation in animals. Caenorhabditis elegans and human, for example, contain only one DICER ( Grishok et al. 2001 ; Ketting et al. 2001 ; Knight and Bass 2001 ; Provost et al. 2002 ; Zhang et al. 2002 ), even though both possess miRNA and siRNA functions. Thus, whereas plants diversified and functionally specialized DCL family members during evolution, animals evolved functionally distinct small RNA systems around one or relatively few DICER activities. Animals, however, evolved relatively large AGO-related families ( Carmell et al. 2002 ), and these may provide modules for functional specialization. Roles of Endogenous siRNA-Generating Systems in Plants Both DCL3 and RDR2 cooperate with AGO4, and possibly also with SDE4 and HEN1, at the AtSN1 locus to initiate or maintain a heterochromatic state ( Hamilton et al. 2002 ; Zilberman et al. 2003 ). Loss of DCL3, RDR2, and AGO4 factors correlates with loss of DNA methylation and histone H3K9 methylation. Interestingly, these factors are also necessary for silencing triggered de novo during the transformation process using transgenic FWA ( Chan et al. 2004 ). Silencing of FWA is due to cytosine methylation of a region in the promoter that contains direct repeats ( Soppe et al. 2000 ). The effect of the rdr2-1 mutation on chromatin structure and gene silencing of AtSN1 and FWA was generally stronger than the effect of the dcl3-1 mutation. This may be explained by the presence of residual siRNAs formed by another DCL activity in the dcl3 mutant (see Figure 1 B). The picture that emerges from these and other results shows that DCL3 and RDR2 function as components of an endogenous siRNA-generating system and that the resulting siRNAs may guide chromatin modification events through effector complexes containing AGO4. Given that AGO proteins are components of RISCs that catalyze sequence-specific RNA degradation ( Carmell et al. 2002 ) and that different AGO proteins have DNA- or RNA-binding activities ( Lingel et al. 2003 ; Song et al. 2003 ; Yan et al. 2003 ), it seems reasonable to speculate that AGO4 engages a chromatin-associated RISC-like complex and interacts with nuclear siRNAs or target sequences. But unlike RNAi events in the cytoplasm, chromatin-associated complexes likely interact with DNA methyltransferase and histone methyltransferase systems. RdDM can occur at CpG and non-CpG sites, but maintenance of non-CpG methylation after DNA replication may generally require the continued activity of the siRNA-guided complex ( Luff et al. 1999 ; Jones et al. 2001 ; Aufsatz et al. 2002 ). Methylation at CpG sites, in contrast, can be maintained by template-driven methylation on hemimethylated products of DNA replication, which explains why CpG methylation frequently persists in subsequent generations after one or more silencing factors or trigger loci are lost. Accumulation of siRNA from endogenous loci and transgenes does not necessarily require AGO4 (D. Zilberman and S. Jacobsen, unpublished data), suggesting that AGO4 acts downstream of siRNA formation to direct DNA methylation. Losses of AGO4 and HEN1 have nearly identical effects on all siRNAs tested, possibly because HEN1 and AGO4 affect a similar point in the pathway. If AGO4 and HEN1 function downstream of siRNA formation, why do siRNAs derived from some sites ( AtSN1 and 5S rDNA) accumulate to such low levels in ago4 and hen1 mutants? One possibility is that heterochromatic marks (DNA and H3K9 methylation) and associated factors serve to recruit RDR2, DCL3, or both to specific sites on chromatin, thus establishing a reinforcement loop. Loss of heterochromatin in an ago4 mutant, for example, would result in failure to recruit the siRNA-generating enzymes to transcripts originating from a target locus and, therefore, the absence of siRNAs. This hypothesis, however, does not hold for some other siRNA-generating sites, such as those that yield cluster2 siRNAs and siRNA02. Accumulation of siRNAs from these sites is unaffected or even enhanced in ago4 and hen1 mutants. In wild-type plants, these loci are both hypomethylated at CpG and non-CpG sites and are associated with histone H3 that largely lacks K9 methylation (data not shown). The siRNAs formed from these loci clearly require RDR2 and DCL3, but they appear not to affect chromatin structure. These siRNAs may be sequestered elsewhere in the cell and unable to interact with chromatin or chromatin-associated factors. The spectrum of naturally occurring siRNAs in Arabidopsis is informative about the roles of these molecules in genome maintenance, genome expression, and defense. The fact that siRNAs from highly repeated sequences, largely retroelements and transposons, are overrepresented compared to unique genome sequences suggests that sequence duplication events are sensed and dealt with through RNA-guided formation of heterochromatin. This is frequently discussed within the context of genome defense, whereby suppression of mobile DNA promotes genome stability ( Plasterk 2002 ; Dawe 2003 ). Indeed, loss of heterochromatin is often associated with increased activity of transposons and retroelements ( Hirochika et al. 2000 ; Miura et al. 2001 ; Singer et al. 2001 ; Gendrel et al. 2002 ). However, it should be appreciated that these and other repeated sequences might also serve as cis -active, epigenetic regulatory modules if positioned near or within functional genes ( Kinoshita et al. 2004 ). The rapidly expanding number of examples, such as vernalization ( Bastow and Dean 2003 ), of cellular memory conditioned by epigenetic events hint that siRNA-directed processes may be embedded broadly as a regulatory mechanism during growth and development ( Goodrich and Tweedie 2002 ). Materials and Methods Plant materials All plants were grown under standard greenhouse conditions. The dcl1- 7, hen1-1 , cmt3-7 , and ago4-1 mutant lines were described previously ( Cao and Jacobsen 2002 ; Golden et al. 2002 ; Park et al. 2002 ; Zilberman et al. 2003 ). Other mutant lines were obtained from the Salk Institute Genome Analysis Laboratory (SIGnAL, La Jolla, California, United States) and Torrey Mesa Research Institute (now a subsidiary of Syngenta, Basel, Switzerland). dcl2-1 has a T-DNA insertion within predicted intron 9 (after nucleotide 2,842 from ATG of the genomic DNA) of DCL2 (At3g03300). dcl3-1 has a T-DNA insertion within predicted exon 7 of DCL3 (At3g43920) at a point 2,136 nucleotides beyond the ATG in genomic DNA. This introduces four codons after the serine 288 codon, followed by a premature stop codon. rdr1-1 has a T-DNA insertion within predicted exon 1 after nucleotide 2,366 beyond the ATG of RDR1 (At1g14790). rdr2-1 has a T-DNA insertion within predicted exon 1 (in front of nucleotide 316 from the ATG) of RDR2 (At4g11130). rdr6-1 has a T-DNA insertion within predicted exon 2 (in front of nucleotide 3,977 from ATG of the genomic DNA) of RDR6 (also known as SDE1/SGS2 ; At3g49500). Each insertion line was backcrossed twice to Col-0 and brought to homozygosity. Additional information about the insertion lines are provided in the supplemental online materials. For analysis of each insertion mutant, Col-0 was the wild-type control plant. For dcl1-7 , hen1-1 , ago4-1 , and cmt3-7 mutants, La- er was the wild-type control. RNA blot analysis Extraction of low- and high-molecular weight RNAs and blot assays were done as described previously ( Llave et al. 2002 a). Low-molecular weight RNA (20 μg) from Arabidopsis inflo-rescence tissue was used for miRNA and endogenous siRNA analysis. Probes for miR-171 and AtSN1 -siRNA analysis were described previously ( Llave et al. 2002 b; Zilberman et al. 2003 ). miR-159 was detected using an end-labeled DNA oligonucleotide AS-159 (5′-TAGAGCTCCCTTCAATCCAAA-3′). siRNA02 and siRNA1003 were detected using the end-labeled DNA oligonucleotides AS-02 (5′-GTTGACCAGTCCGCCAGCCGAT-3′) and AS-1003 (5′-ATGCCAAGTTTGGCCTCACGGTCT-3′), respectively. The probe for cluster2 siRNAs was a random primer-labeled fragment spanning a 235-nucleotide IGR of chromosome I (nucleotides 4,506,544–4,506,778) (see Figure 1 A) and was amplified from genomic DNA using primers AS-285 (5′-TTGCTGATTTGTATTTTATGCAT-3′) and S-786 (5′-CTTTTTCAAACCATAAACCAGAAA-3′). Analysis of DNA and histone methylation Cytosine methylation was analyzed by bisulfite sequencing of genomic DNA or by DNA blot assay following digestion with methylation-sensitive restriction endonucleases, as described elsewhere ( Jacobsen et al. 2000 ; Zilberman et al. 2003 ). The region of AtSN1 analyzed (chromosome III, nucleotides 15,805,617–15,805,773) was treated with sodium bisulfite and amplified using primers AtSN1 -BS1 (5′-GTTGTATAAGTTTAGTTTTAATTTTAYGGATYAGTATTAATTT-3′) and AtSN1 -BS2 (5′-CAATATACRATCCAAAAAACARTTATTAAAATAATATCTTAA-3′). At least 18 independent clones were sequenced for each genotype. ChIP assays were done using antibodies specific for dimethyl-histone H3K4 (Upstate Biotechnology, Lake Placid, New York, United States) or dimethyl-histone H3K9 (kindly provided by T. Jenuwein, Research Institute of Molecular Pathology, Vienna, Austria) as described elsewhere ( Gendrel et al. 2002 ). Methylation of H3K4 and H3K9 at AtSN1 in wild-type Col-0 and rdr2-1 and dcl3-1 mutants was measured relative to that at internal control loci, At4g04040 and At4g03800. The data were then normalized against the values measured in Col-0. Analysis of GFP fusion proteins The 35S:DCL3–GFP construct contained the DCL3 coding region fused to GFP coding sequence, flanked by the cauliflower mosaic virus (CaMV) 35S promoter and terminator sequences. The expression cassette was cloned in pSLJ755I5. All other GFP fusion constructs were made by cloning the coding sequence into pGWB5 (for C-terminal GFP) or pGWB6 (for N-terminal GFP), a set of gateway-compatible binary vectors designed for 35S promoter-driven expression of GFP fusion proteins (kindly provided by T. Nakagawa, Shimane University, Izumo, Japan). Cloning using gateway vectors was done using reagents and protocols from Invitrogen (Carlsbad, California, United States). Constructs were introduced into Agrobacterium tumefaciens strain GV2260 and expressed in N. benthamiana leaves as described previously ( Johansen and Carrington 2001 ). Fusion proteins were detected by confocal microscopy and immunoblot assay using a monoclonal antibody against GFP (Roche, Basel, Switzerland). Virus infection assays Wild-type and mutant Arabidopsis plants (approximately 4 wk old, prior to bolting) were infected with TuMV–GFP, CMV-Y, and TCV as described previously ( Whitham et al. 2000 ; Lellis et al. 2002 ). At 7 and 14 dpi, systemic tissues consisting of inflorescences and cauline leaves were harvested for ELISA and RNA blot assays. Antibodies used for TuMV and TCV ELISAs were as described previously ( Lellis et al. 2002 ). Computational methods Computational identification of repeat sequences, including transposons and retroelements, in the Arabidopsis genome was done using RepeatMasker (http://ftp.genome.washington.edu/RM/RepeatMasker.html) and Repbase ( http://www.girinst.org/index.html ). Further information about Arabidopsis siRNAs and miRNAs, including those that were analyzed in this work, can be found in the Arabidopsis Small RNA Project database ( http://cgrb.orst.edu/smallRNA/db/ ). Supporting Information Figure S1 DCL and RDR Mutant Lines (A) Exon (bars)/intron (lines) organization of the Arabidopsis DCL and RDR genes and location of T-DNA insertion sites in mutant lines. (B) RNA blot analysis (20 μg of total RNA) for DCL2 and DCL3 mRNA in Col-0 and the respective mutants. DNA fragments corresponding to nucleotides 2,652–3,292 of the DCL2 open reading frame and nucleotides 2,805–3,571 of the DCL3 open reading frame were used as hybridization probes. As a control, the blots were stripped and hybridized with a β-tubulin-specific probe ( Kasschau et al. 2003 ). (C) RNA blot analysis (10 μg of total RNA) for RDR1 and RDR2 mRNA in Col-0 and the respective mutants. DNA fragments corresponding to nucleotides 2,900–3,300 of the RDR1 open reading frame and nucleotides 10–271 of the RDR2 open reading frame were used as gene-specific probes. RNA samples from SA-treated leaf tissues were also included in the analysis. (5.9 MB EPS). Click here for additional data file. Figure S2 Immunoblot Analysis of GFP Fusion Proteins The 35S promoter-driven GFP fusion constructs were transiently expressed in N. benthamiana using an Agrobacterium -injection procedure. Leaf tissue from injected zones was excised at 2 dpi for immunoblot assay using a monoclonal antibody against GFP and confocal microscopy (see Figure 4 ). An arrow indicates the position of predicted full-sized fusion protein. (10.8 MB EPS). Click here for additional data file. Table S1 Cloned siRNA Loci in the Arabidopsis Genome (25 KB DOC). Click here for additional data file. Table S2 Cytosine Methylation of Arabidopsis AtSN1 (24 KB DOC). Click here for additional data file. Accession Numbers The GenBank ( http://www.ncbi.nlm.nih.gov/Genbank/ ) accession numbers for the entities discussed in this paper are At1g14790 (NM_101348), At3g03300 (NM_111200), At3g43920 (NM_114260), At3g49500 (NM_114810), At4g11130 (NM_117183), chromosome I (NC_003070.3), chromosome III (NC_003074.4), and siRNA02 (AF501743). The SAIL (formerly Garlic) ( http://signal.salk.edu/cgi-bin/tdnaexpress ) accession numbers for the T-DNA insertion lines discussed in this paper are rdr1-1 (SAIL_672F11), rdr2-1 (SAIL_1277H08), and rdr6-1 (SAIL_388H03). The SIGnAL database ( http://signal.salk.edu/ ) accession numbers for the T-DNA insertion lines discussed in the paper are dcl2-1 (SALK_064627) and dcl3-1 (SALK_005512).

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC350667.xml

536008

Evolution of Sex Chromosomes: The Case of the White Campion

There are many different sex-determining systems in plants and animals with separate sexes (dioecious species). In some species, environmental factors activate sex-determining genes that trigger expression of genes leading to male or female development. Other species have evolved specialized sex chromosomes. In the well-known X-Y system of mammals, individuals inheriting a Y chromosome become males, and XX individuals become females. Sex chromosomes have arisen independently in many taxonomic groups. It is an interesting question whether the same mechanisms were involved each time. Similarities in sex chromosome evolution have been reported between birds and mammals (although in birds, females are the heterozygous sex). In a new study, Michael Nicolas and colleagues uncover striking parallels in the details of sex chromosome evolution between mammals and a far more distant group: plants. Sex chromosomes are an oddity in flowering plants. They are limited to dioecious species, a subset of plants that carry male and female organs (stamens and carpels, respectively) on separate individuals (most flowering plants are hermaphrodites). The genus Silene , which includes the White Campion, includes both dioecious and hermaphrodite species. The authors focus on three dioecious species, Silene dioica , S. latifolia , and S. diclinis , which share an X-Y sex-determination system where Y specifies maleness. The theory of sex chromosome evolution holds that sex chromosomes were once homologs (a pair of equivalent autosomes—the non-sex chromosomes) that evolved different morphology and gene content because they lost their ability to recombine. Suppression of recombination is thought to start around the sex-determining region, but may eventually affect much of the sex chromosomes. Recombination is a key genetic process in which two chromosomes pair and exchange their sequences. In the absence of recombination, the two chromosomes of a pair evolve separately. Flowers of Silene species, clockwise from top left: male flowers of S. latifolia and S. dioica , hermaphrodite flower of S. vulgaris , male flower of S. diclinis In the case of mammals, whose sex chromosomes evolved about 320 million years ago, loss of recombination led to widely diverged X and Y chromosomes that pair only over a very small region, the pseudoautosomal region (PAR; because in this region the X and Y still behave like autosomes). The X and Y chromosomes of dioecious Silene species are morphologically distinct, like those of mammals, and they also have a PAR and a nonrecombining region. Nicolas and colleagues' results shed some light on how recombination suppression evolved on the Silene sex chromosomes. The authors studied four genes outside the PAR on the Silene X chromosomes that are also present on their Y chromosomes. They mapped the genes relative to the PAR and compared the nucleotide sequences of the X and Y version of each gene in each species. As expected of sequences that no longer recombine, the X and Y versions of each gene have diverged. Strikingly, the extent of nucleotide divergence increases with the gene's distance from the PAR. Evolutionary biologists use sequence divergence as a clock: the longer two originally identical sequences have been isolated from one another, the more independent mutations they accumulate. The picture that emerges from the Silene data is one of a progressive suppression of recombination, gradually diminishing the PAR. A similar scenario has been proposed in mammals and birds. However, the authors estimate that the Silene sex chromosomes started diverging only 10 million years ago. The Silene chromosomes might therefore offer a better chance to observe recombination suppression in its early stages, and perhaps to get at its mechanisms. The authors also report evidence for some degeneration of the Silene Y chromosome genes. Y degeneration is well documented in mammals, in which most X-linked genes have no Y-linked counterparts. Understanding X-Y divergence in Silene species may thus shed light on the evolution of sex chromosomes in vertebrates as well.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC536008.xml

545970

A randomized clinical trial comparing hydrocolloid, phenytoin and simple dressings for the treatment of pressure ulcers [ISRCTN33429693]

Background Pressure sores are important and common complications of spinal cord injury. Many preventive and therapeutic approaches have been tried and new trials are evolving. One relatively recent method is application of a hydrocolloid dressing (HD). In this study we compared the therapeutic effects of HD on pressure ulcer healing with two other topical applications, phenytoin cream (PC) and simple dressing (SD). Methods Ninety-one stage I and stage II pressure ulcers of 83 paraplegic male victims of the Iran-Iraq war were randomly allocated to three treatment groups. Mean age and weight of the participants were 36.64 ± 6.04 years and 61.12 ± 5.08 kg, respectively. All the patients were managed in long term care units or in their homes for 8 weeks by a team of general practitioners and nurses, and the ulcer status was recorded as "Complete healing", "Partial healing", "Without improvement" and "Worsening". Results Complete healing of ulcers, regardless of location and stage, was better in the HD group than the PC [23/31(74.19%) vs 12/30(40%); difference: 34.19%, 95% CI = 10.85–57.52, (P < 0.01)] or the SD [23/31(74.19%) vs 8/30(26.66%); difference: 47.53%, 95% CI = 25.45–69.61, (P < 0.005)] groups. Complete healing of stage I ulcers in the HD group [11/13(85%)] was better than in the SD [5/11(45%); difference: 40%, 95% CI = 4.7–75.22, (P < 0.05)] or PC [2/9 (22%); difference: 63%, 95% CI = 29.69–96.3, (P < 0.005)] groups. Complete healing of stage II ulcer in the HD group [12/18 (67%)] was better than in the SD group [3/19(16%); difference: 51%, 95% CI = 23.73–78.26, (P < 0.005)], but not significantly different from the PC group [10/21 (48%); difference: 19%, 95% CI = -11.47–49.47, (P > 0.05)]. We performed a second analysis considering only one ulcer per patient (i.e. 83 ulcers in 83 patients). This "per patient" analysis showed that complete ulcer healing in the HD group was better than in the PC [20/28(71.4%) vs 11/28 (39.3%); difference: 32.1%, 95% CI = 7.4–56.7, (P < 0.01)] or SD [20/28(71.4%) vs 8/27 (29.6%); difference: 41.8%, 95% CI = 17.7–65.8, (P < 0.005)] groups. Conclusion We deduced that HD is the most effective method investigated for treating stage I and II pressure ulcers in young paraplegic men.

Background Skin ulcers caused by pressure and strains are known by various names: decubitus ulcer, bedsore, ischemic ulcer and pressure ulcer. "Pressure ulcer", which indicates the etiology of the lesion, seems the most appropriate name [ 1 ]. An estimated 50–80% of individuals suffering from spinal cord injury develop pressure ulcers at least once in their lifetime. Most of these ulcers occur during the first two years after injury, but even after 3–4 years an incidence of 30% has been reported [ 2 - 4 ]. Although the major challenge is to prevent the occurrence of ulcers [ 5 , 6 ], therapeutic measures merit due attention. Pressure ulcer therapy is among the expensive of medical and surgical interventions [ 5 - 7 ]. In one study in the United Kingdom, data relating to chronic wound management practice obtained from 15 pressure sore studies showed a cost range of 422–2548 pounds per healed wound for primary dressing, nursing time, wound cleansing and debridements [ 8 ]. These figures do not include the much higher costs of hospitalization and plastic surgery. We have tried to find a more effective and cost-efficient method of treatment. Different methods have been used for preventing and treating pressure ulcers. These include various training programs for patients [ 4 , 9 , 10 ]; physiotherapy methods employing ultrasound, ultraviolet irradiation and laser treatment [ 7 ]; good nutrition emphasizing high protein, high calorie diet and more liquid; electrical stimulation; and application of local ointments and creams such as bacitracin, silver sulfadiazine, neomycin, polymixin, phenytoin and hydrocolloid dressings [ 11 - 19 ]. The results of the studies conducted so far are incompatible, even contradictory. Most of them considered too few patients and/or lacked a control group. In Iran, 5000 patients suffer from spinal cord injury (SCI): of these, 2000 are lran-lraq war victims and 3000 were handicapped by other causes. In view of the enormous prevalence of pressure ulcers in war victims and other spinal handicap patients, and the importance of these lesions in terms of morbidity, mortality and cost of treatment, we have compared the efficacies of applying hydrocolloid dressing, phenytoin cream and a simple dressing. The aims were to determine: 1. which is the most effective in terms of complete ulcer healing; 2. whether healing rates differ with respect to the ulcer stage (I and II) or location (gluteal, ischial, sacral) using these three different methods. Methods The study was a randomized single blind clinical trial involving 2015 Iranian spinal cord injury (SCI) victims of the Iran-Iraq war (1980–1988). The SCI victims were accessed through the mediation and assistance of the Jaonbazan Medical and Engineering Research Center (JMERC) , the medical and research section of the official governmental body responsible for SCI war victims. The study proposal was reviewed, approved and granted by JMERC. The medical records of all 2015 subjects were studied to identify cases with pressure ulcers. Where the data were unknown or unreliable, the patients were visited at home or in victims' long term care centers. Finally, 165 pressure ulcers in 151 patients were identified. All relevant data including patient age and weight, the longevity of the ulcer before our intervention, and the size, stage and location of the ulcer, were collected by the general practitioners in the team. Next, all the patients were examined by one of the authors to confirm their eligibility for the study. The eligibility criteria were: A) Inclusion Criteria: 1. Paraplegia caused by spinal cord injury; 2. Pressure ulcer stage I and II according to Shea classification [ 20 ] or National Pressure Ulcer Advisory Panel [ 21 ] (Fig. 1 ); 3. Patient's informed consent; 4. Smoothness of ulcer area to establish whether adhesive could be used at the site. Exclusion criteria: 1. Addiction; 2. Heavy smoking (more than 20 cigarettes a day or more than 10 packs per year; 3. Concomitant chronic disease (e.g. diabetes mellitus or frank vascular disease such as Buerger's disease). Figure 1 Two pressure ulcer classifications. Seventy-four ulcers in 68 patients were excluded because they did not meet these eligibility criteria: 31 ulcers (28 patients) were stage III or higher; 27 ulcers (25 patients) were excluded because of patient's smoking/addiction; 5 ulcers (5 patients) had uneven surfaces; 4 ulcers (4 patients) were excluded because of systemic diseases; and 6 patients with 7 ulcers refused to participate (Fig. 2 ). Thus, the study sample comprised 83 patients with 91 pressure ulcers in the ischial, sacral or gluteal areas. These 91 ulcers were allocated to three different groups (30 ulcers each) by stratified randomization. Three therapeutic methods were applied as follow: simple dressing (SD), hydrocolloid dressing (HD), and adhesive and phenytoin cream (PC). Two general practitioners and nine nurses trained in treatment interventions administered the protocols. Figure 2 Flow diagram of participants through each stage of the study. The SD patients were visited twice a day, the PC patients once a day and the HD patients twice a week. All participants were visited and examined in their family homes or nursing homes by general practitioners every two weeks to ensure that the treatments were being properly applied and were consistent among the three groups. There were no differences in the facilities available for patients in family homes versus nursing homes, and all the patients had free access to victims' long term care centers. In the SD group, the following steps were taken twice a day. The ulcer was cleaned and washed 3 times with normal saline, then dried with a sterile gauze and, depending on the size of ulcer, covered with wet saline gauze dressing. In the PC group, daily dressing and cleaning of ulcer were similar to the SD group, except that a thin layer of phenytoin cream was applied to the ulcer before the dressing was performed. In the HD group, after the ulcer had been cleaned in a similar manner to the SD group, the hydrocolloid adhesive dressing was applied to the ulcer area. The adhesive dressings were changed twice a week. Any necrotic tissue was debrided before treatment; all debridements preceded ulcer tracing and assignment of the participants to the trial groups. No debridement was allowed after treatment had started. No concomitant topical or systemic antibiotic, glucocorticoid or immunosuppressive agent was allowed during the treatment period. Fortunately, none of our patients needed debridement or the aforementioned concomitant therapies during the study period. There were no differences among the trial groups with respect to other concomitant care measures. Every two weeks a questionnaire regarding the ulcer's status was completed by the general practitioners; and at the end of 8 weeks, the ulcers' conditions were examined blind by one author and assessed as "Complete Healing", "Partial Healing", "Without Improvement" or "Worsening". To measure each ulcer's surface area, the ulcer borders were traced on to a paper overlay. This primary schematic representation was then scanned, redrawn and measured by AutoCAD 2000 software. The primary outcome was whether or not the ulcer was completely healed within 8 weeks. "Complete ulcer healing" was defined as: A) For stage I ulcer, intact epidermis, no red area; B) For stage II ulcers, intact dermis and epidermis, no abrasion or ulceration. Other definitions were as follows. "Partial healing" = any decrease in ulcer size compared to the baseline ulcer tracing, excluding complete healing. "Without improvement" = no change in ulcer size compared to the baseline ulcer tracing. "Worsening" = any increase in ulcer size compared to the baseline ulcer tracing. The difference in responses between patients receiving HD and patients receiving the other therapies (PC or SD) were determined [ 22 ]. Before the study, we assumed response rates of 30%, 40% and 80% for SD, PC and HD, respectively. Thus, based on the 40% difference, power of 0.85, 95% confidence level and estimated follow-up loss of 10%, 29 patients were required for each study group. The number of ulcers that met the eligibility criteria totaled 91 and all were enrolled in the study. A random-number table was used to generate the random allocation sequence, and stratified randomization was used to achieve balance between the treatment groups and subgroups (ulcer stages and locations). If a patient had more than one ulcer, all the ulcers were treated by the same method to eliminate the possible complicating factor of treatment interactions. The statistician in the team generated the random allocation sequence. He was informed of the patient list (numbers only) and the ulcer stage and location of each patient. The treatment category for each patient was determined by the statistician and was delivered in an opaque sealed envelope bearing only the number of the patient. These sealed envelopes were delivered to the general practitioners, along with the list of patients' numbers and names. After each patient was visited, the appropriately numbered envelope was opened by the general practitioner to determine whether the SD, PC or HD method would be used, then the appropriate intervention commenced. The authors were blind to the patients' assignment to trial groups. The general practitioners were also blind to the treatment of each patient up to the start of the study, when they opened the sealed envelopes. After intervention began, both the general practitioners and the nurses knew the trial groups, because significant differences among the three treatment methods precluded blinding. The patients were also aware of the treatment methods, although they initially had equal chances of entering any of the trial groups. Thus, the study was single-blinded and the author who enrolled the patients to the study was blind to treatment assignment. The author who finally assessed the outcomes was also blind to the trial group of each patient. To maintain the blinded status on assessment of outcomes, the assessor examined the patients, after the ulcer dressings had been removed by the general practitioners, with no knowledge of the trial groups to which they had been assigned. The gross appearance of the ulcers without dressing, whether healed or not, did not indicate the trial group. The assessor was asked during the 8-week outcome assessment to try to identify which treatment has been administered to each patient. Overall, 27.7% of his guesses were correct (25% in the HD group, 32.1% in the PC group and 25.9% in the SD group), so they were were no better than chance; i.e. there were no significant differences among the three trial groups with respect to proportions guessed correctly (P > 0.2 in all cases). The study proposal was designed in November 2001, and the recruitment of patients began in March 2002 and lasted about 2 months. Then the patients were allocated to the treatment groups and followed-up for another 2 months. Finally, all the collected data were analyzed within 2 months. Thus the study from proposal to final analysis took about 10 months (November 2001-September 2002). At the end of the study, all the data collected from the patients' preliminary and complementary questionnaires were analyzed by SPSS software using ANOVA and Chi square tests, and P-values of <0.05 were assumed significant. The 95% confidence intervals were also calculated and reported [ 23 ]. For rare events (more than 20 percent of cross tabulation cells had values less than 5), Fisher's exact test was used. Based on stage and location of ulcers, subgroup analyses were performed using the same statistical tests. Results Ninety-one ulcers in 83 male patients were treated by one of three methods. The mean age and weight of the patients were 36.64 ± 6.04 years and 61.12 ± 5.08 kg, respectively. Of the 91 ulcers, 33 were stage I and the remaining 58 were stage II. There were no significant differences among the three therapeutic groups in baseline demographic characteristics (table 1 ) or in ulcer location (sacral, gluteal, ischial) or stage (I or II) (Fig. 3 and 4 ). Table 1 Baseline characteristics of study subjects assigned to hydrocolloid, phenytoin and simple dressing groups Variables Mean Age Of patients (yr ± SD) Mean weight Of patients (kg ± SD) Mean duration of ulcer before treatment (wk ± SD) Mean ulcer size (cm 2 ± SD) Stage of ulcer (no) Treatment group; no I II Total n = 83 patients 91 ulcers 36.64 ± 6.04 61.12 ± 5.08 6.25 ± 6.56 7.54 ± 12.99 33 58 Hydrocolloid n = 28 patients 31 ulcers 36.81 ± 6.71 62.26 ± 5.44 7.63 ± 5.59 7.26 ± 15.4 13 18 Phenytoin n = 28 patients 30 ulcers 36.5 ± 4.99 60.07 ± 4.39 5.84 ± 8.04 5.12 ± 3.63 9 21 Simple dressing n = 27 patients 30 ulcers 36.6 ± 6.17 61 ± 5.03 5.25 ± 5.39 10.27 ± 15.32 11 19 P-Value of comparing variables of 3 groups P > 0.10 P > 0.10 P > 0.10 P > 0.10 P > 0.62 Figure 3 Ulcer distribution according to treatment group and location. Figure 4 Ulcer distribution according to treatment group and stage. The numbers of ulcers and the degree of improvement in the three therapeutic groups are shown in table 2 . The completion of healing, regardless of location and stage, was better in the HD than in the PC [23/31(74.19%) vs 12/30(40%); difference 34.19%, 95% CI = 10.85–57.52, (P < 0.01)] or the SD [23/31(74.19%) vs 8/30(26.66%); difference 47.53%, 95% CI = 25.45–69.61, (P < 0.005)] groups. Completion of healing of stage I ulcers in the HD group [11/13(85%)] was also better than in the SD [5/11(45%); difference 40%, 95% CI = 4.7–75.22, (P < 0.05)] or PC [2/9 (22%); difference 63%, 95% CI = 29.69–96.3, (P < 0.005)] groups. Completion of healing of stage II ulcers was better in the HD group [12/18(67%)] than in the SD group [3/19(16%); difference 51%, 95% CI = 23.73–78.26, (P < 0.005)], but there was no significant difference from the PC group [10/21 (48%); difference 19%, 95 CI = -11.47–49.47, (P > 0.05)]. Table 2 Healing status of pressure ulcers in 3 treatment groups (hydrocolloid, phenytoin and simple dressing) healing status Complete Partial Not improved Worsened Total Treatment group; no (%) Hydrocolloid n = 31 23 (74.19%) 4 (12.58%) 2 (6.45%) 2 (6.45%) 31 (100%) Phenytoin n = 30 12 (40%) 4 (13.33%) 12 (40%) 2 (6.66%) 30 (100%) Simple dressing n = 30 8 (26.66%) 5 (16.66%) 8 (26.66%) 9 (30%) 30 (100%) Gluteal ulcers healed more completely in the HD group [6/6(100%)] than in the PC [2/7 (29%); difference 71%, 95% CI = 37.38–100, (P < 0.005)] or SD [1/8(13%); difference 87%, 95% CI = 63.69–100, (P < 0.001)] groups. The corresponding figures for ischial ulcers were: HD group 13/18(72%) and SD group 3/14 (21%); difference 51%, 95% CI = 21.2–80.7, (P < 0.005)]. The PC group was not significantly different from HD: 8/18(44%); difference 28%, 95% CI = -2.9–58.9, (P < 0.1)]. In the case of sacral ulcers, complete healing in HD group did not differ significantly from either of the others. The results were: HD group 4/7 (57%), SD group 4/8(50%); difference 7%, 95% CI = -50–64.15, (P > 0.35), and PC group 2/5(40%); difference 17%, 95% CI = -39.4–73.4, (P > 0.20)]. We performed a second analysis on 83 ulcers in 83 patients. We selected one ulcer per patient using a random number table; 31 of the 83 ulcers were stage I and the remaining 52 were stage II. There were again no significant differences among the trial groups with respect to baseline characteristics (table 3 ). This "per patient" analysis showed that complete ulcer healing, regardless of location and stage, in the HD group was better than in the PC [20/28(71.4%, 95% CI = 54.7–88.1) vs 11/28 (39.3%, 95% CI:21.3–57.3); difference 32.1%, 95% CI = 7.4–56.7, (P < 0.01)] or SD [20/28(71.4%, 95% CI = 54.7–88.1) vs 8/27 (29.6%, 95% CI = 12.4–46.8); difference 41.8%, 95% CI = 17.7–65.8, (P < 0.005)] groups. Table 3 Baseline characteristics of study subjects assigned to three trial groups considering the patient as unit of analysis(one ulcer per patient). Variables Mean duration of ulcer before treatment (wk ± SD) Mean ulcer size (cm 2 ± SD) Stage of ulcer (no) Treatment group; no I II Total n = 83 5.92 ± 6.27 7.78 ± 13.53 31 52 Hydrocolloid n = 28 7.12 ± 5.68 7.47 ± 16.4 12 16 Phenytoin n = 28 6.11 ± 8.4 5.13 ± 3.67 9 19 Simple dressing n = 27 4.47 ± 3.64 10.84 ± 16.32 10 17 P-Value P > 0.20 P > 0.20 P > 0.70 All completely healed ulcer patients were followed up by monthly visits from general practitioners for a further 4 months after the end of the trial. They were also examined by the assessor author. No recurrence of ulceration was observed in any of the trial groups during this period. All patients completed the study and there were no losses to follow up, no treatment withdrawals, no trial group changes and no major adverse events (Fig. 2 ). Discussion Diphenyl hydantoin sodium (phenytoin) is an effective anti-epileptic medication. Its capacity to accelerate ulcer healing was reported more than 40 years ago [ 24 ]. Since then, it has been used topically for different kinds of wounds and ulcers such as war wounds, sores caused by venous stasis, atrophic ulcers and burns, and positive effects have been reported [ 25 - 27 ]. Possible mechanisms of action of phenytoin cream on wound healing are as follows: 1. Decrease in serum corticosteroid; 2. Acceleration of assembly and presence of collagen and fibrin in the ulcer area, and stimulation of alkaline phosphatase secretion [ 28 ]. The use of HD for healing pressure ulcers dates from about 20 years ago. The benefits of this method in comparison with conventional methods include reduction of bacterial contamination, facilitation of patient movement, improvement in patient's psychological condition, more convenience and less pain [ 29 - 34 ]. Hydrocolloid adhesive dressings absorb water and low molecular weight components from ulcer secretions, so they swell to produce a jelly. This jelly protects the ulcer, and new cells proliferate [ 35 ]. Moreover, the jelly stimulates the immune system locally by activating granulocytes, monocytes and the complement system [ 36 ], decreasing the effects of bacterial colonization and ensuring autodebridement of the ulcer [ 1 ]. Bacterial colonization is likely under the HD layer and is responsible for the unpleasant aroma detected when the dressings are changed, but it should not be misinterpreted as clinical infection. In fact, clinical trials of HD on more than 2000 ulcers have shown a much lower incidence of infection than in other treatment methods [ 29 , 30 , 33 ]. Thus, the ulcer dry-out method is not considered as useful as it once was, and the current trend is towards a damp method using HD [ 37 - 41 ]. In this study, the therapeutic effects of HD on gluteal and ischial ulcers were shown to be superior to those of PC and SD. In view of the cost of pressure ulcer management in hospitals and sanitariums and the high expense of plastic surgery [ 42 ], and the psychological problems associated with paralysis and pressure management in SCI victims [ 35 , 43 , 44 ], it seems rational to shift to simpler methods that are more cost efficient and executable by the individual patient [ 31 , 34 , 35 ]. HD treatment of pressure ulcers is less expensive and more comfortable and will ultimately increase the patients' self-confidence [ 8 , 45 ]. These adhesives are available in different sizes and brands convenient for use in ulcers of different parts of body. In the most recent products, the appropriate time for changing the adhesive is indicated by a color conversion. In addition, their transparency makes it easy to observe the ulcer's status without removing the adhesive and dressing [ 46 ]. Although the therapeutic effects of HD on sacral ulcers, in contrast to gluteal and ischial ulcers, did not appear in this survey to be significantly better (p > 0.05) than phenytoin and simple dressings, nor was it less effective. Whether the lesser healing effect of HD on sacral ulcers corresponds to the pressure effects in this area, or to greater bacterial colonization or other factors [ 3 , 4 , 6 , 47 ], needs to be clarified by further studies. Gross differences among the three treatment modalities precluded double blinding. Blinding the authors to the treatment groups minimized this limitation. The major tasks, i.e. defining the study population, enrolling the participants who met the eligibility criteria and assessing the primary and secondary outcomes, were performed blind by the authors. To reduce differences in baseline demographic characteristics among the treatment groups and subgroups and to minimize losses to follow-up, war-related SCI patients were recruited and all the patients who met the eligibility criteria were enrolled in the study. They were all relatively young males (mean age 36.64 ± 6.04 years) and had good motivation to complete the course of treatment. The results of this trial cannot be extrapolated to stage III or stage IV pressure ulcers or to other types of wounds. Furthermore, the small numbers of gluteal and sacral ulcers preclude definitive statements about differences among the treatment subgroups. Conclusion The observed efficacy of HD in the treatment of pressure ulcers suggests that it might be effectively applied to other stage I or stage II pressure ulcers. Competing interests The author(s) declare that they have no competing interests. Authors contributions MTH designed the study and wrote the proposal, and visited all the patients and examined them for eligibility criteria. HKH designed the study, helped in the recruitment of patients, planned the data analyses, assessed the trial groups for primary and secondary outcomes and wrote the paper. FY reviewed the literature, advised on data analysis and contributed to writing the paper. Funding The study was supported by the Jaonbazan Medical and Engineering Research Center, the medical and research section of the official governmental body responsible for SCI war victims. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545970.xml

549192

Effect of urbanization on bone mineral density: A Thai epidemiological study

Background The incidence of fractures in rural populations is lower than in urban populations, although the reason for this difference is unclear. This cross-sectional study was designed to examine the difference in bone mineral density (BMD), a primary predictor of fracture risk, between urban and rural Thai populations. Methods Femoral neck and lumbar spine BMD was measured by dual-energy X-ray absorptiometry (GE Lunar, Madison, WI) in 411 urban and 436 rural subjects (340 men and 507 women), aged between 20 and 84 years. Body mass index (BMI) was calculated from weight and height. Results After adjusting for age and body weight in an analysis of covariance model, femoral neck BMD in rural men and women was significantly higher than those in urban men and women ( P < 0.001), but the difference was not observed at the lumbar spine. After stratifying by sex, age group, and BMI category, the urban-rural difference in femoral neck BMD became more pronounced in men and women aged <50 years and with BMI ≥ 25 kg/m 2 . Conclusions These data suggest that femoral neck BMD in rural men and women was higher than their counterparts in urban areas. This difference could potentially explain part of the urban-rural difference in fracture incidence.

Background Osteoporosis and its ultimate consequence of low traumatic fracture pose a major public health problem, because it incurs significant costs and increased risk of mortality [ 1 - 3 ]. Osteoporosis is sometimes considered a "consequence" of modernization, because the incidence of fractures in urban regions is often higher than in rural regions [ 4 - 10 ], although the underlying reason for this trend is largely unknown. Measurement of bone mineral density (BMD) is considered the primary predictor of fracture risk [ 11 ]. Therefore, it could be hypothesized that the urban-rural difference in fracture incidence is partly explained by the urban-rural difference in BMD. However, such a difference has not been well documented due to limited data available [ 12 - 14 ]. Some previous studies reported that rural subjects had higher BMD or bone mineral content (BMC) than those urban subjects [ 12 , 13 ], but another study found no such difference [ 14 ]. The pace of urbanization in developing countries is more pronounced than in developed countries. Therefore, developing countries are ideal settings for studying the urban-rural difference in BMD. The aim of this study was to examine the difference in BMD between an urban population and a rural population in Thailand. Methods Setting and subjects The present study was designed as a cross-sectional, population-based investigation. The setting was Bangkok city and Khon Kaen province in Thailand. Bangkok is the capital city with a population of 5.7 million and Khon Kaen is a rural province, located 445 km northeast of Bangkok with a population of 1.8 million and is largely an agricultural community. Further details of this study have been described elsewhere [ 15 ] The study included 872 Thai men and women, aged between 20 and 84 years, of whom 422 subjects were from Bangkok and 450 subjects from Khon Kaen. In Khon Kaen, subjects were recruited from 2 villages in the Muang district. There were 14 hamlets in the two villages. In each hamlet, a full list of subjects was obtained, from which 40 subjects were randomly selected by the village's administrator. The selected subjects were then sent a letter of invitation to participating in the study. The response rate was 80.3%. In Bangkok, subjects were recruited via a media campaign, and the sampling technique was similar to the scheme used in Khon Kaen, where subjects were randomly selected from 5 districts within the city of Bangkok. All Khon Kaen subjects were farmers, while Bangkok subjects were office workers, factory workers or house workers. Twenty-one subjects were excluded from analysis because of bone disorders, chronic diseases, history of taking medications that are deemed to affect calcium and bone metabolism, such as the use of steroids or thyroid hormone; and 4 women were excluded on the basis of pregnancy, lactation, delivery or abortion within the previous 3 months, previous history of oophorectomy and premature menopause. The study was conducted in accordance with the Helsinski Declaration in 1975 and as revised in 1983 and was approved by the Ethics Committee of Faculty of Medicine Ramathibodi Hospital Mahidol University (Bangkok) and Khon Kaen University (Khon Kaen), and written informed consent was obtained from all subjects. Measurements BMD at the femoral neck and lumbar spine (L2-4) in g/cm 2 , was measured by dual-energy X-ray absorptiometry with a Lunar DXP-IQ densitometer (GE Lunar Radiation Corp, Madison, WI, USA). The two study sites (Bangkok and Khon Kaen) used the same model of the DXA machine and the same protocol of measurements. The radiation dose with this method is < 0.1 μ Gy. The coefficient of variation of BMD for normal subjects is 0.96 and 0.98 at proximal femur and lumbar spine, respectively. Body weight (including light indoor clothing) was measured using an electronic balance (accuracy 0.1 kg) and standing height (without shoes) with a standiometer (nearest 0.1 cm). Body mass index (BMI) was calculated as ratio of weight (in kg) over height (in meter squared). Statistical analyses Descriptive statistics were computed for each residential region and sex separately. In order to test for difference between urban and rural regions, an analysis of covariance (ANCOVA) model was performed. In this model, BMD was treated as outcome variables; age and weight (or BMI) were treated as covariates; and residence (urban or rural) was the factor. Interactions between age and BMI or age and residence variable were also considered in the model. Estimates of the model parameters were based on the least square method via the SPSS version 9.0 (SPSS, Inc, Chicago). Results Demographic characteristics After excluding 25 subjects, data from 847 subjects (340 men and 507 women) were analysed. There was no significant difference between urban and rural subjects with respect to age or sex distribution. The mean age was 49 and 50 years old in men and women, respectively. However, urban men had higher weight and greater height than rural men ( P < 0.001), whereas urban women had a greater height ( P < 0.001) but equivalent weight ( P = 0.72) compared with rural women (Table 1 ). Table 1 Characteristics of study subjects Urban (Bangkok) Rural (Khon Kean) Mean Difference (95% CI) P value Men Number of Subjects 159 181 Age (years) 49.6 ± 17.5 49.1 ± 17.1 0.5 (-3.2, 4.2) 0.800 Body weight (kg) 64.3 ± 11.1 58.2 ± 8.8 6.1 (3.0, 8.1) <0.001 Height (cm) 165.5 ± 6.3 161.2 ± 5.9 4.3 (3.0, 5.6) <0.001 Body Mass Index (kg/m 2 ) 23.4 ± 3.6 22.4 ± 2.8 1.0 (0.3, 1.7) 0.003 Bone Mineral Density (g/cm 2 ) Femoral neck 0.87 ± 0.16 0.96 ± 0.18 -0.09 (-0.13, 0.05) <0.001 Lumbar spine 1.12 ± 0.17 1.11 ± 0.16 0.01 (-0.03, 0.04) 0.64 Women Number of Subjects 252 255 Age (years) 50.4 ± 15.1 50.6 ± 15.9 -0.2 (-2.9, 2.4) 0.853 Body weight (kg) 55.5 ± 8.9 55.9 ± 10.5 -0.4 (-2.0, 1.4) 0.718 Height (cm) 154.7 ± 5.4 152.1 ± 5.2 2.6 (1.6, 3.5) <0.001 Body Mass Index (kg/m 2 ) 23.2 ± 3.8 24.1 ± 4.0 -0.9 (-1.5, -0.1) 0.017 Bone Mineral Density (g/cm 2 ) Femoral neck 0.79 ± 0.13 0.87 ± 0.19 -0.08 (-0.11, -0.05) <0.001 Lumbar spine 1.05 ± 0.18 1.01 ± 0.21 0.04 (-0.11, 0.08) 0.16 All values are shown in mean ± standard deviation (SD). In the entire sample, higher weight was associated with higher BMD in men ( r = 0.13, P = 0.017 for femoral neck, and r = 0.37, P < 0.001 for lumbar spine) and in women ( r = 0.33, P < 0.001 for femoral neck, and r = 0.33, P < 0.001 for lumbar spine). On the other hand, advancing age was associated with a significant reduced BMD in men ( r = -0.53, P < 0.001 for femoral neck, and r = -0.15, P = 0.007 for lumbar spine) and women ( r = -0.63, P < 0.001 for femoral neck, and r = -0.60, P < 0.001 for lumbar spine). However, the strength of relationship between age and BMD in urban subjects was less pronounced than in rural subjects, such that rural women had a higher cross-sectional "rate of bone loss" than urban women, particularly at the femoral neck. For example, in women, each 5-year increase in age was estimated to associate with a 2.1% and 1.2% decrease in femoral neck BMD for rural and urban group, respectively; in men, the respective rate of decrease was 1.3% and 0.8%. As a result, among those aged 50 + years, BMD in rural subjects tended to be lower than (or converged to) BMD in urban subjects (Figure). Figure 1 Interaction effects of age and residence variable on bone mineral density at the femoral neck in men (A) and women (C), and at the lumbar spine in men (B) and women (D). Urban-rural difference in BMD In both sexes, after adjusting for age and weight, BMD in rural individuals was significantly higher than in urban individuals. For instance, femoral neck BMD in rural men and women was 0.22 and 0.23 g/cm 2 significantly higher ( P < 0.001) than in urban men and women, respectively; but the difference was lower for the lumbar spine BMD (0.12 g/cm 2 in men, P = 0.017 and 0.05 g/cm 2 in women, P = 0.293). The statistical significance of the age-by-residence interaction term in the ANCOVA model suggested that the urban-rural difference in BMD decreased with advancing age (Table 2 ). Table 2 Effects of age, weight and residence on bone mineral density: estimates of parameters of the analysis of covariance stratified by sex and BMD site Effect Estimate ± SE P value Men Femoral neck BMD Age (+5 yr) -0.020 ± 0.003 <0.001 Weight (+5 kg) 0.015 ± 0.004 <0.001 Residence (Rural) 0.222 ± 0.046 <0.001 Age × Residence (Rural) -0.012 ± 0.004 0.008 Lumbar spine BMD Age (+5 yr) -0.001 ± 0.003 0.874 Weight (+5 kg) 0.030 ± 0.004 <0.001 Residence (Rural) 0.122 ± 0.051 0.017 Age × Residence (Rural) -0.010 ± 0.004 0.048 Women Femoral neck BMD Age (+5 yr) -0.026 ± 0.002 <0.001 Weight (+5 kg) 0.026 ± 0.003 <0.001 Residence (Rural) 0.233 ± 0.034 <0.001 Age × Residence (Rural) -0.015 ± 0.003 <0.001 Lumbar spine BMD Age (+5 yr) -0.033 ± 0.003 <0.001 Weight (+5 kg) 0.031 ± 0.003 <0.001 Residence (Rural) 0.047 ± 0.045 0.293 Age × Residence (Rural) -0.009 ± 0.004 0.036 SE, Standard Error. Notes: Since height was not a significant factor in the analysis of covariance model, it was removed from the final model with no significant change of the results. Further analyses stratified by sex, age group, and BMI category indicated that the urban-rural difference in femoral neck BMD was more pronounced in the younger age group (< 50 years old) and higher BMI (≥ 25 kg/m 2 ). This trend was consistent for men and women. However, for lumbar spine BMD, no significant urban-rural difference was observed in most subgroups, with the exception of women aged ≥ 50 years and BMI < 25 kg/m 2 in whom BMD was lower in the rural group compared to the urban group (Table 3 ). Table 3 Bone mineral density in urban and rural men and women by age group and body mass index (a) Lumbar spine BMD Sex Age (years) BMI (kg/m 2 ) Bone mineral density (g/cm 2 ) Urban Rural Mean difference and 95% CI Men < 50 < 25 0.94 ± 0.16 1.05 ± 0.16 -0.11 a (-0.16, -0.06) ≥ 25 0.89 ± 0.12 1.04 ± 0.17 -0.15 b (-0.26, -0.03) ≥ 50 <25 0.80 ± 0.15 0.86 ± 0.15 -0.06 b (-0.12, -0.01) ≥ 25 0.83 ± 0.11 0.92 ± 0.15 -0.09 b (-0.17, -0.01) Women < 50 < 25 0.85 ± 0.11 0.97 ± 0.13 -0.12 a (-0.16, -0.09) ≥ 25 0.92 ± 0.11 1.04 ± 0.13 -0.12 b (-0.20, -0.05) ≥ 50 <25 0.70 ± 0.11 0.71 ± 0.15 -0.01 (-0.05, 0.03) ≥ 25 0.76 ± 0.11 0.82 ± 0.15 -0.06 b (-0.11, -0.01) (b) Femoral neck BMD Sex Age (years) BMI (kg/m 2 ) Bone mineral density (g/cm 2 ) Urban Rural Mean difference and 95% CI Men < 50 < 25 1.12 ± 0.14 1.13 ± 0.15 -0.01 (-0.06, 0.04) ≥ 25 1.11 ± 0.09 1.14 ± 0.16 -0.03 (-0.13, 0.08) ≥ 50 < 25 1.05 ± 0.18 1.06 ± 0.16 -0.01 (-0.07, 0.05) ≥ 25 1.21 ± 0.19 1.19 ± 0.20 0.02 (-0.11, 0.14) Women < 50 < 25 1.14 ± 0.13 1.13 ± 0.14 0.01 (-0.04, 0.04) ≥ 25 1.21 ± 0.15 1.15 ± 0.13 0.06 (-0.02, 0.15) ≥ 50 < 25 0.92 ± 0.14 0.83 ± 0.19 0.09 b (0.03, 0.14) ≥ 25 1.02 ± 0.17 0.97 ± 0.21 0.05 (-0.01, 0.13) Statistical significance at a P < 0.001 and b P < 0.05. Statistical significance is indicated by bold-faced letters. Discussion Osteoporosis has emerged as one of the most common diseases in the aged population, and represents one of the most significant public health problems in Asia [ 2 , 16 , 17 ]. A consistent trend in osteoporosis is that the incidence of fracture is higher in developed countries than in developing countries; and in any country, the incidence is higher in urban than in rural communities [ 5 - 11 ]. While many factors are hypothesized to be responsible for this trend, BMD is thought to be a primary determinant, because it is the most consistent and robust predictor of fracture risk [ 1 , 11 ]. In the present population-based study, we have shown that BMD in a rural Thai population was significantly higher than in urban population, particularly at femoral neck. The magnitude of difference was more than 1 standard deviation which is clinically relevant. It is difficult to compare the present study's results to previous studies' due to differences in methodology and study design. For instance, Sundberg et al [ 12 ] reported that lumbar spine BMD (measured by DXA) in rural adolescents was significantly higher than that in urban adolescents, but there was no significant difference in femoral neck BMD. Furthermore, a study from Southern Sweden suggested that bone mass at the forearm (measured by single-photon absorptiomety) in rural population was significantly higher than in urban population and the difference was more pronounced when comparing a true urban population who had lived their entire life in a city with a true rural population who had never lived in a city [ 13 ]. A study from Eastern Poland found that the mean lumbar spine BMD values in every age range were higher in rural population than in urban population, but the difference was not statistically significant [ 14 ]. Taken together, these results including ours, suggest that rural subjects tend to have higher BMD than in urban subjects. The present study's data and design can not elucidate any underlying factors that are responsible for the difference but some propositions could be put forward. The urban-rural difference in femoral neck BMD could be due to the difference in the peak of bone mass levels. In this study, both rural men and women aged between 20 and 30 years had significantly higher BMD than urban counterparts. For example, young rural men and women had significantly higher than urban subjects (1.17 vs. 1.03 g/cm 2 , [95% CI: 0.07–0.22] in men and 1.02 vs. 0.86 g/cm 2 , [95% CI: 0.10–0.22] in women). This finding was partially consistent with a previous study [ 12 ] and could be explain the fact that the urban-rural difference was mainly found in younger age groups. This study also found that the urban-rural difference in femoral neck BMD decreased with advancing age. The difference may be attributed to the difference in physical activity between the two populations. Rural populations were generally more physically active than urban populations [ 18 , 19 ]. The rural population in this study was mainly farmers who spend most of their time in rice field long hours of physical activity. However, the difference was sex- and site- dependent. The difference in femoral neck BMD was much more pronounced than that in lumbar spine and this was more transparent before the age of 50 in men and before the menopause in women. After this age the difference was no longer significant. The data suggested that the rate of bone loss in rural population may be more rapid than in urban population. However, this finding was not consistent with a previous study which demonstrated that the rate of bone loss was higher in urban population compared with rural population [ 13 ]. The reason(s) for the higher rate of bone loss in rural population in this study is unknown, but low dietary calcium intake could be a contributory factor [ 20 - 22 ]. The present findings must be interpreted in the context of a number of potential strengths and weaknesses. The data were obtained from a large and well-defined rural vs. urban area, which allowed the rural and urban difference to be reliably delineated. The study subjects were Thai, among whom, cultural backgrounds and environmental living conditions are different from Western populations. Thus care should be taken when extrapolating these results to other populations. Conclusions These data have demonstrated that femoral neck BMD in rural men and women was higher than their counterparts in urban areas. This difference could potentially explain part of the urban-rural difference in fracture incidence. List of abbreviations All abbreviations are defined in the text. Competing interests The author(s) declare that they have no competing interests. Authors' contributions Chatlert Pongchaiyakul had an active role in the conduct of this study, obtained and analysed data, and drafted the manuscript. Tuan V Nguyen was involved in the conceptual discussion of this study, and had an active role in data analysis, drafting of the manuscript. Vongsvat Kosulwat, Nipa Rojroongwasinkul, and Somsri Charoenkiatkul had an active role in the study design, and was involved in the conceptual discussion. Rajata Rajatanavin conducted and established this study. All authors contributed to the last version of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549192.xml

517950

Co-administration of a DNA vaccine encoding the prostate specific membrane antigen and CpG oligodeoxynucleotides suppresses tumor growth

Background Prostate-specific membrane antigen (PSMA) is a well characterized prostate-specific tumor associated antigen. Its expression is elevated in prostate carcinoma, particularly in metastatic and recurrent lesions. These observations suggest that PSMA can be used as immune target to induce tumor cell-specific recognition by the host and, consequently tumor rejection. We utilized a DNA-based vaccine to specifically enhance PSMA expression. An immune modulator, such as CpG oligodeoxynucleotides which promote Th1-type immune responses was combined to increase the efficacy of tumor recognition and elimination. Methods A eukaryotic expression plasmid pCDNA3.1- PSMA encoding full-length PSMA was constructed. C57BL/6 mice were immunized with endotoxin-free pCDNA3.1- PSMA alone or in combination with CpG oligodeoxynucleotides by intramuscular injection. After 4 immunizations, PSMA specific antibodies and cytotoxic T lymphocyte reactivity were measured. Immunized C57BL/6 mice were also challenged subcutaneously with B16 cells transfected with PSMA to evaluate suppression of tumor growth. Results Vaccine-specific cytotoxic T lymphocytes reactive with B16 cells expressing PSMA could be induced with this treatment schedule. Immune protection was observed in vaccinated mice as indicated by increased tumor growth in the control group (100%) compared with the groups vaccinated with DNA alone (66.7%) or DNA plus CpG oligodeoxynucleotides (50%) respectively. Average tumor volume was smaller in vaccinated groups and tumor-free survival time was prolonged by the vaccination. Conclusion The current findings suggest that specific anti-tumor immune response can be induced by DNA vaccines expressing PSMA. In addition, the suppression of in vivo growth of tumor cells expressing PSMA was augmented by CpG oligodeoxynucleotides. This strategy may provide a new venue for the treatment of carcinoma of prostate after failure of standard therapy.

Background Carcinoma of prostate is the most common cancer in males in America, ranking as the second most common leading cause of cancer-related deaths, just after carcinoma of the lung. In addition, the incidence and mortality of carcinoma of prostate are increasing in China. Although surgery and radiation therapy remain the primary choice for localized stage of carcinoma of prostate, there is no effective treatment for patients who develop recurrences or those who have metastatic disease at the time of diagnosis. Therefore, there is an urgent need for new types of treatment. Strategies that stimulate the ability of the immune system to recognize and destroy cancer cells via selective killing mechanisms have shown promise in the treatment of cancer. DNA vaccines offer several potential advantages for the immunotherapy of cancer. Proteins encoded by DNA vaccines are expressed in the cytoplasm and presented through the endogenous processing pathway associated with MHC Class I molecules, thereafter leading to the activation of CD8 + cytotoxic T lymphocytes (CTL) [ 1 , 2 ], which act as effectors in the anti-tumor immune response. DNA vaccines are cost-effective since DNA is relatively simple to purify in a large quantities. Another intrinsic advantage consists in the presence in the plasmid itself of un-methylated CpG motifs (immunostimulatory sequences) that may act as a potent immunological adjuvant [ 3 ]. Thus, there is a good rationale for further development of DNA vaccines to immunize against antigens present on cancer cells. Prostate specific membrane antigen (PSMA), a well-established prostate specific tumor associated antigen (TAA), is 100 kD type II transmembrane glucoprotein. It is predominantly expressed in the prostate gland, minimal levels of expression in brain tissue, jejunum and proximal kidney tubules [ 4 , 5 ]. Its expression is significantly elevated in carcinoma of prostate, particularly in metastastic disease and recurrent disease after hormone therapy fails [ 6 , 7 ]. These properties of PSMA propose it as an ideal target of anti-cancer vaccines. A number of strategies are under evaluation to enhance the potency of DNA vaccines, some of which involves broad stimulation of the immune system using immunomodulatory agents. Synthetic CpG oligodeoxynucleotides have immunological effects similar to those seen with bacterial DNA and represent promising vaccine adjuvants, which promote T helper1 (Th1)-type immune responses [ 8 ]. Unmethylated CpG motifs are present at a much higher frequency in the genome of prokaryotes than eukaryotes. The release of unmethylated CpG DNA during an infection provides a 'danger signal' to the innate immune system, triggering a protective immune response that improves the ability of the host to eliminate the pathogen [ 9 ]. CpG oligodeoxynucleotides up-taken by B cells and plasmacytoid dentritic cells (pDCs), which express Toll-like receptror 9 (TLR9) [ 10 , 11 ] initiate an immune stimulatory cascade that culminates in the indirect maturation, differentiation and proliferation of T cells and natural killer (NK) cells[ 12 , 13 ]. Together, these cells secrete cytokines and chemokines that create a pro-inflammatory (IL-1, IL-6, IL-18 and tumor necrosis factor-α) and Th1-polarized (interferon-γ, and IL-12) immune milieu [ 14 ], which further facilitates the development of antigen-specific CTLs [ 15 - 17 ]. These effects indicate that CpG oligodeoxynucleotides could act as vaccine adjuvant. The present study was designed to test the therapeutic efficacy of a PSMA-based DNA vaccine in a mouse model of tumor cell implants expressing PSMA. In addition, the adjuvant role of CpG oligodeoxynucleotides to augment the potency of the constructed DNA vaccine was tested. Methods Mice and Cell lines C57BL/6 mice (H-2b) were bred and kept under pathogen-free conditions. Male mice were used at 12 to 16 weeks of age. All animal experiments were performed in an approved protocol and in accordance with recommendations for the proper care and use of laboratory animals. The murine melanoma B16 cell was purchased from the Type Culture Collection of the Chinese Academy of Sciences and cultured in RPMI-1640 medium (Life Technologies, Gaithersburg, MD) supplemented with 10% (v/v) heat-inactivated fetal bovine serum (Hyclone, Logan, UT), penicillin G (100 U/ml), and streptomycin (100 μg/ml). The COS-7 cell line was cultured in DMEM medium (Life Technologies) supplemented with 10% (v/v) heat-inactivated fetal bovine serum, sodium pyruvate (1 mM), penicillin (100 U/ml) and streptomycin (100 μg/ml). Antibodies and reagents The monoclonal antibody 4A12 specific for an extra-cellular epitope of PSMA was previously described [ 18 ]. The β-actin-specific monoclonal antibody was purchased from Sigma Chemical Co (St. Louis, MO). FITC-labeled goat anti-mouse IgG was purchased from Santa Cruz Biotech (Santa Cruz, CA). Recombinant murine IL-2 was purchased from PeproTech (Rocky Hill, NJ). ConA and mitomycin C were purchased from Sigma Chemical Co. CpG oligodeoxynucleotides 1826 chosen according to published data [ 19 - 21 ] had the following sequence TTCAT GACGTT CCT GACGTT (CpG motifs were shown underlined) with the backbone phosphorothioate stabilized. CpG oligodeoxynucleotides were synthesized by Sangon (Shanghai, China), reconstituted in sterile pyrogen-free water and diluted in phosphate buffered saline for in vivo injections. Transient transfection of COS7 cells The eukaryotic expression plasmid pCDNA3.1- PSMA encoding full-length PSMA was constructed by cloning the BamH I /Xho I fragment of pBluescipt- PSMA as described previously [ 18 ] into the pCDNA3.1 vector (Invitrogen Corp, Carlsbad, CA) cut with identical endonuclease. COS-7 cells were transfected with pCDNA3.1- PSMA or an empty (mock) vector by the mediation of liposome Tfx-20™ (Promega, Madison, WI) according to the manufacturer's instructions. Briefly, COS-7 cells were cultured in six-well tissue culture plates with a coverslip in each well and grown to 50–70% confluence. 1.5 μg plasmid DNA was mixed with 4.5 μl Tfx-20™ and diluted in 1000 μl serum-free DMEM medium before addition to cells. After 20 minutes of incubation, DNA-liposome complex was added to the cells and incubated for 6 hours at 5% CO2, 37°C. Complete DMEM medium containing 10% fetal bovine serum was added to the cells and incubated overnight, and then the medium was replaced with complete DMEM medium. After 2 days, the cells were fixed in cold acetone for 10 minutes at 4°C followed by extensive washing with phosphate buffered saline. The cells were incubated with anti-PSMA monoclonal antibody 4A12 for 1 hour at 37°C, subsequently incubated with FITC-labeled goat anti-mouse IgG (1:40) for 1 hour at 37°C. After thorough washing, the coverslips were mounted and observed with a fluorescence microscope. Stable transfection of B16 melanoma cells with PSMA plasmid The B16 murine melanoma cells were transfected with 2 μg of pCDNA3.1- PSMA or empty vector by the mediation of 6 μl liposome Tfx-20™ as above. After 2 days of culture, the cells were reseeded into a 10 cm-dish and cultured for other 2 days, complete RPMI-1640 medium containing 1000 μg/ml G418 (Life Technologies) was added to the culture. After 20 days of selection, all non-transfected cells died and discrete clones were visible in transfected cells. These clones were expanded in the presence of 400 μg/ml G418, positive cells expressing PSMA were identified as follows. Detection of PSMA mRNA by Reverse Transcriptase-PCR Total RNA was extracted from mock-transfected or transfected B16 cells using Trizol (Life Technologies) and dissolved in RNase free water. 2 μg of total RNA was transcribed into cDNA using AMV reverse transcriptase (Promega). Briefly, the total RNA was mixed with 1 μl oligo (dT) primers (0.1 μg/ μl), 4 μl RT Buffer (5×), 2 μl dNTPs (10 mM), 1 μl AMV reverse transcriptase and diethypyrocarbonate-treated water to a final volume of 20 μl. The cDNA synthesis was performed using the following PCR parameters: 37°C for 1 hour then 10 minutes at 95°C. Synthesized cDNA was used as template for PCR. The sequence of the primers used were 5'- CGAGGAGGG ATGGTG TT-3' (forward) and 5'-TGTTGTGGCTGCTTGAG-3' (reverse). PCR was carried out in a 10 μl aliquot containing 0.5 μl cDNA, 0.5 μl each primer (10 μM), 1 μl dNTPs (2 mM), 1 μl Taq buffer (10×), 0.8 μl MgCl 2 (25 mM), 1 unit of Taq. The PCR reaction conditions included 5 minutes of initial denaturation at 94°C followed by 30 cycles of 30 seconds at 94°C, 1 minute at 62°C, 1 minute at 72°C and 10 minutes of final extension at 72°C. The 358 bp fragment was resolved on 2% agarose gel. GAPDH was also detected as internal reference. Detection of PSMA protein by Western Blot The transfected and mock-transfected B16 cells were harvested and lysed with lysis buffer (50 mM NaCl, 0.01 M Tris-Cl (pH8.0), 5 mM EDTA, 0.5% NP-40, 1 mM phenylmethylsulfonyl fluoride, 1 μg/mL aprotinin, 1 μg/mL leupeptin) for 30 min at 4°C, cell debris were removed by centrifugation. Cell lysates were heated at 100°C for 3 minutes, the samples were loaded on 6% SDS-PAGE for electrophoresis. After electrophoresis, the proteins were transferred to polyvinylidene difluoride membranes (Amersham Pharmacia Biotech, Piscataway, NJ) using semi-humid transferring system (Bio-Rad, Hercules, CA). The polyvinylidene difluoride membranes were blocked with Tris-buffered solution containing 5% (w/v) non-fat milk for 1 hour at room temperature. For detection of protein, the polyvinylidene difluoride membranes were probed with anti-PSMA monoclonal antibody 4A12 and monoclonal antibody to β-actin (1:50) respectively for 1 hour at room temperature then overnight at 4°C, after then the membranes were incubated with horse anti-mouse IgG-HRP conjugate (1:500, Vector, Burlingame, CA) for 1 hour at 37°C, ABC complex (1:500, Vector) for 1 hour at 37°C subsequently. The bands were visualized with 3,3'-diaminobenzidine substrate solution (5 mg diaminobenzidine dissolved in 10 ml Tris-buffered solution, 10 μl 30% hydrogen peroxide). DNA vaccination of C57BL/6 mice Plasmids pCDNA3.1- PSMA and pCDNA3.1 were purified with EndoFree plasmid Maxi Kit (Qiagen, Valencia, CA). Three groups including 6 mice each were immunized: DNA vaccination group, CpG oligodeoxynucleotides and DNA vaccine co-administration group (hereafter referred to as CpG+DNA vaccination) and control group receiving empty plasmid. All mice were injected with 0.25% lidocaine in the quadriceps femoris muscle 3 days before vaccination in order to improve the uptaking of plasmids by muscles. The mice then received bilateral intramuscular injection with 50 μg of plasmid in the regenerating muscles. Mice in the DNA vaccination group were immunized with endotoxin-free pCDNA3.1 -PSMA , mice in CpG+DNA vaccination group were further immunized with 25 μg of CpG oligodeoxynucleotides in the same location 3 days after DNA plasmid immunization, as control, mice were injected with pCDNA3.1 plasmid. All mice were boosted every 4 weeks for 3 times. Measurement of PSMA specific serum antibodies Two weeks after the last immunization, the mice were bled and serum antibodies were measured by solid phase enzyme-linked immunosorbent assay (ELISA). Briefly, bacterially expressed fusion protein containing PSMA-derived fragment was coated on 96-well plates. The plates were blocked with 5% bovine serum albumin in phosphate buffered saline overnight at 4°C. The sera from C57BL/6 mice were serially diluted in phosphate buffered saline with 5% bovine serum albumin and then 100 μl of diluted serum was added into each well. The plates were incubated at 37°C for 1 hour, 100 μl of a 1:3000 dilution of goat anti-mouse IgG-HRP conjugate (Jackson ImmunoResearch, West Grove, PE) was added into each well and incubated for 1 hour at 37°C, then 100 μl tetramethyl benzidine (TMB) chromagen/substrate solution (0.1 mg/ml TMB, 0.1 M citric acid buffer pH6.0, 4 μl 30% hydrogen peroxide per 10 ml) was added to each well. The plates were read and the absorbance at 450 nm (A450) was measured by microplate reader. Cytotoxic T Lymphocyte (CTL) Assay Two weeks after the last immunization, mice were sacrificed. Their spleens were removed and teased apart in serum-free RPMI-1640 media, the lymphocytes were collected and cultured in RPMI-1640 medium supplemented with 10% heat-inactivated fetal bovine serum, 50 IU/ml recombinant murine IL-2 and 2 μg/ml ConA for 2 days. B16 stimulator cells expressing PSMA (B16-PSMA) were prepared in complete RPMI-1640 medium containing 50 μg/ml mitomycin C for 1 hour at 37°C. Two × 10 7 lymphocytes (responders) were incubated in 6-well plates with 2 × 10 6 stimulator cells in the presence of 50 IU/ml recombinant murine IL-2 and 2 μg/ml ConA. After 6 days of culture, the re-stimulated cells were harvested and separated from the dead cells. Target cells (B16-PSMA) and re-stimulated lymphocytes (effector cells) were resuspended in phenol red-free RPMI-1640 medium supplemented with 5% new born calf serum, 5 × 10 3 target cells and various number of effector cells were added into individual flat-bottom wells in 96-well plates. The cells were incubated at 37°C overnight. 50 μl per well of supernatant was transferred to fresh 96-well plates. CTL reactivity was assessed by measuring lactate dehydrogenase (LDH) release using a Cytotox 96 assay kit (Promega). Controls were setup on each plate for spontaneous LDH release by target and effector cells. A parallel experiment using B16 cells transfected with pCDNA3.1 as target cells was also performed to test the specificity of lysis. All experiments were performed in triplicate. Percent lysis was calculated according to manufacturer's instructions. Subcutaneous transplantation of tumor cells C57BL/6 mice were divided into 3 groups and immunized as above. B16-PSMA cells were trypsinized and resuspended in phosphate buffered saline, 2 × 10 5 cells were subcutaneously injected into the left lateral flank of mice. The time to the development of tumor was recorded. After tumors became detectable, their volume was measured two-dimensionally with a caliper along the longest axis (x) and the axis perpendicular to the longest axis (y) every second days. The volume of tumors was estimated by the following formula: Volume = π/6 × x × y 2 After 26 days, when tumor reached 20 mm in their largest axis, the mice bearing tumors were sacrificed. Tumors were removed and weighed. Data Analysis The data from ELISA and CTL assays are expressed as means ± SD and are representative of at least three different experiments. Comparisons between individual data points were made using ANOVA or student's t-test. In the tumor challenge experiment, the primary endpoint was time of tumor appearance. Tumor-free survival time was compared by the Kaplan-Meier method and log-rank statistic. P < 0.05 were considered significant. Results Expression of plasmid pCDNA3.1-PSMA in COS7 cells To confirm the expression of PSMA in mammalian cells, plasmid pCDNA3.1 -PSMA was introduced into COS7 cells. The cells were the incubated with anti-PSMA monoclonal antibody and goat anti-mouse IgG-FITC conjugate. The immunofluorescence assay demonstrated that the reactivity was present in the cytoplasm of COS7 cells transfected with pCDNA3.1- PSMA but not in mock-transfected cells, thus indicating that pCDNA3.1 -PSMA could express protein in mammalian cells (Figure 1 ). We considered that plasmid pCDNA3.1 containing the Simian virus 40 (SV40) origin of replication were rapidly amplified in COS7 cells, which constitutively express SV40 large T antigen (T-Ag), so pCDNA3.1- PSMA underwent multiple rounds of duplication within one cell generation. A large amount of protein therefore was expressed and could not be delivered completely through the transporting machinery to cytomembrane, this may explain why the reactivity was found in cytoplasm. Figure 1 Immunofluorescence staining of COS7 cells. The COS7 cells were transfected with either pCDNA3.1- PSMA or empty pCDNA3.1 and fixed with cold aceton. Fixed cells were incubated with a anti-PSMA monoclonal antibody 4A12, stained with a goat anti-mouse immunoglobulin-FITC conjugate, PSMA immunoreactive cells were visualized with a fluorescent microscope (×40). Cytoplasmatic reactivity was found in COS7 cells transfected with pCDNA3.1- PSMA (A) but not in COS7 cells transfected with pCDNA3.1 (B). Detection of PSMA mRNA The transfected and mock-transfected B16 murine melanoma cells were selected by G418 and, after 3 weeks of selection, clones resistant to G418 were obtained. Total RNA was extracted and reverse-transcribed into cDNA to be used as template for PCR detection of PSMA mRNA. A 358 bp band was present in 3 B16 clones transfected with pCDNA3.1- PSMA but not in clones transfected with pcDNA3.1 (Figure 2 ). GAPDH were detected in all samples (data not shown). Figure 2 Detection of PSMA mRNA by RT-PCR. The murine melanoma B16 cells were transfected with either pCDNA3.1- PSMA or empty pCDNA3.1 and selected by G418. After 20 days of selection, clones resistant to G418 were acquired. Total RNA was extracted, and PSMA mRNA was detected by RT-PCR. 3 clones of B16 cells transfected with pCDNA3.1- PSMA were positive for PSMA mRNA, while B16 cells transfected with pCDNA3.1 were negative. Lanes 1, 2 and 3, B16 clones transfected with pCDNA3.1- PSMA ; lane 4, B16 clones transfected with pCDNA3.1. Detection of PSMA protein The lysates from 3 clones with detectable PSMA mRNA were resolved by 6% SDS-PAGE followed by immunoblotting. A predicted 100 kD band was identified in the 3 clones but not in mock-transfected B16 cells (Figure 3 ). One of the clones was designated as B16-PSMA and used as target cells for cytotoxic T lymphocytes (CTL) assay or in tumor challenge experiment. A B16 cell line transfected with pCDNA3.1 (referred to as B16-pCDNA) was also obtained and used as negative control in CTL analysis. Figure 3 Detection of PSMA protein by Western Blot. Total cell lysates were harvested and presence of PSMA protein was detected by anti-PSMA monoclonal antibody 4A12. A 100 kD band was identified in 3 clones with detectable PSMA mRNA but not in B16 cells transfected with empty vector (A). Lane1, 2 and 3, B16 cells transfected with pCDNA3.1- PSMA . Lane 4 B16 cells transfected with pCDNA3.1. β-actin was used as reference (B). Measurement of PSMA specific serum antibodies ELISA was used to measure PSMA specific serum antibodies in C57BL/6 mice. All mice immunized with pCDNA3.1- PSMA generated low titers of antibodies, while PSMA specific antibodies were not detected in control group. When sera were diluted at 1:10, 1:20,1:40,1:80,1:160, the differences between control group and other two groups were all statistically significant ( P < 0.001). However, titers were similar between the DNA vaccine and the CpG+DNA vaccination groups ( P > 0.05), suggesting that CpG oligodeoxynucleotides did not augment the antigen-specific humoral immunity (Figure 4 ). The experiment was repeated 3 times with sera from independently immunized mice yielding comparable results. Figure 4 Measurement of PSMA specific serum antibodies. Mice were immunized 4 times at 4 weeks intervals by intramuscular injection with pCDNA3.1- PSMA or pCDNA3.1, antibody titers were measured by ELISA. Sera were serially diluted and measured individually. The experiment was performed in triplicate. Shown is the average antibody titer (n = 6) with standard errors. P valure was calculated by ANOVA. Antibody titer was similar in DNA vaccination group and CpG + DNA vaccination group, no antibody was detected in control group. * indicated the difference between control group with other two groups was statistically significant ( P < 0.05). Cytotoxic T Lymphocyte (CTL) Assay Splenocytes from C57BL/6 mice were re-stimulated for 6 days with mitomycin C-treated B16-PSMA cells. Cytotoxicity was measured by LDH release from attacked B16-PSMA cells or B16-pCDNA cells. Splenocytes from mice in the DNA vaccination group exhibited specific lysis against B16-PSMA, whereas those from mice in the control group did not acquire killing activity. The differences between control group and the other two groups were statistically significant at E:T ratios of 40:1, 20:1, 10:1 ( P < 0.01). More importantly, CTL reactivity was significantly enhanced in mice treated with CpG oligodeoxynucleotides compared with DNA vaccine alone at E:T ratios of 40:1, 20:1 (t = 9.737, P < 0.001; t = 2.14, P = 0.021 respectively) (Figure 5A ). However, specific lysis was not observed in all groups when B16-pCDNA cells were used as target cells (Figure 5B ). Figure 5 Cytotoxic T Lymphocyte (CTL) Assay. Splenocytes from C57BL/6 mice were re-stimulated for 6 days with mitomycin C-treated B16-PSMA cells. Cytotoxicity was measured by LDH release assay. The experiment was carried out in triplicate. Shown is the average CTL (n = 6) with standard errors. When B16-PSMA cells were used as target cells, specific lysis was found in DNA vaccination group but not in control group. The CTL reaction was enhanced by CpG oligodeoxynucleotides(Fig. 5A). However, when the mock-transfected B16-pCDNA cells were used as target cells, specific lysis was not observed in all groups (Fig. 5B). * indicated the difference between control group with other two groups was statistically significant ( P < 0.05); △ indicated the difference between DNA vaccination group and CpG + DNA vaccination group was statistically significant ( P < 0.05). Suppression of Tumor Growth in Tumor-bearing Mice by pCDNA3.1- PSMA C57BL/6 mice (n= 6/group) were vaccinated with pCDNA3.1- PSMA or empty vector, and then challenged with B16-PSMA cells. Protection was observed in pCDNA3.1- PSMA vaccinated mice with decrease of tumor incidence. After 26 days, all mice in the control group developed tumors while 2 (2/6) and 3 tumor-free mice (3/6) were observed in the DNA and in the CpG + DNA vaccination groups respectively. Kaplan-Meier curves showed that the tumor-free survival interval was 19.67 ± 2.24 days in the DNA vaccination group, 22.33 ± 1.61 days in the CpG +DNA vaccination group and 13.17 ± 1.01 in the control group (Fig. 6A ). The difference between DNA vaccination group and control group was statistically significant ( P = 0.0161), so was the difference between CpG+DNA vaccination group and control group ( P = 0.0016). Tumor-free survival time was longer in CpG+DNA vaccination group than DNA vaccination group, but the difference was of no statistical significance ( P = 0.49). This observation may be associated with the small number in each group. Figure 6 Suppressive Effects of DNA vaccine to Tumor Growth. C57BL/6 mice (n= 6/group) were vaccinated with pCDNA3.1- PSMA or empty vector, and then challenged with B16-PSMA cells. Kaplan-Meier curves showed the tumor-free survival time of mice was 19.67 ± 2.24 days in DNA vaccine group, 22.33 ± 1.61 days in CpG + DNA and 13.17 ± 1.01 days in control group respectively (Fig. 6A). Some mice still developed tumors after pCDNA3.1- PSMA and CpG oligodeoxynucleotides vaccination, but the individual tumors grew much more slowly than those in control group (Fig. 6B). Two mice in control group died before the termination of the experiment, all mice bearing tumor were sacrificed and tumor tissues were removed, the volumes of tumors in control group were larger than those in other two groups (Fig. 6C). Although mice developed tumors after pCDNA3.1- PSMA vaccination, the individual tumors were consistently smaller than those in the control group and the analysis of tumor growth kinetics indicated that the tumor growth was significantly slower in the CpG +DNA vaccination group compared to the other two groups (Fig. 6B ). The volume of the tumors in the control group was consistently larger than in other two groups (Fig. 6C ) and the average tumor weight was 2.28 ± 0.51 g in the DNA vaccine group, 1.10 ± 0.70 g in CpG + DNA group and 4.75 ± 0.66 g in the control group. The differences between control group and DNA vaccination group, CpG +DNA vaccination group were statistically significant (t = 5.92, P = 0.001; t = 7.062, P = 0.001 respectively). Moreover, the difference between DNA vaccination group and CpG +DNA vaccination group was statistically significant (t = 2.588, P = 0.049). Discussion Although treatments are available for organ-confined carcinoma of prostate, there is no effective approach to treat recurrent disease after androgen deprivation therapy fails. New approaches are required to treat this incurable disease. DNA vaccination enables maintenance of tumor antigen expression at the vaccination site and results in immune responses in the host, therefore, shedding light on the treatment of cancer. It has been reported that tumor growth is suppressed when tumor cells are implanted in mice previously immunized with DNA vaccines encoding tumor antigens [ 22 - 25 ]. PSMA is a well-defined prostate-restricted tumor associated antigen whose expression is significantly elevated in carcinoma of prostate, especially in advanced stages. The expression of PSMA is down-regulated by androgen, after androgen deprivation therapy, its expression is strongly elevated[ 6 , 26 - 28 ], Thus, PSMA is a potential target for the immunotherapy to carcinoma of prostate. Several PSMA-based vaccines had been developed and it has been observed in a phase II trial utilizing MHC Class I-restricted peptides that PSMA can induce immune responses in patients with advanced carcinoma of prostate and alleviate the disease [ 29 - 31 ]. This observation suggests PSMA as an appropriate target of active-specific immunization against carcinoma of prostate. However, standard methods of protein/epitope preparations often coupled to the adoptive transfer of antigen presenting cells are labor-intensive decreasing the widespread use of vaccines in the general cancer patient population. The purpose of our work was to delineate new ways to induce immune responses by DNA vaccination. In this study, all mice immunized with DNA vaccine expressing PSMA generated PSMA specific antibodies at a low level, which may result from the small amount of antigen expressed by plasmid in vivo. What is noteworthy is that all immunized mice developed CTL reactivity to B16-PSMA which led to suppression of tumor growth. In addition, although some tumors developed in some treated mice, they were consistently smaller in the control group. These findings suggest that DNA vaccines expressing PSMA could elicit immune response against tumor cells expressing the target molecule. Although DNA vaccines provide a convenient and effective approach to elicit cellular immunity, clinical outcomes have not been satisfactory, mainly because tumor-specific CTL elicited by the vaccines are insufficient to suppress cancer progression. CD8+ CTLs constitute one of the most important arms of the immune system, exhibiting the capacity of recognizing and destroying cancerous cells[ 32 , 33 ]. A variety of approaches are under evaluation to activate CD8+ CTLs, to that end, vaccines need to be administered in combination with adjuvants of which the most commonly used in experimental models is incomplete Freud's adjuvant (IFA). However, this adjuvant is not widely used in human vaccination protocols due to its undesirable side effects, such as erythema and induration at the injection site, in addition, IFA functions mainly to promote humoral immunity. For these reasons, alternative potent and safe adjuvants need to be identified to enhance cellular immune response against cancer [ 34 , 35 ]. Synthetic CpG oligodeoxynucleotides represent a promising adjuvant. The predominant effect of CpG oligodeoxynucleotides exposure is the promotion of Th1-type immune responses. Professional antigen presenting uptake CpG oligodeoxynucleotides and become activated with increased expression of MHC and co-stimulatory molecules [ 36 - 38 ] that promote antigen presentation to naïve T cells. In addition, dentritic cells are stimulated to secret Th1-biased cytokines, such as interferon-γ and IL-12 particularly desirable in cancer immunotherapy [ 39 , 40 ]. Therefore, CpG oligodeoxynucleotides may be useful vaccine adjuvants. CpG oligodeoxynucleotides 1826 is a potent enhancer of Th1-type immune responses and may benefit anti-cancer therapy [ 41 - 43 ]. We, therefore, hypothesized that the administration of these CpG oligodeoxynucleotides should enhance the cellular immunity elicited by DNA vaccines. However, co-adminstration of CpG oligodeoxynucleotides and DNA vaccines inhibit each other activity because CpG oligodeoxynucleotides may compete with plasmid uptake by antigen presenting cells. Furthermore, IFN-γinduced by CpG oligodeoxynucleotides could inhibit the activity of the CMV promoter utilized by eukaryotic expression vector, thus decreasing antigen expression. To examine the efficacy of CpG oligodeoxynucleotides as vaccine adjuvants, we injected them at the DNA injection site 3 days after vaccination rather than simultaneously. This strategy was based on a previous observation that transfected cells reach maximum yield of antigen expression between day 2 and 3 after vaccination. In this context, delivering the CpG oligodeoxynucleotides at the time of maximal antigen expression may be crucial to optimize the immunogenic boost [ 44 , 45 ]. Consistent with previous reports, this study suggests that CpG oligodeoxynucleotides enhance cellular immunity. The activity of CTL against PSMA expressing cells in the CpG +DNA vaccination group was significantly higher than in the DNA vaccination group. Furthermore, tumor challenge experiments demonstrated a potentiation of the suppressive effects on the growth of tumor cells expressing PSMA. These findings indicate that CpG oligodeoxynucleotides should be a powerful adjuvant in the context of DNA-based vaccination. Conclusions In this study, we designed a DNA vaccine expressing prostate specific membrane antigen (PSMA) and utilized CpG oligodeoxynucleotides to promote Th1-type immune response. We discovered that the constructed vaccine generated anti-tumor reactivity against malignant cells expressing PSMA that was enhanced by CpG oligodeoxynucleotides co-administration. This strategy may provide a new venue for the treatment of carcinoma of prostate, particularly for recurrent disease after hormone therapy fails. Competing interests None declared. Authors' contributions R.J.Q participated in the design of the study and carried out plasmid DNA transfection, RT-PCR, immunofluorescence assay, DNA vaccination, lymphocyte stimulations, cytotoxicity assays, and completed the preparation of the manuscript. Z.L participated in the design of the study and carried out the construction of the expression plasmid, western blotting, and histological analysis and assisted in the preparation of the manuscript. C.Q carried out cell culture, L.H carried out RNA extraction, Z.L carried out ELISA. Z.H.G conceived of the study, participated in its design and coordination, and helped draft the manuscript. All authors read and approved the final manuscript.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517950.xml

545958

Psychometric properties of a Swedish translation of the VISA-P outcome score for patellar tendinopathy

Background Self-administrated patient outcome scores are increasingly recommended for evaluation of primary outcome in clinical studies. The VISA-P score, developed at the Victorian Institute of Sport Assessment in Melbourne, Australia, is a questionnaire developed for patients with patellar tendinopathy and the patients assess severity of symptoms, function and ability to participate in sport. The aim of this study was to translate the questionnaire into Swedish and to study the reliability and validity of the translated questionnaire and resultant scores. Methods The questionnaire was translated into Swedish according to internationally recommended guidelines for cross-cultural adaptation of self-report measures. The reliability and validity were tested in three different populations. The populations used were healthy students (n = 17), members of the Swedish male national basketball team (n = 17), considered as a population at risk, and a group of non-surgically treated patients (n = 17) with clinically diagnosed patellar tendinopathy. The questionnaire was completed by 51 subjects altogether. Results The translated VISA-P questionnaire showed very good test-retest reliability (ICC = 0.97). The mean (± SD) of the VISA-P score, at both the first and second test occasions was highest in the healthy student group 83 (± 13) and 81 (± 15), respectively. The score of the basketball players was 79 (± 24) and 80 (± 23), while the patient group scored significantly (p < 0.05) lower, 48 (± 20) and 52 (± 19). Conclusions The translated version of the VISA-P questionnaire was linguistically and culturally equivalent to the original version. The translated score showed good reliability.

Background Patellar tendinopathy Patellar tendinopathy affects athletes in many sports and at all levels of participation, but is of particular concern for elite jumping athletes [ 1 ]. Many different types of sport activities have an increased risk for overuse of the patellar tendon including endurance sports (e.g. long-distance running and cross-country skiing) and sports with repetitive demands on strength and technique (e.g. tennis, baseball, volleyball, basketball and ballet) [ 2 ]. Athletes who participate in these sports may develop anterior knee pain that presents as tenderness at the inferior pole of the patella. This clinical syndrome is commonly called Jumper's knee, or patellar tendinopathy [ 3 ]. The term tendinopathy is considered to be the most appropriate clinical description for these chronic painful tendon conditions since there is no evidence of an inflammatory reaction in the chronically degenerated tendon [ 4 , 5 ]. The changes in the tendon are mainly due to chronic collagen fiber degeneration [ 6 ], but the cause and source of the pain still remains unclear. There are few studies on non-surgical treatment of patellar tendinopathy and there is a lack of evidence-based knowledge evaluating the therapy [ 7 ]. In-vivo studies in human or animals indicate possible benefits from treatments like heavy pressure [ 8 ], therapeutic ultrasound [ 9 ] and eccentric strength training [ 10 - 14 ]. Self-rated inventories for knee function Patient-administrated questionnaires are frequently applied as primary outcome measures in clinical trials and several inventories have been translated from English into Swedish [ 15 - 17 ]. The WOMAC osteoarthritis index has been tested for reliability and validity in Sweden [ 18 ] and compared to quality of life instruments (SF-36 and NHP) [ 19 ]. The Knee injury and Osteoarthritis Outcome Score (KOOS) is also a self-administrated instrument measuring outcome after knee injury at impairment, disability, and handicap level with five subscales [ 16 ]. Garratt et al determined that KOOS showed good evidence of reliability, validity and responsiveness, and is recommended the score for knee diagnosis like ACL reconstruction, total knee replacements and for arthroplasty patients [ 20 ]. The only published clinical scale for patellar tendinopathy problems (VISA-P) was developed in Australia by the Victorian Institute of Sport Assessment in Melbourne [ 21 ]. The aim was to assess symptoms, simple tests of function and the ability of subjects to undertake sports. This self-administrated questionnaire has been documented as a reliable instrument for monitoring the progress of rehabilitation [ 3 , 7 ]. It has also been shown to be a valuable tool in the assessment and documentation of recovery from patellar tendinopathy [ 22 ]. Even so, the responsiveness and validity of the questionnaire have not yet been fully proven. The purpose of this study was to translate and cross-culturally adapt the VISA-P score for a Swedish population and to perform a psychometric analysis as well as reliability and initial validity testing of the Swedish VISA-P score. Methods Subjects Fifty-one subjects gave informed consent to participate in this study. The VISA-P score was administered to 17 healthy students [9 women, 8 men, mean age (± SD) 24 (± 6)]; a population at risk, the Swedish male national basketball team [17 men, mean age 26 (± 3)], and patients with the diagnosis patellar tendinopathy [17 men, mean age 22 (± 5)]. The study was approved by the Ethical Committee at the Medical Faculty of the Karolinska Institute, Stockholm (Dnr 00-103). The VISA-P score The VISA-P score consists of eight questions [ 21 ], of which six questions concern pain experienced during a range of everyday activities. Two questions deal with the ability to engage in sport activities. All questions are answered on separate scales (0–10), where a higher score indicates a lower level of pain or impairment (Appendix A) [see additional file 1 ]. The maximal total score is 100 points, which would indicate that the person has no knee pain, good function and can perform fully in sports. The theoretical minimum score is 0 points. The original VISA-P score lacks information about the selection of items, weighting of each answer and the ranking of the options in the subscales in question 8. The aim of the present investigation was to get a "working tool" for further studies of the usefulness of the instrument for patellar tendinopathy patients in Sweden. Translation procedure The VISA Tendon Study Group at the University of Melbourne in Australia was informed and gave their consent to a Swedish translation of their original VISA-P score (Karim Khan, personal communication, 2003). The translation process followed the method described by Beaton et al [ 23 ]. This method is currently used by a number of organizations, including the American Association of Orthopaedic Surgeons (AAOS) Outcomes Committee as they coordinate translations of the different components of their outcome batteries [ 23 ]. The translation process is divided into five different stages: (I) Translation, (II) Synthesis, (III) Reverse translation, (IV) Expert committee review and (V) Pre-testing. Initially, two physiotherapists performed two independent translations (I) from English into Swedish. A synthesis (II) of these translations was made, and the consensus of the two translated Swedish versions was documented. Reverse translations (III) were performed independently by three native Anglophones fluent in Swedish. One of the reverse translators was a physiotherapist, one was an economist and the third was a teacher. The three physiotherapists in the expert committee (IV) then made a semantic and idiomatic equivalence analysis between the original source and target Swedish version of the VISA-P questionnaire. The translated questionnaire was pre-tested (V) on 12 individuals, six patients with patellar tendinopathy and six physical education students. Test-retest reliability The Swedish VISA-P score (Appendix B) [see additional file 2 ] was administrated to all 51 participants at Bosön, the Swedish National Sports Confederation Centre (Lidingö, Sweden). The participants completed the questionnaire twice within an interval of one week (range 4–7 days). The principal investigator administrated the questionnaires at all test occasions, with the exception of six of the tendinopathy patients. Validity For validity, the factor structure of the VISA-P score was analyzed with a principal component analysis, Varimax rotation. The number of extracted factors was equal to the number of eigen values above 1.00. Internal consistency of subscales, based on the factor analysis, and the total scale was calculated as a Cronbach α coefficient [ 24 ]. For discriminative validity of the VISA-P questionnaires were compared between three groups, each of which were expected to have different levels of scoring. Statistics All variables were summarized according to standard descriptive methods [mean and standard deviation (SD)] and checked for outliers. No significant deviations from the normal distribution criterion were found. The test-retest reliability was analyzed according to the method described by Bland and Altman, which yields an intra-class correlation (ICC) [ 25 ]. Differences between test occasions and groups were analyzed with an ANOVA (analysis of variance for repeated measurements, group *time). In the post-hoc tests of group differences, Tukey's HSD method was applied. A significance level of five percent was applied (two-tailed). Results Translation The expert committee considered the translation and reverse translation satisfactory. Test-retest reliability The test-retest of the Swedish VISA-P score showed high reliability and significance (ICC = 0.97, p < 0.001). In Figure 1 , the Bland-Altman plot is showing the difference in total score between occasion one (A) and occasion two (B), plotted against the mean value of both test occasions. There were no significant differences for the total VISA-P score between the first and second test occasions. Each question (Q) was analyzed separately regarding the reliability. Seven out of eight questions has a reliability of more than ICC = 0.8 (range 0.68–0.97). The score was easy to use and it took about five minutes to complete. Internal consistency The internal consistency of the total scale was high for the scores both at the first and second occasion, 0.83 and 0.82, respectively. Factor structure The principal component analysis yielded a two-factor solution. The communality, i.e. the degree of explained variance, of one of the questions (Table 1 , "sit pain-free?") was below 0.35, and thus not sufficiently explained by this solution. Thus, a three-factor solution was preferred which explained 85% of the total variance, with all communalities above 0.60. The first component comprised of six questions. The second and third components comprised of one question each. This solution showed high stability, being invariant in a second factor analysis of the scores from the second occasion (the amount of explained variance was 83%). Group differences in the VISA-P score At the first test occasion (A) the mean (± SD) of the VISA-P score in the healthy student group was 83 (± 12), in the basketball players 79 (± 23), and 47 (± 20) in the patient group (Table 2 ). In all questions, the patient group had lower scores as compared to the other two groups and statistical significance (p < 0.05) was observed in all individual questions except the first ("sit pain-free"). In Table 1 the post-hoc tests for group differences are presented. The questions concerning pain ("pain during 10 single leg hops") had the greatest difference between the groups (F = 12.7, p < 0.001). Both activity questions ("currently undertaking sport" and" pain during activity") showed significant (p < 0.001) differences between the groups. Discussion Translation The expert committee of the translation process expressed a general agreement of all the questions except one (Q1). During the translation procedure of the VISA-P score, the translation for "pain" was debated. Different Swedish words were discussed and compared between the different translators. Translations into the mother tongue, or the first language, more accurately reflected the nuances of the language. Reverse translation into English of the Swedish VISA-P version was without remarks. Thus, the original and translated versions were judged by the expert committee to be congruent. Test-retest reliability Over a time interval of one week (range 4–7 days), the Swedish version of the VISA-P score showed high reliability (ICC = 0.97). As compared to other test-retest investigations of this score, this interval is the longest that has been studied [ 21 ]. Validation of the VISA-P A factor analysis yielded three factors, of which the first showed the highest correlations with two questions ("pain during a full weight bearing lunge" and "problems squatting", see Table 1 ). The two other factors comprised only one question each, "currently undertaking sport" and "sitting pain-free", respectively. The separate factor for the question about "sitting pain-free" may be an artefact, as this item was the first one where misperceptions of the response dimension were more likely, thereby increasing the risk of higher error or unique variance. Some subjects in the pre-testing group, reported that they had perceived high scores as more pain. Conceptually, this question is equivalent to the questions of the first component. Experiences from the pre-testing resulted in a more detailed instruction for filling out the Swedish questionnaire. Group difference The patellar tendinopathy patients scored lower for all questions in the VISA-P score. The basketball players scored higher than the healthy students in two questions ("sitting pain-free" and "currently undertaking sport", see Table 2 ). The first question was the only question that did not show any statistical significance between the groups and, noteworthy, the lowest score, i.e. highest degree of problem. The reason given above regarding the risk of misperception of the response dimension might be an explanation. The VISA-P score has not yet been validated for pathological knee conditions other than patellar tendinopathy. Considering the separate questions (Appendix A) [see additional file 1 ] it would be of interest to test the VISA-P score for patients with anterior knee pain other than patellar tendinopathy. The significantly higher scores of the basketball players in question 7, "currently undertaking sport" (Table 2 ) were trivial and obvious, since all of them were active players in Swedish the national team. The standard deviation was nearly twice as high for the patients and basketball players as compared to the healthy students. This reflects the heterogeneity of the first two groups. Generally, there is a debate concerning scores about the relevance of using the total score or dividing the score in different subgroups. A short clinical scale is often an advantage. The factor analysis as well as the analysis of differences between the groups suggests that the VISA-P score could be abbreviated to two or three items without losing significant clinical information (Table 1 ). An important aspect of a clinical scale is its sensitivity for change or its ability to follow amelioration or exacerbation during treatment. The theoretical range of the VISA score, i.e. the floor and ceiling, is 0–100. The mean total score of the patients was approximately 50 (with a minimum value of 16) and for the control groups 80 (with a maximum value of 100. Thus, there seems to be sufficient scope to follow treatment effects, as well as to follow deteriorations of a risk group. It should be noted, however, that the present study was not designed to study treatment effects or development of a pathological process. The conclusion regarding the sensitivity of the VISA score, thus, awaits empirical support. Although the mean VISA-P scoring was significantly different between asymptomatic subjects and patients with patellar tendinopathy, the score is not suggested to be a diagnostic test [ 21 ]. Therefore the score is considered to be suitable for group and intra-individual comparisons but should be avoided in inter-individual comparison. Another limitation of the score has not been shown to be applicable in a non-athletic population. Adaptation of a questionnaire for use in a new setting is time consuming and costly. There are specific criteria that investigators should apply when evaluating patient-based outcome measures [ 26 ]. That being the case, larger international data collections and better correlations can be made when proper translations are performed and evaluations conducted. Additionally, there is a need for international accepted 'golden standards' in outcome scores. In conclusion, the results of the present study suggest that the translated Swedish version (Appendix B) [see additional file 2 ] of the original Australian VISA-P score (Appendix A) [see additional file 1 ] had satisfactory test-retest reliability when used to evaluate symptoms, tests of function and ability to undertake sport in patients with patellar tendinopathy. Authors' contributions AF initiated the study, led the translation process and conducted all test occasions. TS and PR helped with general analysis and writing the article. GE guided and helped the main author with the statistical analyses of the data collected. All four of the authors read and approved the article. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Appendix A. The original VISA-P score. Click here for file Additional File 2 Appendix B. The translated and cross-culturally adapted Swedish VISA-P score. Click here for file

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545958.xml

555587

The discovery of the first human retrovirus: HTLV-1 and HTLV-2

I describe here the history leading up to and including my laboratory's discovery of the first human retrovirus, HTLV-I, and its close relative, HTLV-II. My efforts were inspired by early work showing a retroviral etiology for leukemias in various animals, including non-human primates. My two main approaches were to develop criteria for and methods for detection of viral reverse transcriptase and to identify growth factors that could support the growth of hematopoietic cells. These efforts finally yielded success following the discovery of IL-2 and its use to culture adult T cell lymphoma/leukemia cells.

Background After arriving at NIH in 1965 I spent my first year as a young physician caring and treating (mostly unsuccessfully) acute leukemias in children: a vivid experience and one which made me absolute in a decision to be fully involved in laboratory research and not return to clinical medicine. My research interest almost from the very start was in the biology of blood cells, and I focused on comparisons of human leukemic cells with normal leukocytes. This was mainly limited to comparative biochemistry. Specifically, I studied enzymes of pyrimidine nucleoside and nucleotide metabolism, tRNA species and their corresponding amino acyl-tRNA synthetases, and finally DNA polymerases [see refs.[ 1 - 5 ] as examples] – though now this approach would seem to be empirical in the extreme, because we have so many obvious rational things in cancer research today. However, at that time "fishing expeditions" were "where things were at". I hoped to uncover clues that might help us better understand the nature of leukemic cells and also their origin. Discovery that human T cells made cytokines ("lymphokines") and early hints of human retroviruses The main leukemia I worked on was acute lymphocytic leukemias (ALL). After all, these were the most common of the acute leukemias and gram quantities of these cells were available from my clinical colleagues at NCI. Importantly, these were the only leukemias for which reasonably similar normal control cells were available, namely, normal human lymphoblasts. Scientists in Philadelphia had just discovered that a plant lectin, phytohemagglutinin (PHA), could induce human lymphocytes to become activated and go through a mitotic cycle. These normal lymphoblasts looked like ALL cells, but these were days before most of us would or could know of the great complexity of subtypes of lumphocytes. Functional discriminatory assays were barely available and monoclonal antibodies with their capacity to provide surface markers were yet to come. Thus, we did not then sub-classify lymphocytic leukemias. Herb Cooper of NIH had learned how to purify lymphocytes from columns packed with nylon; myeloid cells would adhere, but lymphocytes passed through. Cooper generously provided this technique to me. During this period (1968–1970) I became very impressed by the studies of Leo Sachs in Israel and later also of Don Metcalf in Australia who where showing that, like some lymphocytes, myeloid cells could also be grown in the laboratory but not in liquid culture. Instead, they used the technique previously applied to virus transformed cells of cell growth on a methylcellulose solid surface in the form of cell colonies. However, growth was transient and the amount of cells quite limited, precluding many types of biochemical, molecular biological, and virological experiments. Nonetheless, from this system, Sachs and his colleagues and Metcalf and co-workers made seminal discoveries, including a growth/differentiation factor, granulocyte macrophage colony-stimulating factor (GM-CSF), which was specific for the myeloid lineage. Sachs logically believed the main production of GM-CSF would be from myeloid cells, i.e., a feedback regulation – granulopoietic progenitors proliferated and formed "dead end" granulocytes, which should produce their own granulopoietic factor [see refs. [ 6 , 7 ] for reviews]. Meanwhile, while comparing ALL cells to normal lymphocytes, I decided to test the conditioned medium of the PHA-stimulated normal cells for growth factor. The late Alan Wu had just joined me from the laboratories of Till and McCulloch in Toronto shortly after the publication of their famous paper describing hematopoietic stem cell assays (in mice) for the first time. Alan and Joan Prival, a post-doctoral fellow, joined me in reporting the then surprising finding that lymphocytes (T cells) made GM-CSF [ 8 ]. This would be the start of my long involvement with "conditioned medium" from PHA-stimulated lymphocytes. Dane Boggs, F. Ruscetti, and co-workers in Pittsburgh had described the same phenomena at almost exactly the same time. These papers were likely among the first to describe lymphokines (lymphocyte-derived cytokines). In this period (early 1970s) I began to study animal retroviruses because in several animals these kinds of viruses caused leukemias. Thus, no matter whether human retroviruses (leukemia-causing or otherwise) existed or not, a study of animal retroviruses, especially focused on learning their mechanisms of leukemia causation, might provide insights into the mechanisms involved in human leukemias. However, my co-workers and I also decided to search for human retroviruses, an unpopular goal at this time, considering the decades of attempts and failures. I was, nonetheless, encouraged by discussions with William Jarrett, the Scottish veterinarian who discovered feline leukemia virus, and by the work of the late Howard Temin. Temin, of course, had predicted that retroviruses of animals replicated by having their RNA genome transcribed into a DNA form, which would then integrate into the DNA of the target cell. He referred to this integrated form as provirus, the name given to his theory. In 1970 Temin and his colleague Mizutani, and separately, David Baltimore, gave credence to the theory with their discovery of the DNA polymerase carried by all retroviruses, reverse transcriptase (RT)[ 9 , 10 ]. For me it also meant a convenient, inexpensive, and extremely sensitive assay for a retrovirus. (This would be one of two technologies that would be key for later discoveries of all human retroviruses). RT forms in virions only upon budding from the cell. Consequently, finding this enzyme in media of cultured cells implied release of retrovirus particles, and finding RT from extracts of cells implied the presence of a cell associated virus particles, as for example, virions associated with the cell surface membrane. We found rare cases of leukemia that scored positive in RT assays. The problem, however, was that RT might be a product of a normal cellular gene. We needed to develop the assay not only as a very sensitive one but also one that would distinguish RT from all of the then known cellular DNA polymerases (alpha, beta, and gamma). This became a major objective [see refs. [ 5 ] and [ 11 - 14 ] for examples]. Armed with these RT assays we did find a few cases of adult lymphocytic leukemias with RT showing all the characteristics of RT from a retrovirus (we had by then purified and characterized RT from many different animal retroviruses). We published on the one best characterized in Nature New Biology in 1972 [ 15 ]. We believed this was a "footprint" of a human retrovirus, but we failed to isolate virus from this patient. (Though we will never know, it is interesting to speculate whether this young adult had ATL because of some clinical similarities to ATL). We also thought it would attract wide interest and excitement in the field. It did not. It was clear that we had to isolate a replicating virus, one we could study, perpetuate, and give to others. The obvious and easiest approach to virus isolation was by using cell lines. Cell culture technology had become widely available by the 1960s, and many cell lines from different species were available. The approach is generally to co-culture the primary cells (in our case the leukemic cells) with a wide variety of such lines and hope virus will take in one or more. This, of course, would be after scoring positive in the RT assay. However, by this period, there was increasing antagonism to research directed toward the finding of human tumor viruses and especially of retroviruses. The NCI had created the heavily funded Virus Cancer Program which was under attack for failing to find clear evidence of tumor viruses. Moreover, by the mid-1970s there had been not only decades of failure to find human retroviruses, there had been many false starts by many investigators utilizing the co-culture system that involved cell lines, including one by me. The usual problem was a cross contamination with an animal retrovirus. For this reason I became convinced that we had to find ways to grow primary blood cells, but not with the systems of Sachs and Metcalf. These methylcellulose colonies of leukocytes provided too few cells and growth of these cells were limited in number and in time. When we had our next hint of an RT positive leukemic sample it turned out to be from a patient with a myeloid leukemia, so we searched for a growth factor that would maintain and promote growth of human myeloid leukemic cells in liquid suspension culture. This had not been achieved before. From an early-term (first few weeks) abortion, we obtained some human embryonic cells that produced a factor that led to the first successful routine growth of these human leukemic cells in liquid suspension(16). We called these HL (human leukemia) cells, with a given sequential number of the samples we had studied. One of these cultured cell populations became an immortalized cell line, HL-60. It was the first human leukemic myeloid cell line [ 17 ] and like almost all the others, the HL-60 cells showed no evidence of virus. However, one growing human leukemic cell culture (not immortalized) did yield virus, and it was anxiously propagated. Unfortunately, the embryonic factor needed to keep these cells alive and growing was lost when the freezer in which they were stored broke down over a long holiday weekend, which was not recognized for some time. (My first lesson in never storing a divisible valuable all in one place!). This led us to a frantic search to find another source. We screened conditioned medium from a wide variety of cell lines and cell strains, including many more fetal cells – all to no avail. One approach was to culture many different types of cells from many different tissue sources (including human embryos) for several days, collect the media (conditioned media or CM), and add it to leukocytes from normal human cord blood, samples of human bone marrow, and myeloid leukemic cells. In this period (early mid-1970s), a post-doctoral fellow, Doris Morgan, joined our group and took part in the search. As would be expected, CM from PHA-stimulated lymphocytes was one of the cell sources I asked to be screened. Doris was succeeding in growing cells from human bone marrow, and was intensely nursing them daily for months. But they were lymphocytes, not myeloid cells. It was neither unique nor interesting to grow human B cells. Even at this time Epstein-Barr virus (EBV) immortalized cell lines were well known to grow often from normal blood or a bone marrow mixed cell population. Indeed, they were the only kind of blood cells that could be routinely grown in long-term culture, but analyses of the cells revealed that they were T cells, which at that time had only recently been clearly delineated from B cells by certain functional assays (the E rosette assay, for example). The factor we had found in the PHA-CM was a new growth factor. Francis Ruscetti had then joined our group and carried out a set of experiments that demonstrated this more fully, and we reported these results in 1976–1977[ 18 , 19 ] and they were to be the first reports of what we termed a T cell mitogenic factor, later called TCGF, and finally interleukin-2 (IL-2). The purification was later [ 20 ]. IL-2 was among the first well-defined cytokines. The combination of IL-2 growth of T cells with sensitive RT assays would be (and still is) the key to the discoveries of human retroviruses in T cell leukemias and AIDS. The debate about the possible existence of human retroviruses In this same period the pressure against attempts to find human retroviruses intensified. It was not only the prevailing atmosphere of failure but also reasonable scientific arguments. For examples: (1) there was little evidence for leukemia viruses in primates. (2) When retroviruses were found in animals they were not difficult to find. Extensive viremia preceded disease, therefore, if they infected humans, they would be easy to find and would have been discovered much earlier. (3) Human sera in the presence of complement lysed animal retroviruses, thereby providing a rational mechanism for the conclusion that humans were protected. Finally, there were technical difficulties such as the ability to culture primary human cells (see Table 1 ). Table 1 Factors that led to consensus that human retroviruses did not exist 1. Failure to discover them after an extensive survey by many investigators in the 1950s, 1960s, and 1970s. 2. Ease of detection in animal models because of extensive virema. 3. Difficulties in growing primary human cells. 4. Results showing human sera with complement lysed animal retroviruses. We reasoned otherwise. Kawakami and colleagues had just discovered gibbon ape leukemia virus, and linked it to chronic myeloid leukemia in that species [ 21 ]. Later, we discovered a variant of that virus which caused T cell leukemia [ 22 ]. Bovine leukemia virus (BLV) was discovered [ 23 , 24 ], and it was noted that BLV replicated at very low levels thus putting to rest the notion of "extensive viremia always precedes animal retrovirus induced leukemias". The biased view came from the fact that the earlier small animal models were naturally selected for their utility . Consequently, models in which virus is difficult to detect would be selected against. As for human sera lysing retroviruses, unfortunately those studies were limited to tests of retroviruses from non-primates. Later, we would learn that many primate retroviruses, including the retroviruses of many, are not susceptible. Our ultimate focus on T cell leukemias was dictated by several factors. First, most animal leukemias caused by retroviruses are lymphocytic leukemias and of these T cell leukemias predominate. Second, the first and to this date only leukemia of non-human primates is caused by a retrovirus [ 21 ], and a particular strain of this virus which we isolated caused T cell leukemia [ 22 ]. Third, fortune dictated that we would end up focusing on human T cell malignancies because of our discovery of IL-2 which allowed us to grow significant numbers of such cells in many but not all instances (not all T cell leukemias or lymphomas respond to IL-2). One other development also influenced our continuation of the pursuit of human retroviruses. This was a documented interspecies transmission of a gibbon ape leukemia virus (GaLV) from a pet old world Gibbon ape to a new world Wooly monkey. It was well known that retroviruses could move from one species to another, but in all cases these were very ancient events only discovered by analyses of cellular DNA of many animals. But in this case the event occurred "right before our eyes", giving rise to the virus from the Wooly monkey known as simian sarcoma virus [ 25 ]. We felt humans could not be excluded, and indeed later we would learn that the first human retrovirus discovered (HTLV-1) has close relatives among many old world primates and may have arisen from an ancient transmission from monkey to man. A more relevant example, of course, is HIV. There is much evidence that it came into humans as a much more recent infection from African primates (see Table 2 ). Table 2 Factors encouraging us to continue searching for human retroviruses 1. The discovery of bovine leukemia virus (minimally replicates, difficult to find) 2. Technological advances – A. A sensitive specific assay for a footprint of a retrovirus, namely, reverse transcriptase. B. Capacity to grow significant numbers of primary human T cells in liquid suspension culture giving us access to virus detection and isolation, namely by using IL-2. 3. Discovery of a retrovirus causing leukemias in a species close to man, namely GaLV. 4. A documented example of a retrovirus transmission from one species of primates to another, namely GaLV from a gibbon ape to a wooly monkey [26]. 5. Purification and characterization of reverse transcriptase from a patient with an adult lymphocytic leukemia (type unknown) 1972 [15]. Discoveries of HTLV-1 and HTLV-2 The first detection and isolation of HTLV-1 was in 1979, and the first detection came from the analysis of a T cell line established by J. Minna and A. Gazdar from a patient these clinicians called a cutaneous T cell lymphoma. Alternatively, such patients were also called mycosis fungoides or Sezary T cell leukemia depending upon clinical nuances. Though IL-2 was supplied by us for them to use in their initial culturing of these cells, the cells rapidly immortalized. An outstanding post-doctoral fellow, Bernard Poiesz, carried out RT assays on these cells with positive results, and we soon arranged for electron microscopic analysis of concentrated RT plus cultures and found retrovirus particles. Because putative human retroviruses viruses had been found many times before by several investigators in established cell lines, only to be subsequently shown to be accidental laboratory contaminants, by the late 1970s I was well aware that much more had to be done before this work was presentable. For instance, we had to (1) show that the same virus could be isolated from primary tissue samples of the same patient by culturing primary T cells with IL-2; (2) demonstrate that the virus was novel, i.e., not any of the known animal retroviruses; (3) show it could infect human T cells in vitro; (4) demonstrate specific antibodies to the virus in the serum of the patient; (5) demonstrate that proviral DNA could be found integrated in the DNA of the cells from which the virus was isolated; (6) provide evidence that this was not a one-time affair by showing serological evidence of specific antibodies not only in the patient but in others as well. These results were successfully obtained in 1979–1980 and available by the time we submitted and published our first report in 1980 [ 27 ], enabling us to follow quickly with several other essential reports [ 28 - 33 ], also including independent isolates from other patients [ 29 , 34 ]. One of these patients was a black woman from the Caribbean, and the second was a white merchant marine who acknowledged sexual contacts in southern Japan and the Caribbean. These and all subsequent isolates of HTLV-1 in our laboratory were from primary cells cultured with IL-2. After an initial struggle to publish in the J. of Virology, fortunately, we were soon able to publish the original report in PNAS, and this opened the door. It soon became clear that HTLV-1 was specifically associated with adult T cell malignancy (usually CD4+ cells) in which the patients frequently had cutaneous abnormalities and hypercalcemia. Clinicians in the United States had not at that time made any distinction of HTLV-1-associated T cell malignancies from other neoplasms, and as noted above collectively referred to these patients with others (non-HTLV associated) as cutaneous T cell leukemia-lyumphomas. However, a few years earlier Kiyoshi Takatsuki and his co-workers Junji Yodoi and Takashi Uchiyama defined clusters of leukemia in southwest Japan with special clinical features and cellular morphology, which when coupled with the geographic clustering, led him to propose in 1977 that this was a distinct form of leukemia. He named it adult T cell leukemia (ATL) [ 35 ]. Two events significantly catalyzed the further development of our work and of our understanding of HTLV-1 and its role in T cell malignancies. The first of these (in the summer of 1980) was information from Drs. Tom Waldmann and H. Uchiyama, who had come to NIH as a visit scientist. They brought to our attention the ATL cluster in Japan so in the fall of 1980 I contacted two Japanese friends, the late Yohei Ito, then Chair of Microbiology at Kyoto University and Tad Aoki for more information and for sera from such patients to test for antibodies to HTLV. This specific clinical entity had been described as early as 1977 by Takatsuki and his co-workers Yodoi and Uchiyama, and was called adult T cell leukemia by him. Aoki and Ito sent sera from such patients to me in 1980, and these sera scored positive for antibodies to HTLV-1. Based on these results Ito organized a small meeting at Lake Miwa outside of Kyoto attended by a few co-workers and myself from the U.S. and Aoki, Ito, and several other Japanese scientists most notably Takatsuki, Y. Hinuma, and T. Miyoshi. The meeting was held in March 1981. Several of my colleagues and I presented our results in detail. This included description of several isolates of HTLV-1, characteristics of purified HTLV-1 p24 as well as reverse transcriptase proteins, evidence of integrated HTLV-1 provirus T cell malignancies and healthy volunteers which provided clear evidence for the linkage of HTLV-1 to certain T cell malignancies, and the positive serological results in Japanese ATL patients. In organizing this meeting the intention of Ito was to foster wide collaboration in Japan with me and my co-workers on this disease. The meeting summary was published in Cancer Research in November 1981 [ 36 ]. It was only at the end of the meeting when we were summarizing and planning for this collaboration with the Japanese investigators, that Dr. Yorio Hinuma "announced" he too had a retrovirus. He presented EM pictures of virus particles from a cell line established by Dr. Miyoshi by co-cultivation of ATL cells and normal human T cells. These results of Miyoshi were the first indication of the transforming capability of HTLV-l because the cell line that was immortalized was from the normal donor [ 37 ]. Later, my colleague M. Popovic was able to make this a routine, that is, we would show that HTLV-1 could routinely immortalize normal human T cells [ 34 ]. It was obvious to all that the virus pictures shown by Hinuma were HTLV-1. By the time of this meeting we had already published a few papers on HTLV-1. Hinuma called his isolate ATLV (adult T cell leukemia virus), but argued against collaboration claiming it was not possible to provide human sera from Japan for "cultural reasons". In June 1982 Hinuma and colleagues published on their isolate of ATLV [ 38 ]. After comparative analyses of isolates of ATLV and HTLV were performed we published with Japanese colleagues M. Yoshida, T. Miyoshi and Y. Ito that HTLV-1 and ATLV were the same virus [ 39 ]. Consequently, we agreed that the virus name should be HTLV to recognize the priority of our virus work, and the disease would be referred to as ATL in recognition of the Japanese priority in distinguishing this malignancy as a specific identity which had been "lumped" with other T cell leukemias/lymphomas in western countries and elsewhere as cutaneous T cell lymphomas [ 40 ]. Yoshida was soon to make many of the major advances in the molecular biology of HTLV-1 but this is another story. The second meeting of considerable importance was in London chaired by the late hematologist Sir John Dacie and attended by Dacie, Drs. Daniel Catovsky, Robin Weiss, Mel Greaves, and William Jarrett among others from Great Britain and by my collaborator in epidemiological studies, Dr. William Blattner, and myself. It was Catvosky who called for this meeting because he noted that we had found HTLV-1 mainly in African Americans and black persons in the Caribbean and he had found an unusual frequency of adult T cell malignancies in Caribbean immigrants to England. He recognized the similarities of their disease to Takatsuki's ATL. Thus, he postulated they were one and the same disease and HTLV-1 would be present in all. He was right. Promptly, Blattner accelerated his studies in the Caribbean and documented that HTLV-1 was endemic in some islands. He and Guy de Thé of France would then show that this result depended upon the particular tribes in Africa from which the individuals descended. Some of these experiences would be a precursor of a persistent pattern, i.e., HTLVs are not easy to transmit, remain within families and regions over long periods of time, and have old-world linkage. Ultimately, related viruses would be found in old-world primates and more distantly related viruses in some ungulates. The modes of transmission would soon be forthcoming as sexual contact, blood, and mother to child via breast feeding. Later in 1981 we isolated HTLV-2 from a leukemia described as "a hairy cell T cell leukemia" [ 41 ], but this strain is far less pathogenic that HTLV-1. Many of the features of these viruses coupled with CD4 T cell tropism would prove to be remarkably similar to those of the virus about to enter our work, HIV. A companion article in Retrovirology by Kiyoshi Takatsuki recounts the events surrounding the discovery of the disease, adult T-cell leukemia [ 41 ].

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555587.xml

555593

Evaluation of SLC11A1 as an inflammatory bowel disease candidate gene

Background Significant evidence suggests that a promoter polymorphism withinthe gene SLC11A1 is involved in susceptibility to both autoimmune and infectious disorders. The aim of this study was to evaluate whether SLC11A1 has a role in the susceptibility to inflammatory bowel disease (IBD) by characterizing a promoter polymorphism within the gene and two short tandem repeat (STR) markers in genetic proximity to SLC11A1 . Methods The studied population consisted of 484 Caucasians with IBD, 144 population controls, and 348 non-IBD-affected first-degree relatives of IBD patients. IBD subjects were re-categorized at the sub-disease phenotypic level to characterize possible SLC11A1 genotype-phenotype correlations. Polymorphic markers were amplified from germline DNA and typed using gel electrophoresis. Genotype-phenotype correlations were defined using case-control, haplotype, and family-based association studies. Results This study did not provide compelling evidence for SLC11A1 disease association; most significantly, there was no apparent evidence of SLC11A1 promoter allele association in the studied Crohn's disease population. Conclusion Our results therefore refute previous studies that have shown SLC11A1 promoter polymorphisms are involved in susceptibility to this form of IBD.

Background Inflammatory bowel disease (IBD) encompasses a number of disorders that are characterized by chronic inflammation of the gastrointestinal tract. The three specific forms of the disease known as Crohn disease (CD), ulcerative colitis (UC), or the less well-characterized indeterminate colitis (IC), are diagnosed based on the results of clinical observations, histology, radiology, and to a lesser extent, serology. Prominent clinical features include profuse diarrhea, abdominal pain, and increased colorectal cancer risk. IBD affects an estimated one million Americans [ 1 ], with UC being the more prevalent form. Current pathogenic models state that IBD develops in a genetically susceptible individual in response to environmental stimuli [ 2 ]. The aim of this study was to investigate the role of the gene SLC11A1 in the susceptibility to IBD. SLC11A1 is a proton-coupled bivalent metal antiporter that is crucial in early macrophage activation [ 3 ]. SLC11A1 encodes a highly hydrophobic 550 amino acid membrane protein and is located on the long arm of chromosome 2 (2q35), a region that has not been identified as being within an IBD susceptibility locus. Genetic studies have shown that different alleles of a (GT) n dinucleotide repeat polymorphism located within the promoter region of SLC11A1 cause susceptibility to different human diseases. Allele 2 of this polymorphism is associated with susceptibility to intracellular pathogen infection [ 4 ], and allele 3 is associated with autoimmune disease susceptibility [ 3 ]. SLC11A1 has been implicated in susceptibility to IBD. Previous work from this laboratory [ 5 ] has shown that two tetranucleotide short tandem repeat (STR) polymorphisms genetically linked to SLC11A1 are associated with CD [ 5 ]. The (GT) n promoter polymorphism, however, was not examined in that study. A recent study by Kojima et al [ 6 ] indicated that promoter polymorphism allele 7 was associated with IBD in a Japanese population. In our current work, we used a case-control association study to define the role of the (GT) n promoter polymorphism in susceptibility to IBD and to further characterize the STR markers previously described [ 5 ]. These markers (D2S434 and D2S1323) are 0.67 Mbp proximal and 3.44 Mbp distal of SLC11A1 , respectively. Methods Population This prospective study was approved by the University of Louisville Institutional Review Board. Written informed consent was obtained from all subjects. Patients were derived from a University-based colon and rectal surgery practice. Initial IBD diagnoses were determined through radiological, endoscopic, and/or histopathological studies. Histology was available in all cases. The study population consisted of a total of 628 Caucasians, including 254 unrelated individuals with CD (63% women), 165 with UC (53% women), 65 with IC (70% women), and 144 population controls (74% women). Demographic data for the IBD population is shown in Table 1 . For the purposes of more accurate phenotyping, the CD group was subdivided based on the Vienna classification [ 7 ]. This categorizes CD patients based on age of onset (A), location of disease (L), and disease behavior (B). With regard to age of onset, 175 of 254 (69%) CD patients were diagnosed at < 40 years of age (A1 group) and 47 of 254 (18%) had disease diagnosed at ≥40 years of age (A2 group). Information regarding age of onset was not available for 32 of 254 patients (13%). Regarding disease location, 70 of 254 (28%) had terminal ileal disease (L1 group), 152 of 254 (60%) had purely colonic and/or ileocolic (ileal and colonic) disease (L2/L3 groups), and 32 of 254 (13%) had disease located proximal to terminal ileum (L4 group). The L2 and L3 groups were combined for the purposes of this study, since colonic disease constitutes the primary focus of our group. Regarding disease behavior, 96 of 254 (38%) had uncomplicated inflammatory disease (B1 group), 28 of 254 (11%) had stricturing disease (B2 group), and 130 of 254 (51%) had penetrating disease (B3 group). A total of 348 non-affected first-degree relatives of IBD patients were available for the study. There were 63 CD families (29 triads, 30 discordant sibling pairs, 4 triad/discordant sibling pair combinations), 43 UC families (11 triads, 24 discordant sibling pairs, 8 triad/discordant sibling pair combinations), and 14 IC families (5 triads, 5 discordant sibling pairs, 4 triad/discordant sibling pair combinations). In addition, there were 17 CD families, 18 UC families, and 9 IC families who could not be classified as either a triad or discordant sibling pairs. Markers and genotyping STRs were amplified from genomic DNA extracted from peripheral leukocytes. Polymerase chain reaction (PCR) primers were custom-synthesized (Proligo, La Jolla, CA). PCR amplification of the SLC11A1 promoter polymorphism was performed using 100 ng template DNA, 2.5 mM MgCl 2 , 50 mM KCl, 10 mM of tris/HCl, 0.2 μM of each dNTP, 200 pmol of each primer, and 1 unit of Taq polymerase (Promega, Madison, WI). Primer sequences were as follows: 5'GACATGAAGACTCGCATTAG3' & 5'TCAAGTCTCCACCAGCCTAGT3'. The product was amplified using the following conditions: 94°C for 10 min, followed by 25 cycles of 94°C for 30 sec, 55°C for 75 sec, and 72°C for 20 sec. A final extension step was run at 72°C for 6 min. Genotyping was performed using gel electrophoresis (Spreadex EL 300 S-100 gel [Elchrom Scientific, Lake Park, FL]). PCR amplification of the STR markers D2S434 and D2S1323 was performed as described previously [ 5 ]. Statistical analyses Case-control tests of association Case-control analyses were performed using STR genotype-allele frequencies in cohorts subdivided on the variables outlined in Table 1 . Correction for multiple testing was performed by using Storey's q-value method, where p-values were adjusted according to the experimental false discovery rate (FDR) [ 8 ]. The global null hypothesis of no difference in genotype-allele frequency between controls and any of the other groups was tested by using Fisher's method [ 9 ], which was used to combine all of the control group comparison global p-values obtained for each of the three markers. Q-values were calculated from these global p-values using the QVALUE program [ 10 ] with the tuning parameter λ = 0, which dictates the assumption π o = 1, where π o is the proportion of tests in which the null hypothesis is true. To limit the number of individual disease-control genotype-allele comparisons, further analyses were only performed in those disease groups possessing a global test q-value <0.05. Therefore, 2 × 2 contingency tables were only constructed for genotypes with a global test q-value meeting this condition. Disease versus control comparisons were performed using Fisher's exact test and corrected for multiple testing using the q-value method as described above. Since IBD affecting the colon is the primary focus of our research, the L2 colonic CD and L3 ileocolonic CD groups were combined, enabling maximization of statistical power. We argue that this approach will not impair the validity of statistical analyses given that the Vienna Classification is somewhat arbitrary and that group genetic homogeneity may be increased by combining all those cases with colonic IBD. Furthermore, owing to the relatively low numbers of individuals with CD proximal to the terminal ileum (L4 group, n = 32), this disease subgroup was excluded from case-control analyses. Family-based tests of association Family-based tests used included the pedigree disequilibrium test (PDT) [ 11 , 12 ] and a likelihood ratio test implemented in the computer program TRANSMIT [ 13 ]. PDT analysis has distinct advantages over other family-based tests of association. PDT allows for analysis of data from related nuclear families and discordant sibling pairs from extended pedigrees. The likelihood ratio test, as implemented in the computer program TRANSMIT [ 13 ], can account for missing parental genetic data by inferring parental genotypes based on the genotypes of their offspring. To maximize power, family-based analyses were not performed at the sub-phenotypic level. Results Allele and genotype frequencies Alleles for the SLC11A1 promoter polymorphism STR D2S434 and STR D2S1323 were named numerically according to molecular weight, with "1" representing a larger allele than "2" and so on. Allele distributions are shown in Table 2 . DNA sequence analysis of the three SLC11A1 promoter polymorphism alleles identified in the study population revealed the following sequences: Allele 1: T(GT) 5 AC(GT) 5 AC(GT) 11 GGCAGA(G) 6 Allele 2: T(GT) 5 AC(GT) 5 AC(GT) 10 GGCAGA(G) 6 Allele 3: T(GT) 5 AC(GT) 5 AC(GT) 9 GGCAGA(G) 6 These sequences corresponded to alleles 1, 2, and 3 as described by Searle and Blackwell [ 14 ]. Alleles 4–7 were not identified in this population. Case-control association studies Results from global disease versus control comparisons of the studied markers are shown in Table 3 . Analysis of genotype and allele frequencies for the (GT) n promoter polymorphism did not reveal any evidence of association. Additionally, we studied STR markers in genetic proximity to SLC11A1 . The STR D2S1323 exhibited evidence of association with the test of the global null hypothesis of homogeneity of all groups exhibiting statistical significance for allele frequencies (q = 0.050, p = 0.008). To limit multiple testing, individual pair-wise disease versus control comparisons were only performed for D2S1323 alleles. These analyses showed over-representation of allele 1 in the UC group (q = 0.022, p = 0.005; UC frequency 78 of 100 alleles [78%] versus control frequency 54 of 92 alleles [59%]). The STR D2S434 did not reveal any evidence of association. Further analyses of genotype and allele data for all markers did not show a correlation with either age of onset or disease behavior. Family-based association studies Family-based tests were performed for each of the three markers. Neither the likelihood ratio test nor the PDT revealed significant evidence of association for any of these markers. Discussion No association was observed between any SLC11A1 promoter polymorphism alleles and any IBD sub-phenotypic group in this population, which unexpectedly refutes the significant evidence of others [ 6 ]. This result was somewhat surprising, given that many studies have shown SLC11A1 promoter polymorphisms to be involved in susceptibility to autoimmune disorders and disorders that are characterized by a high degree of immunological dysregulation, including IBD [ 3 ]. Searle and Blackwell [ 14 ] investigated the enhancer properties of each of the four identified promoter polymorphism alleles. Allele 3 was found to have endogenous enhancer activity, whereas alleles 1, 2, and 4 were poor promoters in the absence of exogenous stimuli. Stimulation of macrophages with bacterial lipopolysaccharide (LPS) had minimal effects on SLC11A1 expression driven by alleles 1, 2, and 4; however, allele 3-driven SLC11A1 expression was enhanced. It has long been recognized that altered expression of SLC11A1 in response to LPS is a key event in the activation of macrophages upon contact with a variety of pathogens [ 15 ]. Based on those observations, we hypothesized that over-representation of promoter allele 3 in CD patients could lead to hyperactivation of bowel wall macrophages that are chronically exposed to high levels of LPS. This could subsequently cause the autoimmune-like phenotype characteristic of IBD. This hypothesis must be rejected in this cohort due to the absence of association with promoter allele 3 in any IBD subgroup. Analysis of intragenic single nucleotide polymorphisms are required to conclusively exclude SLC11A1 as an IBD candidate gene. D2S1323 was associated with UC in this population. Indeed, this was the only evidence for association in any of the studied STR markers in proximity to SLC11A1 . This association was not, however, confirmed upon family-based statistical analysis. The current finding was particularly surprising to us, given that our previous study showed this STR to be associated with CD rather than UC [ 5 ]. The significance of this finding is thus uncertain, and the divergent outcomes from the current and earlier studies could be considered somewhat concerning. We do, however, believe that a number of prominent features of the current study render our results of higher validity than those of our previous study. An important feature of our current work is that all histological specimens from our subjects have been reviewed by a single gastrointestinal pathologist. This was not the case in the earlier study, and diagnostic misclassification (i.e., UC and/or IC incorrectly diagnosed as CD and vice versa ) may have been a confounding variable that may have in part led to what we now consider the false-positive association observed in the CD population. It should also be noted that the current study contains over two-fold the number of subjects as compared with the previous work, and thus statistical power has been significantly increased. Furthermore, our current statistical methodology is now far superior, and we argue that our stringent means of correcting for the effects of multiple testing effectively minimizes the likelihood of detecting a false-positive association. This is a vast improvement over our earlier study in which compounding of a type 1 error was not taken into account. Finally, genotyping in the earlier study was performed on specimens with a rather heterogeneous nature and consisted of genetic data derived from DNA extracted from both surgical specimens (i.e., somatic DNA) and leukocytes (i.e., germline DNA). In our current study, genetic material has only been isolated from leukocytes, thus reducing the possibility of detecting variations within somatic DNA. Conclusion In summary, we have characterized the SLC11A1 (GT) n promoter polymorphism in IBD sub-phenotypic groups and accept the null hypothesis as confirmed by family-based testing. We did, however, identify an association between an allele of the STR D2S1323 and UC upon case-control analysis. This association was not confirmed on family-based and haplotype testing, and its actual significance in IBD remains to be seen. Competing interests The author(s) declare that they have no competing interests. Authors' contributions NPSC performed genotyping, PDT analyses, and drafted the manuscript; MRE performed genotyping and was responsible for coordination of laboratory work; authors DWC and RKL performed genotyping; GAC performed statistical analyses; REP performed re-review of colonic IBD histology; SG collected all patient samples, participated in study design, coordination, and manuscript preparation and obtained funding for this work. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555593.xml

434151

Functional Dissection of an Innate Immune Response by a Genome-Wide RNAi Screen

The innate immune system is ancient and highly conserved. It is the first line of defense and the only recognizable immune system in the vast majority of metazoans. Signaling events that convert pathogen detection into a defense response are central to innate immunity. Drosophila has emerged as an invaluable model organism for studying this regulation. Activation of the NF-κB family member Relish by the caspase-8 homolog Dredd is a central, but still poorly understood, signaling module in the response to gram-negative bacteria. To identify the genes contributing to this regulation, we produced double-stranded RNAs corresponding to the conserved genes in the Drosophila genome and used this resource in genome-wide RNA interference screens. We identified numerous inhibitors and activators of immune reporters in a cell culture model. Epistatic interactions and phenotypes defined a hierarchy of gene action and demonstrated that the conserved gene sickie is required for activation of Relish. We also showed that a second gene, defense repressor 1, encodes a product with characteristics of an inhibitor of apoptosis protein that inhibits the Dredd caspase to maintain quiescence of the signaling pathway. Molecular analysis revealed that Defense repressor 1 is upregulated by Dredd in a feedback loop. We propose that interruption of this feedback loop contributes to signal transduction.

Introduction As a typical metazoan suffers numerous microbial assaults during its lifespan, survival depends on robust defense strategies. Metazoan defenses are classified as either innate or adaptive. Adaptive immunity is characterized by elaborate genetic rearrangements and clonal selection events that produce an extraordinary diversity of antibodies and T-cell receptors that recognize invaders as nonself. While of profound importance, the adaptive responses are slow and limited to higher vertebrates. In contrast, the machinery of innate immunity is germ-line encoded and includes phylogenetically conserved signaling modules that rapidly detect and destroy invading pathogens ( Medzhitov and Janeway 2000 ; Janeway and Medzhitov 2002 ). Model organisms, particularly insects, have played an important role in uncovering the wiring of innate immune pathways ( Hoffmann 2003 ). Importantly, these organisms have provided powerful genetic approaches for identifying molecules that sense pathogens, elucidating steps that trigger innate defenses, and uncovering the weaponry used to kill or divert potential pathogens ( Hoffmann et al. 1999 ). We have further refined the experimental approaches for rapid functional dissection of immune responses and describe new steps in an important pathway of the innate immune response. Signaling in innate immunity consists of three steps: detection of pathogens, activation of signal transduction pathways, and mounting of appropriate defenses. The first step is triggered by the detection of pathogen-associated molecular patterns by host pattern recognition receptors ( Akira et al. 2001 ). Typical pathogen-associated molecular patterns are β-1,3-glucan of fungi, peptidoglycan and lipopolysaccharides (LPS) of bacteria, and phosphoglycan of parasites. Signaling engages several pathways, including Toll, tumor necrosis factor, mitogen-activated protein kinase (MAPK), and Jun kinase pathways. NF-κB–type transcription factors form an important downstream nexus of the signaling pathways, and their activation promotes important defense responses. Although the defense responses are diverse and often tailored to the type of pathogen, some of the defense strategies, such as production of a panel of antimicrobial peptides, activation of phagocytic cells, and production of toxic metabolites, are evolutionarily conserved. Interest in Drosophila as a model for analyzing innate immune signal transduction had a serendipitous origin. The Toll signaling pathway was discovered and characterized in Drosophila because of its role in specification of the embryonic dorsal ventral axis ( Anderson et al. 1985 ). Similarities of pathway components to genes involved in mammalian immunity stimulated a hallmark study showing that the Toll pathway is a central mediator of antifungal and gram-positive bacterial defenses in Drosophila ( Ip et al. 1993 ; Lemaitre et al. 1996 ). It is now recognized that Toll signaling is a conserved mediator of innate immune responses. A combination of classical genetics and molecular approaches has since identified numerous components of Toll signaling in Drosophila immunity, and it has highlighted similarities to mammals at the level of signal transduction and differences at the stage of pathogen detection ( Ip et al. 1993 ; Rosetto et al. 1995 ; Nicolas et al. 1998 ; Drier et al. 1999 ; Manfruelli et al. 1999 ; Meng et al. 1999 ; Rutschmann et al. 2000a , 2002 ; Tauszig et al. 2000 ; Horng and Medzhitov 2001 ; Michel et al. 2001 ; De Gregorio et al. 2002 ; Ligoxygakis et al. 2002 ; Tauszig-Delamasure et al. 2002 ; Gobert et al. 2003 ; Weber et al. 2003 ). A second pathway, the Immune deficiency (Imd) pathway, mediates responses to gram-negative bacterial infection in Drosophila ( Lemaitre et al. 1995 ). Although similar to the mammalian tumor necrosis factor pathway, there are several differences between the two signaling cassettes, particularly at the level of activation. As it is presently understood, the Imd pathway is headed by an apparent pattern recognition receptor, the transmembrane peptidoglycan recognition protein LC (PGRP-LC; Choe et al. 2002 ; Gottar et al. 2002 ; Ramet et al. 2002 ). Although the mechanisms are largely unknown, signaling proceeds through Imd (homolog of mammalian receptor interacting protein), dTAK1 (MAP3K homolog), and a complex of Ird5/Kenny (homologous to the IKKβ/IKKγ kinase). The active IKK complex phosphorylates the p105 homolog Relish, and Dredd (caspase-8 homolog) cleaves Relish, separating an N-terminal NF-κB domain of Relish from a C-terminal ankyrin domain ( Lemaitre et al. 1995 ; Dushay et al. 1996 ; Wu and Anderson 1998 ; Hedengren et al. 1999 ; Hu and Yang 2000 ; Leulier et al. 2000 , 2002 ; Rutschmann et al. 2000b ; Silverman et al. 2000 ; Stoven et al. 2000 ; Georgel et al. 2001 ; Lu et al. 2001 ; Vidal et al. 2001 ; De Gregorio et al. 2002 ; Gottar et al. 2002 ; Khush et al. 2002 ; Naitza et al. 2002 ; Silverman et al. 2003 ; Stoven et al. 2003 ; Ryu et al. 2004 ). The N-terminal domain enters the nucleus and promotes transcription of genes encoding proteins with defense functions such as the antimicrobial peptide Diptericin (Dipt), whose expression provides a signature for activation of the pathway. Unlike the Toll pathway, which was thoroughly studied in its developmental capacities, analysis of the Imd pathway is relatively recent. Its more complete genetic dissection may well define another conserved and fundamental pathway of immune signaling. Of particular interest, a pivotal step in the Imd pathway—the regulation of Dredd-mediated cleavage of Relish—is not understood. To begin to address this, we developed a powerful RNA interference (RNAi)–based approach to functionally dissect the Imd pathway. In collaboration with others at the University of California, San Francisco, we produced a library of 7,216 double-stranded RNAs (dsRNAs) representing most of the phylogenetically conserved genes of Drosophila. We developed a cell culture assay that allowed application of this library to a high-throughput RNAi evaluation of Imd pathway activity. This screen identified numerous components of signal transduction (including negative and positive regulators of innate immune signaling), defined a hierarchy of gene action, and identified a novel gene, sickie (sick), required for activation of Relish. Focusing on regulation of the Dredd caspase, we identified a novel inhibitor of Dredd, Defense repressor 1 (Dnr1), which is upregulated by Dredd in a feedback loop that maintains quiescence. We propose that interruption of this feedback loop contributes to signal transduction. Results A Drosophila Reporter Cell Line of Imd Pathway Activity To facilitate rapid dissection of Imd pathway signaling, we established an S2 reporter cell line that expresses β-galactosidase under control of the promoter from a gene, Dipt, that encodes an antimicrobial peptide, Dipt-lacZ. Commercial preparations of LPS contain bacterial cell wall material capable of activating the receptor PGRP-LC and act as gratuitous inducers of antimicrobial peptide genes in Drosophila tissue culture cells ( Samakovlis et al. 1992 ; Engstrom et al. 1993 ; Dimarcq et al. 1997 ). Consistent with previous studies, 20-hydroxyecdysone enhanced Dipt-lacZ induction by LPS ( Figure 1 A; Silverman et al. 2000 , Silverman et al. 2003 ). Inactivation of critical Imd pathway members (PGRP-LC, Imd, Ird5, and Dredd) by RNAi virtually eliminated Dipt-lacZ induction by LPS ( Figure 1 B). In contrast, inactivation of the Toll pathway members Spaetzle, Tube, or Dif by RNAi had no effect on LPS-dependent induction of Dipt-lacZ. We conclude that LPS-dependent induction of Dipt-lacZ requires an intact Imd signaling pathway. Figure 1 A Cell Culture Screen Identifies Novel Regulators of the Innate Immune Response (A) LPS induces an increase of about 10-fold in the number of Dipt-lacZ cells that stain positively for β-galactosidase. Ecdysone sensitizes the cells and promotes the response. (B) Dipt-lacZ induction by LPS requires known Imd signaling components, but not Tl pathway members. The fraction of β-galactosidase-positive cells was normalized to the induced control (normalized %), and influence of RNAi of Tl pathway members (dif, spz, and tub) or Imd pathway members (PGRP-LC, Imd, Ird5, and Dredd) is shown. (C–H) Activity stain (X-Gal) for β-galactosidase. (C) Untreated cells. (D) Cells treated with ecdysone alone. (E) Cells treated with ecdysone and LPS. About 10% of cells express detectable β-galactosidase. (F) RNAi against the DDRi sick reduces Dipt-lacZ expression in response to LPS. (G) RNAi of a representative EDRi, the Ras signaling pathway component Cnk, enhances Dipt-lacZ induction by LPS. (H) RNAi of a representative CDRi, the actin regulator SCAR induces Dipt-lacZ in the absence of LPS. (I–J) Immunofluorescence of S2 cells with actin in red, tubulin in green, and DNA in blue. Scale bars in (I) and (J) indicate 10 μm. (I) Wild-type cells have a characteristic rounded morphology. (J) RNAi against many CDRi genes disrupts morphological features of wild-type S2 cells. S2 cells are shown treated with MESR4 dsRNA. Cells are significantly larger in appearance and less round, with irregular tubulin and actin networks. To identify additional modulators of Dipt-lacZ expression, we prepared a library of 7,216 dsRNAs representing most of the phylogenetically conserved genes of Drosophila. Using the Dipt-lacZ cell line, we performed a high-throughput RNAi screen for genes whose inactivation impinges on Dipt-lacZ induction. In one screen, we identified dsRNAs that altered Dipt-lacZ induction by LPS, either enhancing or suppressing activation. In a second screen performed without addition of LPS, we identified genes whose inactivation spontaneously activated the reporter. The phenotypes defined three categories of genes, which we named—based on the phenotype of their inactivation—decreased defense by RNAi (DDRi) genes, enhanced defense by RNAi (EDRi) genes, and constitutive defense by RNAi (CDRi) genes ( Figure 1 ). Identification of DDRi, EDRi, and CDRi Genes In an initial visual screen, dsRNAs that altered the induced or constitutive expression of β-galactosidase were selected as candidate innate immunity genes. We subjected all the initial positives to a more stringent retest where we resynthesized the candidate dsRNAs, retested these under identical conditions, and counted the number of β-galactosidase-positive cells. We defined DDRi dsRNAs as reducing the frequency of Dipt-lacZ-expressing cells to below 40% of LPS-treated controls, EDRi dsRNAs as increasing the frequency of Dipt-lacZ-expressing cells more than 2-fold, and CDRi dsRNAs as inducing Dipt-lacZ-expressing cells to a level equal to or higher than that induced by LPS. About 50% of the initial positives met these criteria, yielding 49 DDRi dsRNAs, 46 EDRi dsRNAs, and 26 CDRi dsRNAs ( Figure 2 A– Figure 2 C; Table 1 ). The entire process of screening and retesting was performed without knowing the identity of the dsRNAs. Nonetheless, we successfully identified all of the known Imd pathway components in the library (PGRP-LC, Dredd, and Relish) as DDRi genes, supporting the validity of this approach for identifying genes that affect Imd pathway signaling. Figure 2 List of Modulators of the Immune Response and a False Color Display of Their Influence on Dipt-lacZ Induction The genes identified as DDRi (A), EDRi (B), and CDRi (C) are listed, and the colored bars show the influence of the corresponding dsRNA on Dipt-lacZ expression. The top two entries in (A), (B), and (C) show control cells (no dsRNA) without and with LPS, respectively. The scales for the false colors are given at the bottom left. Dipt-lacZ levels are given in terms of percent positive cells. For exact Dipt-lacZ expression values for each dsRNA refer to accompanying supplemental tables. In (B), the color scale (right) is compressed and extended compared to (A), and an asterisk indicates genes that also caused a CDRi phenotype. In (C), the pound sign indicates morphological defects and an asterisk indicates genes that also caused an EDRi phenotype, and the division of the genes into epistatic groups is shown. To the immediate left a false-color bar (coded as in [B]) indicates the effect of the dsRNAs on Dipt-lacZ expression without LPS addition. The block of colored columns shows the results of epistasis tests. Here, we set the undisturbed level of CDRi activation to 100% (as indicated in the left column in this group and the color code below), and to the right we represent reduction of this activation by prior RNAi of different Imd pathway genes. Five epistatic clusters (I–V) were identified (indicated by the lines to the left). Table 1 Measurement of the Percent of LacZ-Positive Cells After Treatment with dsRNA Against Individual EDRi, DDRi, and CDRi Genes TF, transcription factor Cell culture conditions were as described above, with the exception that CDRi genes were not treated with LPS. For each sample, 350–550 cells were counted. Controls (+ LPS and −LPS) are the average of five independent experiments. EDRi genes were defined as having greater than or equal to twice +LPS control induction levels. DDRi genes were defined as having less than or equal to 40% of +LPS control induction levels. CDRi genes were defined as having greater than or equal to four times −LPS control levels. Results were reproducible for all genes subjected to further analysis The dsRNAs that enhance, and those that constitutively activate, the immune reporter are both expected to target inhibitors of the immune response. Nonetheless, there was only a small overlap between the EDRi genes and CDRi genes. Of the 46 confirmed EDRi dsRNAs, only five caused a CDRi phenotype, suggesting that the mechanisms that silence Imd pathway activity in the absence of infection are largely distinct from those moderating or downregulating the response to infection. We distinguish the five EDRi genes capable of constitutive activation and designate them EDRi C . EDRi C genes are listed as both EDRi and CDRi ( Figure 2 B and Figure 2 C, indicated with an asterisk). Approximately half of the CDRi dsRNAs also caused morphological defects ( Figure 2 C, indicated with a pound sign), i.e., enlarged cells with irregular cytoskeletal structures (see Figure 1 J). While we do not know the basis for the altered morphology, gene expression profiling showed that LPS induces numerous cytoskeletal regulators, suggesting that cytoskeletal rearrangement is a component of the innate immune response ( Boutros et al. 2002 ). We also observed EDRi and CDRi phenotypes upon inactivation of Act5C and Act42A. Due to extensive sequence homology, RNAi against either actin triggers destruction of both transcripts (A. Echard, G. R. X. Hickson, E. Foley, and P. H. O'Farrell, unpublished data). Inactivation of either actin with dsRNA directed to the actin UTRs demonstrated that both actin transcripts must be inactivated for an observable EDRi or CDRi phenotype ( Figure 2 B and Figure 2 C). Epistatic Evaluation of CDRi Genes As RNAi of CDRi genes leads to ectopic Dipt-lacZ induction, we reasoned that CDRi genes are required to maintain quiescence in the absence of LPS and that induction by a CDRi dsRNA corresponds to release of inhibition of the Imd pathway. The large number of CDRi genes makes it likely that individual CDRi genes inhibit distinct steps in the Imd pathway. We sought to determine the position at which the individual CDRi genes impact the Imd pathway. In contrast to Caenorhabditis elegans, several genes can be inactivated by RNAi in Drosophila without an obvious drop in the efficiency of gene inactivation ( Li et al. 2002 ; Schmid et al. 2002 ). The ability to inactivate two different gene products in sequence by RNAi provides a powerful tool to position CDRi genes relative to known Imd pathway components. In a first step, we inactivated one of three known Imd signaling components—either Imd, Dredd, or Relish. In a second step, we inactivated individual CDRi genes and monitored Dipt-lacZ induction. We reasoned that inactivation of Imd, Dredd, or Relish would not block pathway derepression by a CDRi dsRNA if the cognate CDRi impinged on the pathway at a step beyond the actions of Imd, Dredd, or Relish. Using this approach, we subdivided 20 CDRi genes into five epistatic groups ( Figure 2 C; Table 2 ). Group I contained four CDRi dsRNAs whose action was independent of Imd, Dredd, and Relish. Group II contained 12 dsRNAs whose CDRi phenotype was independent of Imd and Dredd, but depended on Relish. Group III contained two dsRNAs whose CDRi phenotype was Dredd-independent, but was reduced in the absence of Imd and Relish. Group IV contained a single dsRNA whose phenotype was independent of Imd, but dependent on Dredd and Relish. Finally, Group V contained three dsRNAs whose ability to activate the immune reporter depended on Imd, Dredd, and Relish. The epistatic relationships demonstrate that genes in Groups II–V have inputs into the known Imd pathway, while Group I might have inputs in independent pathways required for effective Dipt-lacZ expression. Table 2 Measurement of the Percent of LacZ-Positive Cells after Treatment with dsRNA against Imd, Dredd, or Relish Followed by RNAi for the Individual CDRi Cell culture conditions were as described in the body of the manuscript, with the exception that CDRi genes were not treated with LPS. For each sample, 350–550 cells were counted Sick Is a Conserved Gene Required for Relish Activation We are particularly interested in regulators contributing to activation of the Relish transcription factor by the caspase Dredd, because this is such a pivotal step in the Imd pathway and its regulation is not understood. To identify regulators that affect Relish processing, we developed an assay that more directly monitored Relish activation. We produced an S2 cell line that expresses a copper-inducible N-terminal green fluorescent protein (GFP)–tagged Relish (GFP-Relish; Figure 3 ). GFP-Relish is predominantly cytoplasmic in untreated cells ( Figure 3 A) and rapidly translocates to the nucleus upon treatment of cells with LPS or exposure to Escherichia coli ( Figure 3 B and Figure 3 D). Western blot analysis with a monoclonal anti-GFP antibody showed that GFP-Relish is rapidly processed from a full-length form to a shorter form after exposure to LPS ( Figure 3 C). These findings indicate that the GFP-Relish cell line is a reliable reporter for Relish activation. Additionally, inactivation of PGRP-LC by RNAi prevented nuclear translocation of GFP-Relish in response to bacterial exposure ( Figure 3 E), indicating that the reporter can be used to assay function of Imd pathway genes. Figure 3 A GFP-Relish Reporter Cell Line Subdivides DDRi dsRNA into Three Categories (A–B) Immunofluorescence of GFP-Relish cells with GFP-Relish in green, DNA in blue, and actin in red. Relish is predominantly cytoplasmic in untreated control cells and rapidly translocates to the nucleus of cells incubated with LPS. (C) An anti-GFP Western blot of lysates harvested from GFP-Relish cells treated with LPS for different periods. GFP-Relish rapidly shifts from a full-length form to a shorter processed form after exposure to LPS, and full-length Relish gradually reaccumulates. (D–E) Immunohistochemistry of GFP-Relish cells incubated with GFP-expressing E. coli (arrowheads) and treated with (E) or without (D) dsRNA against PGRP-LC. Imd pathway inactivation prevents bacterial-induced Relish nuclear translocation. (F) Shows effects of treatment of GFP-Relish cells with DDRi dsRNAs for 4 d prior to LPS treatment. GFP-Relish was scored as cytoplasmic (uninduced), nuclear (induced), or reduced in amount (abnormal). (G) Shows an epistatic analysis of the DDRi, sick. Suppression of sick interferes with Dipt-lacZ induction by Group III, IV, and V CDRi dsRNAs, but not those of Groups I and II, suggesting that Sick acts downstream of Imd and Dredd, but upstream of Relish in signal transduction. We tested all DDRi dsRNAs for their effects on the response of GFP-Relish to LPS ( Figure 3 F). Most DDRi dsRNAs did not affect GFP-Relish levels or its LPS-stimulated nuclear concentration, suggesting that their effects on Dipt-lacZ are independent of this step of Relish activation. Four DDRi dsRNAs (Relish, ubiquitin, CG8129, and Asph) severely reduced GFP-Relish levels, indicating that these dsRNAs directly or indirectly interfered with Relish expression or stability. The ability of these dsRNAs to block Dipt-lacZ induction suggests that Dipt-lacZ induction requires substantial levels of Relish. We identified four DDRi dsRNAs that prevented LPS-stimulated nuclear translocation of GFP-Relish: PGRP-LC, Dredd, Dox-A2, and CG10662. We named CG10662 sick . While prolonged Dox-A2 RNAi caused cell lethality, cell viability appeared unaffected by sick RNAi for up to 8 d. As sick RNAi prevents nuclear translocation of GFP-Relish and decreases Dipt-lacZ induction after LPS treatment, we propose that the Imd pathway requires sick activity for Relish-dependent Dipt-lacZ induction. Epistasis provides a second approach for positioning a DDRi gene in the hierarchy of gene action. To this end we assessed the relationship of sick to the five CDRi epistatic groups that we defined (above). We inactivated sick by RNAi and subsequently tested dsRNAs representing the five CDRi epistatic subgroupings for their ability to activate Dipt-lacZ expression in the absence of Sick ( Figure 3 G; Table 3 ). Group I and II CDRi do not require Sick, indicating that Sick acts upstream of, or in parallel to, their action, which is at the level of Relish or downstream of Relish. Induction of Dipt-lacZ by Group III and IV CDRi dsRNAs requires Sick, suggesting that Sick is required for the effective induction of Dipt-lacZ by Dredd and Imd. Combined with the observed Sick requirement for Dipt-lacZ induction and the nuclear translocation of Relish by LPS, these data imply that Sick either mediates or supports Relish activation by Dredd and Imd. Table 3 Measurement of the Percent of LacZ-Positive Cells after Treatment With dsRNA against Sick Followed by RNAi for the Individual CDRi Cell culture conditions were as described in the body of the manuscript, with the exception that cells were not treated with LPS. For each sample, 350–550 cells were counted Dnr1 Is a Novel Inhibitor of Dredd Negative regulators are likely to participate in the circuitry that controls Dredd activation of Relish. The key candidate for action at this level was the single Group IV CDRi gene, CG12489, which showed epistatic relationships consistent with a role in inhibiting Dredd. RNAi of CG12489 induced Dipt-lacZ expression without immune stimulus, indicating that CG12489 normally prevents Dipt expression. As CG12489 inactivation fails to induce Dipt-lacZ in the absence of Dredd or Relish (see Figure 2 C), we reasoned that CG12489 normally suppresses Dredd-dependent induction of Dipt-lacZ. We named this gene dnr1 and discuss its actions more fully below. Dnr1 is a conserved protein with an N-terminal ezrin/radixin/moesin domain and a C-terminal RING finger ( Figure 4 A). To confirm that Dnr1 inactivation stimulated Dipt-lacZ production, we measured the β-galactosidase activity of lysates from Dipt-lacZ cells treated with dnr1 dsRNA. Exposure of Dipt-lacZ cells to LPS reproducibly increased Dipt-lacZ production 4- to 5-fold ( Figure 4 B). Importantly, in three independent experiments, Dnr1 RNAi stimulated Dipt-lacZ production to a similar degree in the absence of LPS. Furthermore, Dipt-lacZ activation in response to LPS was essentially reduced to background levels upon inactivation of sick. These findings provide additional support for negative and positive regulation of Relish by Dnr1 and Sick, respectively. While we detected genes with similarity to dnr1 in many higher eukaryotes, we failed to find a homolog of Dnr1 in C. elegans. Interestingly, C. elegans does not rely on an Imd pathway for innate defenses ( Kurz and Ewbank 2003 ). Other RING finger proteins are E3 ubiquitin ligases that target a variety of substrates for proteolytic destruction. The RING finger motif in Dnr1 has greatest sequence homology to the RING fingers found in inhibitor of apoptosis proteins (IAPs; Figure 4 C). IAPs are critical inhibitors of caspase activity that ubiquitinate their targets and promote autoubiquitination ( Bergmann et al. 2003 ). Previous reports demonstrated that caspase inhibitors activate their own destruction and that this activity is RING finger mediated ( Yang et al. 2000 ). Consistent with these reports, we observed surprisingly low levels of accumulation of a hemagluttanin (HA)–tagged Dnr1 in transfected cells. A point mutation in a residue critical for RING finger function resulted in increased accumulation of transfected HA-Dnr1 ( Figure 4 D). We also detected a protein processing event that appears to depend on the RING finger. Upon expression of C-terminally HA-tagged Dnr1, we observed a slightly lower molecular weight isoform, suggesting N-terminal processing of Dnr1 ( Figure 4 E). The absence of this lower molecular weight isoform in cells transfected with the N-terminally HA-tagged Dnr1 ( Figure 4 D) is consistent with processing near the N-terminus. This processed isoform was absent in cells transfected with constructs containing the RING finger mutation ( Figure 4 E). The presence of the RING finger motif, and its apparent role in destabilizing Dnr1, argues that Dnr1 is a caspase inhibitor and that, given its functional role and epistatic position as an inhibitor of Dredd, it is likely to act directly to inhibit this caspase. Figure 4 Dnr1 Is a Conserved Inhibitor of Dredd Activity (A) A comparison of the amino acid sequence of Dnr1 with XP_32149 from Anopheles gambiae and human MIR. Shaded regions indicate the N-terminal ezrin/radixin/moesin domain and C-terminal RING finger. Asterisks indicate conserved residues. (B) Measurements of β-galactosidase activity in lysates prepared from Dipt-lacZ in control cells, LPS-treated cells, dnr1 dsRNA –treated cells, and sick dsRNA–treated cells exposed to LPS, respectively. Each experiment was performed in triplicate. (C) Similarity between the RING finger in Dnr1 and other IAPs. Critical residues are shaded. Asterisks indicate conserved residues. (D) Lysates from S2 cells transfected with equal amounts of N- and C-terminally HA-tagged wild-type Dnr1 (lanes 1 and 3, respectively), or N- and C-terminally HA-tagged C563Y Dnr1 (lanes 2 and 4, respectively), and analyzed by an anti-HA Western blot. Residue C563 is critical for RING finger function and is indicated with an arrowhead in (C). (E) Higher resolution of lysates from C-terminally HA-tagged wild-type or C563Y Dnr1. Mutation of the RING finger prevents accumulation of a lower isoform of Dnr1. (F and G) Subcellular localization of HA-Dnr1 transiently expressed in S2 cells treated without (F) or with (G) LPS, with HA in green, DNA in blue, and actin in red. Dnr1 Protein Levels Are Regulated by Dredd Activity While LPS had no dramatic effect on the subcellular localization of HA-Dnr1 ( Figure 4 F and Figure 4 G), exposure to LPS had a transient effect on the levels of Dnr1 protein. Addition of LPS caused an increase in HA-Dnr1 levels ( Figure 5 A), which rose 4- to 5-fold 2 h after treatment with LPS and then gradually declined. Since LPS-dependent processing of Relish by Dredd proceeded in a similar manner (see Figure 3 C), we tested whether Dredd inactivation affected Dnr1 protein levels. Cotransfection of the caspase inhibitor p35 along with HA-Dnr1 blocked HA-Dnr1 accumulation ( Figure 5 B). Similarly, even transient treatment with the caspase inhibitor z-VAD-FMK at concentrations sufficient to prevent Relish processing ( Figure 5 C) reduced LPS-dependent HA-Dnr1 accumulation ( Figure 5 D). As these data implicated caspase function in Dnr1 accumulation, we tested the five Drosophila caspases represented in our library for their influence on Dnr1 stability. Only Dredd RNAi reproducibly reduced HA-Dnr1 levels ( Figure 5 E and Figure 5 F). Consistent with a role for Dredd as the critical caspase in LPS-dependent Relish activation, of all caspases tested only Dredd inactivation blocked LPS-dependent Dipt-lacZ induction ( Figure 5 G). Figure 5 Dnr1 Protein Levels Are Regulated by Dredd Activity (A) Amounts of HA-Dnr1 transiently increase in S2 cells treated with LPS. Anti-HA Western blot of lysates from HA-Dnr1-transfected S2 cells that were incubated with LPS for indicated periods. (B) Anti-HA Western blot of lysates from S2 cells transfected with HA-Dnr1. Coexpression of the caspase inhibitor p35 dramatically inhibits HA-Dnr1 accumulation in the absence (lanes 1 vs. 3) or presence (lanes 2 vs. 4) of LPS. Actin levels are shown as a loading control. (C) Upper panel shows the percentage of cells with nuclear GFP-Relish after the indicated treatments. The lower panel is an anti-GFP Western blot of lysates from S2 cells treated in the identical manner. z-VAD-FMK prevents nuclear accumulation of GFP-Relish and GFP-Relish processing in response to LPS. (D) Anti-HA Western blot of lysates from S2 cells transiently transfected with HA-Dnr1. While 2 h incubation with LPS normally leads to a 4-fold increase (quantified by titration) in HA-Dnr1 (lanes 1 vs. 2), incubation with z-VAD-FMK prevents the accumulation (lanes 3 vs. 4). (E) Anti-HA Western blot of lysates from S2 cells transfected with HA-Dnr1. Cells had been previously incubated with (lanes 3 and 4) or without (lanes 1 and 2) Dredd dsRNA. Results are shown for two independent experiments. Actin levels are shown as a loading control. (F) Anti-HA Western Blot of lysates from S2 cells transfected with HA-Dnr1 shows that prior RNAi against the caspases Dcp-1, Ice, Nc, and Decay does not substantially affect HA-Dnr1 levels (compare with control without LPS). (G) The number of Dipt-lacZ-expressing cells after LPS treatment is greatly reduced after Dredd RNAi, while RNAi against Dcp-1, Ice, Nc, or Decay has no effect. In summary, addition of LPS to S2 cells activates Dredd and stabilizes Dnr1, while inactivation of Dredd by RNAi or caspase inhibitors reduces Dnr1 protein levels. We conclude that Dnr1 protein levels are regulated by Dredd activity. While it is not presently known how Dredd caspase function might influence Dnr1 accumulation, we note that the data are consistent with a negative feedback loop in which Dredd activity promotes accumulation of its own inhibitor, Dnr1. Discussion It was previously recognized that the Drosophila macrophage-like S2 cell line responds to bacterial cell wall components with the induction of antimicrobial peptide expression. This model lacks the complexities of communication between tissues that drive the spread of the immune response in larvae ( Foley and O'Farrell 2003 ), but it offers an exceedingly powerful system for identification of mediators of antimicrobial peptide induction. To develop a genetic approach to identify novel signal transduction components, we produced reporter cell lines to follow innate immune signaling and a library of 7,216 dsRNAs representing the conserved genes of Drosophila to inactivate genes by RNAi. We focused on a screen for immune response genes in the Imd pathway because it is the less thoroughly understood of the two immune response pathways in Drosophila. A central aspect of our strategy for dissection of the pathway was to identify negative regulators as well as positively acting genes. In addition to modulating signal transduction pathways, negative regulators participate directly in signaling when downregulated by the inducing signal. Beyond the inherent importance of this relatively unexplored group of regulators, we were interested in their potential utility as an experimental lever: Identification of inhibitors acting at numerous levels of the pathway provides tools for ordering the action of the positively acting genes in the pathway and vice versa. The experimental approach and strategy proved highly efficient, yielding numerous regulators and defining a cascade of gene action by epistasis. A secondary test for the influence of positively acting genes on the nuclear translocation of Relish and the epistasis order allowed us to narrow our focus to genes that are centrally involved in the immune response. Focusing on the unresolved issue of Dredd regulation, we characterized a negative regulator, Dnr1, that provides a critical check on unwarranted Dredd activity. Our results suggest that Dredd controls Dnr1 stability in a negative feedback loop that restricts Dredd function. Normal activation of the Imd pathway may include release or bypass of this negative feedback loop. Categories of Innate Immune Inhibitors A priori, we considered two roles for inhibitors of the Imd pathway: either suppression of spontaneous activation of immune responses in the absence of infection, or downmodulation of a response to limit or terminate it. We designed screens for both these types of activities. In conjunction with the RNAi screen for dsRNAs that blocked response to LPS, we identified dsRNAs that enhanced the response—EDRis. This phenotype represents a failure to downmodulate the response. In an independent screen without LPS, we identified dsRNAs that resulted in constitutive activation of the pathway—CDRis. Surprisingly, there was remarkably little overlap in the genes identified in these two screens: Of 26 CDRis and 46 EDRis only five were in common. At present, we do not understand the functional underpinnings of the distinctions between inhibitors that sustain quiescence (CDRis) and those that downregulate an ongoing response (EDRis). Interestingly, groups of inhibitors implicate distinct pathways in immune regulation. For example, of the 17 genes that had the strongest EDRi phenotype, four encode splicing factors and four encode products that appear to interact with RNA. This functional cluster suggests that disruptions to some aspect of RNA processing/metabolism can substantially increase the number of S2 cells that activate expression of the Dipt-lacZ reporter in response to LPS exposure. While we do not know how RNA metabolism contributes to this phenotype, the repeated independent isolation of genes lying in a functional cluster reinforces a conclusion that the process is involved. Several other functional clusters were picked up in our screens. Three genes involved in Ras signaling (MESR4, Ras, and Cnk) were identified as EDRi genes. In addition, we noted weak EDRi phenotypes with three additional Ras signaling components (rolled/MAPK, Dsor1, and Pointed). These findings argue that Ras signaling downregulates responses to LPS. This might represent a negative feedback circuit. However, the finding that MESR4 also has a CDRi phenotype suggests that the Ras/MAPK pathway may also impinge on the maintenance of quiescence. Several genes involved in cytoskeletal structure or regulation were identified among the inhibitors. Genes encoding tubulin (α-Tub84D), a kinesin motor (Klp10A), and microtubule-severing function (CG4448/katanin) were isolated as EDRi genes. Perhaps an event involving microtubule structures helps limit immune responses. The two cellular actin genes (Act5C and Act42A) were individually dispensable, but their joint inactivation produced both EDRi and CDRi phenotypes. A regulator of actin function, SCAR, was also identified as a CDRi, and both actin and SCAR CDRi phenotypes fell into epistasis Group II. This suggests that disruption of the actin cytoskeleton in quiescent cells can activate the immune response in a Relish-dependent fashion. Since S2 cells are induced to phagocytose bacteria, and changes in cell shape are induced in response to LPS, it would not be surprising if cytoskeletal functions contribute to immune responses. Indeed, microarray studies showed induction of numerous cytoskeletal components in S2 cells upon incubation with LPS ( Boutros et al. 2002 ). Our findings, however, suggest a different involvement of the cytoskeleton in which it functions to constrain S2 cells, preventing or limiting their innate immune responses. A previous conventional genetic screen for mutations leading to constitutive action of the Imd pathway in Drosophila larvae demonstrated that Relish basal signaling is maintained at a low level by proteosomal destruction of processed Relish ( Khush et al. 2002 ). A Skp1/Cullin/F-box (SCF) component was identified as involved in ubiquitination of the N-terminal Relish domain. We did not include any genes in the category of ubiquitination and proteasome function in our CDRi group. This might mean that this pathway does not influence the cellular responses in the S2 tissue culture system. However, our first round of screening suggested that RNAi to a Drosophila F-box resulted in increased basal signaling (unpublished data). This and other tentative indications of involvement of this pathway were either not reproduced or fell below the threshold in retesting. We are left uncertain about SCF contributions to immune induction in our system. A GFP-Relish Reporter Line Subdivides DDRi Genes As in the case of the CDRi and EDRi phenotypes, our screen for DDRi phenotypes identified numerous genes falling into functional categories. One potential limitation of our approach for identification of DDRi is that some genes required for ecdysone maturation may be selected as immune deficient. Additionally, one of the largest functional categories was genes involved in translation and included four ribosomal proteins, three initiation factors, two amino acyl-t-RNA synthases, and an elongation factor. It seems likely that RNAi of genes in this category affects translation of the Dipt-LacZ reporter, as opposed to affecting modulation of signaling events. To cull our collection of DDRis of such indirect modulators of the response, we developed a secondary screen that does not rely on de novo gene expression. Based on the previously described phenotypes of Imd pathway members, we reasoned that inactivation of the core components transducing the signal would compromise activation of the Relish transcription factor. To identify DDRi dsRNAs that prevented Relish activation, we prepared a GFP-Relish reporter cell line and rescreened DDRi dsRNAs for loss of GFP nuclear translocation in response to LPS. In addition to confirming a requirement for Dredd and PGRP-LC in Relish activation, we implicated a proteosomal regulatory subunit Dox-A2 and identified a novel gene sick as involved in Relish nuclear translocation in response to LPS. Although cells treated with sick dsRNA failed to mount an immune response, the cells were otherwise healthy through the course of the experiment. Dox-A2 RNAi reduced the survival of cells and was effectively lethal within a few days of the scoring of the immune response. We conclude from this that Sick and Dox-A2 contribute to the central signal transduction process, but it is presently unclear whether Dox-A2 has a significant specific input or if its effects are secondary to a global effect on cell viability. It is notable that only two DDRi genes passed our secondary screen based on GFP-Relish localization. Does this mean that all the other DDRis are not really involved? While we have not yet analyzed all these genes, we suspect that many of them will modify the Imd pathway, either impinging on the pathway at a point beyond Relish translocation, or quantitatively or kinetically modifying Relish translocation in a manner that we did not detect in our screens. Insight into this issue is likely to be derived from further epistasis tests that might place some of these DDRis in the signaling pathway. An Epistatic Network to Position CDRi and DDRi Genes We identified an unprecedented large number of immune response inhibitors (CDRi genes) in our screens. As there are diverse steps within and potentially outside the Imd signaling pathway at which the CDRi inhibitors might act, we sought to position their actions with respect to known Imd pathway functions by RNAi epistasis tests. By sequential inactivation of known Imd pathway components and CDRi gene products, we tested whether constitutive activation of immune reporters by CDRi dsRNAs depends on steps in the signal transduction pathway. In this way, we defined five distinct epistatic categories of CDRi gene products. The four CDRi genes that continue to activate immune responses despite inactivation of Imd, Dredd, or Relish are likely to act on signal-transduction-independent factors that maintain transcriptional quiescence of Dipt. The largest group of CDRis (12) depends on Relish function but not on upstream activators of Relish. These are likely to include two types of regulators: one type that sets the threshold of response so that basal activity of Relish does not trigger pathway activity, and a second type that contributes to suppression of Relish activity. The latter type of regulator might include inhibitors that impinge on the late steps in the signal transduction cascade. For example, genes whose normal function inhibits the activity of the full-length Relish transcription factor might be required to make the pathway activator dependent, and these would be found in this category. The remaining upstream epistasis groups that rely on additional signal transduction components are strongly implicated as significant contributors to the immune induction pathway. As all of the CDRis induced robust immune responses in the absence of ecdysone (unpublished data), we propose that the CDRis have their input into the Imd pathway at a level that is the same or lower than the level of the input from ecdysone. Given that this is true for all five epistatic groups of CDRis, the result suggests that ecdysone has its input at an early level of the Imd pathway. The identification of five epistasis groups of inhibitors also provides reference points for a second round of epistasis tests that position novel DDRi genes within the Imd pathway. We used this approach to show that the novel DDRi sick is required for constitutive activation of the responses by inactivation of CDRi genes in Groups III, IV, and V genes but not for the action of CDRi Group II or Group I genes. If we assume a simple linear pathway, this would indicate that Sick functions upstream of Relish and downstream of Imd and Dredd. It is noteworthy that the epistatic data are consistent with molecular data indicating that Sick is required for Dipt-lacZ induction and the nuclear translocation of Relish in response to LPS. This combination of phenotypic, epistatic, and molecular data argues for participation of Sick in the regulated activation of the Relish transcription factor. Dnr1 Prevents Ectopic Dredd-Dependent Relish Activation One epistatic group struck us as particularly interesting. While Dnr1 inactivation caused ectopic Dipt-lacZ expression, simultaneous loss of Dredd or Relish restored cells to their resting state. These data indicate that the wild-type function of Dnr1 is to prevent Dredd-dependent activation of Relish. Consistent with this hypothesis, we identified a C-terminal RING finger in Dnr1 with greatest similarity to the RING finger motifs observed in the C-terminus of IAP proteins. In addition to regulating caspase activity, IAPs also regulate their own stability through ubiquitin-mediated proteolysis. Similarly, we observed that mutation of a critical RING finger residue greatly stabilized Dnr1. These features suggest that Dnr1 is a caspase inhibitor, suggesting that it might act directly to inhibit Dredd activity. We observed that exposure of cells to LPS transiently stabilized Dnr1 and that this stabilization directly paralleled the period of Dredd-dependent Relish processing. This suggested to us that Dnr1 stability and accumulation might be regulated by its target, Dredd, a regulatory connection that could establish a negative feedback loop. We confirmed that Dredd activity is required for accumulation of Dnr1. These results suggest that Dredd modulates a RING-finger-dependent Dnr1 destruction pathway ( Figure 6 ). Figure 6 A Schematic of the Proposed Relationships of the Novel Immune Regulators, Sick and Dnr1, to Dredd and Rel Pointed and blunt arrows indicate activation and inhibition, respectively. Both Sick and Dredd are required for translocation of Rel to the nucleus and for activation of Dipt expression and they are consequently positioned upstream of Rel as activators. In the absence of Sick or Dredd, Dnr1 function is not needed to maintain pathway quiescence. Thus, Dnr1 is ordinarily required to either inhibit Sick and Dredd functions or to negate their actions, and we have indicated these regulators as being downstream of Dnr1 (A). Although we have no epistatic data that separates the action of Sick and Dredd, Dredd appears to directly cleave Rel and is hence likely to be immediately upstream of Rel. Sick might function in conjunction with Dredd or as an activator of Dredd. Since dnr1 RNAi does not enhance the response to LPS, we suggest that its inhibitor activity is either repressed or bypassed upon exposure to LPS. Consequently, we have shown that treatment with LPS counteracts Dnr1-dependent Dredd inhibition (B). We do not mean to preclude other actions of LPS that might contribute to induction, but it is notable that inactivation of Dnr1 is sufficient to activate signaling. Finally, we have shown that Dnr1 levels are affected by Dredd, and we have indicated this with a positive feedback arrow. Our results are consistent with a feedback inhibitory loop where Dredd activity promotes accumulation of its own inhibitor ( Figure 6 ); however, it is not clear under what circumstance this loop functions. Since Dnr1 inactivation did not enhance Dipt-lacZ production by LPS, we propose that Dnr1 inhibition of Dredd is suppressed or bypassed by LPS treatment and that Dnr1 is not essential for downregulation of an ongoing response. Further, as suppression of Dnr1 by RNAi is sufficient to activate immune responses, Dnr1 functions in the absence of induction and this function is required for quiescence. Thus, LPS inactivation of Dnr1 function ought to be sufficient to trigger Dredd-dependent cleavage of Relish in the Imd pathway, and it could make a significant contribution to pathway activation. In summary, a new and powerful screening approach has provided many candidate regulators of the Imd pathway of the innate immune response, and we suggest that the newly identified contributors Dnr1 and Sick will govern central steps in the regulatory cascade that activates the Relish transcription factor. While our analysis has led to a focus on these two regulators, we suspect that other genes among those isolated will also make important direct contributions to the Imd pathway. Furthermore, some of the groups of genes falling into functional clusters are likely to define physiologically relevant inputs into the induction pathway. Materials and Methods Generation of dsRNA library A library of DNA templates bearing the T7 RNA polymerase promoter at each 5′ end was prepared from genomic DNA in a two-step PCR protocol. In the first round of PCR, targeted regions of DNA were amplified using gene-specific primers (18–22 nucleotides) with a 5′ GC-rich anchor (GGGCGGGT). Primers were designed to amplify a region of nonintronic genomic DNA between 250 and 800 bp with minimal sequence overlap to all other amplimers. Templates from the first step were amplified in a second round using a universal primer containing the T7 RNA polymerase promoter sequence followed by the GC-rich anchor TAATACGACTCACTATAGGGAGACCACGGGCGGGT. dsRNA was generated from templates in in-vitro transcription reactions for 6 h at 37 °C. In vitro transcription products were annealed by heating to 65 °C and slowly cooling to room temperature. All products were tested for yield and size by gel electrophoresis, with 97% giving satisfactory results. Generation of stable cell lines Dipt-lacZ cell and copper-inducible GFP-Relish stable S2 cell lines were generated according to the Invitrogen Drosophila Expression System Protocol using hygromycin B (Invitrogen, Carlsbad, California, United States) as a selection marker. The Dipt-lacZ plasmid has been described previously ( Dimarcq et al. 1997 ). To prepare N-terminally GFP-tagged full-length Relish, the stop codon in enhanced GFP (EGFP) was replaced with a Not site, and EGFP was cloned into pUAST as an EcoRI/NotI fragment. Full-length Relish cDNA was fused in frame to the EGFP coding sequence as a NotI/XbaI fragment, and the entire sequence was confirmed by sequencing. GFP-Relish was cloned into pMT/V5-HisB as an EcoRI/XbaI fragment to allow copper-inducible expression of GFP-Relish under control of the metallothionine promoter. Cell culture S2 cells were plated into glass-bottomed 96-well microplates (BD Biosciences Pharmingen, San Diego, California, United States) with 40,000–50,000 cells in 200 μl of Schneider's Drosophila medium (GIBCO, San Diego, California, United States) supplemented with 10% heat-inactivated fetal calf serum, penicillin, streptomycin, and hygromycin per well. dsRNA was added to each well at a final concentration of 10 μg/ml. Cells were cultured for 4 d at 25 °C and incubated an additional 24 h in 1 μM 20-hydroxyecdysone (Sigma, St. Louis, Missouri, United States). LPS (Calbiochem, San Diego, California, United States) was added at a final concentration of 50 μg/ml for 12 h. RNAi protocols were as previously described ( Clemens et al. 2000 ). β-galactosidase assays To measure β-galactosidase in S2 cells, medium was aspirated from the wells, and cells were fixed in 0.5% glutaraldehyde in PBS for 30 s. Cells were then incubated in X-Gal staining buffer overnight at 37 °C (10 mM phosphate buffer [pH 7.2], 150 mM NaCl, 1 mM MgCl 2 , 3.5 mM K 3 Fe(CN) 6 , 3.5 mM K 4 Fe(CN) 6 , and 0.2% X-Gal in DMF). β-galactosidase activity assays were performed as described previously ( Dimarcq et al. 1997 ). Microscopy, immunofluorescence, and image processing β-galactosidase induction in S2 cells was observed with a Leica (Wetzlar, Germany) IMRB microscope. Immunofluorescent images were taken on an Olympus (Tokyo, Japan) IX70 driven with DeltaVision software (Applied Precision, Issaquah, Washington, United States). S2 cells were deposited on Superfrost Plus Gold slides (Fisher Scientific, Hampton, New Hampshire, United States) for immunofluorescence. Cells were fixed for 10 min in 4% formaldehyde (Sigma). Tubulin was detected with mouse anti-α-tubulin (Sigma). DNA was visualized with Hoechst 33258, and actin was visualized with rhodamine-coupled phalloidin (both from Molecular Probes, Eugene, Oregon, United States). Images were processed with Adobe Photoshop 5.5, and figures were assembled with Adobe Illustrator 9.0. Western blotting Dnr1-expressing vectors were prepared by cloning Dnr1 cDNA into pAc5/V5HisA (Invitrogen). The C365Y mutant form of Dnr1 was prepared with the Stratagene point mutation protocol using a TTCAATCCGTACTGTCACGTC sense primer and a GACGTGACAGTACGGATTGAA antisense primer. The mutation was confirmed by sequencing. For experiments with z-VAD-FMK, S2 cells were incubated in 100 μM z-VAD-FMK for 4 h at room temperature. Cells were harvested by centrifugation at 1,000 g for 3 min, washed in PBS and lysed on ice for 10 min in lysis buffer (0.5 M HEPES [pH 7.5], 150 mM NaCl, 5 mM EDTA, 0.2% NP40, PMSF, leupeptin, pepstatin, NaF, and microcystine LR). Lysate was spun for 10 min at maximum speed, and the supernatant was added to sample loading buffer. Samples were separated by SDS-PAGE and analyzed by Western blotting. Anti-GFP antibody was purchased from BabCO (Richmond, California, United States), and HA and actin antibodies were purchased from Sigma. Table 1 Continued Table 1 Continued Table 1 Continued

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC434151.xml

545780

Assembly and characterization of heterochromatin and euchromatin on human artificial chromosomes

An assay of the formation of heterochromatin and euchromatin on de novo human artificial chromosomes containing alpha satellite DNA revealed that only a small amount of heterochromatin may be required for centromere function and that replication late in S phase is not a requirement for centromere function.

Background In the post-sequencing phase of genome characterization, it is important to understand the contribution of non-coding sequences to higher-order genome structure and stability. Maintenance of genome integrity and the faithful transmission of genetic information in mitosis and meiosis are essential to organism survival and are critically dependent on two repetitive chromosomal elements. Telomeres protect against chromosomal truncation or fusion events [ 1 ], while centromeres ensure faithful chromosome segregation through cell division [ 2 - 4 ]. Failure in the function of these elements can lead to genomic instability, with often catastrophic consequences in humans such as miscarriage, congenital birth defects or cancer. In contrast to the telomere, whose properties have been well explored at the genomic and molecular levels [ 5 ], the human centromere remains relatively poorly characterized, and experimental systems for the genomic study of centromere formation and behavior are only just being developed and optimized [ 6 - 14 ]. Defining the minimal DNA sequences required for centromere function on a normal human chromosome has proved challenging, owing to the complex nature of inter- and intra-chromosomal homology and variability in genomic DNA content near the primary constriction. Common to all normal human centromeres are large amounts of alpha-satellite DNA, which is comprised of a family of diverged 'monomers' of around 171 base-pairs (bp) that have been amplified in multimeric groups (higher-order repeats) on different chromosomes to form chromosome-specific arrays typically megabases in length [ 15 - 17 ]. In addition, the core of higher-order repeat alpha-satellite is, where examined in detail, surrounded by other alpha-satellite sequences that fail to form a recognizable higher-order structure (so-called 'monomeric' alpha satellite) [ 10 , 18 - 20 ]. Together, the two types of centromeric repeat span up to several megabases of genomic DNA at each centromere region and account for much of the largest remaining gaps in the human genome sequence assembly [ 21 , 22 ]. Support for a critical role for alpha-satellite DNA in centromere function comes from recent studies on the human X chromosome, where the most abundant alpha-satellite sequence at this centromere, DXZ1, has been shown to be sufficient for centromere function [ 10 , 23 ] and, more generally, from studies demonstrating the formation of de novo centromeres on human artificial chromosomes following transfection of some types of alpha-satellite sequences into human cells [ 6 - 14 ]. Paradoxically, despite conservation of the functional role of the centromere in every eukaryotic cell, DNA sequences at eukaryotic centromeres are quite divergent in sequence even between closely related species [ 24 , 25 ]. Although primary genomic sequence has not been conserved at eukaryotic centromeres, they do, nonetheless, share features in common such as a structure based on tandem repeats, overall AT-rich composition, and packaging into specialized centromeric chromatin marked by the presence of centromere-specific histone H3 (CenH3) variants (reviewed in [ 4 , 26 , 27 ]). The ability of different genomic sequences to fulfill centromeric requirements in different species is in accord with data showing that the DNA normally associated with the genetically mapped centromere on normal human chromosomes is not always sufficient or necessary for centromere function. Rare chromosomal rearrangements can result in either dicentric chromosome formation, where one centromere is typically inactivated [ 28 , 29 ], or in the formation of neocentromeres, where a centromere assembles on DNA that is not associated with the normal centromere genomic locus (reviewed in [ 3 ]). Together, these observations suggest that epigenetic factors are critical for centromere function [ 30 ] and point to the as-yet incompletely understood interplay of underlying genomic DNA sequences located in the centromeric region and their ability to package into specialized centromeric chromatin [ 2 , 4 , 27 ]. Recent evidence suggests that a complex system of epigenetic modifications based on histone variants and histone tail modifications is important for centromere activity (reviewed in [ 4 , 31 ]), in much the same way as a histone code is involved in determining the transcriptional competence of DNA [ 32 ]. Although the epigenetic basis of centromere function is not yet fully defined, a strong candidate for specifying the site of the functional centromere (kinetochore-forming region) is the family of CenH3 variants, which are conserved from yeast to humans and are essential to viability of the organism (reviewed in [ 2 ]). In humans and flies, CenH3 is restricted to the centromere where CenH3- and typical H3-containing nucleosomes exist in an alternating arrangement, generating a unique chromatin structure that may be important for centromere function [ 33 , 34 ]. The most completely studied complex eukaryotic centromere at the molecular level is that of the fission yeast Schizosaccharomyces pombe . Detailed analyses of a 40-kilobase (kb) S. pombe centromere revealed that it encompasses both the kinetochore, as defined by the exclusive association of Cnp1, the fission yeast CenH3, with the central core element [ 35 ] and adjacent repeats enriched for heterochromatin-associated factors [ 36 ] that are important for centromeric cohesion [ 37 - 40 ]. Within the heterochromatic domains, histone H3 is methylated at lysine 9 (H3MeK9), resulting in the recruitment of the heterochromatin protein HP1-homolog Swi6 [ 41 ]. There is substantial evidence that HP1 is involved in setting up and/or maintaining a repressed chromatin state in several epigenetic systems (reviewed in [ 42 ]). HP1 proteins are conserved and localize to centromere regions in human and mouse cells [ 43 - 45 ]. Human cells express three HP1 isoforms, HP1α, HP1β and HP1γ. HP1α and HP1β localize primarily to pericentromeric regions, while HP1γ is dispersed at sites along chromosome arms [ 43 ]. Furthermore, modified H3MeK9 nucleosomes, which create a binding site for HP1 (reviewed in [ 46 ]), have also been localized cytologically to centromere regions in flies and mice [ 44 , 47 - 53 ]. These observations suggest a model in which local modifications of chromatin composition represent a crucial and highly conserved element necessary for the specification and/or maintenance of complex eukaryotic centromeres [ 2 ]. Consistent with these models, chromatin immunoprecipitation assays with highly specific antibodies have shown that both mouse minor and major satellite DNA sequences exhibit trimethylation of histone H3 at lysine 9 [ 51 , 53 ]. However, while the association of histone modifications typical of repressive heterochromatin has been clearly demonstrated for sequences that flank the functional centromere, it is less certain what modifications, if any, may characterize the CenH3-containing chromatin of the functional centromere itself. Indeed, many of the characteristics historically assigned to pericentromeric DNA (that is, repressive heterochromatin and late-replication in S phase [ 54 , 55 ]) may be features of the surrounding heterochromatin, more so than of the functional centromere per se . One way to address the interacting and complementary role(s) of DNA sequence and trans -acting chromatin factors in human centromere function is through the construction of detailed genomic maps of human centromeric regions and evaluation of their associated proteins [ 10 , 19 , 56 , 57 ]. An alternative empirical approach is to construct minimal human artificial chromosomes from defined alpha-satellite DNA sequences [ 6 - 14 ] as tools for evaluating the essential genomic requirements of centromere specification. Indeed, previous studies have shown that the human CenH3 - centromere protein A (CENP-A) - is deposited at the centromere on artificial chromosomes constructed from alpha-satellite DNA [ 12 , 13 , 58 ]. However, it is not known whether heterochromatin formation is required for centromere establishment and propagation and/or whether de novo centromeres on human artificial chromosomes without large amounts of adjacent heterochromatin demonstrate the same chromatin characteristics as either normal human centromeres or human artificial chromosomes with large amounts of heterochromatin. In the present study, we have characterized the nature of heterochromatin and euchromatin formed on a series of human artificial chromosomes derived from higher-order repeat alpha satellite from chromosomes X or 17 [ 12 , 14 ]. While large artificial chromosomes contain substantial amounts of heterochromatin (characterized by the presence of modified H3MeK9 nucleosomes and HP1α) and replicate later in S phase, small artificial chromosomes show features more consistent with the euchromatin of the chromosome arms, including the presence of histone variants typical of expressed euchromatin and replication earlier in S phase. These data suggest that the chromatin environment required for de novo centromere formation and function is likely to be generally conducive to gene expression, as will probably be required for either gene-transfer experiments and/or functional genomic applications of the artificial chromosome technology. Further, the data raise the possibility that functional centromeres may adopt a novel chromatin state that is, contrary to what has been long assumed, quite distinctive from that of conventional heterochromatin. Results To examine the chromatin composition of human artificial chromosomes, we used a panel of artificial chromosomes formed after transfection with vectors containing either synthetic chromosome 17 (D17Z1) or cloned X chromosome (DXZ1) alpha-satellite sequences [ 12 , 14 ]. Each of the artificial chromosomes tested contains a functional de novo centromere assembled from the transfected DNA, as well as at least one copy of a functioning gene used as a selectable marker. Together, this panel of artificial chromosomes provides an opportunity to examine the nature of heterochromatin and euchromatin assembled on the transfected DNA sequences. The high mitotic stability and de novo composition of artificial chromosomes generated from D17Z1 (17-E29, 17-D34 and 17-B12) or DXZ1 (X-4 and X-5) have been described [ 12 , 14 ]. As a more direct measure of artificial chromosome segregation errors, we have used an assay that allows cells to undergo anaphase but cannot complete cytokinesis [ 14 ]. Using fluorescence in situ hybridization (FISH), artificial and host chromosome segregation products can be measured and nondisjunction or anaphase lag defects recorded. In X-4 and X-5, artificial chromosomes mis-segregated in 1.8% and 2.4% of cells, respectively ([ 14 ] and Table 1 ). Similar analyses of artificial chromosome segregation errors in 17-B12 revealed that they mis-segregated in 2.4% of the cells (Table 1 ). This segregation error rate is comparable to that found for the majority of other human artificial chromosomes previously characterized [ 14 ]. Artificial chromosomes in 17-E29 and 17-D34 have segregation efficiencies corresponding to more than 99.9% per cell division, using metaphase analyses [ 12 ]. For comparison, we also examined an additional cell line, 17-C20, which contains highly mitotically unstable D17Z1-based artificial chromosomes. In 17-C20, artificial chromosome copy number was high (average 4.7 per cell) and artificial chromosomes were lost from the cell population by 30-40 days of culture without selection, despite containing both inner (CENP-A) and outer (CENP-E) kinetochore proteins (data not shown). In the anaphase assay, 12.2% of artificial chromosomes in 17-C20 were mis-segregating (at 12 days without selection) and the predominant defect was anaphase lag (Table 1 ). Sizes of D17Z1-containing artificial chromosomes were based on comparison of the signal intensity on the approximately 3 Mb D17Z1 array on chromosome 17 to intensities on the artificial chromosomes using FISH analyses with a D17Z1 probe (Table 2 ; see also Figures 2 and 3 in [ 12 ]). Artificial chromosomes that had signal intensities several-fold less than the endogenous D17Z1 signals were estimated to be 1-3 Mb in size, whereas artificial chromosomes that produced signals similar to or several-fold more intense than those of the endogenous D17Z1 arrays were estimated to be in the 3-10 Mb size range. Similar comparisons of the signal intensities on the DXZ1-based artificial chromosomes with those of the host DXZ1 signals were used to estimate the sizes of the DXZ1-based human artificial chromosomes (Table 2 and data not shown). Properties of artificial chromosomes used in the present study are summarized in Tables 1 and 2 . Variation in levels of heterochromatin-associated factors correlates with artificial chromosome size To test whether human artificial chromosomes were capable of forming heterochromatin, we first examined several established markers of heterochromatin on the artificial chromosome panel. Indirect immunofluorescence with an antibody recognizing histone H3 modified by trimethylation at lysine 9 and lysine 27 (H3TrimK9/K27) was applied to metaphase spreads. Methylation of lysines at these sites has been associated with formation of repressive chromatin, including pericentric heterochromatin in mouse cells [ 32 , 51 - 53 , 59 , 60 ]. As shown in Figure 1a and 1b , small D17Z1-based artificial chromosomes, estimated to be in the 1-3 Mb size range (Table 2 ), do not stain detectably with the H3TrimK9/K27 antibody, in contrast to the centromeric regions of the natural human chromosomes that stain, in some cases intensely, with this antibody. On the other hand, larger artificial chromosomes, estimated to be in the 3-20 Mb size range (Table 2 ), stained strongly for H3TrimK9/K27 modifications (Figure 1c-g ), often at levels greater than those of many endogenous centromeric regions (Figure 1g ). It is clear that at least large amounts of transfected alpha satellite are capable of assembling into heterochromatin in the context of human artificial chromosomes. Whether small artificial chromosomes are truly negative for this marker of heterochromatin, or whether they assemble only small amounts of heterochromatin below the level of detection, cannot be assessed with this assay. Nonetheless, they clearly have assembled far less of this epigenetically modified heterochromatin than exists at the relevant endogenous 17 centromeric regions (Figure 1 ). In a parallel approach, we examined the distribution of HP1α in four lines containing D17Z1-based artificial chromosomes. Each line was stably transfected with a Myc-epitope tagged form of HP1α (see Materials and methods) to permit detection of HP1α using an anti-Myc antibody. The smaller artificial chromosomes stained very weakly (at a level similar to that of the staining on the euchromatic chromosome arms), well below the levels of HP1α detected at the centromeric region of the endogenous chromosome 17s (Figure 2a,b ). As seen with the H3TrimK9/K27 antibody, the larger artificial chromosomes stained strongly for HP1α (Figure 2c,d ), at levels comparable to the endogenous chromosome 17s. The intensity of HP1α-Myc staining was variable at endogenous human centromere regions (Figure 2d ); similar results were obtained using a primary anti-HP1α antibody (data not shown). This contrasts with the amount of CENP-A, which appears to be present at consistent levels at all normal human centromeres [ 61 ] and artificial chromosomes tested (Figure 2d ) [ 12 , 13 , 58 ]. Notably, the CENP-A signal is localized to a discrete subdomain within the larger artificial chromosomes, whereas HP1α covers a much larger area of the artificial chromosome (Figure 2d ). This suggests that HP1α may be a marker for generalized pericentromeric heterochromatin that flanks the kinetochore-associated alpha satellite of the functional centromere, rather than a marker of the functional centromere per se . Such a model [ 2 , 3 ] is also consistent with the observation that small artificial chromosomes, which contain little if any of the flanking heterochromatin, do not contain elevated levels of HP1α (Figure 2a,b ; Table 2 ). Euchromatin forms on artificial chromosomes For their potential use as gene-transfer vectors or as general vehicles suitable for interrogation of genome function, human artificial chromosomes must also be capable of forming euchromatin to support gene expression. Indeed, one would hypothesize that at least small amounts of transcriptionally active chromatin must form during artificial chromosome formation to permit expression of the selectable marker gene(s) contained on the transfected constructs [ 10 , 12 , 14 ]. It has previously been shown using immunocytochemical methods [ 62 , 63 ] that methylation of histone H3 at lysine 4, an epigenetic modification associated with transcriptionally permissive chromatin [ 64 - 66 ], is generally enriched on autosomes and depleted at the repressed inactive X chromosome and human centromere regions. As a test for formation of permissive chromatin on artificial chromosomes, we stained metaphase spreads with an antibody that recognizes histone H3 dimethylated at lysine 4 (H3DimK4). All artificial chromosomes tested stained positively for H3DimK4 modifications (Figure 3a-f ; Table 2 ). In contrast, the endogenous centromeric regions were depleted for H3DimK4 staining, although, as noted above for markers of heterochromatin formation, this depletion may reflect the state of the surrounding heterochromatin, rather than that of the functional centromere per se . Previous structural analyses of artificial chromosomes indicate that they consist of input DNA multimers arranged as blocks of alpha-satellite DNA interspersed with vector sequences [ 7 , 11 , 12 ]. This structural organization is consistent with the presence of multiple selectable marker genes and differs from the large uninterrupted blocks of alpha-satellite DNA found at all human centromeres that are typically under-represented for this active chromatin mark (Figure 3 ). Because mitotically stable artificial chromosomes can have permissive as well as repressive chromatin present, these data suggest that this chromatin configuration does not significantly disturb mitotic centromere function. Two modes of artificial chromosome replication timing While the genomic determinants of potential origins of DNA replication in the human genome, as well as of their timing of replication during S phase, are still not well understood, the generally accepted paradigm is that expressed sequences replicate in the first half of S phase, while non-expressed sequences replicate in the second half [ 67 ]. Consistent with this pattern, alpha-satellite DNA, as well as constitutive heterochromatin (such as that found on the Yq arm), replicate in the mid to late S phase period [ 54 , 55 , 68 , 69 ]. In the present study, we have asked whether D17Z1-based artificial chromosomes replicate at a similar time to endogenous chromosome 17 alpha-satellite DNA. To determine the time of replication, unsynchronized cells were pulsed with bromodeoxyuridine (BrdU) for 2 hours, followed by a thymidine chase for varying lengths of time before harvesting cells in metaphase (see Materials and methods). Detection of BrdU incorporation at sites of DNA replication was performed using indirect immunofluorescence with an anti-BrdU antibody on metaphase spreads. While there was overlap between artificial chromosome replication timing patterns and those of the host 17 centromere regions during mid S phase (Table 3 ), we found two modes of artificial chromosome replication timing. The heterochromatin-enriched artificial chromosomes (17-B12 and 17-C20; see Table 2 ) commenced replication in mid S phase (2-4 hours into S phase) and completed replication by 6 hours into S phase (Figures 4 and 5c ; Table 3 ). In contrast, the heterochromatin-depleted artificial chromosomes (17-D34 and 17-E29; see Table 2 ) started replicating within the first 2 hours of S phase (early S phase) and their replication was completed by 4 hours into S phase (Figure 5a,b ; Table 3 ). That these differences are characteristic of each particular artificial chromosome is suggested by the observation that, in all lines, when multiple artificial chromosomes were present in a given cell, they are frequently replicated synchronously (Figures 4c and 5a,c ). From these data, it is tempting to propose that the presence of large amounts of heterochomatin in the larger artificial chromosomes may have influenced replication timing on these artificial chromosomes and promoted a shift towards later in S phase. Discussion Human artificial chromosomes provide a novel system for analyzing cis - and trans -acting factors necessary for chromosome segregation and offer potential for both functional genomics and gene-transfer applications. The artificial chromosomes we used contain defined alpha-satellite DNA sequences [ 12 , 14 ]. Studying how epigenetic components assemble with alpha satellite to form a de novo centromere on artificial chromosomes may reveal the critically important components and may help distinguish between those features that are characteristic of the functional centromere itself and those that are markers of the surrounding heterochromatin. Such a distinction is extremely difficult in normal human chromosomes but should be enhanced by the ability to generate a variety of different artificial chromosomes made with different input sequences. Recent detailed molecular studies in the fission yeast have revealed that such epigenetic factors are critical for centromere function. The fission yeast CenH3, Cnp1, is deposited only at the central core domain, while heterochromatin (marked by methylation of histone H3 at lysine 9 and by binding of the HP1 homolog, Swi6) forms on the surrounding inverted repeats [ 35 , 36 , 41 ]. The yeast data, together with the observations that CenH3s are conserved and that H3K9-modified nucleosomes and HP1 proteins are often found close to the centromere in higher eukaryotes, have contributed to the development of models for centromere packaging in the larger chromosomes of multicellular eukaryotes, including mammals. In these models, a specific centromeric chromatin configuration, in which CenH3-containing chromatin is surrounded by pericentric heterochromatin, is conserved and may be an important determinant of centromere function [ 2 - 4 ]. While the data presented here are largely consistent with these models, they permit two important refinements. First, large amounts of heterochromatin (containing alpha satellite and marked by H3TrimK9/K27 staining, HP1α binding and late replication) are not required for effective chromosome segregation during mitosis; indeed, the small artificial chromosomes examined here do not contain detectable amounts of H3TrimK9/K27 (Table 2 ). Second, the cytological characteristics of heterochromatin (repressive chromatin and later replication in S phase), classically attributed to the centromere [ 54 , 55 ], may instead reflect features of the surrounding heterochromatin and do not appear to define critical properties of the functional centromere. Our own data would argue that the functional centromere - at least as assembled on the smaller D17Z1-based human artificial chromosomes - is instead characterized by a distinctive chromatin containing CenH3 (CENP-A) that can form within regions epigenetically modified with markers of euchromatin (Tables 1 and 2 ). This conclusion is consistent with parallel work on the organization of centromeric chromatin of normal Drosophila and human chromosomes [ 34 ]. The finding that CENP-A-containing chromatin can be deposited within euchromatin-rich artificial chromosomes that are highly mitotically stable (more than 99.9 % segregation efficiency per cell division) yet depleted for heterochromatin modifications, suggests that only a very small amount of heterochromatin may be required on an artificial chromosome (from observations in yeast [ 37 - 40 ] and chicken DT40 cells [ 70 ] this is presumably for assembling the cohesin complex), and that this could also be true for human centromeres. This study also addresses the question of timing of replication of D17Z1-based artificial chromosomes. The smaller artificial chromosomes that completely overlap with CENP-A [ 12 ] and euchromatic modifications (Figure 3 ) replicate early in S phase whereas the larger artificial chromosomes that have assembled heterochromatin (H3TrimK9/K27 and HP1α) in addition to euchromatin replicate later in S phase (Table 3 ). The later onset of replication on the larger artificial chromosomes is similar to that of host chromosome 17 centromere regions that are also enriched for H3TrimK9/K27 and HP1α (Figures 1 and 2 , Tables 2 and 3 ). With the caveats that higher-resolution methods will be required to determine the precise replication timing of the CENP-A domain on the artificial chromosomes, and that differences in vector DNA content may be influencing origin establishment and/or usage, our observations are consistent with local chromatin modification being an important factor influencing artificial chromosome replication. Chromatin composition as a factor in determining replication timing has also been implicated in a study of a Drosophila minichromosome deletion series. In this study, replication timing was shifted to an earlier point in mid-S phase following deletion of large amounts of pericentromeric heterochromatin from the minichromosomes [ 71 ]. Support for a direct role of chromatin composition in replication timing comes from studies in budding yeast, where regions associated with acetylated histones (an epigenetic mark of active chromatin) replicate earlier than those depleted for this histone modification [ 72 ]. However, unexpected recent evidence from fission yeast has shown that centromeric heterochromatin replicates early in S phase, suggesting that chromatin composition is not a uniform determinant of replication timing in lower eukaryotes [ 73 ]. As the euchromatin-rich and highly mitotically stable artificial chromosomes replicate in the first half of S phase (in 17-E29, the majority of artificial chromosomes (75%, n = 20) replicated in the first 2 hours of S phase (Table 3 )) these findings challenge the current dogma that replication later in S phase is an obligatory function of the centromere. The present findings are also supportive of earlier studies suggesting that replication timing of CenH3-containing chromatin is not a determinant of the functional centromere [ 69 , 71 ]. Cytological data indicate that the amount of CENP-A modified chromatin (in addition to several other kinetochore-associated CENPs) is similar on endogenous human chromosomes and on all artificial chromosomes regardless of the amount of total alpha satellite present. This suggests that the amount of CENP-A chromatin and/or the size of the kinetochore is regulated and/or limited in some manner [ 6 - 14 , 58 , 61 ]. In contrast, the results of the present study indicate that the heterochromatic fraction of centromeric DNA (on both endogenous chromosomes and artificial chromosomes) is highly variable. In line with current models, we did detect elevated levels of H3TrimK9/K27 modifications and HP1α, diagnostic of heterochromatin on large artificial chromosomes generated from chromosome 17 (D17Z1) or X (DXZ1) alpha-satellite DNA. However, no immunocytochemically detectable heterochromatin (H3TrimK9/K27) was associated with the smaller artificial chromosomes. To evaluate their potential for characterization of genome sequences and, eventually, for gene transfer or gene therapy applications, we sought to determine the extent of transcriptionally competent chromatin formation in artificial chromosomes. Epigenetic modification of histone H3 by dimethylation at lysine 4 (H3DimK4), a marker of transcriptionally competent chromatin, was present on all artificial chromosomes tested. This contrasts with the staining pattern associated with the centromere regions on human metaphase spreads, where this modification is largely undetectable, probably reflecting the general absence of genes mapping to centromere regions (Figure 3 ). As selectable marker genes are expressed on artificial chromosomes, it may be presumed that at least a portion of the artificial chromosome chromatin structure is transcriptionally permissive, consistent with the positive staining for H3DimK4. In line with these observations, large human transgenes have been expressed from artificial chromosomes [ 74 - 76 ] and selectable marker genes on artificial chromosomes assemble acetylated histones, another marker of euchromatin [ 77 ]. Furthermore, detection of transcription of genes within the CenH3 domain of a human neocentromere [ 78 ] and a rice centromere [ 79 ] suggests that CenH3-containing chromatin can be transcriptionally competent. The relationship between active and repressive chromatin and underlying genomic sequences on the larger artificial chromosomes is not known and will require more detailed follow-up analyses. As other detailed chromatin immunoprecipitation studies have shown that methylation of histone H3 at lysine 4 or lysine 9 seem to be mutually exclusive [ 64 , 65 ], it will be interesting to find out how the two types of chromatin are assembled during artificial chromosome formation and to find out if there is a mechanism that prevents spreading of chromatin between the heterochromatic and euchromatic sub-domains. An advantage of the artificial chromosome system is the capacity to manipulate sequence content and to test directly the involvement of candidate sequences in gene expression, chromatin establishment or timing of DNA replication. In this study we included one line, 17-C20, that contains de novo D17Z1-based artificial chromosomes that retain both inner and outer kinetochore components yet are highly mitotically unstable as a result of their rapid loss in the absence of selection and the very high segregation error rate (12.2%) detected in the anaphase assay (Table 1 ). The artificial chromosomes in this line have a global chromatin composition indistinguishable to that of similar-sized D17Z1-based mitotically stable artificial chromosomes, as both H3DimK4- and H3TrimK9/K27-modified nucleosomes and HP1α are assembled (Table 2 ). Our study has not revealed the cause of the segregation defect of artificial chromosomes in 17-C20, and so a more extensive examination of additional epigenetic markers or centromere-associated factors may be informative. Detailed anaphase segregation analyses of D17Z1- and DXZ1-based artificial chromosomes have revealed that there is a range of mitotic stability among artificial chromosomes [ 14 ]; future studies will aim to characterize the mechanistic basis of the segregation defects and the relative contribution of genomic and/or epigenetic factors to chromosome behavior. Conclusions In summary, we have shown that artificial chromosomes assemble transcriptionally permissive chromatin and that there is a link between artificial chromosome size and the assembly of heterochromatin. Our results with the artificial chromosome panel are largely consistent with current models proposing that the formation of heterochromatin within the vicinity of CENP-A chromatin is functionally important, although the amount of heterochromatin assembled is quite variable, suggesting either that it is required only in small amounts or that it perhaps could even be dispensable. Strikingly, the studies here on the chromatin composition of artificial chromosomes, in combination with studies on normal human centromeres [ 34 ], strongly suggest that the chromatin state of the functional centromere region (as defined by CenH3 association) is quite distinct from pericentric heterochromatin. The artificial chromosome system provides a new set of reagents for investigating the role of both defined alpha-satellite DNA sequences and trans -acting epigenetic factors that cooperate to form a functional human centromere. A fuller understanding of the structure-function relationships of the chromatin and DNA composition of artificial chromosomes is important not only to further our understanding of the role of centromeres in genome stability, but also for the potential development of artificial chromosomes for gene transfer applications. Materials and methods Cell lines Characterization of cell lines containing mitotically stable human artificial chromosomes formed after transfection with either synthetic D17Z1 arrays (PAC17HT1.E29 (17-E29), PAC17HT1.D34 (17-D34), BAC17HT4.B12 (17-B12) or cloned DXZ1 sequences (X-4, X-5) have been described previously [ 12 , 14 ]. The artificial chromosomes in 17-C20 were generated using VJ104-17α32 [ 12 ], hybridize with both D17Z1 and BAC vector probes, are de novo in composition and assemble CENP-A and CENP-E (data not shown). All artificial chromosomes were formed in human HT1080 cells. Cell lines were grown as described [ 12 ] and supplemented with either 100 μg/ml G418 (Gibco) (17-B12, 17-C20) or 2 μg/ml Blasticidin S HCl (ICN) (17-E29, 17-D34, X-5, X-6), as described [ 12 ]. Anaphase assays Anaphase assays used to directly measure chromosome segregation defects in 17-B12 and 17-C20 (Table 1 ) were carried out as previously described [ 14 ]. Assays were carried out at either 45 days (17-B12) or 12 days (17-C20) culture without selection. The spectrum orange-labeled D17Z1 probe (Vysis) hybridized with host 17 centromere regions and artificial chromosomes, whereas the spectrum green-labeled BAC vector probe VJ104 [ 6 ] hybridized exclusively with the artificial chromosomes. Co-localization of vector and D17Z1 probes produced yellow fluorescence on the artificial chromosomes, which allowed them to be distinguished from the host D17Z1 sequences (data not shown). Generation of clonal lines expressing Myc-tagged HP1α The nucleotide sequence of human HP1α (NCBI Nucleotide database: S62077) was used in BLAST searches against entries in the human expressed sequence tag (EST) database using the NIH BLAST server [ 80 ]. A representative HP1α cDNA clone (IMAGE 627533) was obtained from Research Genetics. DNA was prepared with the Wizard-plus mini-prep DNA purification system (Promega), and the cDNA was sequenced on an ABI 373 (Perkin-Elmer) with a fluorescence labeled dye-terminator cycle sequencing kit according to the manufacturer's instructions (PRISM Ready DyeDeoxy Terminator Premix from Applied Biosystems). The full coding sequence of IMAGE 627533 was PCR-amplified with primers incorporating an Eco RI restriction enzyme recognition site (HP1α forward primer, 5'-GGAATT CTGATGGGAAAGAAAACCAAGCG-3'; reverse primer, 5'-GGAATTCGCTCTTTGCTGTTT CTTTC-3') and subcloned using standard techniques [ 81 ] into pcDNA3.1-CT-Myc-His (Invitrogen). Subclones were sequenced to verify sequence integrity and orientation as above. The HP1α-Myc tagged construct (pHP1α-Myc) was transfected into 17-C20, 17-B12, 17-E29 or 17-D34 cell lines using lipofectamine (Invitrogen), resulting in the formation of clonal lines (17-C20-1.B22, 17-B12-1.B10, 17-E29-1.C23 and 17-D34-1.A2, respectively) that stably express Myc-tagged HP1α. G418 selection at 400 μg/ml was applied to select clonal lines 17-E29-1.C23 and 17-D34-1.A2. Since 17-C20 and 17-B12 cells are G418-resistant, pHP1α-Myc was co-transfected in the presence of a second construct, pPAC4 [ 82 ] that carries a bs r marker gene. Clonal lines (17-C20-1.B22 and 17-B12-1.B10) resistant to 4 μg/ml Blasticidin S HCl (ICN) were selected and expanded. Confirmation of Myc-tagged HP1α expression was by immunofluorescence using a mouse monoclonal anti-Myc antibody (Invitrogen). Immunofluorescence and fluorescence in situ hybridization (FISH) Metaphase spreads were prepared for immunofluorescence using previously described protocols [ 29 ]. Primary antibodies to the dimethylated form of histone H3 at lysine 4 (anti-H3DimK4) were purchased from Upstate Biotechnology (anti-dimethyl-histone H3 (Lys4)). Modification of histone H3 by trimethylation at lysine 9 (H3TrimK9) was detected using an antibody to the tri-methylated form of histone H3 at lysine 9 purchased from Abcam (anti histone H3-tri methyl K9). This antibody cross-reacts with lysine 27 on histone H3 and is termed anti-H3TrimK9/K27 in the present study. The CENP-A antibody was a generous gift from Manuel Valdivia (Cadiz University, Spain) [ 83 ]. Antibodies to H3DimK4, H3TrimK9/K27 and CENP-A were raised in rabbits. Primary and secondary antibody incubations were in 1× PBS supplemented with 1% BSA (Sigma). Secondary antibodies were purchased from Jackson ImmunoResearch. After immunofluorescence detection, 20-50 spreads were captured and their positions on the slide recorded. Slides were subsequently hybridized with an appropriate alpha-satellite probe to detect transfected and endogenous alpha-satellite sequences. FISH was carried out using standard protocols. Replication timing assay Cells were pulsed with 10 μM BrdU (Roche) in T25 cm 2 flasks for 2 h intervals. Following three PBS washes, medium supplemented with 50 μM thymidine (Sigma) was added. Cells were left in thymidine-containing medium until chromosome harvest. At appropriate intervals, colcemid was added to block cells in mitosis. Cells were harvested and fixed in 3:1 methanol/acetic acid. Metaphase spreads on microscope slides were baked for 1 h at 60°C. Primary anti-BrdU antibody (Roche) was added for 1 h at room temperature. Rhodamine-donkey-anti-mouse secondary antibodies (Jackson ImmunoResearch) were used to visualize sites of BrdU incorporation. Typically, 25 metaphase spreads were captured following BrdU detection and their coordinates on the slide recorded. Subsequent analyses with a D17Z1 probe were used to confirm identity of artificial or endogenous chromosomes.

/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545780.xml